You are on page 1of 314

Validation of Accident Models for Intersections

PUBLICATION NO. FHWA-RD-03-037 MAY 2005

Research, Development, and Technology Turner-Fairbank Highway Research Center 6300 Georgetown Pike McLean, VA 22101-2296

FOREWORD
The existing crash prediction models for rural intersections developed for the Interactive Highway Safety Design Model (IHSDM) require validation and recalibration to improve their credibility and enhance their applicability. This report describes the results of validation and calibration of motor vehicle crash models for rural intersections. Both the validation and recalibration activities were conducted in pursuit of one overriding research objective, which was to make marginal improvements to an existing set of statistical models for predicting crashes at two- and four-lane intersections, with the primary intent to be used in the IHSDM. The five types of intersection models for which conclusions are drawn and recommendations are made are: three-legged stop controlled intersections of two-lane roads; four-legged stop controlled intersections of two-lane roads; three-legged stop controlled intersections with two lanes on minor and four lanes on major road; and four-legged stop controlled intersections with two lanes on minor and four lanes on major road, and signalized intersections of two-lane roads. Copies of this report can be obtained from the Research and Technology Product Distribution Center, 9701 Philadelphia Court, Unit Q, Lanham, MD 20706; telephone: 301–577–0818; fax: 301–577–1421; or the National Technical Information Service (NTIS), 5285 Port Royal Road, Springfield, VA 22161; telephone: 703–487–4650; fax: 703–321–8547.

Michael F. Trentacoste Director Office of Safety Research and Development Notice This document is disseminated under the sponsorship of the U.S. Department of Transportation in the interest of information exchange. The U.S. Government assumes no liability for the use of the information contained in this document. This report does not constitute a standard, specification, or regulation. The U.S. Government does not endorse products or manufacturers. Trademarks or manufacturers' names appear in this report only because they are considered essential to the objective of the document. Quality Assurance Statement The Federal Highway Administration (FHWA) provides high-quality information to serve Government, industry, and the public in a manner that promotes public understanding. Standards and policies are used to ensure and maximize the quality, objectivity, utility, and integrity of its information. FHWA periodically reviews quality issues and adjusts its programs and processes to ensure continuous quality improvement.

1. Report No.

FHWA-RD-03-037
4. Title and Subtitle

2. Government Accession No.

3. Recipient's Catalog No.

Validation of Accident Models for Intersections

5. Report Date

May 2005

6. Performing Organization Code 7. Author(s) 8. Performing Organization Report No.

Simon Washington, Bhagwant Persaud, Craig Lyon, and Jutaek Oh
9. Performing Organization Name and Address

None

University of Arizona Department of Civil Engineering Tucson, AZ 85721--0072 Ryerson University Department of Civil Engineering 350 Victoria Street Toronto, ON Canada M5B 2K3
12. Sponsoring Agency Name and Address

10. Work Unit No. (TRAIS) 11. Contract or Grant No.

DTFH61-00-C-00073

13. Type of Report and Period Covered

Department of Transportation Federal Highway Administration 400 7th Street, SW Room 4110 Washington, DC 20590-0001
15. Supplementary Notes

Research Report 11/2000 to 12/2003

14. Sponsoring Agency Code

Joe Bared, HRDS-05, Contracting Officer’s Technical Representative

16. Abstract

This report describes the results of validation and calibration of motor vehicle crash models for rural intersections. Both the validation and recalibration activities were conducted in pursuit of one overriding research objective, which was to make marginal improvements to an existing set of statistical models for predicting crashes at two and four lane intersections, with the primary intent to be used in the Interactive Highway Safety Design Module (IHSDM). The five types of intersection models for which conclusions are drawn and recommendations are made are: Three-legged stop controlled intersections of two-lane roads; four-legged stop controlled intersections of two-lane roads; three-legged stop controlled intersections with two lanes on minor and four lanes on major road; and four-legged stop controlled intersections with two lanes on minor and four lanes on major road, and signalized intersections of two-lane roads.
17. Key Words 18. Distribution Statement

Accident modification factors, traffic safety, signalized intersections, crash models, crash model validation, Interactive Highway Safety Design Model
19. Security Classif. (of this report) Unclassified 20. Security Classif. (of this page) 21. No. of Pages Unclassified 311 22. Price

SI* (MODERN METRIC) CONVERSION FACTORS
APPROXIMATE CONVERSIONS TO SI UNITS
Symbol
in ft yd mi in 2 ft 2 yd ac 2 mi fl oz gal 3 ft 3 yd
2

When You Know
inches feet yards miles square inches square feet square yard acres square miles fluid ounces gallons cubic feet cubic yards

Multiply By LENGTH
25.4 0.305 0.914 1.61

To Find
millimeters meters meters kilometers square millimeters square meters square meters hectares square kilometers

Symbol
mm m m km mm 2 m 2 m ha 2 km mL L 3 m 3 m
2

AREA
645.2 0.093 0.836 0.405 2.59

VOLUME
29.57 milliliters 3.785 liters 0.028 cubic meters 0.765 cubic meters 3 NOTE: volumes greater than 1000 L shall be shown in m

MASS
oz lb T
o

ounces pounds short tons (2000 lb) Fahrenheit

28.35 0.454 0.907 5 (F-32)/9 or (F-32)/1.8

grams kilograms megagrams (or "metric ton") Celsius

g kg Mg (or "t")
o

TEMPERATURE (exact degrees)
F

C

ILLUMINATION
fc fl lbf 2 lbf/in foot-candles foot-Lamberts poundforce poundforce per square inch 10.76 3.426 4.45 6.89 lux 2 candela/m newtons kilopascals lx 2 cd/m N kPa

FORCE and PRESSURE or STRESS

APPROXIMATE CONVERSIONS FROM SI UNITS
Symbol
mm m m km mm 2 m 2 m ha 2 km mL L 3 m 3 m g kg Mg (or "t")
o 2

When You Know
millimeters meters meters kilometers square millimeters square meters square meters hectares square kilometers milliliters liters cubic meters cubic meters grams kilograms megagrams (or "metric ton") Celsius lux 2 candela/m newtons kilopascals

Multiply By LENGTH
0.039 3.28 1.09 0.621

To Find
inches feet yards miles square inches square feet square yards acres square miles fluid ounces gallons cubic feet cubic yards ounces pounds short tons (2000 lb) Fahrenheit foot-candles foot-Lamberts poundforce poundforce per square inch

Symbol
in ft yd mi in 2 ft 2 yd ac 2 mi fl oz gal 3 ft 3 yd oz lb T
o 2

AREA
0.0016 10.764 1.195 2.47 0.386

VOLUME
0.034 0.264 35.314 1.307

MASS
0.035 2.202 1.103 1.8C+32

TEMPERATURE (exact degrees)
C F

ILLUMINATION
lx 2 cd/m N kPa 0.0929 0.2919 0.225 0.145 fc fl lbf 2 lbf/in

FORCE and PRESSURE or STRESS

*SI is the symbol for the International System of Units. Appropriate rounding should be made to comply with Section 4 of ASTM E380. (Revised March 2003)

ii

TABLE OF CONTENTS
EXECUTIVE SUMMARY ...............................................................................................1 VALIDATION FINDINGS ............................................................................................ 2 RECALIBRATION FINDINGS..................................................................................... 3 RECOMMENDATIONS................................................................................................ 3 1. INTRODUCTION..........................................................................................................5 1.1 VARIABLE ABBREVIATIONS ............................................................................. 6 1.1.1 Model Types ...................................................................................................... 6 1.1.2 Accident Models ................................................................................................ 6 1.1.3 Definitions of Variables..................................................................................... 7 2. VALIDATION OF ACCIDENT MODELS ..............................................................11 2.1 VALIDATION APPROACH AND PRELIMINARIES ........................................ 11 2.2 DATA SOURCES AND ISSUES .......................................................................... 13 2.3 MODEL PERFORMANCE MEASURES ............................................................. 15 2.4 VALIDATION ACTIVITY 1: VALIDATION USING ADDITIONAL YEARS OF ACCIDENT DATA .............................................................................................. 18 2.4.1 Model I............................................................................................................. 19 2.4.2 Model II ........................................................................................................... 26 2.4.3 Model III .......................................................................................................... 32 2.4.4 Model IV.......................................................................................................... 38 2.4.5 Model V ........................................................................................................... 43 2.5 VALIDATION ACTIVITY 2: VALIDATION WITH GEORGIA DATA ........... 51 2.5.1 Model I............................................................................................................. 51 2.5.2 Model II ........................................................................................................... 56 2.5.3 Model III .......................................................................................................... 61 2.5.4 Model IV.......................................................................................................... 71 2.5.5 Model V ........................................................................................................... 83 2.6 VALIDATION ACTIVITY 3: VALIDATION OF THE ACCIDENT PREDICTION ALGORITHM .................................................................................... 99 2.6.1 Validation of the Base Models....................................................................... 100 2.6.2 Validation of the Accident Prediction Algorithm.......................................... 104 2.7 DISCUSSION OF RESULTS............................................................................... 111 2.7.1 Model I........................................................................................................... 112 2.7.2 Model II ......................................................................................................... 114 2.7.3 Model III ........................................................................................................ 116 2.7.4 Model IV........................................................................................................ 118 2.7.5 Model V ......................................................................................................... 119 3. RECALIBRATION ...................................................................................................123 3.1 RESEARCH APPROACH ................................................................................... 123 3.1.1 Model Functional Forms................................................................................ 125 3.1.2 Goodness-of-Fit Evaluation ........................................................................... 125 3.2 DESCRIPTION OF DATA .................................................................................. 127 3.2.1 Summary of Datasets ..................................................................................... 127 3.2.2 Type I ............................................................................................................. 129 3.2.3 Type II............................................................................................................ 131 iii

3.2.4 Type III .......................................................................................................... 134 3.2.5 Type IV .......................................................................................................... 139 3.2.6 Type V ........................................................................................................... 145 3.3 AADT MODEL ESTIMATION RESULTS ........................................................ 153 3.3.1 Description of Base Conditions ..................................................................... 153 3.3.2 Model Results ................................................................................................ 157 3.4 FULLY PARAMETERIZED STATISTICAL MODEL ESTIMATION RESULTS ................................................................................................................................... 173 3.4.1 Type I Models ................................................................................................ 174 3.4.2 Type II Models............................................................................................... 181 3.4.3 Type III Models ............................................................................................. 186 3.4.4 Type IV Models ............................................................................................. 188 3.4.5 Type V Models .............................................................................................. 191 3.5 ESTIMATION OF ACCIDENT MODIFICATION FACTORS ........................ 194 3.5.1 AMFs Derived from Recalibration Full Models............................................ 196 3.5.2 AMFs Derived from Regression Models....................................................... 199 3.5.3 Summary of AMFs ........................................................................................ 204 3.6 SENSITIVITY ANALYSIS RESULTS ............................................................... 208 3.6.1 Type I Intersections........................................................................................ 208 3.6.2 Type II Intersections ...................................................................................... 216 3.6.3 Type III Intersections..................................................................................... 225 3.6.4 Type IV Intersections..................................................................................... 236 3.6.5 Type V Intersections ...................................................................................... 245 3.7 SUMMARY, DISCUSSION, AND CONCLUSIONS......................................... 249 3.7.1 Model Recalibration....................................................................................... 250 3.7.2 Summary of AMFs ........................................................................................ 257 3.7.3 Conclusions and Recommendations .............................................................. 262 APPENDIX A: ALL MODELS IN THE REPORTS OF RELEVANCE.................265 APPENDIX B: MODELS VALIDATED.....................................................................267 APPENDIX C: DESCRIPTIVE STATISTICS FOR FULL DATASETS BY STATE ........................................................................................................................................269 APPENDIX D: CURE PLOTS FOR TYPE I, II, III, IV, AND V AADT MODELS293 REFERENCES.............................................................................................................. 297

iv

81 Figure 12.... CURE Plot for Alternate TOTACC Type I Group A Model ... 69 Figure 8............................... 90 Figure 15................. Observed versus Predicted Accident Frequency: TOTACCI................. Observed versus Predicted Accident Frequency: TOTACC Variant 1 ......... 178 Figure 23.............. Observed versus Predicted Accident Frequency: Total Accidents Type II.................... CURE Plot for Type IV TOTACC AADT Model ....... CURE Plot for Type V TOTACC AADT Model... 76 Figure 10. Observed versus Predicted Accident Frequency: Injury Variant 2 ........ Observed versus Predicted Accident Frequency: INJACCI..... Observed versus Predicted Accident Frequency: TOTACC ................ CURE Plot for Type I TOTACC AADT Model .... CURE Plot for TOTACC Type I Group A Model ............. 71 Figure 9.................... 83 Figure 13... Observed versus Predicted Accident Frequency: TOTACC Main Model .............. 127 Figure 21....... 59 Figure 4.... 88 Figure 14.............. 295 v ......... Observed versus Predicted Accident Frequency: Injury Variant 1 ....... 126 Figure 20... CURE Plot for Type III TOTACC AADT Model........ 64 Figure 6.................................... 177 Figure 22....................... Observed versus Predicted Accident Frequency: Total Accidents Type I ................................... CURE Plot for Type III TOTACC AADT Model Using the CURE Method: Alternative Model . Observed versus Predicted Accident Frequency: INJACCI........ 294 Figure 26.................................. CURE Plot for Type II TOTACC AADT Model................. Observed versus Predicted Accident Frequency: TOTACC Type II ............................... 56 Figure 3. 66 Figure 7......... CURE Plot for Type III TOTACC AADT Model...................................................................... 95 Figure 17... 97 Figure 18............... 99 Figure 19....... Observed versus Predicted Accident Frequency: Injury Accidents Type I.............. Observed versus Predicted Accident Frequency: TOTACCI Type III .......... 92 Figure 16................... Observed versus Predicted Accident Frequency: INJACC ..................LIST OF FIGURES Figure 1..................... 54 Figure 2.......... 293 Figure 24.. Observed versus Predicted Accident Frequency: TOTACCI Main Model..... Observed versus Predicted Accident Frequency: INJACC .... Observed versus Predicted Accident Frequency: Injury Accidents Type II ... Observed versus Predicted Accident Frequency: TOTACCI Variant 3. 79 Figure 11.... 61 Figure 5........................ 295 Figure 27................ 294 Figure 25.................................

......................... Parameter Estimates for Type I Total Accident Model Using Additional Years of Data........ 25 Table 10.......................................................................................................................................................................... Validation Statistics for Type I Total Accident Model Using Additional Years of Data.......................................... Parameter Estimates for Type II Injury Accident Model Using Washington Data .......................................................................................................................... 35 vi .............................................. Accident Summary Statistics for Type II Sites.............................................. 30 Table 18....................................................................................................................... 29 Table 16........................ Parameter Estimates for Type I Injury Accident Model Using Additional Years of Data.................................... 24 Table 9... Parameter Estimates for TOTACC Type III Model Using Additional Years of Data......... 26 Table 12..................................................................................................................................................... Parameter Estimates for TOTACCI Type III Model Using Additional Years of Data.... Validation Statistics for TOTACC Type III Model Using Additional Years of Data................. 26 Table 13..... 34 Table 23............................................................................................................... 33 Table 22... Validation Statistics for Type I Injury Accident Model Using Washington Data . 21 Table 5...... 34 Table 24........................................................................................................ 23 Table 7.......................................... 22 Table 6................................................................................................................................................................................................. Accident Summary Statistics of Type III Sites.................................................. Validation Statistics for Type I Injury Accident Model Using Additional Years of Data ........................ Validation Statistics for Type I Total Accident Model Using Washington Data .................................................... 25 Table 11......................................................................................................................... 14 Table 2.................................................................................................................................................. 27 Table 14........................................... 32 Table 21............................................................................................. Basic Statistics for the Data Sources ........................................................... Parameter Estimates for Type I Injury Accident Model Using Washington Data ......................... 28 Table 15......................................... 23 Table 8................................................... Validation Statistics for Type II Total Accident Model Using Washington Data ..... Validation Statistics for Type II Injury Accident Model Using Additional Years of Data ............... Parameter Estimates for Type II Injury Accident Model Using Additional Years of Data .....................LIST OF TABLES Table 1................... 18 Table 3............................................ 29 Table 17.......... Parameter Estimates for Type II Total Accident Model Using Additional Years of Data ..................................................................................................... Validation Statistics for Type II Total Accident Model Using Additional Years of Data .................................................................... 32 Table 20............................................................................................................................. Validation Statistics for Type II Injury Accident Model Using Washington Data .................................................................... Parameter Estimates for Type I Total Accident Model Using Washington Data ............................ 20 Table 4................ Validation Statistics from Original Report1 ................ Accident Summary Statistics for Type I Sites .......................... Parameter Estimates for Type II Total Accident Model Using Washington Data ....................................................................... 31 Table 19..........................................

........... 44 Table 41........... Validation Statistics for TOTACC Type V Model Using Additional Years of Data: Main Model......................................... 47 Table 46....................... Parameter Estimates for TOTACCI Type V Model Using Additional Years of Data: Variant 3................................. Parameter Estimates for INJACC Type IV Model Using Additional Years of Data.......................................................................... Parameter Estimates for INJACC Type III Model Using Additional Years of Data: Variant 2................................................. Parameter Estimates for TOTACC Type V Model Using Additional Years of Data: Main Model................................................................................. Validation Statistics for TOTACCI Type III Model Using Additional Years of Data........................................................................Table 25............................................................................ 47 Table 45................................................................................................... 45 Table 43............................. Parameter Estimates for TOTACC Type IV Model Using Additional Years of Data......... Parameter Estimates for TOTACCI Type V Model Using Additional Years of Data: Main Model.................... 38 Table 31............................ 37 Table 28.................................................................... 49 vii .................................................................................................................................................................................... 43 Table 40................................ Validation Statistics for TOTACCI Type V Model Using Additional Years of Data: Variant 3................... 37 Table 29................. Parameter Estimates for INJACC Type III Model Using Additional Years of Data: Variant 1..... Parameter Estimates for TOTACC Type V Model Using Additional Years of Data: Variant 1................................................................................................................................................... Validation Statistics for INJACC Type IV Model Using Additional Years of Data...................................................................................... Accident Summary Statistics of Type V Sites.... Accident Summary Statistics of Type IV Sites ......... 38 Table 30................................. Parameter Estimates for INJACC Type V Model Using Additional Years of Data ............................................................................................................................ 40 Table 34............................................................................................................ 39 Table 33....................................................................................................... 36 Table 27.......... Validation Statistics for TOTACC Type V Model Using Additional Years of Data: Variant 1.................................... 46 Table 44....................................................................................................... 43 Table 39......................................................................... 42 Table 37.................................................................................. Validation Statistics for INJACC Type III Model Using Additional Years of Data: Variant 2........... 39 Table 32........................................................................... Validation Statistics for TOTACC Type IV Model Using Additional Years of Data........................................... 35 Table 26................................. 41 Table 36.................................................. Validation Statistics for TOTACCI Type V Model Using Additional Years of Data: Main Model................................. Validation Statistics for INJACC Type III Model Using Additional Years of Data: Variant 1.............. 40 Table 35............................................................... Parameter Estimates for INJACCI Type IV Model Using Additional Years of Data...................... 48 Table 47................... 45 Table 42......................................................................................................................................................................................................................... Parameter Estimates for TOTACCI Type IV Model Using Additional Years of Data.......... 49 Table 48.......................................... Validation Statistics for TOTACCI Type IV Model Using Additional Years of Data............................................................ 42 Table 38........................................................ Validation Statistics for INJACCI Type IV Model Using Additional Years of Data..........

............................................................................................ Summary Statistics of California...................................... Validation Statistics for Type I Total Accident Model Using Georgia Data.................................... Validation Statistics for Type II Total Accident Model Using Georgia Data ............................ 59 Table 60...................... Parameter Estimates for Type II Total Accident Model Using Georgia Data............. 82 Table 81........................... Validation Statistics for INJACCI Type IV Model Using Georgia Data ............ Georgia......................... Summary Statistics for Three States: California....... Michigan............................................. 70 Table 71........................... 76 Table 76............ 78 Table 78................................................... Summary Statistics of Georgia Data: Type V .......... Parameter Estimates for INJACC Type III Model Using Georgia Data: Variant 2.......................... 79 Table 79..... 82 Table 82. 50 Table 51. Validation Statistics for TOTACCI Type III Model Using Georgia Data ........... Validation Statistics for INJACC Type III Model Using Georgia Data: Variant 2....... Parameter Estimates for Type I lnjury Accident Model Using Georgia Data .............................................................. Parameter Estimates for INJACCI Type V Model Using Additional Years of Data ............................. 73 Table 73.......... Validation Statistics for TOTACC Type III Model Using Georgia Data........................... Validation Statistics for TOTACCI Type IV Model Using Georgia Data ..................... Parameter Estimates for TOTACCI Type IV Model Using Georgia Data............... 70 Table 70... 77 Table 77...... Validation Statistics for INJACC Type V Model Using Additional Years of Data ......... 85 Table 84.. Summary Statistics of Georgia Data: Type IV .. 62 Table 63..... 64 Table 65..... Validation Statistics for Type I Injury Accident Model Using Georgia Data ......... 61 Table 62.......................... 50 Table 50...... 58 Table 59.......................... Summary of Georgia versus Minnesota Data for Type II Sites............... 52 Table 53.......... 75 Table 75........... Parameter Estimates for TOTACC Type IV Model Using Georgia Data ....... 67 Table 68...................... Parameter Estimates for Type II Injury Accident Model Using Georgia Data 60 Table 61.......... Validation Statistics for INJACC Type III Model Using Georgia Data: Variant 1................ 56 Table 57....... 72 Table 72... Parameter Estimates for INJACC Type III Model Using Georgia Data: Variant 1....... Validation Statistics for TOTACC Type V Model Using Georgia Data: Main Model........................ 63 Table 64......................... 73 Table 74............................................................................... 68 Table 69.... Parameter Estimates for INJACC Type IV Model Using Georgia Data ..... and Georgia............. Validation Statistics for INJACCI Type V Model Using Additional Years of Data ..... 66 Table 67.................... Parameter Estimates for Type I Total Accident Model Using Georgia Data ................... Validation Statistics for TOTACC Type IV Model Using Georgia Data...................................................................... Validation Statistics for INJACC Type IV Model Using Georgia Data..... Michigan................ Summary Statistics of Georgia Data: Type III Sites........................................ 51 Table 52................ Summary of Georgia versus Minnesota Data for Type I Sites ............................................. Validation Statistics for Type II Injury Accident Model Using Georgia Data....... 84 Table 83................. Parameter Estimates for TOTACC Type III Model Using Georgia Data ...... 86 Table 85. 80 Table 80.......... Pearson Correlations: Original.............. 53 Table 54............... 87 viii ........................................................ Parameter Estimates for INJACCI Type IV Model Using Georgia Data... Parameter Estimates for TOTACC Type V Model Using Georgia Data: Main Model............. and Georgia1 ............................. and California........... 57 Table 58....... 65 Table 66.................................. 55 Table 56.............. 53 Table 55.................................. Parameter Estimates for TOTACCI Type III Model Using Georgia Data ...Table 49...

. Summary Statistics for Type IV Sites................ Summary Statistics for Type I Sites ........................................................ 101 Table 98. Summary Accident Statistics for Sites Meeting Base Conditions............................................ 106 Table 103......... 111 Table 108................................... 131 Table 112............................ 109 Table 106....... 107 Table 104............................... 102 Table 100.......................................................................... Summary Statistics for Type V Sites........ Summary of Correlations for Independent Variables for Type IV Sites .................................... 135 Table 116........... 140 Table 119.......................................... Summary of Correlations for Independent Variables for Type I Sites ......................... 100 Table 97........................ Correlation Between Crashes and Independent Variables for Type I Sites...................... Correlation Between Crashes and Independent Variables for Type IV Sites143 Table 120.......................................................................... 93 Table 91........... Validation Statistics for the Accident Prediction Algorithm: Type V.......... Validation Statistics for Type V Base Model Intersections.............................. II.................. Validation Statistics for the Accident Prediction Algorithm: Type V Sites with a Major Road Right-Turn Lane ..... 96 Table 93....... 104 Table 101...... 96 Table 94..... Parameter Estimates for TOTACC Type V Model Using Georgia Data: Variant 3........... Sources of Data............................................................. 98 Table 95......... Summary Accident Statistics for Sites Meeting Base Conditions...... 91 Table 89................................ Summary of Correlations for Independent Variables for Type II Sites............................ 134 Table 115........................................................................................................ 89 Table 88... Validation Statistics for the Accident Prediction Algorithm: Type I Sites with a Major Road Turning Lane... 144 Table 121.............. 92 Table 90....................................................... 108 Table 105................... Validation Statistics for INJACC Type V Model Using Georgia Data..... Validation Statistics for TOTACCI Type V Model Using Georgia Data: Main Model.................. Parameter Estimates for INJACCI Type V Model Using Georgia Data ........................... The AMFs for Type I........ Validation Statistics for Type I Base Model Intersections .................................... 128 Table 109... Validation Statistics for the Accident Prediction Algorithm: Type II .......... Validation Statistics for the Accident Prediction Algorithm: Type II Sites with a Major Road Turning Lane.......................... Correlation Between Crashes and Independent Variables for Type V Sites 149 ix ............. 130 Table 111.............Table 86. Validation Statistics for TOTACCI Type V Model Using Georgia Data: Variant 3......... Summary Statistics for Type II Sites .. Parameter Estimates for TOTACCI Type V Model Using Georgia Data: Main Model.............. Validation Statistics for Type II Base Model Intersections............................. Validation Statistics for INJACC Type V Model Using Georgia Data.... 88 Table 87...................................... 138 Table 118........ 129 Table 110............. 94 Table 92..... 110 Table 107..................................... 145 Table 122.............................................. Correlation Between Crashes and Independent Variables for Type III Sites137 Table 117............................................. 98 Table 96........... 132 Table 113................ Validation Statistics for TOTACC Type V Model Using Georgia Data: Variant 1.......... and V Intersections .......................... Validation Statistics for the Accident Prediction Algorithm: Type I ............................ Correlation Between Crashes and Independent Variables for Type II Sites 133 Table 114.................................... 102 Table 99.......................... Parameter Estimates for INJACC Type V Model Using Georgia Data ....................... Summary of Correlations for Independent Variables for Type III Sites ............................................... Summary Statistics for Type III Sites.... 105 Table 102.................. Parameter Estimates for TOTACC Type V Model Using Georgia Data: Variant 1..............................

............................................................... Parameter Estimates for Recalibrated TOTACC Type II: Main and Group B AADT Models .............................. 167 Table 148...................................................................... 157 Table 130...................................................................................................... 160 Table 136........... 167 Table 147..................................................... 162 Table 139.......... 164 Table 141............................................................. Number of Sites Used for Type V Main AADT Model ................................ 158 Table 133.. 164 Table 142..... Summary Statistics for Type IV Sites: Main and Group B Base AADT Models.......................... 163 Table 140.......................................................................... 157 Table 132. 159 Table 135........................................... Summary Statistics for Type III Sites: Main and Group B Base AADT Models..........................Table 123................................... 153 Table 125............... 169 Table 151.... Summary Statistics for Type II Sites: Main and Group B Base AADT Models......................... Parameter Estimates for Recalibrated INJACC Type II: Main and Group B AADT Models .... Group A Base Conditions for Type IV INJACC AADT Models....................... Group A Base Conditions for Type IV TOTACC AADT Models........................................................ Summary of Correlations for Independent Variables for Type V Sites ................................... 169 Table 150....................... Group A Base Conditions for Type V TOTACC AADT Models ............. 165 Table 144....................................................... Parameter Estimates for Recalibrated TOTACC Type III: Main and Group B Base AADT Models........ Group A Base Conditions for Type III TOTACC AADT Models ............................ Number of Sites Used for Type IV Main and Group B Base Condition AADT Models ............................................. 157 Table 131........................................... Number of Sites Used for Type II Main and Group B Base AADT Models 160 Table 137.............................................. 150 Table 124......................................................................................................................... Parameter Estimates for Type IV Group A TOTACC and INJACC Base Models...................................... 166 Table 146............... Summary Statistics for Type IV Sites: Group A Base AADT Models ....... 168 Table 149............... Summary Statistics for Type I Sites: Main and Group B Base AADT Models......... 156 Table 127............... Summary Statistics for Type III Sites: Group A Base AADT Models................ Parameter Estimates for Recalibrated INJACC Type I: Main and Group B AADT Models ..................................................... 158 Table 134.................... 165 Table 143...... 156 Table 129............. 166 Table 145............................. 161 Table 138.............................. Group B Base Conditions for Type I to IV AADT Models....................................................................... Group A Base Conditions for Type V INJACC AADT Models ........ 170 x . 170 Table 152............................................................................... 156 Table 128........................... Summary of AADT Models Recalibrated and Data Used................................. Parameter Estimates for Type III Group A TOTACC and INJACC Base Models......... Number of Sites Used for Type I Main and Group B Base AADT Models........... Number of Sites Used for Type III Main and Group B Base Condition AADT Models ................................................. Parameter Estimates for Recalibrated INJACC Type III: Main and Group B Base AADT Models.......... Parameter Estimates for Recalibrated TOTACC Type I: Main and Group B AADT Models ...... Group A Base Conditions for Type III INJACC AADT Models ........................................................................................................ 155 Table 126.............. Parameter Estimates for Recalibrated INJACC Type IV: Main and Group B Base AADT Models..................................... Parameter Estimates for Recalibrated TOTACC Type IV: Main and Group B Base AADT Models....................................

................... Base Model for Type IV Sites ................ 185 Table 167....................... Parameter Estimates for Recalibrated TOTACC Full Models: Type I Group B............................................................................. Parameter Estimates for Recalibrated TOTACC Type V: Main AADT Model ........................... Base Model for Type V Sites................................. 206 Table 187...................................... Comparison of AMFs for INJACC............................................................................................................................................ Parameter Estimates for Type V Group A TOTACC and INJACC Base Models................................................ 187 Table 168......... 183 Table 165................. Parameter Estimates for Recalibrated INJACC Full Models: Type II Group A ....... Parameter Estimates for Recalibrated INJACC Full Models: Type III ..............................................Table 153.................. 193 Table 172......................... 203 Table 183.............................................. Parameter Estimates for Recalibrated INJACC Full Models: Type I Group B.. and V Full Models ... 200 Table 177..... Parameter Estimates for Recalibrated INJACC Type IV: Full Models........................... Base Conditions for Type III Sites ....... 182 Table 164.................................. Summary Statistics for Type V Sites: Group A Base AADT Models.................................... Summary Statistics for Type V Sites: Main AADT Model......................................................... 196 Table 174....................... 198 Table 176.............................................................. 194 Table 173...... Parameter Estimates for Recalibrated TOTACC Full Models: Type V ..... Parameter Estimates for Recalibrated TOTACC Full Models: Type II Group B............... AMFs for Type IV Sites ............................................................................ 201 Table 180......................................................... 171 Table 155........................... 190 Table 170......................................................................... 172 Table 157................................... 173 Table 159..................... Parameter Estimates for Recalibrated INJACC Full Models: Type I Group A ............................... AMFs Derived from Type III..... 201 Table 178............................. 204 Table 185.......... 205 Table 186....................... Base Conditions for Type IV Sites ......................................... Base Conditions for Type V Sites.................................. 172 Table 156............................... 176 Table 161........ AMFs Derived from Type I and II Full Models ............................... Parameter Estimates for Recalibrated INJACC Type V: Main AADT Model ....................................................... Parameter Estimates for Recalibrated INJACC Full Models: Type II Group B...................................................... Parameter Estimates for Recalibrated TOTACC Full Models: Type IV................................................................................................................. 191 Table 171.. Parameter Estimates for Recalibrated TOTACC Full Models: Type III ....... 201 Table 179.............................................. Parameter Estimates for Recalibrated TOTACC Full Models: Type I Group A ....... 184 Table 166.............. Base Model for Type III Sites........................................... AMFs for Type III Sites........ 173 Table 158........................................ 202 Table 181.................. IV....... Parameter Estimates for Recalibrated TOTACC Full Models: Type II Group A ......................................................... Parameter Estimates for Recalibrated INJACC Full Models: Type V ...................... AMFs Estimated from Regression Procedure ................................... Comparison of Type I–II AMFs for TOTACC ..................... 202 Table 182.............. 175 Table 160............................ 179 Table 162.............................................................. 180 Table 163..................................................................................................................................... 197 Table 175......................... Summary of Models Recalibrated and Data Used................................. Comparison of Type III–V AMFs for TOTACC.... 207 xi ...................... 171 Table 154................. 188 Table 169.......................................................................................... AMFs from Previous Studies.................. 204 Table 184.

.. 215 Table 196....................... Sensitivity of Safety to HI1 for Type II INJURY Accidents Per Year.... Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type III TOTAL Accidents Per Year ............. Sensitivity of Safety to RT MAJ for Type II INJURY Accidents Per Year..... Sensitivity of Safety to LT MAJ for Type II TOTAL Accidents Per Year ........ Sensitivity of Safety to HI1 for Type I INJURY Accidents Per Year.......................................................................... Sensitivity of Safety to Commercial Driveways on Major Road for Type III TOTAL Accidents Per Year .................... 209 Table 190............................................................. 230 Table 211... Sensitivity of Safety to DRWY1 for Type II TOTAL Accidents Per Year......................Table 188...... 211 Table 192. 220 Table 201.................................... Sensitivity of Safety to Major Road Left-Turn Lane for Type III TOTAL Accidents Per Year .......................... 237 Table 218........................................................................................ Sensitivity of Safety to RT MAJ for Type I INJURY Accidents Per Year ................................. Sensitivity of Safety to LT MAJ for Type II INJURY Accidents Per Year............................... 229 Table 210.................................................................. 238 xii ............. 224 Table 205....... 208 Table 189......... Sensitivity of Safety to HI1 for Type I TOTAL Accidents Per Year . Sensitivity of Safety to Vertical Curves on Major Road (VEI1) for Type III TOTAL Accidents Per Year .. Sensitivity of Safety to Median Width on Major Road (MEDWIDTH1) for Type III TOTAL Accidents Per Year ............................ 217 Table 198............................... Sensitivity of Safety to LT MAJ for Type I TOTAL Accidents Per Year ....... Sensitivity of Safety to Painted Median Type on Major Road (MEDTYPE1) for Type III TOTAL Accidents Per Year ......................................... Sensitivity of Safety to Peak Turning Percentage (PKTURN) for Type III TOTAL Accidents Per Year .. 210 Table 191................................... 222 Table 203................. Sensitivity of Safety to MEDIAN for Type II TOTAL Accidents Per Year 219 Table 200.............. 216 Table 197.......................................... 223 Table 204................ 231 Table 212...... Sensitivity of Safety to DRWY1 for Type II INJURY Accidents Per Year................... Sensitivity of Safety to Skew Angles for Type III INJURY Accidents Per Year............... 232 Table 213................................................................................. 218 Table 199.................................................. 235 Table 216.......... Sensitivity of Safety to Right Sight Distance from Minor Road (SDR2) for Type IV TOTAL Accidents Per Year ................................................................................................. Sensitivity of Safety to Skew Angles for Type III TOTAL Accidents Per Year...................... Sensitivity of Safety to MEDIAN for Type II INJURY Accidents Per Year................................... Sensitivity of Safety to Skew Angles for Type IV TOTAL Accidents Per Year.. 233 Table 214.......... Sensitivity of Safety to RT MIN for Type I TOTAL Accidents Per Year .............................. 214 Table 195........ 212 Table 193.. 236 Table 217....... Sensitivity of Safety to Hazard Rating on Major Road (HAZRAT1) for Type III INJURY Accidents Per Year ......... Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type III TOTAL Accidents Per Year ............. 234 Table 215................... 228 Table 209........... 221 Table 202..... Sensitivity of Safety to Commercial Driveways on Major Road for Type III INJURY Accidents Per Year ... Sensitivity of Safety to RT MAJ for Type I TOTAL Accidents Per Year ........ Sensitivity of Safety to RT MIN for Type I INJURY Accidents Per Year ......................................... 226 Table 207.......................... 213 Table 194............................... 227 Table 208............................................ Sensitivity of Safety to LT MAJ for Type I INJURY Accidents Per Year ............................... Sensitivity of Safety to RT MAJ for Type II TOTAL Accidents Per Year.................................................................. 225 Table 206.......

Table 219. Sensitivity of Safety to Peak Through Percentage on Minor Road (PKTHRU2) for Type IV TOTAL Accidents Per Year................................................ 239 Table 220. Sensitivity of Safety to Peak Left-Turn Percentage (PKLEFT) for Type IV TOTAL Accidents Per Year ......................................................................................... 240 Table 221. Sensitivity of Safety to Skew Angles for Type IV INJURY Accidents Per Year............................................................................................................................... 241 Table 222. Sensitivity of Safety to Major Right-Turn Lane for Type IV INJURY Accidents Per Year ....................................................................................................... 242 Table 223. Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type IV INJURY Accidents Per Year ........................................................................................ 243 Table 224. Sensitivity of Safety to Peak Left-Turn Percentage (PKLEFT) for Type IV INJURY Accidents Per Year ........................................................................................ 244 Table 225. Sensitivity of Safety to Commercial Driveways on Major Road for Type V TOTAL Accidents Per Year ......................................................................................... 245 Table 226. Sensitivity of Safety to Horizontal Curve Combinations on Major and Minor Roads (HEICOM) for Type III TOTAL Accidents Per Year ............................ 246 Table 227. Sensitivity of Safety to Light for Type IV TOTAL Accidents Per Year...... 247 Table 228. Sensitivity of Safety to Horizontal Curves on Minor Road (HEI2) for Type IV TOTAL Accidents Per Year ........................................................................... 248 Table 229. Sensitivity of Safety to Light for Type IV TOTAL Accidents Per Year...... 249 Table 230. Summary of Models Recalibrated and Data Used........................................ 251 Table 231. Comparison of Type III–V AMFs for TOTACC.......................................... 258 Table 232. Comparison of Type I–II AMFs for TOTACC ............................................ 259 Table 233. Comparison of Type III–V AMFs for TOTACC.......................................... 260 Table 234. Comparison of AMFs for INJACC............................................................... 261 Table 235. AADT Models Recommended for Use in IHSDM ...................................... 262 Table 236. Full Models Recommended for Use in Crash Prediction ............................. 263 Table 237. All Models in the Reports of Relevance........................................................265 Table 238. Models Validated...........................................................................................267 Table 239. Summary Statistics for Type I Sites by State ................................................269 Table 240. Summary Statistics for Type II Sites by State ...............................................271 Table 241. Summary Statistics for Type III Sites by State: California (CA), Michigan (MI), and Georgia (GA) .................................................................................................273 Table 242. Summary Statistics for Type IV Sites by State: California (CA), Michigan (MI), and Georgia (GA) .................................................................................................278 Table 243. Summary Statistics for Type V Sites by State: California (CA), Michigan (MI), and Georgia (GA) .................................................................................................285

xiii

EXECUTIVE SUMMARY
The research described in this report consists of two separate yet complimentary activities—validation and calibration of crash models for rural intersections. Both the validation and recalibration activities were conducted with one overriding research objective: Given existing database limitations, make marginal improvements to an existing set of statistical models for predicting crashes at two- and four-lane intersections, with the primary intent to provide robust predictive models for use in the Interactive Highway Safety Design Module (IHSDM). The five types of intersection models addressed in this research effort include: Type I: Type II: Type III: Type IV: Type V: Three-legged stop controlled intersections of two-lane roads. Four-legged stop controlled intersections of two-lane roads. Three-legged stop controlled intersections with two lanes on minor and four lanes on major roads. Four-legged stop controlled intersections with two lanes on minor and four lanes on major roads. Signalized intersections of two-lane roads.

The models that are the focus of this research are presented in three different Federal Highway Administration (FHWA) reports: Vogt and Bared (Types I and II);(1) Vogt (Types III, IV, and V);(2) and Harwood et al. (Types I, II, and V).(3) Each report presents several variants of the models for each type of intersection. The first two reports include models for total as well as injury accidents and present what are referred to as full models. The Harwood et al. report presents base models for Types I, II, and V intersections.(3) These base models included variables that were statistically significant at the 15 percent level and are at the backbone of an algorithm for predicting accidents at intersections that are different in one or more features from the specified base conditions. Specifically, accident modification factors (AMF) for the features of interest are applied to the base model prediction to estimate accidents per unit of time for a specific intersection. This algorithm is intended for use in the Crash Prediction Module of FHWA’s IHSDM. The anticipated practical application of these models has motivated research directions taken throughout the course of this investigation. The data in support of this research were derived from three sources: 1. The original data used for the calibration of the main models for total accidents were obtained from the researchers who developed those models. 2. Highway Safety Information Systems (HSIS) data were obtained for additional years for the same intersections used in the calibration and for injury accidents for the original and additional years. 3. An independent validation data set of intersections and their relevant crash, traffic, and geometric data in Georgia was specially assembled for this project.(4) 1

The research team faced a number of challenges while conducting this research, including data collection, independent variable characteristics, and the models’ intended end-use: 1. The observational data on which the statistical models are based suffered from intercorrelation. 2. The interactions among variables had to be carefully considered in model estimation. 3. The need to forecast crashes across States posed significant difficulties. 4. The observational data limited the amount of variation in independent variables, reducing overall model precision. 5. Resource and data reliability restrictions prohibited a sufficiently large, randomly selected, fully comprehensive data set on which to estimate statistical models. 6. Incongruencies between data sets across States and across time periods posed serious challenges. Despite these challenges, the research team conducted a model validation and then recalibrated the five intersection models.

VALIDATION FINDINGS
The four sets of validation activities were: 1. Re-estimation of the model coefficients using the original data. This validation activity was used to determine the reproducibility of the published results and to ensure an “equivalent” launching point for all validation and calibration activities. This activity represented a logical starting point for the research effort and was successful in that modeling results were reproduced satisfactorily. 2. Validation of the models against additional years of accident data for the same intersections used in the calibration. Because the crash models were developed as direct inputs into the IHSDM’s Accident Analysis Module, the models will be used by highway agencies to estimate the safety performance of an existing or proposed roadway. Therefore, the models should be able to forecast crashes across time and space. This validation activity was used to assess the models’ ability to forecast crashes across time, determining the models temporal capability and stability. 3. Validation of the models against Georgia data. This validation activity assessed the models’ ability to forecast crashes across space. This activity tested numerous aspects of model prediction: comparing data from different jurisdictions; capturing variables that describe regional and jurisdictional differences; and consistency of crash processes across space.

2

4. Validation of the Accident Prediction Algorithm. While this research’s primary focus was to validate crash models, validating the Accident Prediction Algorithms based on these models was also important. Two basic sets of performance tests were employed. First, the models were re-estimated using the same variables and functional forms as those published in the original reports; the parameters for the original and re-estimated models were then compared, using a level of alpha = 0.10 to establish statistical significance. Second, the model (or algorithm) was used to predict accident frequencies at individual intersections, from which the following summary statistics were calculated: • • • • • Pearson product-moment linear correlation coefficients. Mean Prediction Bias (MPB); MPB/year. Mean Absolute Deviation (MAD); MAD/year. Mean Squared Error (MSE); MSE/year2. Mean Square Prediction Error (MSPE); MSPE/year2.

The details of validation activities 1 through 4 are presented in sections 3.4 through 3.7 respectively, while the results are discussed in section 3.8.

RECALIBRATION FINDINGS
Model recalibration was focused on improving the existing set of intersection crash models through use of an improved and expanded database and through lessons learned in the validation and recalibration activities. For each the five intersection types, the research team developed and/or refined three different sets of models, described in detail in chapter 3. The first type is Annual Average Daily Traffic (AADT) Models, which represent base models for predicting crashes as a function of major and minor road AADT. The analytical results of these models can be found in subsequent sections of this report. The second type of model is Full Models. These statistical models forecast crashes as a function of a relatively large set of independent variables. Details of the Full Models can be found in section 3.4 of this report. The third type of model is AMF. These models, better described as countermeasure correction factors, represent our best efforts to estimate the effect of geometric countermeasures on safety relative to base model predictions. AMF details can be found in section 3.5 of this report. Sensitivity analyses—tables of AMFs as a function of AADT and other factors are provided in section 3.6.

RECOMMENDATIONS
The research supported the proposed IHSDM accident prediction algorithm. An updated set of base models for predicting crashes using only AADT are recommended (see Summary, Discussion, and Conclusions section and Table 235). The updated statistical 3

models are based on larger sample sizes and, in some cases, resulted in slightly modified sets of independent variables compared to the originally estimated models. AMFs should be selected on a case-by-case basis, and should be updated continually to improve the predictive ability of the crash models. Expert opinion derived AMFs should be replaced with the results of state-of-the-practice before-after studies as time progresses and research allows. If expert opinion accurately reflects safety conditions, then carefully conducted future studies should reveal general agreement with expert expectation. When expert opinions are not confirmed over time, then empirical results should replace expert opinion. Full regression models are recommended for crash forecasts and find logical applications in the Highway Safety Manual and SafetyAnalyst (see Summary, Discussion, and Conclusions section and Table 236).

4

a reliable method of estimating safety performance is required. and four-legged signalized intersections of two-lane roads.1. and AMF1 AMF2 … AMFn = AMFs for various intersection features.and four-legged intersections of two-lane roads.(3) Ongoing efforts aim to produce similar documents for other types of facilities. INTRODUCTION Effective safety management of a highway system requires that engineers know the present safety performance of a roadway and how it will perform if contemplated actions are taken. while Vogt documents models for three other types of rural intersections: three.and fourlegged stop controlled with four lanes on the major and two on the minor. Nb = predicted number of total intersection-related accidents per year for base conditions.2) Those projects also resulted in regression models with additional variables indented for use as AMFs in IHSDM.(3) These base models were the best of available accident prediction models developed in earlier FHWA projects and retained model variables that were statistically significant at the 15 percent level. FHWA and its contractors have developed a new approach that combines historical accident data. Type III: Three-legged stop controlled intersections with two lanes on the minor and four lanes on the major road. (1. are presented in two FHWA reports: Vogt and Bared present models for three. and expert judgment to make safety performance predictions that are expected to be better than those obtained by any of the individual approaches. A recent report documents an accident prediction algorithm for implementing the new approach for two-lane rural highway sections that include road segments and five types of intersections. To this end. there are five types of intersection accident prediction models pertaining to the research efforts described: Type I: Three-legged stop controlled intersections of two-lane roads. The accident prediction algorithm has been developed for incorporation in the IHSDM as the Crash Prediction Module. 5 . Type II: Four-legged stop controlled intersections of two-lane roads. regression analysis. Harwood et al. before-and-after studies.2) In summary. but is suitable for stand-alone applications. The full models. along with several variants.(1. The structure of the accident prediction algorithm for the five types of rural at-grade intersections is as follows: Nint = Nb (AMF1 AMF2 … AMFn) (1) where Nint = predicted number of total intersection-related accidents per year after application of AMFs.and four-legged intersections of two-lane rural roads with STOP control. In effect. and signalized intersections of two-lane roads. presented base models and AMFs for three. Type IV: Four-legged stop controlled intersections with two lanes on the minor and four lanes on the major road.

The third chapter presents the recalibration efforts for the five types of intersections. the definitions of model types. Type II: Four-legged stop controlled intersections of two-lane roads.25 meters (m) (76. An additional report provides additional AMFs for left. II and V sites.1. AADT models and fully parameterized models were recalibrated.Type V: Signalized intersections of two-lane roads.(5) These AMFs are also included in the validation effort.1 VARIABLE ABBREVIATIONS This section provides. Thus. and V. This report presents recalibration results for the five types of rural intersections that were the subject of the validation exercise that was undertaken in the first part of the project.2 Accident Models Models I and II Total: Total number of police-reported intersection-related accidents within 76. in the case of Types III. and variables applied in this research investigation. Injury: Total number of police-reported intersection-related injury accidents within 76. sample size. and their results are discussed in the recalibration chapter. 1. It was also of interest to validate the accident prediction algorithm as a whole. 1.1 Model Types Type I: Three-legged stop controlled intersections of two-lane roads. The second chapter describes four different sets of activities conducted to assess the validity of prediction models for the five types of rural intersections.1. IV. for ease of reference. 6 . Type V: Signalized intersections of two-lane roads. Type IV: Four-legged stop controlled intersections with two lanes on the minor and four lanes on the major road. This model recalibration effort complemented the comprehensive model validation conducted as part of a larger technical evaluation of crash prediction models. A natural follow-on to model validation is model recalibration using validation data and findings to improve the specification of the intersection models. This first chapter provides an introduction and presents the description of the variables used in this research. The model recalibration efforts described in this chapter complement the comprehensive model validation exercise presented in the previous chapter as a part of larger technical evaluation of the models.25 m (250 ft) of the intersection. Type III: Three-legged stop controlled intersections with two lanes on the minor and four lanes on the major road. validation of these models across both space and time has become of paramount importance. accident types.25 m (250 ft) (ft)) of the intersection. given its novelty and the fact that it relies on expert judgment for deriving the AMFs.and right-turn lanes for at-grade intersections at type I. 1. The models were developed using data that were limited in terms of geographical diversity and. This report consists of three chapters.

HAZRAT2: Roadside hazard rating on minor road within 76. least hazardous case. and V TOTACC: Total number of accidents within 76. • Four-legged intersections: (right angle . to 7. estimated AADT.25 m (250 ft) of the intersection. in fact.25 m (250 ft) of the intersection.25 m (250 ft) of the intersection center (from 1. COMDRWY1: Commercial driveways on major roads within 76.(6) HEI1: Sum of degree of curve in degrees per hundred feet of each horizontal curve on major road. to 7. respectively. This variable is identical to ADT2 in the original models.25 m (250 ft) of the intersection center. HAU: Intersection angle variable defined where the angle between the major and minor roads is measured from the far side of the minor road: • Three-legged intersections: Angle minus 90 if minor road is to the right of the major road in the increasing direction. This variable is identical to ND and NODRWY1 in the original models for Type I–II and III– V intersections. TOTACCI: Only those crashes considered intersection-related and within 76. INJACC: Total number of injury crashes within 76. least hazardous case. HEICOM: (1 / 2) (HEI1 + HEI2). HAZRAT1: Roadside hazard rating on major road within 76. any portion of which is within ±244 m (±800 ft) of the intersection center divided by the number of such curves. 1. This variable is identical to NODRWYC1 in the original models. most hazardous case). This variable is identical to ADT1 in the original (published) models.25 m (250 ft) of the intersection center (from 1.25 m (250 ft) of the intersection.25 m (250 ft) of the intersection. This variable is identical to NODRWYC2 in the original models.1.25 m (250 ft) of the intersection center. most hazardous case). any portion of which is within ±244 m (±800 ft) of the intersection center divided by the number of such curves. GRADE2: Average absolute grade on minor road within ±244 m (±800 ft) of the intersection center (percent).3 Definitions of Variables AADT1: Average daily traffic on major road (vehicles per day). DRWY2: Driveways on minor roads within 76.left angle)/2. 90 minus angle if minor road is to the left of the major road in the increasing direction.25 m (250 ft) of the intersection. HEI2: Sum of degree of curve in degrees per hundred feet of each horizontal curve on minor road.(6) This variable is identical to RHRI in the original models for Type I and II intersections. IV. GRADE1: Average absolute grade on major roads within ±244 m (±800 ft) of the intersection center (percent). This variable is identical to NODRWY2 in the original models. This change was made after determining that the traffic flow variables were. AADT2: Average daily traffic on minor roads (vehicles per day). INJACCI: Only those injury crashes considered intersection-related and within 76.Models III. DRWY1: Driveways on major roads within 76. COMDRWY2: Commercial driveways on minor roads within 76.25 m (250 ft) of the intersection. 7 .

1 for one approach. 0 otherwise. PKLEFT2: Peak left-turn percentage on minor roads (percent). and 2 for both approaches). PROT_LT: Protected left lane (0 = no. any portion of which is within 76. This variable is identical to LTLN2 in the original models for Type III–V intersections. L3LT (Type III-V): Left-turn lane on minor roads (0 = no.25 m (250 ft) of the intersection center divided by the number of such curves.25 m (250 ft) of the intersection center. 0 otherwise. HI1: Sum of degree of curve in degrees per hundred feet of each horizontal curve on major road. any portion of which is within 76. PKTHRU1: Peak through percentage on major roads (percent). LIGHT: Light at intersection (0 = no. PKTHRU2: Peak through percentage on minor roads (percent). This variable is identical to LTLN1 in the original models for Type III–V intersections. 1 = yes). LT MIN (Type I-II): 1 if left-turn lane exists on at least one approach of minor roads. L1RT (Type III-V): Right-turn lane on major roads (0 = no. 1 = yes). LEGACC1: Acceleration lane on major roads (0 = no.25 m (250 ft) of the intersection center divided by the number of such curves. 1 for one approach.HI: Sum of degree of curve in degrees per hundred feet of each horizontal curve on major road. This variable is identical to HI in the original models for Type I and II intersections. PKLEFT: Peak left-turn percentage (percent). 1 = yes). RESDRWY1: Residential driveways on major roads within 76. This variable is identical to RTLN1 in the original models for Type III–V intersections. 0 otherwise. MEDWDTH1: Median width on major roads (feet). and 2 for both approaches). PKLEFT1: Peak left-turn percentage on major roads (percent). HICOM: (1 / 2) (HI1 + HI2). 2 = curbed.25 m (250 ft) of the intersection center divided by the number of such curves. 1 = painted. MEDWDTH2: Median width on minor roads (feet). L1LT (Type III-V): Left-turn lane on major roads (0 = no. 3 = others). 1 for one approach. MEDTYPE (Type III-V): Median type (0 = no median. LEGACC2: Acceleration lane on minor roads (0 = no. PKTRUCK: Peak truck percentage passing through the intersection (percent). any portion of which is within 76. and 2 for both approaches). LT MAJ (Type I-II): 1 if left-turn lane exists on at least one approach of major roads. HI2: Sum of degree of curve in degrees per hundred feet of each horizontal curve on minor road. 1 = yes). PKTURN: Peak turning percentage (percent). L3RT (Type III-V): Right-turn lane on minor roads (0 = no. 1 for one approach. This variable is identical to NODRWYR1 in the original models. This variable is identical to RTLN2 in the original models for Type III–V intersections. and 2 for both approaches). LTLN1S (Type III-V): Left-turn lane on major roads (0 = no. MEDIAN (Type I-II): 1 if median exists on major roads. 1 = yes). 8 . This variable is identical to MEDWIDTH1 in the original models.

2 if rolling. divided by the number of such curves. SPD2: The average posted speed on minor roads in vicinity of the intersection (miles per hour). VEI2: Sum of absolute change of grade in percent per hundred feet for each curve on minor roads. 1 = rolling. This variable is identical to NODRWYR2 in the original models. SHOULDER1: Shoulder width on major roads (feet). TERRAIN1 (Type III-V): Terrain on major roads within 76. VCEI1: Sum of absolute change of grade in percent per hundred feet for each crest curve on major roads. any portion of which is within 800 feet of the intersection center. any portion of which is within 76. any portion of which is within 800 feet of the intersection center.25 m (250 ft) of the intersection center. divided by the number of such curves. This variable is identical to VCI in the original models for Type I and II intersections. This variable is identical to SPDI in the original models for Type I and II intersections. SDL2: Left-side sight distance on minor roads (feet). any portion of which is within 76. divided by the number of such curves. RT MIN (Type I-II): 1 if right-turn lane exists on minor roads. SPD1: The average posted speed on major roads in vicinity of the intersection (miles per hour). 9 . 0 otherwise.25 m (250 ft) of the intersection center. divided by the number of such curves. SD1: Longitudinal sight distance on major roads (feet). SDR2: Right-side sight distance on minor roads (feet). any portion of which is within 76. VEI1: Sum of absolute change of grade in percent per hundred feet for each curve on major roads. VCEI2: Sum of absolute change of grade in percent per hundred feet for each crest curve on minor roads. any portion of which is within ±244 m (±800 ft) of the intersection center. VICOM (Type III-V): (1 / 2) (VI1 + VI2). VCI1: Sum of absolute change of grade in percent per hundred feet for each crest curve on major roads. divided by the number of such curves. VCI2: Sum of absolute change of grade in percent per hundred feet for each crest curve on minor roads. VEICOM: (1 / 2) (VEI1 + VEI2). 1 = rolling.RESDRWY2: Residential driveways on minor roads within 76. SHOULDER2: Shoulder width on minor roads (feet). or 3 if mountainous terrain.25 m (250 ft) of the intersection center. divided by the number of such curves.25 m (250 ft) of the intersection center. 2 = mountainous). divided by the number of such curves. 0 otherwise. VI2: Sum of absolute change of grade in percent per hundred feet for each curve on minor roads. any portion of which is within 76. divided by the number of such curves.25 m (250 ft) of the intersection center (0 = flat.25 m (250 ft) of the intersection center. 2 = mountainous). RT MAJ (Type I-II): 1 if right-turn lane exists on major roads.25 m (250 ft) of the intersection center (0 = flat. TERRAIN (Type I-II): 1 if flat. TERRAIN2 (Type III-V): Terrain on minor roads within 76. any portion of which is within ±244 m (±800 ft) of the intersection center. VI1: Sum of absolute change of grade in percent per hundred feet for each curve on major roads.

10 .

The first aspect of model validation consists of qualitative assessments of statistical models. Because the existing accident models were developed as direct inputs into the IHSDM’s Crash Prediction Module. Validation of the models against Georgia data. or some combination of thereof. Therefore. 2. the data sources and issues are discussed. in fact. and assessing the implications of the models regarding an underlying theory of the data generating process—motor vehicle crashes. Determine whether the existing accident models developed by FHWA were “over-fit” to the estimation data. This validation was used to assess the models’ ability to forecast accidents over space. they do not. External model validation is more quantitative and is focused on various quantitative measures of a statistical model’s prediction ability. including examining the collection of variables used to identify missing or irrelevant variables. 4. This validation was used to assess the models’ ability to forecast accidents across time. the individual validation activities and their results are presented. if important.2. Finally. 3. are included in the model by some relevant explanatory variable. in this case. the effect of time. To meet these objectives. Second. Determine whether important variables were omitted from the models due to lack of representative data. lack of power. Determine whether the models are valid at other locations (across space) and at the same intersections in the future (across time). thus making elements (factors) appear to affect accidents when. inspecting the functional form of the models. 2. Determine whether the functional forms of the models have been properly specified. The primary application of the models is to forecast the impact of design considerations and countermeasures in regions and jurisdictions not represented in the calibration data. highway agencies will use the models to estimate the safety performance of an existing or proposed roadway. researchers conducted two aspects of model validation: internal and external model validations. It first provides a description of the overall validation approach. or covariates that are influenced by time. the models should be able to forecast accidents over time and space. Three model validation activities comprising both internal and external validation activities were undertaken: 1. 2. are either not important or. Temporal stability of a model suggests that it will predict accident frequencies well across time. insufficient sample size. Third. a discussion of these results is provided. this 11 . Validation of the models against additional years of accident data for the same intersections used in the calibration. VALIDATION OF ACCIDENT MODELS This chapter presents validation results for the five types of rural intersections. Thus.1 VALIDATION APPROACH AND PRELIMINARIES Several objectives were identified to guide the validation efforts: 1.

However. all else being equal. it does not. In the documentation to follow. but the precise and correct conditional probability interpretation is cumbersome and. Thus. however. This means that the analyst would conclude. In general. while committing a type II error results in failing to apply an effective countermeasure. With a type I error.” and “marginal” denote subjective evaluations of GOF comparisons between models. Serious differences in GOF are suggestive of noteworthy or significant model deficiencies. To summarize. Several goodness-of-fit (GOF) statistics to assess model fit to validation data were employed. in reality. This translation is not precise. but improvements might be difficult to obtain. structure. A type II error occurs with beta (β) probability. the analyst concludes that the null hypothesis is false. committing a type I error results in applying an ineffective countermeasure. Comparisons between models. all model specifications were validated “as is. Validation of the Accident Prediction Algorithm as a whole was considered to be important. applying a liberal α equal to 0.” “moderate. As a result of this conclusion. This validation activity addresses the logical defensibility of the algorithm used for predicting accidents. The validation exercises in this chapter focus primarily on external validation—validation concerned with assessing performance of the models when compared to external data. no changes to model specifications were considered or assessed. In statistical models of accident occurrence.validation exercise attempts to assess the ability of the models to capture differences across regions as reflected through relevant variable expressions in the models. one might install left-turn lanes without realizing a reduction in crashes. And more focus is given to internal validation in the recalibration activities documentation. Throughout the report the subjective criteria of alpha (α) equal to 0. Computing the actual β in negative binomial models is extremely difficult. for practical purposes. a type II error can be argued to be more serious than a type I error. and plausibility of the models proposed.” that is. that the presence of a left-turn lane reduces accidents when. which follows the report on these validation efforts. the terms “serious. Moderate differences in GOF suggest cases where models could be improved. choosing a larger α results in a smallerβ. continuing with the previous example. for example. are generally subjective. when in fact it is true with α probability. In these validation activities. does not lend any additional insights. The discussion section mentions internal validation concerns—the internal coherence. theoretical soundness. 3. 12 . making a type II error results in concluding that the presence of a left-turn lane does not reduce crashes when in fact it does.10 is applied. The risk is in failing to install an effective countermeasure. Validation of the Accident Prediction Algorithm. The support for this level of α is as follows.10 suggests that this study has simultaneously accepted a smaller level of beta. and provides quantitative evidence of how well it is predicting accidents. Marginal differences in GOF are thought to be negligible and are potentially explained by random fluctuations in the observed data.

2. or broadside/angle. 13 . Special criteria were needed for the latter case because California data does not include a variable indicating if an accident was “intersection-related. The first validation activity employed the original datasets with the objective of reproducing the original models. and V were each developed using Michigan and California data for two dependent variables. The second used only those accidents considered to be “intersection-related” and occurring within 76. The Washington data collected. IV.(3) Models I and II. 2. these crashes typically represent about 6 percent of the total crashes occurring within 76. An independent validation data set was assembled specifically for this project. and run-off-the-road crashes may could possibly be classified as intersection related but were not included in the analysis in order to maintain consistencyand comparability with previously estimated models from which this research was based. IV.” An accident was considered to be intersection related accidents if it was: 1. For the third validation activity.(2) According to discussions with FHWA. For models III. right turn or Uturn before the crash. and V. these crashes typically represent about 13 percent of the total crashed occurring within 76.25 m (250 ft) of the intersection. including recommended base models. The fourth validation activity assessed the “Red Book” accident prediction algorithm. data from Georgia were assembled during this project and used to further assess the models’ ability to predict accidents over space (in a different jurisdiction). modeled police reported “intersection” or “intersection-related” accidents which occurred within 76. additional years of accident data (1996 and 1997) from California and Michigan were used to validate the models over time.25 m (250 ft) of the intersection. An accident where one vehicle involved was making a left turn. In the second validation activity. The first used all accidents occurring within 76.2 DATA SOURCES AND ISSUES This section presents the data sources used for the validation and some general data issues. for the final development of models I and II. HSIS data were obtained for additional years for the same intersections used in the calibration for the additional years available.25 m (250 ft) of the intersection. A multivehicle accident in which the accident type is either a sideswipe. 3. but not used.25 m (250 ft) of the intersection. Models III. The original data used for the calibration of the main models for total accidents were obtained from the researchers who developed those models. Vehicle-bicyclist. 2. plus accident data for one more year (1996) also were utilized. head-on. for four-leg STOP-controlled intersections. Vehicle-pedestrian accident. rearend.25 m (250 ft) of the intersection. additional years of accident data (1990 to 1998) from Minnesota were used to validate models I and II over time. The data used came from three sources: 1. which were developed using Minnesota data. For three-leg STOP-controlled intersections. 3.

25 m (250 ft). 60–67) 2 Vogt. 1999. consistency of accident location definitions had to be resolved. TOTACCI: Only those crashes considered intersection-related and within 76.06 or . 14 . (pp.05 miles. Similar distinction between INJACC and INJACCI The Georgia data specially collected for this validation also do not include a variable coded as “intersection” or “intersection-related” by the police. 53–64) 3 MN: Minnesota.04. mainly to check the sensitivity of the results to this definition. i. CA: California.05-mile buffers were used in separate validation efforts.04 or 0.25 m (250 ft). the accident data recorded in Georgia measures the distance of an accident from an intersection within two decimal places of a mile.4 ft). MI: Michigan. As such. within 8. or 211 or 264 ft).05 m (26. In the end. To use the Georgia data to validate the original models. Basic Statistics for the Data Sources Basic statistics for the data sources (Blank cells indicate accident types not validated. x indicates accident types validated) Sample size Original Years Georgia Data States3 Original years Subsequent Years Georgia Data Model I1 389 121 MN MN and WA Model II1 327 114 MN MN and WA Model III2 84 52 CA and MI CA and MI Model IV2 72 52 CA and MI CA and MI Model V2 49 51 CA and MI CA and MI GA GA GA GA GA 5 5 3 3 3 Original Years (1985 to 1989) (1985 to 1989) (1993 to 1995) (1993 to 1995) (1993 to 1995) 9 9 (1990 to 1998 (1990 to 1998 for for MN MN Subsequent 1993 to 1996 1993 to 1996 for 2 2 2 for WA) WA) (1996 to 1997) (1996 to 1997) (1996 to 1997) Years covered Years 2 2 2 2 2 Georgia Data (1996 to 1997) (1996 to 1997) (1996 to 1997) (1996 to 1997) (1996 to 1997) Total Accident types Injury validated TOTACC4 TOTACCI INJACC 1 4 X X X X X X X X X X X X X INJACCI X X X Vogt and Bared 1998.08 kilometers (km) (0. (pp.Table 1.e.and 0. 4 TOTACC: Total number of accidents within 76. WA: Washington. The original models included accidents only within 76. However.25 m (250 ft) from the intersection center. the above criteria for determining an “intersection-related” accident were used. both the 0. GA: Georgia. The issue was whether or not to include accidents within 0..

The GOF measures used were: Pearson’s Product Moment Correlation Coefficients Between Observed and Predicted Crash Frequencies Pearson’s product moment correlation coefficient. Mean Prediction Bias (MPB) The MPB is the sum of predicted accident frequencies minus observed accident frequencies in the validation data set. The expectation during model validation is a high correlation coefficient. The smaller the average prediction bias. and is given by: ⎞ ⎜Y − Y ⎟ ∑⎛ ⎝ ⎠ ∧ i =1 i i n MPB = where n = validation data sample size. and will result in a correlation coefficient of exactly 1. a model must be internally plausible. which is expected.2. Conversely. usually denoted by r. and n (3) 15 . A low coefficient suggests that the model is not performing well and that variables influential in the calibration data are not as influential in the validation data.Y 2) 1/ 2 (2) where Y = the mean of the Yi observations. It is important to note at the outset that only after an assessment of many GOF criteria is made can the performance of a particular model or set of models be assessed. the better the model is at predicting observed data. This statistic provides a measure of the magnitude and direction of the average model bias as compared to validation data. divided by the number of validation data points. The MPB can be positive or negative. is a measure of the linear association between the two variables Y1 and Y2 that have been measured on interval or ratio scales. A model that predicts observed data perfectly will produce a straight line plot between observed (Y1) and predicted values (Y2). will not reduce the correlation coefficient significantly. Pearson’s product moment correlation coefficient is given as: r= 2 2 ⎡ ⎣ ∑ (Yi1 − Y 1 ) ∑ (Yi 2 − Y 2 ) ⎤ ⎦ ∑ (Y i1 − Υ 1 )(Yi 2 .3 MODEL PERFORMANCE MEASURES Several GOF measures were used to assess model performance. In addition. a linear correlation coefficient of 0 suggests a complete lack of a linear association between observed and predicted variables. and agree with known theory about crash causation and processes. A different correlation coefficient is needed when one or more variable is ordinal. Random sampling error.

It differs from MPB in that positive and negative prediction errors will not cancel each other out. MSE is typically a measure of model error associated with the calibration or estimation data. the predicted response. divided by the sample size minus the number of model parameters.Y = the fitted value Yi observation. Mean Squared Prediction Error (MSPE) and Mean Squared Error (MSE) MSPE is the sum of squared differences between observed and predicted crash frequencies. Conversely. Unlike MPB. ∑ Y −Y i =1 i n ∧ i n (4) The MAD gives a measure of the average magnitude of variability of prediction. Mean Absolute Deviation (MAD) MAD is the sum of the absolute value of predicted validation observations minus observed validation observations. and so degrees of freedom are lost (p) as a result of producing Yhat. A positive MPB suggests that on average the model overpredicts the observed validation data. Smaller values are preferred to larger values. divided by sample size. ∧ MAD = where n = validation data sample size. MSE is the sum of squared differences between observed and predicted crash frequencies. divided by the number of validation observations. The magnitude of MPB provides the magnitude of the average bias. MAD can only be positive. ∧ ⎛ ⎞ ∑ ⎜ Yi − Yi ⎟ ⎠ MSE = i =1 ⎝ n1 − p n 2 (5) ∧ ⎛ ⎞ ∑ ⎜ Yi − Yi ⎟ ⎠ MPSE = i =1 ⎝ n2 n 2 (6) 16 . MSPE is typically used to assess error associated with a validation or external data set. a negative value suggests systematic underprediction.

K The overdispersion parameter. As overdispersion gets larger. since MSPE and MSE are the mean values of the squared errors (MPB or MAD were squared and divided by n or n-p). GOF measures were computed on a per year basis. in the negative binomial distribution has been reported in different forms by various researchers. K is reported from the variance equation expressed as: Var{m} = E{m} + K * E{m}2 where Var{m} = the estimated variance of the mean accident rate. and K = the estimated overdispersion parameter. An MSPE that is higher than MSE may indicate that the models may have been overfit to the estimation data.e. A comparison of MSPE and MSE reveals potential overfitting or underfitting of the models to the estimation data.where n1 = estimation data sample size. K. This finding could also indicate that important variables were omitted from the model or the model was misspecified. 17 . resulting in a fair comparison of predictions based on different numbers of years. To normalize the GOF measures to compensate for the different numbers of years associated with different data sets. E{m} = the estimated mean accident rate. a model with smaller overdispersion (i. However.. data inconsistencies could cause a relatively high value of MSPE. and n2 = validation data sample size. For MPB and MAD per year. Overdispersion Parameter. Because the Poisson rate is overdispersed. and that some of the observed relationships may have been spurious instead of real. Typically this is the desired result. all else being equal. and consequently all of the standard errors of estimates become inflated. In the model results presented in this report. so does the estimated variance. the estimated variance term is larger than the same under a Poisson process. (7) Variance overdispersion in a Poisson process can lead to a negative binomial dispersion of errors. particularly when the Poisson means are themselves approximately gamma distributed. MSPE and MSE were divided by the square of number of years to calculate MSPE and MSE per year. a smaller value of K) is preferred to a model with larger overdispersion. or possess gamma heterogeneity. As a result. Values of MSPE and MSE that are similar in magnitude indicate that validation data fit the model similar to the estimation data and that deterministic and stochastic components are stable across the comparison being made. Finally. The negative binomial distribution has been shown to adequately describe errors in motor vehicle crash models in many instances. MPB and MAD were divided by number of years.

(p. the lesser quality of the collected Washington data … we take the Minnesota models as fundamental. only data from Minnesota were used for the final calibration. Data Limitations in the Minnesota Data Because site characteristics change over time. For the Type III.2. Of the 327 original four-legged sites.02 Model II (90-93 Minnesota data) 1. Only the models developed using Minnesota data were validated because a report by Vogt and Bared states that “in view of the small size of the Washington State sample … the non-random and ad hoc character of the Washington intersections.” (p. although data from both Minnesota and Washington were collected. The sites that were excluded typically changed from being a rural to a suburban environment. The validation undertaken for this activity included applying the original models to the new data and assessing various measures of GOF. 1990 to 1993 data were used to validate the models in addition to the Washington data. 123)(1) However. and V models. Validation Statistics from Original Report1 Measure MAD 1 Model I (90-93 Minnesota data) 1.28 Model I (93-95 Washington data) 1. Table 2. the original models were developed with accident data that were collected between 1993 and 1995.4 VALIDATION ACTIVITY 1: VALIDATION USING ADDITIONAL YEARS OF ACCIDENT DATA The acquisition of subsequent years of accident data allowed for the validation of the models across time. 1998. In the original report. The Type I and II models were originally calibrated on accident data from 1985 to 1989.68 Vogt and Bared. accident data from 1990 to 1998 were obtained to expand the validation dataset. the Washington data were still obtained to further assess the transferability of these models over time and across jurisdictions. The original report performed a similar validation exercise with Washington data but with fewer years of accident data. It also included recalibrating the models using the additional years of data and comparing the parameter estimates with those of the original models. the original sites were examined on important variables to determine which ones were no longer suitable for inclusion. IV. For the type I and II models. any sites where any year of accident data was missing was not included.17 Model II (93-95 Washington data) 2. Subsequent to the analysis and draft report. Although errors will exist in the accident counts used at these 18 . changed in the number of legs. 131–132) For the new validation of Type I and II models. The results for the MAD are given in Table 2. it was discovered that the mile log information that identifies the location of an intersection for some sites had changed over the 1990 to 1998 period. 315 were retained while 367 of the original 389 three-legged sites were retained. Accident data collected between 1996 and 1997 at the same locations were acquired to validate the models. and changed traffic control or had missing years of accident data. Also.

1 Model I The summary statistics shown in table 3 indicate that there are similar accident frequencies between the original (1985–89 data for 389 sites) and the additional (1990– 98 data for 367 sites) years.(2) In the original data. accidents are mileposted to the mainline. for Type I sites. Subsequent to the analysis and draft report it was discovered that for type V sites from Michigan.25 and 0.26 and 0. At non-State route crossroads. Therefore in the report. minor road AADT from 1993 to 1995 was not available for Michigan intersections. However. respectively. the later year crossroad accident numbers at these sites should be systematically lower than expected.11 for total and injury data. For example. the conclusions drawn from the analysis are not affected and the analysis has not been redone. Note that the statistics compare 5 years of accident data in the original set to 9 years of accident data for validation over time. The Washington data shows a higher average accident frequency and a lower variability between sites compared to the original Minnesota data. Data Limitations in the Michigan Data Traffic volumes from 1993 to 1995 were not updated for this validation exercise because AADT information for major and minor roads for 1996 and 1997 were not available. major AADT plus peakhour turning movement counts were used to estimate missing years of minor road AADT. 2. the average number of accidents per site per year in the validation data was 0. The latter years of data exhibit an increased variability amongst the sites. the Michigan accident data for 1996 and 1997 years obtained did not include the crossroad accidents at intersections with a State route crossroad because the validation team did not have the crossroad milepost information at that time. Therefore. these averages are 0. In the corrected data. the original researchers manually identified crossroad mileposts for about 40 percent or more of these intersections (State routes) and counted accidents that occurred on the crossroads. 19 . One of the complexities regarding AADT acquisition is outlined in the report published originally by Vogt. these errors were found to be negligible.4.sites. As a result.11.

00 0. This indicates that variables with large standard errors compared to the parameter estimates should not be included in the model even if engineering “common sense” suggests that they should.89 1.12/year) 2. 39 17 32 13 14 10 Vogt and Bared.79 2. 20 . VCI1.48 1.25/year) 0.36/year) 0.35 (0. which reveals differences in the parameter estimates of the variables between the two time periods. 1998.33 Min. did not exactly match those statistics given in the report.00 1.Table 3. The parameter estimates. 0 0 0 0 0 0 Max.17/year) Median 0.00 0. K.27/year) 0. their standard errors. (p.43 (0. VCI1 and RT were estimated with opposite signs to the original models. Recall that the intersection related variable for vertical curvature.66 (0.88 N/A2 3.00 0. was estimated with a similar magnitude. although in the original calibration these parameter estimates had large standard errors and were not statistically significant. The other parameters were estimated with the same sign as originally and in some cases similar magnitude and significance to the original models.59 (0.00 Std. and p-values are provided in table 4. 60) 2 N/A: not available Total Accident Model The model was recalibrated with the additional years of accident data for the original Minnesota sites.11/year) 1.21 (0.95 (0. The overdispersion parameter. Accident Summary Statistics for Type I Sites Dataset Minnesota Total (85-89)1 Minnesota Injury (85-89)1 Minnesota Total (90-98) Minnesota Injury (90-98) Washington Total (93-96) Washington Injury (93-96) 1 No.00 0. Deviation 2. of Sites 389 389 367 367 181 181 Mean 1.

0.0130..4991 (0.0339 (0.1420.234) -0. 1998.0177. 0. 0.513) 0. p-value) -13. 0.1726 HAZRAT1 (0.1900 (0.90 (1. <0.001) 0.0001) AADT2 0..001) 0.124) 0. 115) 2 K: Overdispersion value Recalibrated Estimates (s.485 Additional Years Estimates (s. The Pearson product-moment correlation coefficient was marginally higher for the original data than when the model based on that data were used to predict the additional years of data.1511.0327.0144.001) 0.e. The MPB is higher for the prediction of the additional years of data. 0.001) 0. 0.2671 RT MAJ (0.9922 Constant (1.48 (1. The MADs per year are similar for the predictions for the two time periods..0677.1072) 0.0784.8051 (0.07.001) 0. <0.4808 (0.0339 HI1 (0. 21 .0660.16.0206.077 (0.0779 (0. 0.0043 (0.269.0655.1398.0045 HAU (0.0001) AADT1 0.245 (0.402) 0. <0.0375 (0.0584.0032. 0.Table 4. 0. 0.058) 0.3004) 0.058) 0. p-value) -12.1578) K2 0. 0. 0. 0.0285 SPD1 (0. 0. <0.0108) 0. 0.0024.2935.0708.0754. p-value) -12.0001) 0. 0. <0.0639. <0.5037 Log of (0.3229) 0. 0.2690 (0.539) 0.481 1 Vogt and Bared.500 GOF measure comparisons are shown in table 5. 0.0220.00145 (0.8199 (0.075) 0.e.2901 VCI1 (0.001) 0.363) 0.004) 0.00222.1806 (0.e. The MSPE per year squared is lower than the MSE per year squared. 0. 0.8052 Log of (0.0714.126. 0.0273 (0.0145 (0.0561) 0.017) 0.481) -0. Parameter Estimates for Type I Total Accident Model Using Additional Years of Data Original Estimates1 Variable (s. (p.2260.

01 0.13 Recalibrated 1985-89 Model 1990-98 367 0.06 1.612 -0.20 N/A1 10. A comparison of validation measures for the model recalibrated with the original data and the original model applied to the Washington data is shown in table 7. was higher than the original Minnesota model. All parameters were estimated with the same sign as the original models.19 N/A1 Original 1985-89 Model 1990-98 367 0.93 0. For the new Washington data HAZRAT1 had a large standard error and the overdispersion parameter. and p-values are provided in table 6.Table 5. and in some cases with similar magnitude and significance.00 1.52 -0. 22 . The parameter estimates. mean absolute deviations and mean squared prediction errors were similar in magnitude.64 0.050 -0.84 0. For VCI1 both the original and newly estimated parameters had large standard errors. Validation Statistics for Type I Total Accident Model Using Additional Years of Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE 1 Recalibrated 1985-89 Model 1985-89 389 0.84 0.662 -0.94 0. their standard errors.21 4.14 MSPE/yr N/A: not available 2 The model was also recalibrated with the Washington data. The Pearson product-moment correlation coefficient was slightly higher for the original data as compared to Washington data.03 0. which reveals differences in the parameter estimates of the variables between the two locations.614 -0. K. The MPB.20 N/A1 10.06 1. Notable exceptions are VCI1 and HAZRAT1 which were estimated with the same sign but with a large difference in magnitude.

1900 (0.0754.2260.1010.2750.0108) 0. 115) 2 K: Overdispersion value Table 7.1511.0639.1726 (0. 0.2901 (0.001) 0. p-value)1 -12. 0.058) 0.309) 0..017) 0.0356.027 0. <0.0032 (0. 0. 0.1750.495) 0.9922 (1.0043 (0.1030 (0.1420.162) 0.08 1.0708. 0.8052 (0.632) 0. <0. 0.0144.075) 0.769 (0.5037 (0.001) 0. Parameter Estimates for Type I Total Accident Model Using Washington Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU 2 Original Estimates (s.0273 (0.28 23 .Table 6.124) 0. 0. <0. 0.90 (1.0327. p-value) -12.311) 0.662 -0.0339 (0.e.0285 (0.205 4.0561) 0.1806 (0.01 0.2780 (0.001) 0.001) 0. p-value) -12.4858 (0.8051 (0.1578) K 0.0179. 0. Validation Statistics for Type I Total Accident Model Using Washington Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 1 N/A: not available Recalibrated 1985-89 Model 1985-89 (Minnesota) 389 0.485 Washington Data Estimates (s. 0.35 N/A1 4.0001) 0. 0.481 1 Vogt and Bared. 0. 0. 0.4991 (0.0001) 0.0127.058) 0.2671 (0. 1998.802) 0.0170 (0.2935. <0. 0.0032.1398..0177.39 0.30 -0. <0.16.4440.0339 (0.0122 (0.0001) 0.0220.01.64 0.579 -0.6200 (0.0024.0805.e. 0. 0.00 1.1072) 0.001) 0.3004) 0.402) 0.0677.e. (p. <0.2690 (0.0784. 0. 0.001) 0.45 0. 0.19 N/A1 Original 1985-89 Model 1993-96 (Washington) 181 0.3229) 0.59 (2. 0.0660.0045 Recalibrated Estimate (s.8730 (0..

<0. The parameter estimates.e.163.988) 0. The Pearson product-moment correlation coefficient was lower (0.5618) 0. (p.106) was about one half of that for total accidents.0001) 0. and HAU were estimated with opposite signs than the original model.0051 (0. 1998. 116) 2 K: Overdispersion value Original Estimates (s.Injury Accident Model Table 8 shows the parameter estimates for the type I injury accident model using additional years of data.8671) 0.3047) 0.1869 (0.0930. 0.159) -0.00109 (0.0008 (0. The overdispersion parameter.44.004) 0.2594) -0.0327.553) than that for total accidents (0.059) -0.4551 (0.6092) 0.001) 0.0268. Table 8.0263) 0. <0.7774 (0.3620 (0.0857. their standard errors.260 (0. K. 0.e.0714. 0.0973. 0. p-value)1 -13.0514 (0. 0.0067 (0.3657. 0.0001) 0. 0.5815 (0.41 (1. The model was also recalibrated with the Washington data.0001) 0.494 Additional Years Estimates (s. 0.0374 (1. 0.699) -0.001) 0. and in some cases similar magnitude and significance as the original model. Parameter Estimates for Type I Injury Accident Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU DRWY1 K2 1 Vogt and Bared.0760. VCI1. The other parameters were estimated with the same direction of effect. 0. RT MAJ.00282.1814. 0.0460) 0.802) -0.0120 (0. p-value) -15.359 (0.0335 (0.7908. 0.0269.1618 (0.526 Validation statistics are shown in table 9 for the additional years of injury accident data. 0.2065 (0.. Since the original injury accident counts were not obtained these measures are not provided for the original years. and p-values are provided in table 10. was estimated to be slightly higher than that in the original model.361) 0. 0.001) 0.0045. <0.0977.614 for the recalibrated model) and the MAD per year (0. 0. but this is not surprising when the large standard errors are considered..393.0179.0541. 0. 0.0932.0156 (0. 24 . which reveals differences in the parameter estimates of the variables between the two States.8122 (0.

024 0.0045. 0.0695.578 (0.553 -0.4551 (0.3620 (0.190 (0. 25 . 0.0977.0055 (0.552) 0.0269.2594) -0.Table 9.0930. (p. The overdispersion parameter.219 -0. 0.0001) Log of AADT2 0. 0.8122 (0.0460) RT MAJ 0.494 1 Vogt and Bared. was estimated with a similar magnitude.0374 (1. K. 116) 2 K: Overdispersion value Variable Constant Washington Data Estimates (s. 0. 0.0118 (0. 0.305.031 Table 10.119.0120 DRWY1 (0.e.505) than that for total (0. 0.526. 1998. <0.873) 0.0227.7908. Validation Statistics for Type I Injury Accident Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Additional Years 1990-98 367 0.0404.6092) VCI1 0.0973.272) 0.2065 (0.012 (0. 0.513 The parameters were all estimated with the same sign but in several cases the difference in magnitude was large. 0. 0.49.0263) HAZRAT1 0..955 0.1814. Parameter Estimates for Type I Injury Accident Model UsingWashington Data Original Estimates (s.0968.3657.222.e.3047) HI1 0.106 2. p-value) -14.579) accidents.0327.0714. p-value) 1 -13.64 (2. 0. 0. 0.0163. 0.8671) K2 0.0335 (0.071 (0. 0.0156 (0. <0.5618) SPD1 0.1869 (0.0001) 0. and the MADs per year are about one half that for total accidents.001) 1.0111 (0. The Pearson productmoment correlation coefficient was lower (0.0103 (0.530 0. 0. Validation measures for the Washington data are shown in table 11.0001) Log of AADT1 0.001) 0.534) 0. <0.0051 HAU (0.892) 0..5237 (0.602) 0.526) -0.001) 0.

VCI1 and HAU were estimated with a 26 . The parameter estimates.084 -0.021 0.46 (0.31/year) 1. (p.38 Minimum 0 0 0 0 0 0 Maximum 16 9 67 36 20 14 Vogt and Bared.30/year) 0.00 1.00 0.95 5.343 0.00 1.178 1.00 2.36 N/A2 5.07 3. Validation Statistics for Type I Injury Accident Model Using Washington Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Washington Data 1993-96 181 0.Table 11.97 (0. 1998.17/year) 3.4. Table 12.712 0.62/year) Median 1. 64) N/A: not available The parameters were estimated with the same sign as the original model and in several cases with similar magnitude and significance. Deviation 2.2 Model II The summary statistics shown in table 12 indicate that there is little difference in the mean accident frequencies between original (1985–89) and additional (1990–98) years for the Minnesota sites.83 (0. Recall that the intersection related variables did not exactly match those statistics given in the report.00 Std. Total Accident Model The model was recalibrated with the additional years of accident data for the Minnesota sites. The Washington sites exhibit a large increase in the mean accident frequency and a larger variability between sites.00 1. of Sites 327 327 315 315 90 90 Mean 1. although there is more variation in the latter data.51 (0. their standard errors. and p-values are provided in table 13. Accident Summary Statistics for Type II Sites Dataset Minnesota Total (85-89)1 Minnesota Injury (85-89)1 Minnesota Total (90-98) Minnesota Injury (90-98) Washington Total (93-96) Washington Injury (93-96) 1 2 No.50 (0.77 (0.08 2.505 -0.17 2.99/year) 2. which reveals differences in the parameter estimates of the variables between the two time periods.15/year) 2.

0001) 0.0449 (0.0049 (0.018) -0. 0. 0.00006 (0. <0.4260 (1..e. 0.6091 (0.0159. 0.2875) 0.large difference in magnitude.001) 0.001) 0. 0. The Pearson product-moment correlation coefficient was higher for the original data than for the original model applied to the additional years.00141.0001) 0.2628) 0.001) 0.066 (0.0176.123) 0. <0. 0.741) 0.6673 (0.0519.6026 (0.967) 0. 1998.0836.0033.. 0.0694.385 A comparison of validation measures for the original data additional years is shown in table 14. 0.08.0001) 0.0702 (0.2576. was almost twice as large as the original model. <0. The overdispersion parameter. 0. p-value) 1 -10. The MADs per year are similar.0768. the MSPE per year squared is higher than the MSE per year squared. p-value) -10.0187 (0.6135 (0. Parameter Estimates for Type II Total Accident Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 DRWY1 HAU K Vogt and Bared.74 (1. 0. For the additional years model.0622.0130 (0. 0. (p. 0. 0.e. K.3167.3431) 0.415) 0.0417. indicating that the model is not performing as well on the additional years of data.205 Additional Years Estimates (s. Table 13. 115) 2 K: Overdispersion value 1 2 Original Estimates (s.199.0173) -0. 27 .1235 (0.0988 (0.0473.0456.1341) 0.2885 (0.

207 2. 28 .392 -0.017 -0. VCI1. The parameter estimates.364 0. Validation Statistics for Type II Total Accident Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE 1 Original Data 1985-89 327 0. and p-values are provided in table 15. The other parameters were estimated with the same sign but with generally large differences in magnitude.000 0. their standard errors.003 1.185 MSPE/yr N/A: not available 2 The model was also recalibrated with the Washington data.060 0. was estimated to be over three times as large as that for the original model. HI1. K.043 2. which might be expected on the basis of the large standard errors both in the original model and the model estimated using Washington data. which reveals differences in the parameter estimates of the variables between the two States. this would seem to indicate that variables with large standard errors in the parameter estimates should not be included in the model even if engineering common sense suggests they should.229 N/A1 15. and HAU were estimated with opposite signs.034 0.Table 14. The overdispersion parameter. Again.668 -0.095 N/A1 Additional Years 1990-98 315 0.760 -0.

4260 (1.017 -0.26 MSPE/yr N/A: not available 29 . Parameter Estimates for Type II Total Accident Model Using Washington Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 ND HAU 1 Original Estimates (s.517 -0. The MAD per year is much higher for the Washington data.0118 (0.095 N/A1 Washington Data 1993-96 90 0.0694. 0. <0. 0.0187 (0. 0.2875) 0.0473.2230.0173) -0.0001) 0.884) 0.64 (2.032) -0.051) 0. 0. 0.2576.001) -0.1646 (0..e.034 0.2030.19 0.0836. The Pearson product-moment correlation coefficient was higher for the original data as compared to the Washington data.454) 0.0519.7340 (0.667 K2 Vogt and Bared.1235 (0.205 Washington Data Estimates (s.Table 15. 0.0185.0176.0001) 0.2885 (0.3431) 0.364 0.0033.11 3.3680 (0. 1998.0251 (0.1080. 0.15 1.001) 0. Validation Statistics for Type II Total Accident Model Using Washington Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/year MSPE 1 2 2 Original Data 1985-89 327 0. 0.0140 (0. Table 16. 0. 0.207 2. (p. 0.760 -0. p-value) -8.0001) 0.0766. 0. 0. p-value)1 -10. 115) 2 K: Overdispersion value A comparison of validation measures for the original data and the original model applied to the Washington data is shown in table 16.1341) 0. 0. indicating that the model is not performing well on these data.003 1.6091 (0.2628) 0.80 N/A1 20.0129.0815.42 -0.071) 0.3167.6026 (0.0049 (0.950) 0..0139 (0. <0.06.e.0449 (0.

148.0948.2513) 0.668).0889) 0.6727 (0.014 (0. p-value)1 -10. 0. 117) 2 K: Overdispersion value Validation measures for the additional years of data are provided in table 18.2789 (0.671) than that for total accidents (0. The Pearson product-moment correlation coefficient was lower (0. HAU was estimated with the opposite sign to the original model and with a large standard error.7656.0001) 0. was estimated to be more than twice that of the original model.3258) 0.6567) -0.6497 (0.19 (1.0870. 30 .e.0549.924) 0.0815.0513. 0.0639. 0. but in some cases with large differences in magnitude.0635. Table 17.0043 (0.Injury Accident Model For the injury accident model..650) 0.0287 (0. 0.0749. the parameter estimates. 0.159) -0.1799) 0. and p-values are provided in table 17..7829 (1. Because the original injury accident counts were not obtained these measures are not provided for the original years.0251. and the MAD per year is about twice that for total accidents. 0. (p.240.0857 (0.e. 0.6339 (0. K.0935 (0. Parameter Estimates for Type II Injury Accident Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU ND 1 Original Estimates (s.1665. The other variables were estimated with the same sign as the original model.00176. their standard errors. 0.4623.080) 0.6229 (0.0001) 0. 0.1811 Additional Years Estimates (s.0720. 1998. The overdispersion parameter. <0.109 (0.001) 0. <0.00105 (0.0044.1225 (0.001) 0. 0.0791 (0.1313 (0. <0.123) 0. 0.1055.5464) 0. 0.089) 0.0729 (0.552) 0. 0.0001) 0.0204. 0. 0.39. 0.0451 (0. p-value) -12. 0.435 K2 Vogt and Bared.7865) -0.001) 0.0112 (0.

Table 18. and p-values are provided in table 19.170 1.230 0.130 0. K.152 -0. The overdispersion parameter. Validation Statistics for Type II Injury Accident Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Additional Years 1990-98 315 0. and HAU were estimated with opposite signs and large differences in magnitude compared to the original estimates. The parameter estimates.517) and the MADs per year are about sixty percent of that for total accidents. The other variables were estimated with the same sign but with generally large differences in magnitude. The Pearson product-moment correlation coefficient was lower (0.482) than that for total accidents (0. their standard errors. which reveals differences in the parameter estimates of the variables between the two States. RT MAJ. Table 20 shows the validation statistics for the original injury accident model applied to the Washington data.641 -0.137 5. HI1. was estimated to be more than three times that of the original model. 31 .063 The model was also recalibrated with the Washington data.

29 and 0.0589 (0.345 0.080) 0.0639.008) 0.86 0.1055. 1998.0044.025 (0.130.0220 (0. For example. 0.7829 (1. p-value) -11. The means 32 .299 (0.0870.181 Washington Data Estimates (s.071 0.775 (0.0251.025) -0.7865) -0.0112 (0.223.62 (2. <0.0043 (0.845 (0.1799) 0.55 2. 0.0001) 0.486) 0. Note that the statistics compare 3 years of accident data in the original data to 2 years of accident data for validation using the additional years of data. 0.2789 (0. Parameter Estimates for Type II Injury Accident Model Using Washington Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU ND 1 Original Estimates (s. 117) 2 K: Overdispersion value Table 20.0001) 0. Validation Statistics for Type II Injury Accident Model Using Washington Data Measure Number of sites Pearson product-moment correlation coefficient MPB MPB/yr MAD MAD/yr MSPE MSPE/yr 2 Washington Data 1993-96 90 0..3 Model III The summary statistics shown in table 21 indicate that there are differences in the accident frequencies between the original (1993–95) and additional (1996–97) years. 0.001) 0.0729 (0. 0. 0.911) 0.4623. (p. 0.024) 0. <0. 0.086 2. 0.0720. p-value)1 -10..87. 0.135.0136. respectively.0451 (0.7656.292.1225 (0.504 (0.022) -0.001) -0.e.1740 (0. 0.6567) -0.482 0.556 K2 Vogt and Bared. 0.0463 (0.Table 19. 0.0207.39. 0. 0.518 8.0857 (0.4.3258) 0. 0. 0.2513) 0.e.6229 (0. 0.0992.6339 (0.223.5464) 0.1665.0889) 0.105) 0.0001) 0.0845.0635. the means per year of TOTACC and TOTACCI for 1993–95 are 1.

per year of TOTACC and TOTACCI for 1996–97 are 0.75 and 0.62, respectively. Note that the 1993–95 data for the original INJACC and INJACCI models could not be obtained. Table 21. Accident Summary Statistics of Type III Sites
Dataset TOTACC (93-95)1 TOTACC (96-97) TOTACCI (93-95)1 TOTACCI (96-97) INJACC (96-97)
1

Mean 3.88 (1.29/year) 1.50 (0.75/year) 2.62 (0.87/year) 1.23 (0.62/year) 0.55 (0.28/year) 0.46 (0.23/year)

Median 2 1 1 0 0 0

Std. Deviation 4.33 2.42 3.36 2.09 1.19 1.02

Minimum 0 0 0 0 0 0

Maximum 19 12 13 10 7 7

INJACCI (96-97) Vogt, 1999, (p. 53)

Total Accident Model (TOTACC) The models were recalibrated with the additional years of accident data. The parameter estimates, their standard errors, and p-values are provided in table 22. All of the variables were estimated with the same sign as in the original model, but the constant term and AADT2 were estimated with larger differences in magnitude than the other parameters. MEDWDTH1 and DRWY1 became insignificant for the recalibration with the additional years of data, and the overdispersion parameter, K, was almost twice as large as for the original model (for a discussion of K, see section 2.3). Table 23 shows a comparison of GOF measures between the original main models in the Vogt report (1999) applied to the original 1993–95 data and to the 1996–97 data. The Pearson product-moment correlation coefficient was higher for the original years of data than for the additional years of data. The MAD per year was similar, but the MPB per year was larger for the model for the additional years of data. The MSPE per year squared for the additional years was higher than the MSE per year squared for the original years, but the difference was not great.

33

Table 22. Parameter Estimates for TOTACC Type III Model Using Additional Years of Data
Variable Constant Log of AADT1 Log of AADT2 MEDWIDTH1 DRWY1 Original Estimates (s.e., p-value)1 -12.2196 (2.3575, 0.0001) 1.1479 (0.2527, 0.0001) 0.2624 (0.0866, 0.0024) -0.0546 (0.0249, 0.0285) 0.0391 (0.0239, 0.1023) 0.3893 Additional Years Estimate (s.e., p-value) -14.4477 (3.3378, 0.00001) 1.1597 (0.3677, 0.0016) 0.5214 (0.1286, 0.00001) -0.0515 (0.0440, 0.2414) 0.0127 (0.0439, 0.7719) 0.6356

K2 Vogt, 1999, (p. 111) 2 K: Overdispersion value
1

Table 23. Validation Statistics for TOTACC Type III Model Using Additional Years of Data
Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE
1 2

Original Data (1993-95) 84 0.66 -0.01 0.0 2.26 0.75 11.01 1.22 N/A1

Additional Years (1996-98) 84 0.54 1.07 0.53 1.75 0.88 N/A1 5.57 1.39

MSPE/yr N/A: not available

2

Intersection Related Total Accident Model (TOTACCI) The parameter estimates, their standard errors, and p-values are given in table 24. All of the variables were re-estimated with the same direction of effect as the original model, but the constant term and AADT2 were estimated with larger differences in magnitude. MEDWDTH1 and DRWY1 became statistically insignificant for the additional years data, and the overdispersion parameter was slightly higher than that for the original model.

34

Table 24. Parameter Estimates for TOTACCI Type III Model Using Additional Years of Data
Variable Constant Log of AADT1 Log of AADT2 MEDWIDTH1 DRWY1 Original Estimates (s.e., p-value)1 -15.4661 (3.4685, 0.0001) 1.4331 (0.3608, 0.0001) 0.2686 (0.0988, 0.0065) -0.0612 (0.0360, 0.0888) 0.0560 (0.0289, 0.0525) 0.5118 Additional Years Estimate (s.e., p-value) -16.7738 (3.8495, 0.00001) 1.3497 (0.4098, 0.0010) 0.5439 (0.1434, 0.0001) -0.0308 (0.0465, 0.5074) 0.0358 (0.0486, 0.4608) 0.7220

K2 Vogt, 1999, p112 2 K: Overdispersion value
1

Table 25 shows a comparison of GOF measures between the original main model in the Vogt report applied the original data and to the 1996–97 data.(2) The Pearson product-moment correlation coefficient for the additional years of data was lower than that for the original years. The MAD per year was similar, but the MPB per year was slightly larger for the additional years of data. The MSPE per year squared for the additional years was higher than the MSE per year squared. Table 25. Validation Statistics for TOTACCI Type III Model Using Additional Years of Data
Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE
1

Original Data 1993-95 84 0.67 -0.005 -0.002 1.76 0.59 6.50 0.24 N/A1

Additional Years 1996-97 84 0.52 0.52 0.26 1.29 0.65 N/A1 3.54 0.89

MSPE/yr N/A: not available

2

35

Injury Accident Model (INJACC) There were two variants of the original model for injury accidents. These were validated separately. The original injury accident counts were not obtained. Thus, a comparison of prediction performance measures for the original data and additional years could not be accomplished. Variant 1 For the recalibration for the additional years of data, the parameter estimates, their standard errors, and p-values are given in table 26. All of the variables were estimated with the same sign as the original model, but that the constant term and AADT2 were estimated with larger differences in magnitude than the other parameters. HAU became insignificant for the additional years of data, and the overdispersion parameter was about half of that for the original model. Table 26. Parameter Estimates for INJACC Type III Model Using Additional Years of Data: Variant 1
Variable Constant Log of AADT1 Log of AADT2 HAU Original Estimates (s.e., p-value)1 -12.3246 (2.8076, 0.0001) 1.1436 (0.2763, 0.0001) 0.1357 (0.1029, 0.1872) 0.0230 (0.0131, 0.0790) 0.3787 Additional Years Estimate (s.e., p-value) -15.7264 (4.3853, 0.0006) 1.2659 (0.4347, 0.0036) 0.3883 (0.1574, 0.0136) 0.0140 (0.0165, 0.3969) 0.1740

K2 Vogt, 1999, (p. 113) 2 K: Overdispersion value
1

Table 27 shows the GOF measures for the original injury accident model (Variant 1) in the Vogt report applied to the additional years of data.(2) The Pearson product-moment correlation coefficient (0.37) was lower than that (0.54) for TOTACC. However, the MPB, MAD, and MSPE per year squared were smaller than those for TOTACC. Variant 2 The parameter estimates, their standard errors, and p-values are given in table 28, which reveals that the variables AADT1, AADT2, HAU, and ABSGRD1 were estimated with the same sign as for the original model, but DRWY1 was estimated with an opposite sign, although its estimate was statistically insignificant. HAU and ABSGRD1 were also statistically insignificant for the additional years. AADT2 was estimated with a higher magnitude and level significance than it was for the original model. The overdispersion parameter was much higher than that for the original model.

36

Table 27. Validation Statistics for INJACC Type III Model Using Additional Years of Data: Variant 1
Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr
2

Additional Years 1996-97 84 0.37 -0.15 -0.07 1.20 0.60 3.76 0.94

Table 28. Parameter Estimates for INJACC Type III Model Using Additional Years of Data: Variant 2
Variable Constant Log of AADT1 Log of AADT2 HAU DRWY1 ABSGRD1 Original Estimates (s.e., p-value)1 -11.0061 (2.6937, 0.0001) 0.9526 (0.2843, 0.0008) 0.1499 (0.0916, 0.1018) 0.0289 (0.0105, 0.0061) 0.0481 (0.0262, 0.0664) 0.1838 (0.1130, 0.1038) 0.2588 Additional Years Estimate (s.e., p-value) -14.3764 (4.5820, 0.0028) 1.0147 (0.5101, 0.0467) 0.5327 (0.2667, 0.0458) 0.0202 (0.0155, 0.1936) -0.0493 (0.0853, 0.5633) 0.2565 (0.1987, 0.1967) 0.9259

K2 Vogt, 1999, (p. 113) 2 K: Overdispersion value
1

Table 29 shows the GOF measures for the original injury accident model (Variant 2) in the Vogt report applied to the additional years of data.(2) The Pearson product-moment correlation coefficient (0.38) was lower than that (0.54) for TOTACC. However, the MPB, MAD, and MSPE per year squared were smaller than those for TOTACC.

37

Table 29. Validation Statistics for INJACC Type III Model Using Additional Years of Data: Variant 2
Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr
2

Original 1993-95 Model 1996 to 1997 84 0.38 -0.16 -0.08 1.17 0.58 3.73 0.93

2.4.4 Model IV The summary statistics shown in table 30 reveal that the accident frequencies were lower during the additional years. Mean, median, and maximum of the accident frequencies were almost half of those for the 1993–1995 period. Recall that the original 1993–95 INJACC and INJACCI data were not obtained. Table 30. Accident Summary Statistics of Type IV Sites
Dataset TOTACC (93-95)1 TOTACC (96-97) TOTACCI (93-95)1 TOTACCI (96-97) INJACC (96-97)
1

Mean 5.53 (1.84/year) 2.67 (1.34/year) 4.13 (1.38/year) 2.33 (1.17/year) 1.43 (0.72/year) 1.26 (0.63/year)

Median 3.5 1 2 1 1 1

Std. Deviation 6.52 3.60 5.37 3.21 1.96 1.80

Minimum 0 0 0 0 0 0

Maximum 38 16 27 13 9 8

INJACCI (96-97) Vogt, 1999, (p. 57)

Total Accident Model (TOTACC) The parameter estimates, their standard errors, and p-values are shown in table 31, which reveals that the constant term and all of the variables were estimated with the original sign, but some parameters had somewhat large differences in the magnitude compared to the original parameter. The overdispersion parameters were similar.

38

0.4841 (0.13 30.e.0432.5603 (0.0003) 0.1255.62 3.56 versus 0.. 0.0022 ) 0.2311.3342.11 N/A1 9.2779. 0.58).07 -0. 0.7258 (0.2803.53 2.56 -0.4968 (0.0456) 0. 116) 2 K: Overdispersion value Table 32 shows a comparison of the GOF measures for the original main model in the Vogt report and this model applied to the additional years of data. Table 32. The MPB per year was slightly larger for the model based on additional years.4631 (2. p-value)1 -9. p-value) -9.3294 (0.40 N/A1 Additional Years 1996-97 72 0. 0. 0. 0.0145) -0. (p. 0.6398 (3.0076) -0.8503 (0. while the MAD per year was similar.e.02 3.56 2. 1999.58 -1.Table 31..1100 (0.39 MSPE/yr N/A: not available 2 39 .1691. Validation Statistics for TOTACC Type IV Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE 1 Original Data 1993-95 72 0. The MSPE per year squared was lower than the MSE per year squared.22 1.06 -0.1056 (0. Parameter Estimates for TOTACC Type IV Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 LTLN1S Original Estimates (s.0909.0362) 0.4312 1 K2 Vogt.0087) 0.0038) 0.4578 Additional Years Estimate (s. 0.38 1. indicating that the model is performing fairly well on the additional years of data.0412.(2) The Pearson product-moment correlation coefficients of the original and additional years of data are similar (0.0299) 0.0033) 0.5991. 0.

The overdispersion parameter was slightly lower than that for the original model. the effect of the log of AADT2 on accident frequency was almost twice as large as for the original model. p-value)1 -11.3433.3764. Table 34.. The MAD per year was similar.1096 (3. In particular.7096 Additional Years Estimate (s.17 -0. 0.00 1.92 2.1673. 0.0339) 0.06 3. 0. but that there were some differences in magnitude.0067 ) 0. their standard errors. (p. 0. but the MPB per year was somewhat larger for the additional years of data.00 24.53 -0. 1999.6624 (0. Parameter Estimates for TOTACCI Type IV Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 Original Estimate (s.0110) 0. Validation Statistics for TOTACCI Type IV Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE 1 Original Data 1993-95 72 0.0024) 0.11 1.0586. 0.27 2.07 MSPE/yr N/A: not available 2 N/A1 40 .8796 (3.77 Additional Years 1996-97 72 0. 117) 2 K: Overdispersion value 1 Table 34 shows a comparison of GOF measures between the original main model in the Vogt report and for this model applied to the 1996–97 data. 0.6262 K2 Vogt. which reveals that the parameters were estimated with the same sign as the original model.e.7982 (0.3345.6865.e. 0. and p-values are shown in table 33.Intersection Related Total Accident Model (TOTACCI) The parameter estimates.0509) 0.53 -0. The MSPE per year squared was again lower than the MSE per year squared of the original model.1163.0024) 0. Table 33.1491 (0.29 2.1100 (0.(2) The Pearson productmoment correlation coefficient for the additional years of data was estimated as slightly higher than the original years. indicating a good fit to the additional years of data..0008) 0.3536 (0. 0.05 N/A1 8.9299 (0.0001) 0.0563.47 -0. p-value) -11.

Injury Accident Model (INJACC) Because the original injury accident counts were not obtained.0491) 0.2920 ) 0.3237 (0. The parameter estimates. 0.e.48) was lower than that (0. Table 35. and p-values are given in table 37. and p-values are given in table 35.9908.0557. 0.1672 (3. The parameter estimates. However.0577) 0. p-value)1 -12.58) for the TOTACC model. 0. the constant term and AADT1 were estimated with somewhat larger differences in magnitude than the other parameters. 0. a comparison of prediction performance measures for the original data and additional years could not be accomplished. (p.9505 (0.4344. their standard errors.0179.e.3825 (0.1645. which reveals that all of the variables were estimated with the same sign as the original model.5296 (2. their standard errors. but that the constant term and AADT1 were estimated with somewhat larger differences in magnitude than the other parameters. 0.4308 Additional Years Estimate (s.. The overdispersion parameter for the additional years was approximately half that of the original years.3720 K2 Vogt.0231) 0. 1999. 118) 2 K: Overdispersion value 1 Table 36 shows the GOF measures for the original injury accident model (Variant 1) in the Vogt report applied to the additional years data.0676) 0. and MSPE per year squared were smaller than those for TOTACC.3284..4074 (0. p-value) -8.1050 (0. which reveals that the constant term and all of the variables were estimated with the same direction of effect as that in the original model. Intersection Related Injury Accident Model (INJACCI) Since the original injury accident counts were not obtained. Parameter Estimates for INJACC Type IV Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 SPD2 Original Estimate (s.0594) 0.1793.0433.0038 ) 0. 41 .0001) 0. The overdispersion parameter was slightly lower than that for the original model.0216) 0. 0.(2) The Pearson product-moment correlation coefficient (0.0295) 0. 0.0402 (0.0994 (0.3630. However. 0. 0.0339 (0.0220. MAD. 0. a comparison of prediction performance measures for the original data and additional years could not be accomplished. the MPB.

4536 (0.7178 Additional Years Estimate (s.0791) 0.0815) 0. Validation Statistics for INJACC Type IV Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Additional Years 1996-97 84 0.66 3. However.1868.0614.4707 (0. 1999.e.Table 36. 0.9918 (0. 0. 0. p-value) -9..0613.0152) 0..3873 1 K2 Vogt. 0.1894.e.0008) 0.(2) The Pearson product-moment correlation coefficient was the same as that for the TOTACCI model.16 1.4268. 0. 0. and MSPE per year squared were smaller than those for the TOTACCI model. MAD.0240.1077 (0.0429 (0.48 -0. p118 2 K: Overdispersion value Table 38 shows the GOF measures for the original intersection related injury accident model (Variant 1) in the Vogt report applied to the additional years data.33 -0. 42 .5576 (3.13 0. 0.0201) 0.4112 (3.6620. 0.0229. Parameter Estimates for INJACCI Type IV Model Using Additional Years of Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 SPD2 Original Estimate (s.0173) 0.9998.1228 (0.0740) 0.78 Table 37.32 0. 0.3896.2270 ) 0. the MPB.0805) 0.3310 (0.0457) 0. p-value)1 -13. 0.0399 (0.

58 1.40/year) Median 21 7 17 6 2 2 Std.86 (3.5 Model V The summary statistics shown in table 39 indicate that fewer accidents per year occurred during the additional years (1996–97) than during the original years (1993–95).62 2.70 2.87 2. Deviation 11. both of which were validated.58 Minimum 2 0 1 0 0 0 Maximum 48 27 37 23 10 9 INJACCI (96-97) Vogt.53 -0. VEICOM was estimated with an opposite sign to that of the original model. but was not statistically significant in the recalibration. Main Model The parameter estimates.80 (1.82 2. and p-values are given in table 40.93/year) 3.25 0. Accident Summary Statistics of Type V Sites Dataset TOTACC (93-95)1 TOTACC (96-97) TOTACCI (93-95)1 TOTACCI (96-97) INJACC (96-97) 1 Mean 20. Table 39.76 (6. The constant term and PKLEFT2 also became 43 .37/year) 7. Validation Statistics for INJACCI Type IV Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr 2 Additional Years 1996-97 84 0. (p.27 0. note that the original years INJACC and INJACCI data were not obtained.92/year) 8. 1999.4.80 0.66 6.014 1. 61) Total Accident Models (TOTACC) The original report contained a main model and a variant.65 (4.16 (1.12 (5.58/year) 2. which reveals varying degrees of differences in magnitude and significance for the parameters for additional years data compared to those of the original model.33/year) 16.03 -0. their standard errors. Again.Table 38.

0.1857.(2) The Pearson product-moment correlation coefficient for the additional years of data was significantly lower than the original years.2392) -0.1152) 0.1824. 0.3948 (0. p-value) -7.0830) -0.0023) 0.0039) 0. 0.3635) 0. The MSPE per year squared for the additional years of accident data was much higher than the MSE per year squared. 0. Parameter Estimates for TOTACC Type V Model Using Additional Years of Data: Main Model Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK Original Estimate (s. 0.0143.0002) -0.3221 (0. 0. 0. 0..0262 (0.0098.2688. 122) 2 K: Overdispersion value 1 Table 41 shows a comparison of GOF measures between the original main models in the Vogt report applied to the original data and the 1996–97 data.1161 Recalibrated Estimate (s.. Table 40.9536 (2. 0.0894) 0. The MPB and MAD per year were larger for the additional years of data.0142 (0.2651 K2 Vogt. 44 .0115 (0. (p.4489.8238 (0. 0.0275) 0.0133) 0.2504.7911.e.1299 (0. 0.0625 (0.0022) -0.7450 (4.e. 0. p-value)1 -6.6754 (0. The overdispersion parameter was almost twice as large as for the original model.1737. 0.0047.0688.insignificant for the additional years of data.0154.0315 (0. 1999.7625 (0.7450.6199 (0. 0.045.0133) -0.0874) 0.0132) 0.

e.0236 (0.53 2. p-value) -6.3474. Validation Statistics for TOTACC Type V Model Using Additional Years of Data: Main Model Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original Data 1993-95 49 0.0102.1507.Table 41.0153.1186 Recalibrated Estimate (s. 0.1243 (0.42 N/A1 84. PKTRUCK became insignificant for the additional years of data. Parameter Estimates for TOTACC Type V Model Using Additional Years of Data: Variant 1 Variable Constant Log of AADT1*AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK K2 Vogt. 0. 1999. 0.1245) 0.0184) 0. 0.0300 (0.(2) The Pearson product45 .0507.3249) -0.0100 (0. (p. PKLEFT2. VEICOM was estimated with an opposite sign to that for the original model..18 77. 0.40 -5. 0. Table 42. their standard errors. p-value)1 -6. 0.75 21.4643 (0.0043) -0. 0.1236 (2.1091) 0.0142) 0.7273 (0.2801 Table 43 shows the GOF measures for the original accident model (Variant 1) in the Vogt report and the model applied to the additional years of data.3093) 0. 0.0052) 0..0001) -0.40 -0.4546 (0.0048. The constant term. 0.e.13 6.83 3. 0.0681.04 8.19 MSPE/yr N/A: not available 2 Variant 1 The parameter estimates.56 N/A1 Additional Years 1996-97 49 0.1940. VEICOM.0141.6110 (0. The overdispersion parameter was almost twice as large as for the original model. but it was not statistically significant in the recalibration.5973. 122) 2 K: Overdispersion value 1 Original Estimate (s.0566 (3.73 -0.0017) -0.2546.73 6.0331) 0.0692 (0.45 -2. 0.1483. and p-values are given in table 42.0191) -0.0134 (0.

93 MSPE/yr N/A: not available 2 Intersection Related Total Accident Model (TOTACCI) The main model and one variant were validated.70 20.moment correlation coefficient for the additional years of data was significantly lower than for the original years. their standard errors. The MPB and MAD per year were larger for the additional years of data. The MPB and MAD per year were larger for the additional years of data. and VEICOM became insignificant for the additional years of data. Validation Statistics for TOTACC Type V Model Using Additional Years of Data: Variant 1 Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original Data 1993-95 49 0.39 -5. The constant term. PKLEFT2. The MSPE per year squared was also higher than the MSE per year squared.15 N/A1 Additional Years 1996-97 49 0. Since the base model in the accident prediction algorithm is identical to Variant 3 of the Vogt model for TOTACCI.73 -0.43 -2. while VEICOM was estimated with an opposite sign to that of the original model. which again reveals differences in magnitude and significance in the parameter estimates. Table 45 shows a comparison of GOF measures between the original main model in the Vogt report for the original data and the original model applied to the 1996–97 data.16 73. Main Model The parameter estimates. 46 . The MSPE per year squared with the additional years of accident data was also higher than the MSE per year squared.31 8. The overdispersion parameter was almost twice as large as for the original model.37 -0.12 6.72 6.84 3. Table 43. but was not statistically significant in the recalibration. the Variant 3 model was also validated.42 N/A1 83.(2) The Pearson product-moment correlation coefficient for the additional years of data was significantly lower than for the original years.48 2. and p-values are given in table 44.

0.17 11. 0. 0..0165 (0.0937) 0. 0.54 4.1004) -0.5951 (0.28 -0.95 2.04 MSPE/yr N/A: not available 2 Variant 3 The parameter estimates.3834 (4.0365. 0.0685.7381 (0. and p-values are given in table 46.1972. p123 2 K: Overdispersion value Table 45.2935 (0. Validation Statistics for TOTACCI Type V Model Using Additional Years of Data: Main Model Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original Data 1993-95 49 0.47 N/A1 44. The overdispersion parameter was almost twice as large as for the original model.0036) 0.0057. 1999.e.08 -1.2000.7249 (0.0139.88 58.Table 44.1166) 0. 0.0020) 0. 0.2799) 0.4332.37 -3.0116 (0. p-value) -7. 0.24 6.0289 (0. their standard errors. but they were not statistically significant for the recalibrated model. 0.0740 (0. 0. 47 . p-value)1 -6.0095..0186) -0.0841 (3. 0.4708 (0.0131. Parameter Estimates for TOTACCI Type V Model Using Additional Years of Data: Main Model Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK Original Estimate (s. 0.62 -0.2254) -0. AADT2 and PKLEFT2 also became insignificant for the additional years.0724) 0.1366) -0.09 5.3110 (0.0366) 0. 0.3865. 0. VEICOM and DRWY1 were estimated with an opposite sign to those for the original model.2847.1126 (0.63 1.2702.47 N/A1 Additional Years 1996-97 49 0. which reveals that most of the variables showed some differences in magnitude and significance for the additional years.2640.2433 1 K2 Vogt.0943) 0.1313 Recalibrated Estimate (s.0276) 0.e.0233 (0.0063) -0.1893.

0144. is significantly lower than that for the original years.1114 (0. which reveals that the variables AADT1*AADT2 and PKLEFT2 were estimated with a similar degree of magnitude and significance as the original model.0177 (0.0098.Table 46.0326. 0. 0. 48 . but that the other variables showed larger differences in magnitude or significance.2015 (0.0005) 0.0463.2802.7009) 0.1145 Recalibrated Estimate (s.2527) -0.0256 (0. 0. 0.0685.0319) -0.1118) -0. and PKTRUCK became statistically insignificant.0909) 0. 0. their standard errors.4443. p-value)1 -5.4581 (3.e. VEICOM was estimated with an opposite sign to that for the original years. 0.0287) 0.7354 (0. 0.0117. 0. The overdispersion parameter for the additional years was higher than that for the original years.1883.0065) -0.3031) 0. 1999. and p-values are given in table 48. 0.0246. The MSPE per year squared with the additional years data was almost twice as high as the MSE per year squared.4041 (0. Parameter Estimates for TOTACCI Type V Model Using Additional Years of Data: Variant 3 Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK DRWY1 Original Estimate (s. which reveals that the Pearson product-moment correlation coefficient for the additional years of data.2412 1 K2 Vogt. 0.3760.0006) 0.0050.0112 (0.0247 (0.0319) 0.1038) 0.0705 (0..0407 (0.2235.1937.8110 (4. while PRO_LT turned out to be significant in the recalibration.3553 (0. suggesting a general lack-of-fit to the additional years of data.2932) -0.0873) -0. p123 2 K: Overdispersion value A comparison of GOF measures is given in table 47. p-value) -7. 0.. Injury Accident Model (INJACC) The parameter estimates. 0.2795.5995 (0. The MPB and MAD per year were larger for the models based on additional years of data. 0. 0.e.0874) 0. 0.1917. 0.7622 (0.0178 (0.0983) 0.

0678) 0.0822 (0. the MPB..4380 (3.1144) -0.1864.3093 (0.23 2.3907) 0. 1999.0267) 0.(2) The Pearson product-moment correlation coefficient was similar to that for the TOTACC model.2767) 0. 0.2303) 0.2943 (0.57 5.1408) 0. 0.0789) -0.10 -1.62 N/A1 45.67 -0.55 5.1722.31 -0. MAD.1630 Recalibrated Estimate (s.0203 (0. 0.2562 (2.0319 (0.0217.73 N/A1 Additional Years 1996-97 49 0.1707) -0.1219.2419. 0. 49 . 0.0113 (0.78 51. 0.0642 (0.10 5. However.2358 (0.0101. 0. 0. p-value) -4. 0.0323 (0.4734 (0.75 11.0443) -0.e.36 -3.0748.e.44 MSPE/yr N/A: not available 2 Table 48. 0. and MSPE per year squared were significantly smaller than those for the TOTACC model.9932.0504) -0.0062.1358) 0.0146. p124 2 K: Overdispersion value Original Estimate (s. 0.Table 47.34 1.1760. Parameter Estimates for INJACC Type V Model Using Additional Years of Data Variable Constant Log of AADT1*AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK K2 Vogt.0551. p-value)1 -3. 0.2124 1 Table 49 shows the GOF measures for the original injury accident model (Variant 1) in the Vogt report applied to the additional years of data.. Validation Statistics for TOTACCI Type V Model Using Additional Years of Data: Variant 3 Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE 1 Original Data 1993-95 49 0.

0066. 0.0250) 0. and MSPE per year squared were smaller than those for the TOTACCI model.5475 (3.3280) 0.(2) The Pearson product-moment correlation coefficient was similar to that for the TOTACCI model. The overdispersion parameter for the additional years of data was over twice as large as for the original years. However.0261.e.2000.0183 (0.0692.39 10. 0.0827.1757.5075) 0.1164) -0. 0. (p.4627) -0.0298.e.0282 (0.66 2. the MPB. their standard errors..1290 (0.84 -0.1858) 0.41 -1.. 0.92 2.67 Intersection Related Total Injury Accident Model (INJACCI) The parameter estimates.0149 (0. 0. p-value) -2. and p-values are given in table 50.6095) 0. 0. 0.0686 (0. Table 50.0548 (0. 0.1433 Recalibrated Estimate (s. 50 . 0.3554) -0. MAD.79 1. 1999. Parameter Estimates for INJACCI Type V Model Using Additional Years of Data Variable Constant Log of AADT1*AADT2 PKLEFT2 VEICOM PKTRUCK K2 Vogt. 124) 2 K: Overdispersion value 1 Original Estimate (s.Table 49. which reveals that all of the variables were insignificant for the additional years. VEICOM was estimated with an opposite sign for the recalibration. Validation Statistics for INJACC Type V Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr 2 Additional Years 1996-97 49 0.0152.1849 (0.0116.0628) 0.5686 (3.3496 Table 51 shows the GOF measures for the original intersection related injury accident model (Variant 1) in the Vogt report applied to the additional years of data. 0.5994) 0.5706. p-value)1 -1.0255 (0.

K.23 7.06 km (0. had. on average.45 1. Other variables used. higher values for the variables related to horizontal curvature.Table 51. HAZRAT1 was estimated with a similar degree of magnitude and significance as the original model. and p-values are provided in table 53. 51 . for example. HAU.04 miles).47 2. vertical curvature and roadside hazard rating.08 km (0. HI1.96 1. two sets of accidents were extracted—those within 0.99 2.05 miles) of the intersection and those within 0.04 mile limit) were estimated with opposite signs and large differences in magnitude. The overdispersion parameter.39 -0. the models were used to predict accidents for the Georgia data that also were used to re-estimate the models. The summary statistics reveal that Georgia sample had more accidents per year than the original Minnesota data.5 VALIDATION ACTIVITY 2: VALIDATION WITH GEORGIA DATA For this validation activity. This difference in underlying safety may be explained by the fact that Georgia sites. which reveals differences in the parameter estimates between the two States. were based on the 1997 road characteristic files maintained by the Georgia Department of Transportation and on data collected in the field during the summer of 2001. their standard errors. Total Accident Model The model was recalibrated with both sets of the Georgia accident data.1 Model I The summary statistics in the original report and for the Georgia data are given in table 52.95 -0. Recall that for Georgia data. The constant term. was much smaller for the Georgia data. AADT1. and RT (for the 0. The parameter estimates. AADT2. and SPD1 were estimated with the same sign but a larger difference in magnitude. all of which increase accident risk according to indications from the original model. Data from 1996 and 1997 in Georgia were used for accident related variables.5. VCI1. Validation Statistics for INJACCI Type V Model Using Additional Years of Data Measure Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr 2 Additional Years 1996-97 49 0. 2. such as roadway geometrics and traffic volumes.

04 MI) Georgia Total (0.88 55 45 2.644 (0.0 7.595 (0.12 1.2 N/A1 N/A1 N/A1 N/A1 N/A1 37.12/year) 1.Table 54 shows a comparison of validation measures between the original data and the Georgia data.00 1.9 33.13 Median 0.6 N/A1 N/A1 N/A1 N/A1 N/A1 Original Data Georgia Original Data Georgia N/A1 N/A1 N/A1 N/A1 N/A: not available 52 .31 52.59 (0.00 1. Summary of Georgia versus Minnesota Data for Type I Sites Variable and Abbreviation No.1 30.05 MI) Original Data Georgia VCI1 Original Data Georgia SPD1 Original Data Georgia HAZRAT1 Original Data Georgia DRWY1 Original Data Georgia RT MAJ Original Data Georgia HAU Original Data Georgia AADT1 on Major Road AADT2 on Minor Road 1 N 389 389 121 121 121 121 389 121 389 121 389 121 389 121 389 121 389 121 389 121 389 121 389 121 Mean 1.4%) with RT MAJ 117 (96.55 (0.73/year) 1.04 MI) Georgia Injury (0.45 (0.7%) without RT MAJ.00 0.00 3.00 23 25 1.00 0 0. 524 229 176 187 72 78 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 % Zero 51.3%) with RT MAJ -0.09 3687 3565 413 616 0 0 2313 3000 240 430 -90 -65 201 420 4. On a per year squared basis the mean squared prediction errors are much higher than the MSE indicating that the model is not performing well on the Georgia data.53 0.5 27.00 0.0 9 8 Freq.35 (0.53 70 85 60 19413 16900 4206 6480 N/A1 N/A 1 50.3 HI1 224 (57.32/year) 1.515 -3.2 60.00 2.5 0 0 Maximum 39 17 7 7 4 4 29 23. 165 (42.21 2. Table 52.00 0. 4 (3. The MPB and mean absolute deviations are also higher than for the original Minnesota data. of Crashes Original Data Total Original Data Injury Georgia Total (0.0 N/A1 53.30/year) 0.64 0.00 Minimum 0 0 0 0 0 0 0 0.11 2.57 1.00 55 55 5.26 2.3 54.00 0.6 61.27/year) 0.6%) without RT MAJ.00 0. The Pearson product-moment correlation coefficient was much higher for the original data as compared to Georgia.9 69.0 1.75 47.50 1.11 3.00 0.05 MI) Georgia Injury (0.00 4 14.78/year) 0.

04 Mile 1996 to 1997 121 0.0294 (0.03 0.0285 (0.130.0001) 0.2671 (0. <0.0177.108.05 Mile (s.088) 0.163) 0.9922 (1.060) -0.0001) 0.66 -0. 0.0677.0032.00326.0708.484 (0. p-value) -6.1398. 0.0477. It is quite evident that the original model failed to account for higher accident frequencies in most sites in the Georgia data. 0.00455 (0.0108) 0.0223 (0.21 0. p-value) -6.00 1.293) 0.0327.21 4.14..00686 (0.89 MSPE/yr N/A: not available 2 Figure 1 depicts the prediction performance of the original model for individual sites in the Georgia 0.0639.001) 0.127.15 0. 0. 0.64 0. 1998.2935.99 (1.64 N/A1 3.19 N/A1 Georgia Data 0. 0.443) 0.239 (0.23 1.e. (p. <0.001) 0.072) -0.0001) 0.272 (0.0914.56 0.00546 (0..0045 Georgia Data 0.05 Mile 1996 to 1997 121 0.55 0. 115) 2 K: Overdispersion value Table 54.2901 (0. 0.17.00320. <0.00947. 0.e. 0.625) -0. 0. 0.0273.28 1. p-value) -12. <0.580.481 1 Vogt and Bared.28 0.490.283 (0.427) 0. Parameter Estimates for Type I Total Accident Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU 2 Original Estimate1 (s. 53 ..389) 0.32 0.2048 (0.0561) 0.185 (0.5037 (0. 0.60 N/A1 3.111.0339 (0.0413 (0.0890.036) -0.04 Mile (s.Table 53.1072) 0.05-mile data.3004) 0. 0.192 Georgia Data 0.537) 0.0281.748) -0.001) 0.1642 (0.31 0.8052 (0. Validation Statistics for Type I Total Accident Model Using Georgia Data Measure Years used for the validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original Data 1985 to 1989 389 0.79 Georgia Data 0.443) 0.00894. 0.0209 (0.00995 (0.158 (0. 0. 0.0480. 0. 0.3229) 0.497 (0.47 0. 0. 0.1726 (0.e.021) 0.84 (1.01 0.1511.1578) K 0. 0. 0.001) 0. 0.

Observed versus Predicted Accident Frequency: Total Accidents Type I Injury Model The injury model was recalibrated with both sets of the Georgia accident data. none of the variables were estimated with satisfactory significance for the Georgia data. The overdispersion parameter. HI1. and SPD1 (for the 0. HAU. K. of Accidents 4 3 2 1 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101 105 109 113 117 121 Sites Figure 1. RT MAJ. 54 . which reveals differences in the parameter estimates of the variables between the two States. Aside from the AADT variables. Perhaps this should not be surprising given that only two years of accident data were available and injury accidents are relatively few compared to total accidents. The parameter estimates. was estimated to be approximately one half of that for the original model.8 Predicted Accidents Observed A id t 7 6 5 No. their standard errors. and p-values are provided in table 55.05-mile buffer only) were estimated with the opposite sign.

00189 (0. p-value) -7.. 0.0418.07.60 (1. 0.611 (0.464) -0. 0.0001) 0.174. 0.138.0714.1869 (0.270 (0. 1998.0048 (0.001) 0.04 Mile (s.563) 0.001) 0.3047) 0.439) -0. 0.0973. p-value) -13. 0.00492.< 0.001) 0.0374 (1.0694.362) 0.0001) 0.2065 (0.193. 0. 0. p-value) -7.459) -0.2594) -0.0335 (0.147 (0.171.0414.e.0102 (0.0244.5618) 0.00. Figure 2 depicts the prediction performance of the original model for individual sites in the Georgia 0.0977.768) 0.0714. 55 .0061 (0.806) 0.e.0269.94.494 1 Vogt and Bared.098 (0.. 0. 0. < 0.913) -0.0460) 0. 0. 0.05 Mile (s.0039 (0.e.186.00354 (0.101 (0.< 0..299 Georgia Data 0. (p. 0.Table 55.0051 (0. Parameter Estimates for Type I lnjury Accident Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU DRWY1 Original Estimate1 (s. It is quite evident that the original model failed to account for higher accident frequencies in most sites in the Georgia data. The Pearson productmoment correlation coefficients were quite low while the MAD was roughly half that for total accidents. 0.0001) 0.001) 0.0120 Georgia Data 0.3620 (0.8671) K2 0. 0. 0. 0. 116) 2 K: Overdispersion value The validation measures for the Georgia data is shown in table 56.0091 (0.828) 0. 0.8122 (0. 0. 0.0632 (0.0790.788.4551 (0.81 (1.869) 0.1814.7908.0233.842) 0.0045.0233 (0.56 (2.05-mile data.133.599) -0.0930. 0.937) -0.00478. < 0.6092) 0.269) -0.701) -0. 0.0413 (0.0327.087 (0. 0. 0.699 (0. 0.0778.149 (0.3657.0263) 0.0156 (0. 0.450) -0.

04 Mile 1996 to 1997 116 0.2 Model II The summary statistics in the original report and Georgia data are given provided in table 57. 56 .29 0.88 0.22 Georgia Data 0.Table 56. Validation Statistics for Type I Injury Accident Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr 2 Georgia Data 0.31 0.25 0.5 3 No.66 0.5 2 1.23 0.5 1 0.33 1.61 0.03 0. Observed versus Predicted Accident Frequency: Injury Accidents Type I 2. of Accidents 2.24 0.12 0.5 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101 105 109 113 117 121 Sites Figure 2.5.5 Predicted Accidents Observed Accidents 4 3.26 4.14 0.05 Mile 1996 to 1997 116 0. The summary statistics again reveal that the Georgia sample had more accidents per year than the original data.

50 55 55 6 6 150 50 14611 12300 3414 7460 Freq.49 1.03 0. K.8 59. was estimated to be more than twice as large that of the original model.05 Mile) Georgia Injury (0.5 N/A1 N/A1 N/A1 N/A1 Total Accident Model The model was recalibrated with both sets of the Georgia accident data.00 0. their standard errors. The constant term and the other variables were estimated with the same sign but with varying differences in magnitude and significance.00 30 30 0 0 -120 -58 174 420 7 80 Maximum 16 9 12 7 12 7 9 14.00 1.00 1.50 0.4 37.55 2 7.9 50.9 3.04 MI) Georgia Total (0.3 59.00 0.51 (0.98 (0. and p-values are provided in table 58. 494 253 256 112 258 112 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 204 136 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 % Zero 39.00 0.19 -0.66 0.00 1.25 0.13/year) 0.3 28.49/year) 2.8 55.49/year) 0.00 1742 2000 192 430 Minimum 0 0 0 0 0 0 0 0.00 0 0.Table 57.13 1.02 0. VCI1. 57 .09 53.31 0. The parameter estimates.04 MI) No.0 45. The variables HI1.00 0.9 29. of Crashes Georgia Injury (0.27 2238 3073 308 614 Median 1.25 (1. which reveals differences in the parameter estimates of the variables between the two States.6 N/A1 N/A1 67. Summary of Georgia versus Minnesota Data for Type II Sites Variable and Abbreviation Original Data Total Original Data Injury Georgia Total (0.98 (0.00 0. and SPD1 were estimated with opposite signs while HAU was estimated to have no effect on safety for the Georgia data.89 55 47.15/year) 2.26 (1.05 Mile) Original Data HI1 Georgia Original Data VCI1 Georgia Original Data SPD1 Georgia Original Data DRWY1 Georgia Original Data HAU Georgia Original Data AADT1 on Major Road Georgia Original Data AADT2 on Minor Road Georgia 1 N/A: not available N 327 327 114 114 114 114 327 114 327 114 327 114 327 114 327 114 327 114 327 114 Mean 1. The overdispersion parameter.62 1.9 55.00 0 0.0 48.13/year) 0.6 40.30/year) 0.77 (0.97 48.

0. 0.134.493 (0.0449 (0.932) 0.00450.0473.32.154. 0.0158 (0.0707.627 (0.05-mile data.0940 (0.0104.0187 (0.e.184) 0.0165 (0. The MPBs and MADs are also higher than for the original Minnesota data.000) 0.0710.455 Georgia Data 0. 0. 0.0049 Georgia Data 0.4260 (1.0176.1235 (0.154.e. p-value) -7.627 (0.32.001) 0..133.192) -0.00000 (0.0519. 0.3167.0781. Parameter Estimates for Type II Total Accident Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 NODRWYS HAU 2 Original Estimate1 (s. p-value) -10. 0. 0.205 1 Vogt and Bared.001) 0.0836. <0. 0.00448.001) -0.0104.0402.695) -0.0033.0927 (0.e.0403. The Pearson product-moment correlation coefficient was much higher for the original data as compared to Georgia. 0. <0. p-value) -7.087) 0. indicating that the model is not performing well on the Georgia data..0438 (0.00038 (0. 0.500 (0.2576.0001) 0. Figure 3 depicts the prediction performance of the original model for individual sites in the Georgia 0. 0. The MSPEs are much higher than the MSE.0172 (0. 0.001) 0.455 (0. It is quite evident that the original model failed to account for higher accident frequencies in most sites in the Georgia data. 0.0694.6091 (0.445) -0. 58 . 0..0001) 0.1341) K 0.2628) 0. 0.0606 (0.21 (1.001) 0.0177 (0.3431) 0.04 Mile (s.05 Mile (s. 1998.2875) 0. (p.6026 (0. 1.0794.681) -0.Table 58.26 (1. 115) 2 K: Overdispersion value Table 59 shows a comparison of validation measures between the original data and the Georgia data.575) -0. 0. 0.0001) 0.001) -0.097) 0. <0. <0. 0.0173) -0.2885 (0.

04 Mile 1996 to 1997 114 0.94 1.73 14 Predicted Accidents Observed Accidents 12 10 No. Validation Statistics for Type II Total Accident Model Using Georgia Data Measure Number of years Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE MSPE/yr2 1 N/A: not available 2 Original Data 1985 to 1989 327 0.82 0.77 0.38 0.39 0.94 1.72 0.35 1.10 N/A1 Georgia Data 0.05 Mile 1996 to 1997 114 0.82 0.01 0.20 2.Table 59.73 Georgia Data 0. Observed versus Predicted Accident Frequency: Total Accidents Type II 59 . of Accidents 8 6 4 2 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101 105 109 113 Sites Figure 3.39 0.004 0.91 N/A1 6.36 1.00 1.91 N/A1 6.70 0.

0112 (0.0870. VCI1.Injury Accident Model The parameter estimates. Figure 4 depicts the prediction performance of the original model for individual sites in the Georgia 0.1055. 0.2789 (0. 0.0988.0635.0224 (0. which reveals that the variables HI1. 0.1665.85 (2. was estimated to be over twice that of the original model.7865) -0. HAZRAT1.0001) 0. 0.1225 (0. 0. Only one Georgia model is shown since.e. 0.and 0.1811 Georgia Data (s.339) 0.869 (0.0529.0096 (0. SPD1.0451 (0.915) 0. <0.397) -0. 0.0001) 0.0857 (0.00631. <0.1799) 0.7829 (1.094 (0.4623. and p-values are provided in table 60.00603 (0.0043 (0.181. The overdispersion parameter.04.05-mile buffers. 0.0251.001) 0.130. 0.111.070 (0. their standard errors.392 K2 Vogt and Bared.0001) 0. 0. 0. <0. p-value)1 -10.0114 (0.2513) 0.3258) 0. 115) 2 K: Overdispersion value The validation measures for the Georgia data are shown in table 61. 0. 0.766) 0. and HAU were estimated with the opposite signs. With the exception of the AADT variables none were estimated to be highly significant statistically.6229 (0. It is quite evident that the original model failed to account for higher accident frequencies in most sites in the Georgia data. p-value) -10. 0.0205.702 (0.660..908) 0. 1998.001) -0.6567) -0.0044. 0. Table 60.856) -0.e.0639.01.0729 (0. as indicated in table 57.05-mile data. The MAD was roughly half that for total accidents.180.6339 (0. the observed number of accidents at each of the Type II sites was equal for the 0.7656. 0.0889) 0. 60 . The Pearson productmoment correlation coefficients were higher than for total accidents but still quite low.. 0.274) 0.5464) 0.039 (0. Parameter Estimates for Type II Injury Accident Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 HI1 VCI1 SPD1 HAZRAT1 RT MAJ HAU DRWY1 1 Original Estimate (s. K.0720.001) 0. (p.

50 8 Predicted Accidents Observed Accidents 7 6 5 No.05 Mile 1996 to 1997 114 0.44 0.33 0.Table 61. Validation Statistics for Type II Injury Accident Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE MSPE/yr2 1 N/A: not available 2 Georgia Data 0.33 0.00 0.48 N/A1 2.95 0.44 0. of Accidents 4 3 2 1 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 Sites 65 69 73 77 81 85 89 93 97 101 105 109 113 Figure 4.04 Mile 1996 to 1997 114 0.17 0.17 0. which reveals that the Georgia sample had on average fewer accidents per year than the 61 .48 N/A1 2.95 0.5.50 Georgia Data 0.3 Model III The summary statistics in the original report and Georgia data are given in table 62.00 0. Observed versus Predicted Accident Frequency: Injury Accidents Type II 2.

05 Mile) Georgia (0.6 1. Table 62.7 36.0 34. and p-values are shown in table 63. emergency response services.04 Mile) Original Median Width Data on Major Road (MEDWIDTH1) No.08 1.0 3.7 61.5 32. The Georgia sites may also be safer because they have.9 42. etc. improved vehicles. which reveals that the constant term.).original data.6 5.05 Mile) Georgia (0.5 1. This implies that either the Georgia sites were relatively more safe than the sites selected for the original model.2 25.5 12870 13100 596 892 Median 2 1.02 0.5 44. of Driveways on Major Road (DRWY1) AADT1 on Major Road AADT2 on Minor Road 1 N 84 52 52 84 52 52 52 52 52 52 84 52 84 52 84 52 84 52 Mean 3.77 3. on average. 326 124 116 135 85 80 56 53 42 40 N/A1 N/A 1 % Zero 21.5 1 1 1 1 1 1 0 0 0 20 1 1 12050 12200 349 430 Minimum 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2367 6500 15 80 Maximum 19 12 12 13 11 11 8 8 8 8 36 63 15 9 33058 28600 3001 9490 Freq.04 Mile) Georgia (0.4 2. their standard errors.81 0.88 2. 62 .2 48. In addition.05 Mile) Georgia No.74 27. more than 50 percent of the sites in the original data had no median.1 57.3 N/A1 N/A1 N/A1 N/A1 Georgia Original Data Georgia Original Data Georgia Original Data Georgia 259 77 N/A1 N/A1 N/A1 N/A1 N/A: not available Total Accident Models (TOTACC) The model was recalibrated using the Georgia data.1 1.2 2. of (0.04 Mile) IntersectionGeorgia Type Crashes (0.05 Mile) (TOTACCI) Georgia (0.4 21.04 Mile) Original Data Georgia (0. while only 5 percent of sites in Georgia were without a median.62 1.8 42. or that the passage of time between the period for the original calibration (1993–95) and that for the validation data (1996–97) had resulted in an overall improvement in safety (due to many factors including improved roadway design. of Crashes Georgia (TOTACC) (0. Summary Statistics of Georgia Data: Type III Sites Variable and Abbreviation Original Data No. The parameter estimates. wider medians on major roads and fewer numbers of driveways than the original intersections.5 53.

945. 111) 2 K: Overdispersion value Table 64 shows the prediction performance statistics for Model III for TOTACC.163.300 DRWY1 K2 1 Vogt. The overdispersion parameters were lower than that for the original model. AADT1 was also estimated as insignificant.1479 (0.0024) -0. It is quite evident that the original model does not do a good job of predicting accidents at the Georgia intersections.e. 0. this finding was expected on the basis of the low Pearson product-moment coefficients for the Georgia data.3575.0750) 0. (p. p-value) -8. Other validation statistics also suggest a poor fit of the original model to the Georgia data. but the difference was not great.0001) 1.374 Georgia Data 0. 63 .AADT1 and AADT2. 0. MEDWDTH2 and DRWY1 were estimated with an opposite sign.690 (4.05-mile data. Parameter Estimates for TOTACC Type III Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 MEDWIDTH1 Original Estimate1 (s.585.0249.2527.. p-value) -12.1059) 0. 0.459.0010) 0. 0.179.e.. 0.e. 0. 0.2196 (2.551 (0. The MSPE per year squared was almost twice as high as the MSE per year squared.011 (0. were estimated with the same sign but with large differences in magnitude.004 (0.9101) 0.9156) 0.013.009 (0.1737) 0.7748) -0.0285) 0.426. 0.0546 (0.2624 (0.1023) 0. 0. 0.0391 (0.2434) 0. Table 63.012.857 (4. Low Pearson product-moment correlation coefficients with the Georgia data indicate that the accident predictions by the original model are marginally correlated with the observed number of accidents in the Georgia data.0239. 0. p-value) -8. The MPB and MAD per year were larger than those for the original model.04 Mile (s.088.0021) 0. Figure 5 depicts the prediction performance of the original model for individual sites in the Georgia 0.094. 1999.0002 (0.580 (0.0001) 0. 0. 0. although they were not statistically significant.05 Mile (s. 0.536 (0.3893 Georgia Data 0.0866. 0.536 (0.9894) 0.

14 2.98 N/A1 9.36 2.01 0.93 1.66 -0.00 Predicted Accidents Observed Accidents 6.01 1. of Accidents 8. Observed versus Predicted Accident Frequency: TOTACC Type III 64 .00 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 5.05 N/A1 9. Validation Statistics for TOTACC Type III Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original Data 1993 to 1995 84 0.49 -0.22 Georgia Data 0.50 6.34 -0.09 -1.34 0.03 -1.00 2.45 5.41 MSPE/yr N/A: not available 2 N/A1 14.Table 64.00 4.00 10.05 Mile 1996 to 1997 52 0.00 2.00 No.04 Mile 1996 to 1997 52 0.00 0.75 11.26 0.64 2.00 12.

0012) 0.774 (4.5264) 0. AADT2. the variables AADT1. 0.012.0683) 0.. were lower than that for the original model.002 (0.4504) 0. The MSPE per year squared was almost twice as high as the MSE per year squared.3747) 0. 0. this finding was expected on the basis of the low Pearson product-moment coefficients for the Georgia data. MEDWITH1.0001) 0.097.0560 (0.e.740 (0. 0. A plot of the predicted versus actual accidents using Georgia data will help to understand prediction performances of the original model for the Georgia data.163 (4.4661 (3.262. p-value) -15.095 (0. 0. K.8682) 0. 0.232 (0. indicating a general lack-of-fit to the Georgia data. Other validation statistics also suggest lack-of-fit to the Georgia data. As shown in figure 6.2686 (0.7719) 0. 0. 0.339. The MPB and MAD per year were larger than those for the original model. The overdispersion parameters.301 (0. it is quite evident that the original model does not do a good job of predicting accidents at the Georgia intersections.4685.5118 Georgia Data 0. 0.126. The constant term and AADT1. 0.1165) 0.0065) -0. Low Pearson productmoment correlation coefficients with the Georgia data indicate that the accident predictions by the original model are marginally correlated with the observed number of accidents in the Georgia data. their standard errors. 0. (p. 65 . p-value) -7. Table 65.0612 (0. 0.4980) 0. 112) 2 K: Overdispersion value Table 66 shows the GOF statistics for Model III for TOTACCI.04 Mile (s.272 DRWY1 K2 1 Vogt. and DRWY1 were estimated as statistically insignificant.511. 0.e..004 (0.764 (0.Intersection Related Total Accident Model (TOTACCI) The parameter estimates.352 Georgia Data 0.0035) 0.e.0001) 1.0289.090 (0.0888) 0. 0.0988.05 Mile (s.133.4331 (0.229. and DRWY1 were estimated with the same direction of effect but with large differences in magnitude. 0. 0.3608.013.366. p-value) -8.0360. Similar to the model of TOTACC. and p-values are given in table 65. Parameter Estimates for TOTACCI Type III Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 MEDWIDTH1 Original Estimate1 (s. 1999.0525) 0.

002 1.04 Mile 0.Table 66.03 -0.48 0.95 0.67 -0.50 0.08 -1. Validation Statistics for TOTACCI Type III Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 1 N/A: not available Original Data 1993 to 1995 84 0. Observed versus Predicted Accident Frequency: TOTACCI Type III 66 .005 -0.14 1.76 0.93 1.97 N/A1 6.52 1.54 12 10 8 No.72 N/A1 Georgia Data 0.97 N/A1 5.13 -0. of Accidents 6 Predicted Accidents Observed Accidents 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 6.59 6.95 0.10 -1.05 Mile 1996 to 1997 1996 to 1997 52 52 0.56 1.

their standard errors.04 Mile (s. 0. 0.958 (5.531. and MSPE per year squared were smaller than those for the TOTACC model.3246 (2. 0. 113) 2 K: Overdispersion value 67 . 0.0131.000 (0.2774) 0.3787 Georgia Data 0.05 Mile (s. The overdispersion parameters.2923) 0.e. Table 67.0230 (0. 0. AADT1. and HAU became insignificant with the Georgia data. Parameter Estimates for INJACC Type III Model Using Georgia Data: Variant 1 Variable Constant Log of AADT1 Log of AADT2 Report Estimate1 (s.001 (0. 0.4602) 0. Table 68 shows the GOF measures for the original injury accident model (Variant 1) in the Vogt report applied to the Georgia data.255.0790) 0..010. 0.e.642 (6.553 HAU K2 1 Vogt.8076.1029.8886) 0. 0.009. the MPB.0001) 0. were higher than that for the original model. Variant 1 The parameter estimates.397.573. which reveals that the constant term and all of the variables were estimated with the same sign as in the original model.e. and p-values are given in table 67. 0.9743) 0. The constant term.0838) 0. 0.423 (0. p-value) -7.1872) 0. p-value) -6.1357 (0. 0.454 (0.Injury Accident Model (INJACC) The two original variants for model III were validated.. but there were large differences in their magnitudes.4730) 0. 0. However.420 (0.682 Georgia Data 0.949.0001) 1. (p.(2) The Pearson product-moment correlation coefficients were similar to those for the TOTACC model.381 (0.243.0752) 0. K. 1999. MAD.2763.1436 (0. p-value) -12.

11 0.08 0.05 Mile 1996 to 1997 52 0.58 0.39 2.Table 68.23 0.05-mile data. Validation Statistics for INJACC Type III Model Using Georgia Data: Variant 1 Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Georgia Data 0.58 Figure 7 depicts the prediction performance of the original model for individual sites in the Georgia 0. It is quite evident that the original model does not do a good job of predicting accidents at the Georgia intersections.19 0. 68 .09 0.39 2. a finding that would have been expected on the basis of the low Pearson product-moment coefficients for the Georgia data.30 0.78 0.10 0.04 Mile 1996 to 1997 52 0.30 0.78 0.

The overdispersion parameter. Figure 8 depicts the prediction performance of the original model for individual sites in the Georgia 0. It is quite evident that the original model performs poorly when applied to the Georgia data. 69 . the MPB. which reveals that the constant term and all of the variables were estimated with the same sign as in the original model. was almost twice as high as for the original model. a finding that would have been expected on the basis of the low Pearson product-moment coefficients for the Georgia data.(2) The Pearson product-moment correlation coefficient was similar to that for TOTACC. Observed versus Predicted Accident Frequency: Injury Variant 1 Variant 2 The parameter estimates.05-mile data. of Accidents 5 Observed Injury Predicted Injury(Variant1) 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 7. and MSPE per year squared were smaller than those for TOTACC. Table 70 shows the GOF measures for the original injury accident model (Variant 2) in the Vogt report applied to the Georgia data. and there were large differences in the magnitudes of the parameters. their standard errors. However.9 8 7 6 No. all of the variables except AADT2 became insignificant. However. K. MAD. and p-values are given in table 69.

1499 (0. 0. 0. 0.11 0.39 2.e.0061 (2. p-value) -7.4663) 0.120.04 Mile 1996 to 1997 52 0. p-value) -8.0481 (0. 0.08 0.05 0.258.278.001 (0.0920) 0. 0.06 0. 0. 0.04 0.5734) 0.65 70 .15 0. 0.0008) 0. 0.223.61 0.77 0.7488) 0.e.9526 (0.1038) 0.0664) 0. 0.1838 (0.0916. Parameter Estimates for INJACC Type III Model Using Georgia Data: Variant 2 Variable Constant Log of AADT1 Log of AADT2 HAU DRWY1 Report Estimate1 (s.1130.415.4678) 0.2588 Georgia Data 0.225 (0.666 Georgia Data 0.0262. 0.0001) 0. p-value) -11. 0.0061) 0.65 0.6937.Table 69..627.39 2.2803) 0.2962) 0.457 (0.457 (0.2843. 113) 2 K: Overdispersion value Table 70.05 Mile 1996 to 1997 52 0.8764) 0. 0.0771) 0. Validation Statistics for INJACC Type III Model Using Georgia Data: Variant 2 Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE MSPE/yr2 Georgia Data 0.61 0.010.410 (0.468 (0.04 Mile (s.571.002 (0.038 (0. 0.786 (6.e.501 ABSGRD1 K2 1 Vogt.05 Mile (s.151.1018) 0. 0. 0.238 (7.. 0.167 (0. (p.0289 (0.0105. 1999.9046) 0.010.565.5871) 0.439. 0.7042) 0.085 (0.77 0.

All of the sites in Michigan had no LTLN1S. Michigan. Pearson correlations of the original data. and Georgia. of Accidents 5 Predicted Injury(Variant1) Observed Injury 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 8. Georgia. Recall that the Pearson correlation reflects the degree to which the two variables are linearly related. while 17 percent in the Georgia data were without a left-turn lane.5. Since the variable was not present in the Georgia data. and California are given in table 73. The summary statistics for all of the three States (California. Pearson correlations for these variables with the response (accident frequency) were computed for all three States—California. Unlike the 71 . Michigan. The observation that the coefficients for AADT 2 and LTLN1S estimated using Georgia data resulted in opposite signs than the original model required further investigation.4 Model IV The summary statistics are provided in table 71. and Georgia) were also compared (refer to table 72). modifications to the validation procedure had to be performed. Observed versus Predicted Accident Frequency: Injury Variant 2 2. while frequencies of TOTACC and TOTACCI for Georgia were higher than for the California data. Peak left-turn percentage on major road was not available in the Georgia data. The summary statistics showed that about 31 percent of the sites in the original data had no left-turn lane. The variable was removed from the original model by dividing both sides of the model equation by the exponential value of the coefficient of the variable times its average effect (the average effect of PKLEFT1 is the average value of PKLEFT1 in the calibration data). since this variable would be too costly to collect in the field.9 8 7 6 No.

5 13.04 Mile) Georgia (0.7 38.original data.5 13.04 Mile) Original Left-Turn Lanes Data on Major Road (LTLN1S) Georgia Peak Left-Turn Original Percentage on Data Major Road (PKLEFT1) Georgia AADT1 on Major Original Data Road Georgia Original AADT2 on Minor Data Road Georgia 1 N/A: not available N 72 52 52 72 52 52 52 52 52 52 72 52 72 Mean 5.0 2 3.6 17.8 Median 3. The variable LTLN1 is positively related with TOTACC and TOTACCI in Georgia and California (not significant).0 3.5 30.9 36.5 32.67 0.05 Mile) No.05 Mile) Injury Crashes Georgia (INJACCI) (0.1 3.04 Mile) Original Data No.0 9.8 2. 398 222 217 297 160 159 107 104 87 87 N/A1 N/A1 N/A1 % Zero 12.5 38.0 1.2 26. of Crashes (TOTACC) Georgia (0.05 Mile) No.06 2. of Intersection-Type Georgia Crashes (0. but is negative and significant for the Michigan data.06 2.08 3. AADT2 in Georgia is estimated as negative linearly related with TOTACC and TOTACCI.5 4.5 22.0 1 1 1 1.0 9 1 1 13. Summary Statistics of Georgia Data: Type IV Variable and Abbreviation Original Data Georgia (0.27 4. Table 71.17 4. of Injury Georgia Crashes (INJACC) (0.7 0.0 9. but these correlations are marginal and statistically insignificant.3 5.67 1.51 Minimum 0 0 0 0 0 0 0 0 0 0 0 0 0 Maximum 38 13 13 27 11 11 9.5 4.6 N/A1 72 52 72 52 13018 13100 559 892 11166 12200 410 430 3350 6500 21 80 73000 28600 2018 9490 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 72 .0 1.04 Mile) No.0 3.0 2.96 Freq.7 32. of Georgia Intersection-Type (0.0 2.05 Mile) (TOTACCI) Georgia (0.

035 -0.000 N/A1 0.961 0. Max.558 AADT2 -0.000 N/A1 -0. TOTACC 4.461 0. (2-tailed) Pearson Correlation Sig. (2-tailed) Pearson Correlation Sig.5 0 38 4.Table 72.077 1.480 0.074 AADT1 0.096 0.6 N/A N/A N/A N/A LTLNS 0.000 N/A1 0.000 N/A1 0.000 0.096 0.4 3.250 0.000 -0.499 -0.4 8. Mean Median Min.169 0.035 TOTACCI 0.25 1 0 14 4.164 0.461 0.1 0.203 0.203 0.168 0.294 0. Max.05 mile Variable Pearson Correlation TOTACC Sig.108 0. Summary Statistics of California.365 0. (2-tailed) Pearson Correlation TOTACCI Sig. (2-tailed) 1 N/A: not available TOTACC 1.143 0.152 0.096 0.9 1 0 1 1 Summary Statistics for California and Michigan were produced using the obtained original data 2 Used TOTACC and TOTACCI for 0.294 0.096 0.035 0.210 0.247 0.219 0.293 0.3 4 0 13 TOTACCI 3.083 0.313 73 .008 0.152 0.000 N/A1 -0.000 N/A1 -0.956 0. (2-tailed) Pearson Correlation Sig. and Georgia California (N=54)1 Michigan (N=18)1 Georgia (N=52)2 Variable Mean Median Min.480 0.164 0.000 -0.5 0 27 3.5 2 0 21 6 4.499 0. (2-tailed) Pearson Correlation AADT2 Sig. (2-tailed) TOTACC 1.1 3 0 11 AADT1 13788 11250 3350 73000 10707 10550 5967 19383 12631 12831 5300 25800 AADT2 441 301 21 1850 913 733 254 2018 706 463 300 2990 PKLEFT1 2.934 0.156 AADT1 0.365 1.000 1.077 -0.501 -0.000 N/A1 0. Max.008 0.168 1.018 TOTACCI 0. and California The original data (N=72) TOTACC TOTACCI AADT1 AADT2 LTLN1 Variable Pearson Correlation Sig.956 1.934 0.8 11.247 0. (2-tailed) Pearson Correlation AADT1 Sig. (2-tailed) Pearson Correlation LTLN1 Sig.000 N/A1 0.000 1.93 1 0 1 0 0 0 0 0.961 0.108 0.279 0.05 mile Table 73. Pearson Correlations: Original. Mean Median Min.077 AADT2 0.501 0.2 3 0 22 9.000 -0. (2-tailed) Pearson Correlation Sig. Georgia.000 0.000 0.064 Georgia (N= 52): 0. Michigan.

052 0. (2-tailed) LTLN1 N/A: not available Total Accident Models (TOTACC) The parameter estimates. modifications to the validation procedure had to be performed as described earlier.146 0.987 0 1 N/A 1 AADT1 0. (2-tailed) Pearson Correlation Sig.804 0.035 0.508 0 -0. the same parameter estimates in the originally published report were used. For the Georgia data.074 0. (2-tailed) Pearson Correlation Sig. The effect of AADT1 became smaller.201 0.494 0 0. However. and California (Continued) California (N=54) Variable Pearson Correlation TOTACC TOTACCI AADT1 AADT2 Sig.804 1 N/A1 0.266 0.138 0. their standard errors. although AADT2 was insignificant. but the difference was not great.046 0.146 1 N/A 1 AADT2 0. The overdispersion values with the Georgia data were higher than for the original models. 74 .742 -0.319 Sig. (2-tailed) Pearson Correlation Sig. and the parameter estimates were also reproduced without PKLEFT1 for the revised original model (“Revised Estimates” in table 74).Table 73. (2-tailed) Pearson Correlation 1 TOTACC 1 N/A1 0.596 TOTACCI 0.201 0. In the validation. Pearson Correlations: Original.118 0.215 0. The values of the overdispersion parameter K for the Georgia data were lower than those for the original data. Georgia. AADT2 and LTLN1S were estimated with an opposite sign to the original model.494 0 0.987 0 0. while that of AADT2 became larger. In the revised original model. and p-values are given in table 74.035 0.215 0. all of the variables were estimated with the same sign but with large differences in magnitude.118 0. the constant term and AADT1 were estimated with the same sign as for the original models.508 0 0. Since the variable PKLEFT1 (peak left-turn percentage on major road) is not present in the Georgia data.

0. 0.417 LTLN1S K3 1 Vogt.097 (0. Pearson product-moment correlation coefficients. 0. 0.Table 74.599 (3..1255.380.110.241. 1999. Parameter Estimates for TOTACC Type IV Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 Original Estimates1 (s. the average value of PKLEFT1. The MPB and MAD per year for the Georgia data were larger than those for the original year data.501 (0. a finding that would have been expected on the basis of the low Pearson product-moment coefficients for the Georgia data.0860) -0.365.e.0362) 0. GOF measures of the revised original model.273 (0.0087) 0.2174) 0.0181) 0.097. 0. Values of 0.4578 Revised Estimates2 (s.373.382 Georgia Data 0. 0.0003) 0.8503 (0.229. 116) 2 Coefficient estimates of the variables were reproduced without PKLEFT1 using the original data 3 K: Overdispersion value 4 N/A: not available Since PKLEFT1 was not available in the Georgia data. It is quite evident that the original model does not fit the Georgia data well. The MPB per year was higher than that for the original model. 0. 0. indicate that it could be a good alternative to the original model. 0. since PKLEFT1 was not available.624 (0..0022 ) 0. 75 .2311.e.. 0.478 (0.377.e.3294 (0.0032) 0.085 (0.05-mile data.05 Mile (s. PKLEFT1 was removed from the original model by dividing by the exponential value of the coefficient of this variable times its average effect. p-value) -5.04 Mile (s. Figure 9 depicts the prediction performance of the original model for individual sites in the Georgia 0. 0. 0.245. 0. For the original model.0412.4841 (0. 0.5991. two models (original model and revised original model) were used for the validation activity to determine GOF measures.4631 (2.05 and 0.0875) -0.2173) 0. MAD per year. i. p-value) -5. 0.1100 (0.231.112 (0.653 (0. p-value) -9.0301) 0..e.0040) 0. p-value) -6. the same parameter estimates in the report were used.504 (0. but the difference was not great. 0.705 (2.0000) N/A4 -0.977.08 of the Pearson product-moment correlation coefficient indicate that the accidents in the Georgia data are not linearly related with the model-predicted values.6253) N/A4 1. For the revised original model. This could be the result of a significant nonlinearity in the data and original model. The MSPEs per year squared were also higher than the MSEs per year squared.2779.432. (p.553 Georgia Data 0. 0.e.6867) N/A4 1. shown in table 75. and MSE per year squared were similar to those for the original model.764 (4.0393) 0.0076) -0.

13 30.04 Mile) 1996-1997 51 0. Observed versus Predicted Accident Frequency: TOTACC 76 .56 -0.11 1.51 4.47 3.05 Mile) 1996-1997 51 0.56 1.Table 75.25 1.05 2. of Accidents 8 Predicted Accidents Observed Accidents 6 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 9.55 N/A4 18. Validation Statistics for TOTACC Type IV Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 2 1 2 Original Model1 1993-1995 72 0.12 3.47 4.62 3.86 Georgia3 (0.09 1.27 1.07 -0.66 3.13 3. This model includes PKLEFT1 2 Used the same coefficients in the original model.54 N/A4 17.16 32.49 1.08 2.63 MSPE/yr Used the original main model in the report.41 0.02 3.63 N/A4 Georgia3 (0.40 N/A4 Revised Original Model2 1993-1995 72 0. but PKLEFT1 was removed from the model by dividing by the exponential value of the coefficient of this variable times its average effect 3 Used the revised original model 4 N/A: not available 14 12 10 No.38 1.

Table 76.0130) (3.482.495.562 (0. As before. 0. 0.e.4582 0. 0.857 -7.603 (5.2501 -11. the same parameter estimates in the report were used.2568) -0.. The effect of AADT1 became smaller.0586.e. p-value) -4.2564) -0.497. 0.Intersection Related Total Accident Model (TOTACCI) The parameter estimates.857 1 Vogt.0110) PKLEFT1 N/A4 N/A4 3 K 0. which model (“Revised Estimates” in table 76) was included for the validation.0067 ) 0. For the original model.043 (0. In the alternative original model the constant term and parameter estimates of AADT1 and AADT2 were estimated with the same sign but with some difference in magnitude. 1999.1491 (0. For the Georgia data.325. The overdispersion values for the Georgia data were similar to that for the revised original model. and p-values are given in table 76. AADT2 was estimated with an opposite sign to that of the original models. it was statistically insignificant. p-value) Georgia Data 0.498.1071) (0.7096 0. the revised model showed similar prediction performance measures to the original model.0996.4755) 0..9094.1163. The constant term and AADT1 were also estimated as insignificant for the Georgia data.05 Mile (s.8957) N/A4 0.5311 (0.8996) Georgia Data 0.3433. p-value) -4. 0.326. 0. Since the report also developed a model with AADT1 and AADT2 only. the two models (original model and revised original model) were used for the validation.e.3536 0.604 (5. As was the case for the TOTACC models. 117) 2 The report presents this model developed with AADT1 and AADT2 only 3 K: Overdispersion value 4 N/A: not available 77 .e.9299 (0.2844. p-value) Revised Estimates2 (s.1096 (2.563 (0.0001) 0..3345. The overdispersion value was slightly higher than for the original model.041 (0. their standard errors. 0. 0. However.4770) 0. 0.8814 0.04 Mile (s. 0. 0. Parameter Estimates for TOTACCI Type IV Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 Original Estimates1 (s. while that of AADT2 became larger.0008) 0. 0. The prediction performance measures are shown in table 77.0024) (0. (p. 0. and the impact of the variable on the accident prediction was marginal.

05-mile data. The MSPEs per year squared were also slightly higher than the MSEs per year squared.29 N/A4 12.06 3.l7 of the Pearson product-moment correlation coefficients indicate that the accident predictions by the original models are not strongly linearly correlated with the observed number of accidents in the Georgia data.32 Georgia3 (0.16 and 0.85 2.17 -0. It is quite evident that the original model does not fit the Georgia data well.16 1.47 -0. there are several possible explanations for this.43 3.05 MSPE/yr Used the original main model in the report.28 0.54 1.76 0.88 2.Table 77.81 0. Validation Statistics for TOTACCI Type IV Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 2 1 2 Original Model1 1993-1995 72 0.90 2.00 24.77 N/A4 Revised Original Model2 1993-1995 72 0. The MPBs and MAD per year was larger than those for the original models.05 Mile) 1996-1997 51 0.27 N/A4 12.00 24. but PKLEFT1 was removed from the model by dividing by the exponential value of the coefficient of this variable times its average effect 3 Used the revised original model 4 N/A: not available Values of 0.17 1. This model includes PKLEFT1 2 Used the same coefficients in the original model. Figure 10 depicts the prediction performance of the original model for individual sites in the Georgia 0.18 3. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.00 1. 78 .08 3.76 N/A4 Georgia3 (0. Again.04 Mile) 1996-1997 51 0.92 2.59 1.00 1.47 1.

. 118) 2 PKLEFT1 was not included in the model 3 K: Overdispersion value 4 N/A: not available N/A4 0.3392) 0.4308 Georgia Data 0. Again. Parameter Estimates for INJACC Type IV Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 Original Estimates1 (s. (p.1990 ) -0.467.04 Mile2 (s.0001) 0.6100) Georgia Data 0..(2) 79 .031.6586) N/A4 0. Table 78. 0. 1999.912. and AADT2 was estimated with an opposite sign to that of the original model.162 (0.e. p-value) -5.778.12 10 8 No.e.3284.260 (4.0491) 0.0038 ) 0. 0.599 (0. and p-values are given in table 78. 0. 0.374. of Accidents 6 Predicted Accidents Observed Accidents 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 10.005 (0. p-value) -4.9505 (0.4018) 0.5296 (2.8732) 0. 0. Observed versus Predicted Accident Frequency: TOTACCI Injury Accident Model (INJACC) The parameter estimates.010 (0. 0.0994 (0.645 SPD2 K3 1 Vogt.. 0.191 (0.3237 (0.0179. 0.e.652 (0.0433. 0. p-value) -12. 0.9908.031.457. The overdispersion values for the Georgia data were higher than that for the original model.0216) 0.05 Mile2 (s.649 Table 79 shows the GOF measures for the original injury accident model (Variant 1) in the Vogt report applied to the Georgia data. 0. 0. all of the variables including the constant term were insignificant for the Georgia data.0577) 0. 0.367.1543 ) -0.1645.0339 (0.811 (4. their standard errors.7379) 0.

02 0.81 0.05 Mile 1996-1997 52 0.18 0. Validation Statistics for INJACC Type IV Model Using Georgia Data Measure Years used for the validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE 2 1 Georgia1 0.18 0.05-mile data.04 Mile 1996-1997 52 0.31 Used Variant 1.86 5. 80 . a finding that would have been expected on the basis of the low Pearson product-moment coefficients. and the MPB. MAD. but PKLEFT1 was removed from the model by dividing by the exponential value of the coefficient of this variable times its average effect Figure 11 depicts the prediction performance of the original model for individual sites in the Georgia 0.40 1.75 0.The Pearson product-moment correlation coefficients were higher those for the TOTACC models.38 1.23 MSPE/yr 1.67 0. and MSPE per year squared were smaller than those for the TOTACC models.84 5.73 0. Table 79.26 1. It is quite evident that the original model generally does not fit the Georgia data well.

and MSPE per year squared were smaller. The overdispersion values with the Georgia data were similar to that for the original model.10 9 8 7 No. but the MPB. The variable AADT2 was estimated with an opposite sign to that for the original model.(2) The Pearson product-moment correlation coefficient was similar to that for the TOTACCI model. Observed versus Predicted Accident Frequency: INJACC Intersection Related Injury Accident Model (INJACCI) The parameter estimates. MAD. their standard errors. Table 81 shows the GOF measures for the original intersection related injury accident model (Variant 1) in the Vogt report applied to the Georgia data. of Accidents 6 Predicted Injury(Variant1) Observed Injury 5 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 11. 81 . all of the variables were insignificant for the Georgia data. As was the case for INJACC. and p-values are given in table 80.

04 Mile2 (s.475 (5.57 1. 82 .189 (0.e.72 0.499. 0.0457) 0. 0.78 0.2590 ) -0. 0.15 1.475 (5. 0.86 1996-1997 52 0. Parameter Estimates for INJACCI Type IV Model Using Georgia Data Variable Constant Log of AADT1 Log of AADT2 PKLEFT1 Original Estimates1 (s.86 5.89 5.0429 (0. 118) 2 PKLEFT1 was not included in the model 3 K: Overdispersion value 4 N/A: not available Table 81.6430) N/A4 0.789 SPD2 K3 1 Vogt. 0. 0.0805) 0. It is quite evident that the original model generally does not fit the Georgia data well. 0.47 Used Variant 1. but PKLEFT1 was removed from the model by dividing by the exponential value of the coefficient of this variable times its average effect Figure 12 depicts the prediction performance of the original model for individual sites in the Georgia 0.8892) 0.9998.4803) 0. 0. Validation Statistics for INJACCI Type IV Model Using Georgia Data Measure 0.357.40 1. p-value) -4.Table 80.0201) 0.54 1.1894.005 (0.0740) 0.1228 (0.407.0240.14 0. (p.6430) N/A4 0..9918 (0.357. p-value) -13.05 Mile2 (s.564 (0.564 (0. 0.7178 Georgia Data 0.05-mile data. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.0008) 0.5576 (3.2590 ) -0.8892) 0. 0.407.005 (0. 0. 0.15 1.04 Mile Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE 2 1 Georgia Data1 0.499..59 MSPE/yr 1.3310 (0..789 Georgia Data 0.05 Mile 1996-1997 52 0.036.e. p-value) -4. 1999.e.036. 0.4268.08 0.189 (0.0614.4803) 0.

PKLEFT2 and PKTRUCK were not included in the Georgia data. some of the data acquired did not exactly match the summary statistics given in the report. Observed versus Predicted Accident Frequency: INJACCI 2. The summary statistics indicate that the Georgia intersections had fewer accidents. there was a problem reproducing vertical alignment related variables: VEI1. 83 . For example. The majority of the Georgia sites did not have a protected left-turn lane on the major road (PROT_LT). and VEICOM. about 10 percent of the Georgia sites did not have an accident during the period of 1996 and 1997.(2) Specifically. As mentioned previously. of Accidents 6 Predicted IINJACCI(Variant1) Observed INJACCI 5 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 12.5 Model V The summary statistics of the variables used in this model are provided in table 82. on average. while all of the original sites experienced at least one accident. while PROT_LT was present at almost a half of the sites in the original models. VEI2.5.10 9 8 7 No. than those in the original data.

71 2. 61-64) 2 N/A: not available N Mean 49 20.1 8. of Crashes Georgia (TOTACC) (0.05 Mile) Georgia (0.8 N/A2 N/A2 N/A2 N/A2 Separate summary statistics for three States.69 7.05 Mile) Georgia (0. % Zero 48 1017 0.1 25. 84 . were examined to see if there were differences in the variables between States.2 6.6 9.8 51 51 49 51 51 49 51 51 49 51 51 49 9. of Injury Crashes Georgia (INJACC) (0.5 27. shown in table 83.3 2.0 11.0 53 51 37 52 50 25 13 13 21 13 13 37.2 Maximum Freq.Table 82.07 489 473 790 445 433 366 118 113 301 110 106 N/A2 11.8 11. Summary Statistics of Georgia Data: Type V Variable and Abbreviation1 Original Data No.79 N/A2 73000 N/A2 28600 N/A2 2018 N/A2 9490 N/A2 7. (p.96 7. higher accident frequencies than California.7 8.5 4.05 Mile) (INJACCI) Georgia (0.7 4.05 Mile) (TOTACCI) Georgia (0. while the Michigan and Georgia data had no sites with this feature present. 1999.69 13018 13100 559 892 1. on average.2 Median Minimum 21 2 7 7 17 7 7 7 1 1 6 1 1 17. They also reveal that the majority of the sites in California had protected left-turn lanes (PROT_LT=1).97 0 0 1 0 0 0 0 0 0 0 0 4.04 Mile) Original Data No.04 Mile) Peak Left-Turn Original Data Percentage on Minor Road Georgia (PKLEFT2) Peak Truck Percentage Original Data (PKTRUCK) Georgia Original Data Protected Left Turn 0=NO (PROT_LT) 1=YES Georgia 0=NO 1=YES Combined VEI1 and Original Data VEI2 (VEICOM) Georgia AADT1 on Major Road Original Data Georgia AADT2 on Minor Road Original Data Georgia 1 Vogt.14 2.5 N/A2 N/A2 N/A2 N/A2 49 49 28 21 51 42 9 51 49 51 49 51 8.60 11166 12200 410 430 0 3350 6500 21 80 4.04 Mile) Original Data No. These summary statistics indicate that the Michigan sites had.8 13.5 25.47 2.1 18.2 2. of IntersectionGeorgia Type Injury Crashes (0. of IntersectionGeorgia Type Crashes (0.04 Mile) Original Data No.5 7.3 16.1 27.71 N/A 2 N/A2 N/A2 1.8 0.

the constant term. For the original model. K. The overdispersion parameter.5 18 1 37 8.13 1.65 9.18 0 0 1 VEICOM 1. and VEICOM were estimated with the same sign as the reported model.6 0 4.6 7 0 53 TOTACCI 13. and the parameter estimates were also reproduced without PKLEFT1 and PKTRUCK for the revised original model.43 2. AADT2 and VEICOM were also insignificant. was substantially higher than that for the original models. but all of them except PROT_LT were insignificant.91 1. 85 .94 1 0 1 0. and Georgia1 California (N=18)1 Georgia (N=51) Michigan (N=31)1 Variable Mean Median Min. For the revised original model. The validation addressed the main model and one variant. modifications to the validation procedure had to be performed as described earlier.54 1. Max.8 15 2 30 17.79 1 Summary Statistics for California and Michigan were produced using the obtained original data 2 N/A: not available Total Accident Models (TOTACC) Since the variables PKLEFT1 and PKTRUCK were not present in the Georgia data. AADT1.e. their average values.5 68. i.57 28. The overdispersion parameter.13 0 0 1 0.36 6.48 0 6. Two models (original model and revised original model) were used for the validation activity to determine GOF measures. PROT_LT was estimated with an opposite sign. the same parameter estimates in the report were used.7 7 0 52 AADT1 13048 12484 7500 25133 9007 8435 4917 17483 7798 7400 430 15200 AADT2 3630 3026 940 10067 4796 4434 1961 12478 2749 2200 420 10400 PKLEFT2 28.91 75.08 0 8. AADT2.75 1.07 2.17 25. Max. these variables were removed from the original published model by dividing by the exponential value of their coefficients times their average effects.69 15.97 45.89 8. although it was statistically insignificant.Table 83.45 9. was almost twice as high as that for the original model. Summary Statistics for Three States: California.73 N/A2 PKTRUCK 7. Max. Mean Median Min. without the variables PKLEFT2 and PKTRUCK.. In the validation. Main Model The model re-estimation results are shown in table 84.69 1.37 2. the same parameter estimates in the originally original report were used. Mean Median Min.83 25. For the Georgia data. For the revised original model. since PKLEFT2 and PKTRUCK were not available in the Georgia data.43 N/A2 PROT_LT 0. but there were differences in magnitude. the constant term and all of the variables were estimated with the same sign as the reported model.2 16 2 32 24 25 2 48 9. Michigan. K. TOTACC 15.

05 Mile (s. p-value) -6.e.222.264.1824. indicating a poor linear fit.755 (3.222 (0.0002) -0. The revised original model was estimated with the same Pearson product-moment correlation coefficient as that for the original model. 0. 0.1299 (0.2813) N/A4 0.. 0.2504.437.0039) 0. 0.308.0275) 0.432. 86 .0133) -0.1412) 0.e.04 Mile (s.9484) N/A4 0.045.18 of the Pearson correlation coefficient indicates that the accident predictions by the original model and the Georgia data are marginally correlated at best.0047.0372) N/A4 0. 0. p-value) -5.219 (0.Table 84. 0. (p.3761) 0.156.0023) 0.012 (0. 0. indicating a general lack-of-fit.0142 (0.4146) 0.084 (3. 0.179. The MSPEs per year squared were significantly higher than the MSEs per year squared. 0. 0. 0.0133) 0.463.1737.3948 (0. 0.731 Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK 1 K3 Vogt.0143.151.606 (0.1161 Revised Estimates2 (s.1922) N/A4 0.589 (0.0132) 0.216 Georgia Data 0. A value of 0.011 (0.430 (3.0315 (0. 0.2144) 0. 0. 1999. 0.e.1620) 0.325.094 (0.9522) N/A4 0..462 (0.604 (0.. 0.1036) 0.087. The MPB per year was larger than that for the original model.354.0623) 0.1782) N/A4 0.9536 (2. The MPB and MAD per year were larger than those for the original models.7911. The MSE per year was slightly higher than that for the original model.730 Georgia Data 0. Parameter Estimates for TOTACC Type V Model Using Georgia Data: Main Model Original Estimate1 (s.422 (0. 0. p-value) -4. 0.272 (0. The Pearson correlation coefficient for the Georgia data was relatively low. 122) 2 Coefficient estimates of the variables were reproduced without PKLEFT2 and PKTRUCK using the original data 3 K: Overdispersion value 4 N/A: not available The validation statistics are shown in table 85.6199 (0. 0.188. 0.1403) 0.6754 (0. p-value) -5.816.659. 0. while the MADs were similar. 0.1106) -0.e.575 (0.

53 2. was almost twice as high as that for the original model.56 N/A3 31. and p-values are given in table 86. AADT1*AADT2. It is quite evident that the original model generally does not fit the Georgia data well.06 -1.82 2. In the revised original model. 87 . and VEICOM were estimated with the same sign as the reported model. The overdispersion parameter K was significantly higher than that for the original models.53 8.04 2 Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 Revised 1993-95 Model2 1993-1995 49 0.38 -1. the constant term. the constant term.73 -0.59 4. The constant term and VEICOM were statistically insignificant. For the Georgia data. their standard errors.18 -3.06 -1.30 N/A3 130.50 MSPE/yr2 The original main model in the report.02 6. PROT_LT was estimated with an opposite sign.05-mile data. K. but the constant term and VEICOM were insignificant.05 Mile 1996-1997 51 0. but with similar degree of magnitude.18 77.40 -0.66 N/A3 Georgia2 0.01 8.44 0.69 8. but there were differences in magnitude.27 95. Variant 1 The parameter estimates.Table 85.50 4. Validation Statistics for TOTACC Type V Model Using Georgia Data: Main Model Original 1993-95 Model1 1993-1995 49 0. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 3 N/A: not available Figure 13 depicts the prediction performance of the original model for individual sites in the Georgia 0. The overdispersion parameter.13 6. AADT1*AADT2 and PROT_LT were estimated with the same sign as the reported model. This model includes PKLEFT2 and PKTRUCK 2 Used the same coefficients in the original model.25 N/A3 126.18 -3. without the variables PKLEFT2 and PKTRUCK.61 32.73 -3.92 10. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.04 Mile 1996-1997 51 0.

0.04 Mile (s.009 (0..141.e.0001) -0. 0.181.e.082. Observed versus Predicted Accident Frequency: TOTACC Main Model Table 86. 0.9601) PKTRUCK N/A5 N/A5 N/A5 4 0.0792) -0. of Accidents 30 Predicted Accidents Observed Accidents 20 10 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 13.0142) 0.766 (2. p-value) -4.1483. 0.682 (0.0052) 0. 0.0048.0051) N/A5 0.1186 1 Revised Estimates (s.05 Mile (s.60 50 40 No.309 (0.589 (3.e.0184) 0. 0.766 K 0.501 (0.1668) Georgia Data3 0. 0.415. Parameter Estimates for TOTACC Type V Model Using Georgia Data:Variant 1 Variable Constant Log of AADT1*AADT2 PROT_LT PKLEFT2 VEICOM Original Estimate (s.. 0. 0.373 (0.2365) 2 Georgia Data3 0.1236 (2. 0.0141.5973.4643 (0.0343) 0.9648) 0.174.34125) 0. 122) 2 Coefficient estimates of the variables were reproduced without PKLEFT2 and PKTRUCK using the original data 3 PKLEFT2 and PKTRUCK were not included in the model 4 K: Overdispersion value 5 N/A: not available 88 .386. p-value) -3.315 (0.146. 0.097 (0.763 1 Vogt.405.e.0507.1978) 0. p-value) -3.217 0. p-value) -6.0300 (0. 0.891 (2.6110 (0. (p. 0.1008) N/A5 0. 0.0017) -0.008 (0.1243 (0.0917) N/A5 0.684 (0. 0. 1999. 0. 0.0331) 0.313.212.1507.179.0256) 0. 0.0134 (0..669.

76 -1.18 N/A3 123.26 N/A3 126.16 73.67 2. Table 87.80 31.79 N/A3 Georgia data2 0. indicating a general lack-of-fit.19 of the correlation coefficient indicates that the accident predictions by the original model and the Georgia data are marginally correlated at best.12 6.73 -2.73 -0.19 -3. 89 .19 -2. The MPBs and MAD per year was almost twice as large as those for the original model. Validation Statistics for TOTACC Type V Model Using Georgia Data: Variant 1 Revised Variant 12 1993-1995 49 0.73 MSPE/yr The Variant 1 in the report. The MPB per year was larger than that for the original model.81 -0.12 9.36 4.05 Mile 1997-1997 51 0. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.91 Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 2 1 2 Variant 11 1993-1995 49 0.18 0. while the MADs and MSEs per year were similar.94 6.05-mile data.38 8.48 2. A value of 0.15 N/A3 30.04 Mile 1996-1997 51 0. It is quite evident that the original model generally does not fit the Georgia data well.31 8.The validation statistics are shown in table 87. This model includes PKLEFT2 and PKTRUCK 2 Used the same coefficients as Variant 1. For the revised original model the Pearson correlation coefficient was the same as that for the reported model.37 -0.51 4. The MSPEs per year squared were also significantly higher than the MSEs per year squared. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 3 N/A: not available Figure 14 depicts the prediction performance of the original model for individual sites in the Georgia 0.54 8.22 88.08 -1.

Observed versus Predicted Accident Frequency: TOTACC Variant 1 Intersection Related Total Accident Model (TOTACCI) As before.e. the estimation of GOF measures. Main Model The parameter estimates.. of Accidents 30 Predicted Accidents Observed Accidents 20 10 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 14. and p-values are given in table 88.60 50 40 No. their average values. In addition. K. used a revised original in which these variables were removed from the original published model by dividing by the exponential value of their coefficients times their average effects. In the revised original model all of the variables were estimated with the same direction of effect as for the original model. models were re-estimated without PKLEFT1 and PKTRUCK for the revised original model. their standard errors. was somewhat higher than that for the original model. i. The estimates for all of the variables and the constant term were statistically insignificant. 90 . since the variables PKLEFT1 and PKTRUCK were not present in the Georgia data. The overdispersion parameter. but there were sizeable differences in magnitude and significance. The validation addresses the main model and one variant.

The MPB per year was somewhat larger.2847. and VEICOM were estimated with the same sign as the reported model.154.158.0151) 0. but there were differences in the magnitude and significance.0131.663.9078) N/A5 K 0.1490) N/A5 0. 0. 0. 0.637 (0. 0.610 (0.0485) 0.1445) N/A5 0. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.4708 (0.021 (0.0186) -0.0487) 0.245 (0.1236) 0. 0. 0.. PROT_LT was estimated with an opposite sign.2505) N/A5 0.1813) 0.206 (0. p-value) -6.418. are significantly higher than for the original models.0036) 0.730 0. 0.071.e. AADT2.337 (0. 123) 2 Coefficient estimates of the variables were reproduced without PKLEFT2 and PKTRUCK using the original data 3 PKLEFT2 and PKTRUCK were not included in the model 4 K: Overdispersion value 5 N/A: not available For the Georgia data.3037) N/A5 2 Georgia Data3 0.644 (0.5281) 0. 0. The MPB and MAD per year were larger than those for the original models.0165 (0. 0.073 (0.9050) N/A5 Georgia Data3 0.1742) -0.441. The MSPEs per year squared were significantly higher than the MSEs per year squared.23 of the Pearson correlation coefficient indicates that the accident predictions by the original model are marginally linearly correlated with observed number of accidents in the 1996 to 1997 period.04 Mile (s.3865.315.1126 (0. AADT2 and VEICOM also became insignificant.e.. 0. 0.173.694 (0.222. 0.1313 0.0020) 0. p-value) -6. the constant term.551 (2. The Pearson product-moment correlation coefficient of the revised original model was estimated to be the same as that for the original model.0057. 0. 0.0366) 0..486. but the difference was not great.4373) 0. AADT1. 91 .179.0841 (3. although it was statistically insignificant. p-value) -6.708 Vogt.05-mile data.968.286.0365. 0.327.e.1981) 0. K.e. indicating a general lack-of-fit. 0.248. 0. A value of 0.231 0.2000.5951 (0. while the MAD per year was similar to that for the original model. p-value) -3.021 (0.0724) 0. The overdispersion parameters. 1999.Table 88. 0.2935 (0. Parameter Estimates for TOTACCI Type V Model Using Georgia Data: Main Model Original Estimate (s. 0. 0.0289 (0.410 (3. It is quite evident that the original model generally does not fit the Georgia data well. 0.1366) -0. 0. The validation statistics are shown in table 89. (p.05 Mile (s.0276) 1 Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK 4 1 Revised Estimates (s.061 (3.256 (0.1972. The MSE per was higher than that for the original model.204 (0. Figure 15 depicts the prediction performance of the original model for individual sites in the Georgia 0.

63 1.28 -0.23 -3.04 -1.70 7.53 3.53 N/A3 Georgia data2 0.23 -3.58 7. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 3 N/A: not available 60 50 40 No.33 2.77 N/A3 98.39 -1. of Accidents 30 Predicted Accidents Observed Accidents 20 10 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 15.16 -1.18 The original main model in the report. Validation Statistics for TOTACCI Type V Model Using Georgia Data: Main Model Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 1 2 Original 1993-95 Model1 1993-95 49 0.05 Mile 1996-97 51 0.Table 89.35 6.77 N/A3 100.11 85.09 5.04 Mile 1996-97 51 0.53 3.88 58.71 MSPE/yr2 24.36 0. This model includes PKLEFT2 and PKTRUCK 2 Used the same coefficients as the original model.24 6.59 25. Observed versus Predicted Accident Frequency: TOTACCI Main Model 92 .61 -4.81 9.62 -0.47 N/A3 Revised 1993-95 Model2 1993-95 49 0.

400.6011) Georgia Data3 0. 0.298.219 (0. 0.021 (0.2397) 0. The overdispersion parameter.255. p-value) -2. but there were slight differences in magnitude.0407 (0..05 Mile (s. VEICOM was also insignificant.725 0.175. but these were insignificant.1937.0177 (0.472.e. 93 . Parameter Estimates for TOTACC Type V Model Using Georgia Data: Variant 3 Original Estimate (s.9023) N/A5 -0.704 Vogt.1883.0319) 0. 0.006 (3.8998) N/A5 -0.5995 (0. PROT_LT and DRWY1 were estimated with an opposite sign to that in the original model. indicating lack-offit to the Georgia data. their standard errors.663 (0. 0. AADT1.0137) 0.591 (0.0256 (0.e. For the Georgia data. The overdispersion parameter K was significantly higher than that for the original model. and p-values are provided in table 90.208 0. p-value) -5.0287) 0.070 (0. 0. 0.0405) 0.1437) N/A5 0. In the revised original model all of the variables were estimated as insignificant. 0. 0.2291) 0.0005) 0. 0.367.199.047 (0.872. while only AADT2 was insignificant in the original model. 0.3049) N/A5 0.Variant 3 The parameter estimates.187 (0. 0.159. K.156. AADT2.188 (0.021 (0.4041 (0. 1999.1145 0.422.265 (0.617 (0.0319) -0. 0.062. p-value) -6.1917.033 (0. and VEICOM were estimated with the same sign as the reported model.0050.036.289. Table 90.713 (0.475 (2.324. 0. 0.0117. 0.2932) -0. p-value) -6.031 (0.1402) N/A5 0.4581 (3.0983) 1 Variable Constant Log of AADT1 Log of AADT2 PROT_LT PKLEFT2 VEICOM PKTRUCK DRWY1 1 Revised Estimates (s.0326.2015 (0.0246. the constant term.064. 0. 0.222 (0. 0. The Pearson product-moment correlation coefficient of the revised original model was estimated to be the same as the original model. 0. was almost twice as high as that for the original model.6277) 0.1146) 0.1114 (0.068. 0.0874) 0. 0.170.6153) K4 0.. 0.2795.783 (3. 0.04 Mile (s. 0.2666) N/A5 0.3732) 0. 0.0441) 0. There were also differences in the magnitude of the parameters.1874) 2 Georgia Data3 0.e.0006) 0.3911) -0.e. 123) 2 Coefficient estimates of the variables were reproduced without PKLEFT2 and PKTRUCK using the original data 3 PKLEFT2 and PKTRUCK were not included in the model 4 K: Overdispersion value 5 N/A: not available Table 91 shows the validation statistics. (p. The MSE per was also higher than that for the original model.. The MPB and MAD per year were somewhat larger than for the original model. 0.

67 2.20 10. It is quite evident that the original model generally does not fit the Georgia data well.25 -2.91 Georgia2 0.41 -1. a finding that would have been expected on the basis of the low Pearson product-moment coefficients. this model includes PKLEFT2 and PKTRUCK 2 Used the same coefficients as Variant 3.34 N/A3 0.65 29.Table 91.22 98.04 Mile 1996-97 51 0. 94 .05-mile data.85 N/A3 N/A3 29.31 -0. Validation Statistics for TOTACCI Type V Model Using Georgia Data: Variant 3 Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr MSPE 2 Variant 31 1993-95 49 0. The MSPEs were also significantly higher than the MSEs. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 3 N/A: not available A value of 0. which suggests lack-of-fit to the Georgia data.34 1.80 6.60 119.10 5.22 of the Pearson correlation coefficient indicates that the accident predictions by the original model are marginally linearly correlated with observed number of accidents in the 1996 to 1997 period. Figure 16 depicts the prediction performance of the original model for individual sites in the Georgia 0.63 8.57 5.34 N/A3 118.22 -5.68 4.96 MSPE/yr2 1 Variant 3 in the report.22 -5.05 Mile 1996-97 51 0.67 -0.49 -2. The MPBs and MADs per year for the Georgia data were larger than those for the original models.68 4.78 51.67 -5.74 8.73 Revised Variant 32 1993-95 49 0.

Table 93 shows the GOF measures for the original injury accident model in the Vogt report applied to the Georgia data. PROT_LT and VEICOM were estimated with an opposite sign to those in the original model. and MSPE per year squared were smaller. The models estimated with the Georgia data generally showed differences in sign. and significance of the parameter estimates.(2) The Pearson product-moment correlation coefficient was similar to that for the TOTACC model. the MPB. However. their standard errors. Observed versus Predicted Accident Frequency: TOTACCI Variant 3 Injury Accident Model (INJACC) The parameter estimates. MAD. The constant term and AADT1*AADT2 were estimated with the same direction of effect and in general a similar degree of magnitude and significance to the original model. 95 . and p-values are provided in table 92. magnitude.60 50 40 No. The overdispersion parameter K was significantly higher than that for the original model. of Accidents 30 Predicted Accidents Observed Accidents 20 10 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 16. although they were insignificant.

150. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 2 K: Overdispersion value Figure 17 depicts the prediction performance of the original model for individual sites in the Georgia 0.9641) N/A4 0.82 1. 124) 2 PKLEFT2 and PKTRUCK were not included in the model 3 K: Overdispersion value 4 N/A: not available Table 93.1358) 0.04 Mile2 (s. 0.177.9683) N/A4 0.9932. 0.2358 (0. 0.89 1. p-value) -3. 0..1845) Georgia Data 0.1722.04 Mile 1996-1997 51 0.99 -1. Validation Statistics for INJACC Type V Model Using Georgia Data Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE 1 Georgia data1 0.1144) -0.3640) N/A4 0.0323 (0.1100) 0. 96 .234 (0.05-mile data.e.45 11.41 11. 0.0113 (0. It is quite evident that the original model generally does not fit the Georgia data well.0822 (0.05 Mile 1996-1997 51 0. 0.952 (2.2562 (2.398.0062.437..647 PKTRUCK K3 1 Vogt.455.1630 Georgia Data 0.24 0.e. 0.0551.05 Mile2 (s. 1999.0267) 0.Table 92.439 (0.179.361 (0.2700) N/A4 -0.007 (0.1864. 0.15 -1.00 MSPE/yr2 2. 0. (p.00 2.1707) -0.149.2943 (0. p-value) -3.0678) 0.e.75 Used the same coefficients as the original model. 0.95 2. 0.239 (0. 0.397. 0.0146.815 (2. a finding that would have been expected on the basis of the low Pearson product-moment coefficients. p-value) -3.2002) 0.1153) 0.662 0.89 -0.81 2..2767) 0. Parameter Estimates for INJACC Type V Model Using Georgia Data Variable Constant Log of AADT1*AADT2 PROT_LT PKLEFT2 VEICOM Original Estimate1 (s.15 -1. 0.008 (0.

Observed versus Predicted Accident Frequency: INJACC Intersection Related Total Injury Accident Model (INJACCI) The parameter estimates. The constant term and AADT1*AADT2 were estimated as significant.(2) The Pearson product-moment correlation coefficient was slightly higher than for the TOTACCI model. and p-values are provided in table 94. and MSPE per year squared were smaller.14 12 10 No. the constant term. were significantly higher than that for the original model. of Accidents 8 Predicted Injury Observed Injury 6 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 17. 97 . their standard errors. MAD. Table 95 shows the GOF measures for the original intersection related injury accident model (Variant 1) in the Vogt report applied to the Georgia data. while they were insignificant in the original model. The overdispersion parameters. For the Georgia data. and VEICOM were estimated with the same sign but with differences in magnitude and significance. AADT1*AADT2. and the MPB. K.

59 12.7731) N/A4 0. 0.904..25 1. 0. 0. 98 . Parameter Estimates for INJACCI Type V Model Using Georgia Data Original Estimate1 (s.6095) 0.127.0250) 0. 0.56 1996-1997 51 0. 0.1290 (0.60 -1.e.05 Mile 1996-1997 51 0. 1999.1757.302 (0.0628) 0. p-value) -1.127. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 2 K: Overdispersion value Figure 18 depicts the prediction performance of the original model for individual sites in the Georgia 0.05 Mile2 (s.054 (0.26 3.0066.188. 124) 2 PKLEFT2 and PKTRUCK were not included in the model 3 K: Overdispersion value 4 N/A: not available Table 95. 0. (p.4627) -0. 0.84 MSPE/yr 3.1858) 0.754 Variable Constant Log of AADT1*AADT2 PKLEFT2 VEICOM PKTRUCK K3 1 Vogt. Validation Statistics for INJACC Type V Model Using Georgia Data Georgia1 Measure 0.0149 (0.04 Mile2 (s.777 (2.52 -1.21 3.1433 Georgia Data 0.752 Georgia Data 0. 0.5475 (3. 0.e.27 -2. p-value) -5.18 1.04 Mile Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSPE 2 1 0. p-value) -4..14 Used the same coefficients as the original model.0237) N/A4 0. 0.0282 (0.62 12.029 (2.7420) N/A4 0.e.0518) 0.0686 (0..0384) 0.30 3.187.Table 94.05-mile data.288 (0.062 (0.100.0176) N/A4 0.27 -2.0152. a finding that would have been expected on the basis of the low Pearson product-moment coefficients.0692. It is quite evident that the original model generally does not fit the Georgia data well. 0.0298.

Although the GOF measures should improve. Validate the accident prediction algorithm against all Georgia sites and California and Michigan datasets for Model V. 2. The various measures of GOF statistics would be also improved if the higher numbers were used.14 12 10 No. 99 . Observed versus Predicted Accident Frequency: INJACCI 2. was not known at these sites. of Accidents 8 Predicted INJACCI Observed INJACCI 6 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 Sites Figure 18.6 VALIDATION ACTIVITY 3: VALIDATION OF THE ACCIDENT PREDICTION ALGORITHM Two tasks were undertaken to validate the accident prediction algorithm in Harwood et al. The Minnesota and Washington data could not be used for validating the algorithms for models I and II because sight distance. Data Limitations in the Michigan Data As previously discussed for validation activity 2. it was discovered subsequent to the analysis and draft report that the later year crossroad accident numbers at the Michigan Type V sites should be systematically higher than the values used. one of the variables for which an AMF is applied. Validate the recommended base models for sites meeting “base” conditions. the conclusions drawn from these data and the Georgia data would not change since Model V is in need of recalibration in any case.:(3) 1.

2.6.1 Validation of the Base Models Model I The recommended base model for type I intersections is:

Nbi = exp(-10.9 + 0.79lnAADT1 + 0.49lnAADT2)
where Nbi = the expected number of annual intersection-related collisions; AADT1 = average daily traffic volume on the major road; and AADT2 = average daily traffic volume on the minor road. The following are the base conditions to which the model applies: • • • • • Roadside hazard rating = 2. Presence of right-turn lane on major road = none present. Presence of left-turn lane on major road = none present. Skew = none. Sight Distance = no restrictions.

(8)

The roadside hazard rating is not a restriction for a “base” intersection because no AMF is provided for roadside hazard rating. Only 11 sites in the Georgia data met the base conditions. Summary accident statistics for these sites are shown in table 96.

Table 96. Summary Accident Statistics for Sites Meeting Base Conditions
No. Sites = 11 No. Years = 2 Georgia Data 0.04 Mile Georgia Data 0.05 Mile Mean 1.00 (0.50/year) 1.00 (0.50/year) Median 1.00 1.00 Std. Deviation 0.77 0.77 Minimum 0 0 Maximum 2 2

Table 97 shows the validation statistics for the 11 Type I base model intersections.

100

Table 97. Validation Statistics for Type I Base Model Intersections
Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 Georgia Data 0.04 Mile 1996-1997 11 -0.26 0.36 0.18 0.94 0.47 N/A1 Georgia Data 0.05 Mile 1996-1997 11 -0.26 0.36 0.18 0.94 0.47

MSPE 1.29 1.29 MSPE/yr2 0.32 0.32 1 MSE is unknown since these statistics were not given in the report (Harwood et al., 2000) 2 K: Overdispersion value

The Pearson correlation coefficients indicate a negative correlation between the base model predictions and the observed number of accidents. The other statistics also indicate a relatively poor fit, although it should be considered that only eleven sites were available for this validation activity. Model II The recommended base model for type II intersections is:

Nbi = exp(-9.34 + 0.60lnAADT1 + 0.61lnAADT2)
where Nbi = the expected number of annual intersection-related collisions; AADT1 = average daily traffic volume on the major road; and AADT2 = average daily traffic volume on the minor road. The following are the base conditions to which the model applies: • • • • • No driveways within 76.25 m (250 ft) of the intersection on the major road. Presence of right-turn lane on major road = none present. Presence of left-turn lane on major road = none present. Skew = none. Sight Distance = no restrictions.

(9)

The number of driveways is not a restriction for a ‘base’ intersection because no AMF is provided for driveway density. Only nine sites in the Georgia data met the base conditions. Summary accident statistics for these sites are shown in table 98. 101

Table 98. Summary Accident Statistics for Sites Meeting Base Conditions
No. Sites = 9 No. Years = 2 Georgia Data 0.04 Mile Georgia Data 0.05 Mile Mean 3.00 (1.50/yr) 3.00 (1.50/yr) Median 1.00 1.00 Std. Deviation 4.09 4.09 Minimum 0 0 Maximum 12 12

Table 99 shows the validation statistics for the nine Type II base model intersections. Table 99. Validation Statistics for Type II Base Model Intersections
Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr Georgia Data 0.04 Mile 1996-1997 9 0.81 1.78 0.89 Georgia Data 0.05 Mile 1996-1997 9 0.81 1.78 0.89 2.15 1.08

MAD
MAD/yr MSE

2.15 1.08

N/A1 MSE/yr2 MSPE 12.90 12.90 MSPE/yr2 3.23 3.23 1 MSE is unknown since these statistics were not given in the report (Harwood et al., 2000) 2 K: Overdispersion value

The Pearson correlation coefficients indicate a high correlation between the base model predictions and observed number of accidents. The other statistics, however, indicate a poor fit, although it should be considered that only nine sites were available for validation. Model V The base model in the accident prediction algorithm is identical to the Variant 3 of the Vogt model for TOTACCI.
Nbi = exp(-5.46 + 0.60 lnAADT1 + 0.20 lnAADT2–0.40 PROT_LT – 0.018PKLEFT2 + 0.11 VEICOM + 0.026 PKTRUCK + 0.041DRWY1)

(10)

102

where Nbi = predicted number of total intersection-related accidents per year for nominal or base conditions; PROT_LT = presence of protected left-turn signal phase on one or more major-road approaches; = 1 if present; = 0 if not present; PKLEFT2 = percentage of minor-road traffic that turns left at the signal during the morning and evening hours combined; VEICOM = grade rate for all vertical curves (crests and sags) within 76.25 m (250 ft)) of the intersection along the major and minor roads; PKTRUCK = percentage of trucks (vehicles with more than four wheels) entering the intersection for the morning and evening hours combined; and DRWY1 = number of driveways within 76 m (250 ft) of the intersection on the major road. With the nominal conditions of PRTO_LT, PKLEFT2, VEICOM, and PKTRUCK, the base model reduces to:

Nbi = exp(-5.73 + 0.60 lnAADT1 + 0.20 lnAADT2)
The following are the base conditions to which the model applies: • • • • • No PROT_LT. 28.4 percent of PKLEFT2. No VEICOM. 9.0 percent of PKTRUCK. No DRWY1.

(11)

Because none of the Georgia sites met the base conditions for the base model in equation (11), and the independent variables in equation (10) are available in the Georgia data, the base model in equation (10) was used for validation. Table 113 shows the validation statistics for this base model. Because PKLEFT2 and PKTRUCK were not included in the Georgia data, they were removed from the models by dividing both sites of models by the exponential values of the coefficients of the variables times their average effects (the average effects of PKLEFT2 and PKTRUCK are the average values of PKLEFT2 and PKTRUCK). The validation statistics are shown in table 100. Lower Pearson product-moment correlation coefficients (0.36 and 0.22 in table 100) for the additional years of accidents and the Georgia data indicate that the accident predictions by the base model are not strongly correlated with additional years of accidents and are marginally correlated with the Georgia data, at best. The MPBs and MAD per year was larger than those for the original years. The MSPEs per year squared were higher than the MSEs per year squared. In particular, the MSPEs per year squared with the Georgia data were more than twice as high as the MSEs per year squared, which indicates that the model is performing poorly against the Georgia data.

103

Table 100. Validation Statistics for Type V Base Model Intersections
California and Michigan1 Measure 93-95 Year 1993-1995 49 96-97 Year 1996-1997 49 Georgia1 0.04 Mile 1996-1997 51 0.05 Mile 1996-1997 51

Years used for validation Number of sites Pearson product-moment correlation coefficients 0.68 0.36 0.22 0.22 MPB -5.22 -6.37 -5.85 -5.18 MPB/yr -1.74 -3.18 -2.93 -2.59 MAD 6.58 7.39 9.06 8.63 MAD/yr 2.19 3.70 4.53 4.31 MSE 85.24 N/A2 N/A2 N/A2 MSE/yr2 10.58 MSPE 86.93 129.00 118.80 N/A2 21.73 32.25 29.70 MSPE/yr2 1 Used the same coefficients in the Variant 3, but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects. 2 MSE is unknown since these statistics were not given in the report (Harwood et al., 2000)

2.6.2 Validation of the Accident Prediction Algorithm The accident prediction algorithm was validated against the data collected in Georgia using the recommended base models and AMFs provided in Harwood et al. in 2000.(3) These AMFs are shown in table 101. The AMFs provided in a 2002 report by Harwood et al. also were validated.(5) These AMFs are shown in table 101 in brackets.

104

Table 101. The AMFs for Type I, II, and V Intersections
AMF Variable Intersection Skew Left-Turn Lane on Major Type I Intersections exp(0.004*SKEW) 1.00 if none exist 0.78 if at least one exists (0.56 if at least one exists) Type II Intersections exp(0.0054*SKEW) 1.00 if none exist 0.76 on one approach (0.72 on one approach) 0.58 on both approaches (0.52 on both approaches) Type V Intersections 1.00 1.00 if none exist 0.82 on one approach 0.67 on both approaches

Right-Turn Lane on Major

Sight Distance

1.00 if none exist 0.95 on one approach (0.86 on one approach) 0.90 on both approaches (0.74 on both approaches) 1.00 if adequate in all quadrants 1.05 if limited in one quadrant 1.10 if limited in two quadrants 1.15 if limited in three quadrants 1.20 if limited in four quadrants

1.00 if none exist 0.975 on one approach (0.96 on one approach) 0.95 on both approaches (0.92 on both approaches) 1.00

Intersection skew is defined as the deviation from an intersection angle of 90 degrees and carries a positive or negative sign that indicates whether the minor road intersects the major road at an acute or obtuse angle. Left-Turn Lane on Major takes a value of one if no left-turn lane exists on the major road and the values in table 101 where one or more exists. Right-Turn Lane on Major takes a value of one if no right-turn lane exists on the major road and the values in table 101 where one or more exists. Sight Distance in a quadrant is considered limited if the available sight distance is less than that specified by the American Association of State Highway and Transportation Officials (AASHTO) policy for left and right turns from the minor road for a design speed of 20 km/hour (km/hr) (12.5 miles/hr) less than the major road design speed.(7)
In the Georgia data, design speed of the major road was not known. The posted speed limit of the major road was used in lieu of subtracting 20 km/hr (12.5 miles/hr) from the design speed when using the AASHTO warrant for sight distance. In applying the algorithm, a calibration factor is applied to the model, calculated as the ratio of the observed number of accidents to the predicted number of accidents prior to the calibration factor being applied. Harwood et al. recommend that the sample for estimating this calibration factor be such that the distribution of traffic volumes is similar to that in the data used for the original calibration.(3) This was not possible for the Georgia data due to the small sample size. It was felt that this should not create a deterrent, because the calibration factor, at least in the procedure as proposed, is 105

as shown in table 103.66) (2.72) MSPE/yr2 2.57 2.57 (1.(3..00 0. For comparison. (2002).0 indicate whether intersections will experience more or less accidents than the intersections used in the development of the base models for the accident prediction algorithm.59 0.46) 0.59 (1.5) Low Pearson product-moment correlation coefficients with the Georgia data indicate that the accident predictions by the algorithm are not correlated strongly with the observed number of accidents in the Georgia data. 2002 report are in brackets.13 0.00 0. The AMFs from Harwood et al. 2000) 2 K: Overdispersion value To compare further the two sets of AMFs for turning lanes.65 0.76) 1 MSE is unknown since these statistics were not given in the report (Harwood et al.89) MSPE (0. There is little difference between the two sets of AMFs for turning lanes. Model I Table 102 shows the validation statistics for Type I intersections. the results for the AMFs provided in the Harwood et al.58) N/A1 Georgia Data 0.70 (2.independent of traffic volume and is applied to all intersections regardless of the distribution of traffic volumes in the jurisdiction.60) 2. Table 102. In the tables of results.15) (0. Other validation statistics also suggest a lack-offit to the Georgia data. the calibration factors greater or less than 1. although the differences in validation statistics and sample sizes are small.00 1.04 Mile 1996-1997 121 0.00 1.47 (0.73 Calibration Factor (2. and the results for AMFs provided in Harwood et al.18 0. Validation Statistics for the Accident Prediction Algorithm: Type I Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 Georgia Data 0.44) 0.(3. Therefore. it is of interest to examine if the procedure would work for the likely situation where the distribution of traffic volumes in the new jurisdiction is different from the distribution in the calibration data.67) (0.60) (2.82 2.45 (0.5) 106 .20) (0. (2000) provided better validation statistics than those in Harwood et al. the validation statistics were calculated for only those sites with a major road turning lane. 2000 report are not bracketed.05 Mile 1996-1997 121 0.

052) (0.021) (0.04 Mile 1996-1997 7 3 with left-turn lane 4 with right-turn lane 0.024) 2.00 0..00 0.5) Low Pearson productmoment correlation coefficients with the Georgia data indicate that the accident predictions by the algorithm are not correlated strongly with the observed number of accidents in the Georgia data.05 Mile 1996-1997 7 3 with left-turn lane 4 with right-turn lane 0. Validation Statistics for the Accident Prediction Algorithm: Type I Sites with a Major Road Turning Lane Measure Years used for validation Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 Georgia Data 0.52) 0. 2000) 2 K: Overdispersion value Model II Table 104 shows the validation statistics for Type II intersections.030) 0.66 (0. Other validation statistics also suggest lack-of-fit to the Georgia data. (2000) are not bracketed.78) 0.059) (0.026) N/A1 Georgia Data 0.60) (2.011 0.022 (0.(3. There is little difference between the two sets of AMFs for turning lanes.084) (0.Table 103.00 0.018 (0.059 0.73 (2.015 (0.043 0. the results for the AMFs provided in Harwood et al.57 2.036 0.00 0.76) Calibration factor 1 MSE is unknown since these statistics were not given in the report (Harwood et al. (2002) are in brackets. For comparison.85 (0. 107 .045 0. and the results for AMFs provided in Harwood et al.097) MSPE MSPE/yr2 (0.

53 (0.85 (1. the results for the AMFs provided in Harwood et al.60) (1. (2000) provided better validation statistics than those in Harwood et al. (2002) are in brackets.72) (0.04 Mile 2 114 0. although the differences in validation statistics and sample sizes are small..85 (1.36 (5.5) The validation statistics suggest a lack-of-fit to the Georgia data.51) 0.86) N/A1 Georgia Data 0. (2000) are not bracketed.18) (2.69 0.45 5.00 0.51) 0.00 0. the base model in the equation (10) was used for the accident prediction algorithm.20) 1 MSE is unknown because these statistics were not given in the report (Harwood et al.15 2.70 0. 2000) 2 K: Overdispersion value To compare further the two sets of AMFs for turning lanes. and the results for AMFs provided in Harwood et al. Validation Statistics for the Accident Prediction Algorithm: Type II Measure Number of years Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 Georgia Data 0. 108 . For comparison.46 1.36 1.59) (5.(3.17 Calibration Factor (2. (2002).00 1. because none of the sites met the nominal base conditions.53 (0. PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential values of the coefficients of these variables times their average effects.40) 2.72) (0.5) Model V Again.Table 104.05 Mile 2 114 0. the AMFs from Harwood et al.(3.00 1. Table 106 shows the validation statistics for Type V intersections.86) 5. the validation statistics were calculated for only those sites with a major road turning lane. As the results in table 105 indicate.40) (1.

18 (0.00 0.00 0.73 0.00 0.158 0.19 0.47) 0.18) (2.158 0.04 Mile 2 7 6 with left-turn lane on both approaches 3 with right-turn lane 0.079 (0.89) (0. Validation Statistics for the Accident Prediction Algorithm: Type II Sites with a Major Road Turning Lane Measure Number of years Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 2 7 6 with left-turn lane on both approaches 3 with right-turn lane 0.00 0.05 Mile 0.175) (0. 2000) 2 K: Overdispersion value 109 .88) (0.47) 0.22) (0.15 2..52 (0.20) 1 MSE is unknown since these statistics were not given in the report (Harwood et al.52 (0.088) N/A1 Georgia Data 0.174) (0.079 (0.087) Georgia Data 0.22) 2.17 Calibration Factor (2.Table 105.74 0.

83) (0.05) 0.31 23.Table 106. not surprising given the closeness of their estimates.05 (0. There is little difference between the two AMFs.18) (22.89 22.57 (7.05) 0. 110 .00 0.94) (3. These results are shown in table 107.09) (3.04 Mile 2 51 0. Validation Statistics for the Accident Prediction Algorithm: Type V Measure Number of years Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 Georgia Data 0.00 0.85) 1 Used the same coefficients as Variant 3. but PKLEFT2 and PKTRUCK were removed from the model by dividing by the exponential value of the coefficient of these variables times their average effects 2 MSE is unknown since these statistics were not given in the report (Harwood et al.23 93.81 0.05 Mile 2 51 0.00 6.55) 0.49 (6..56) (94. 2000) To compare further the two AMFs for right turn on major road.47 (89.14 3.00 7.99 3.04 (0.39) (23.83 Calibration Factor (0.55) 89. the validation statistics were calculated for only those sites with a major road right-turn lane.47) N/A1 Georgia Data 0.

and translating these results into meaningful observations and conclusions. The intent of the discussion is to provide insight and lay groundwork for the recalibration of the models to follow.Table 107. The validation exercises in the body of this report focused primarily on external validation—validation concerned with assessing performance of the models compared to external data.10 is used. The support for this level of alpha is as follows.34) 1. It should be noted throughout the discussion that the subjective criteria of alpha equal to 0.43 Calibration Factor (1. In statistical models of 111 .82 12. should refer to the body of the report.04 Mile 2 21 (13 with right-turn lane on one approach. Validation Statistics for the Accident Prediction Algorithm: Type V Sites with a Major Road Right-Turn Lane Measure Number of years Georgia Data 0. This discussion is focused on internal validation concerns—the internal coherence.07) (53. structure. such as sources of original results and comparison tables.7 DISCUSSION OF RESULTS Although the body of this report presented validation results in order of validation exercises.37) (12.11) 0.00 0. The discussion provided here focuses primarily on summarizing the results that are detailed and discussed in the body of this report. The reader interested in additional details. 8 with right-turn lane on both approaches) 0.12) 0. More focus is given to internal validation in the recalibration research task scheduled to follow this validation.05 Mile 2 21 (13 with right-turn lane on one approach. theoretical soundness.21 (50.41 4.44) (4. 8 with right-turn lane on both approaches) 0.22) Number of sites Pearson product-moment correlation coefficients MPB MPB/yr MAD MAD/yr MSE MSE/yr2 MSPE MSPE/yr2 49.12) N/A1 Georgia Data 0.00 0. the results here are presented by model. 2000) 2.24) (4.52) (13.00 8. beginning with Model I results and ending with Model V results. and plausibility of the models proposed.43) (1.12 (0. Descriptions of all variable abbreviations and definitions used in this report are found at the beginning of this document.40 1.13 (0.46) 1 MSE is unknown because these statistics were not given in the report (Harwood et al.11 (8.11 (8.21 4.39 13.55 52.00 8..

and these were considered to be marginal. Marginal differences in GOF are thought to be negligible and are potentially explained by random fluctuations in the observed data. With a Type I error. 2. Before discussing the results. however. making a Type II error results in concluding that the presence of a left-turn lane does not reduce crashes when in fact it does. In addition. but improvements might be difficult to obtain.7.crash occurrence. Thus. are generally subjective. To summarize.10 has been used suggests that a smaller level of beta has also been selected. In the following analyses the terms “serious. However. the vertical-alignment related variables VEI1. all else being equal. the relatively small sample sizes and preponderance of engineering theoretical support for variable selection decisions also supports a relatively larger alpha. although they may contribute to an explanation for some individual results. it should be noted that an attempt to reproduce variables in data acquired from the original research revealed some definitional problems. committing a Type I error results in applying an ineffective countermeasure. Serious differences in GOF are suggestive of noteworthy or significant model deficiencies. Small differences in the variable VCI1 for Models I and II are thought to stem from an inability to reproduce “exceptional” cases of vertical curvatures encountered in the database. Moderate differences in GOF suggest cases where models could be improved. So choosing a larger alpha means that a smaller beta has been chosen. The risk is in failing to install an effective countermeasure. and VEICOM computed in the original models could not be reproduced precisely in data acquired from the original researchers. VEI2. For Model V. For Models III and IV. but the precise and correct conditional probability interpretation is cumbersome and for practical purposes does not lend any additional insights).” and “marginal” are applied to denote a subjective evaluation of GOF comparisons between models. that the presence of a left-turn lane reduces crashes when in reality it does not. This means that the analyst would conclude. for example.1 Model I 112 . while committing a Type II error results in failing to apply an effective countermeasure. A Type II error occurs with beta probability. continuing with the previous example. the variables AADT1 and DEV had small differences in the median values. GOF statistics provide an ability to objectively assess the fit of a model to data. on the whole. a Type II error can be argued to be more serious than a Type I error. the analyst concludes that the null hypothesis is false when in fact it is true with alpha probability (this translation is not precise. These definitional problems resulted in values of some variables that are different than they would have been in the original research.” “moderate. Computing the actual beta in negative binomial models is extremely difficult. These differences. For Model II. As a result of this conclusion one might install left-turn lanes without realizing a reduction in crashes. are unlikely to materially affect the overall conclusions. the larger the alpha. In general. the smaller is the beta. knowing that a fairly liberal alpha equal to 0. all of the variables in data acquired from the original researchers were reproduced within rounding-error precision. Comparisons between models.

indicating a marginal to moderate model deficiency—downgraded from serious due to the small sample size. Despite the small sample size. and RT MAJ had the opposite sign for the Georgia data. the base model predicts as a function of minor and major road AADTs only. while. and SPD1 had the opposite signs than those in the original model (table 55). conversely.66) between observed and predicted crashes is reduced by about 50 percent (0.26 (see table 97). with the former having the opposite sign. the correlation coefficient between predicted and observed crashes for the base condition model was -0.89. The variable SPD1 became a highly significant variable. all model variables. Table 4 shows a comparison of the parameter estimates for the original published total accidents model and one based on the additional data. The evidence in the table suggests that the original model is predicting future crashes as well as for the calibration data. The validation statistics for the TOTAL crashes model (see table 54) show that the correlation coefficient (0. new data were collected from later time periods to validate the crash models (see table 3). For INJURY crashes. Comparison of the recalibrated and original models for TOTAL crashes (table 53) shows that HI1. variables that were not statistically significant in the original model were not statistically significant in the model based on additional years of data. 113 . As might be expected. with the exception of the main road AADT. Predictive Ability Across Space Georgia data were used to investigate the model’s performance across jurisdictional boundaries. Crash Prediction Algorithms Eleven sites in Georgia were available for validating the base condition for the crash prediction algorithm (see table 96).19 for the recalibrated model compares to an MSPE per year squared of 0. the RT MAJ and HAZRAT1 became statistically insignificant. compared to marginally significant in the original model. HAU. Data used to estimate Model I were obtained from Minnesota. Other statistics also show considerable lack-of-fit of the Georgia data relative to the Minnesota data.31) on Georgia data—a rather serious drop. Table 5 shows the GOF statistics of Model I (total accidents model) on the additional years of data. Because of the small sample size and the homogenous nature of the intersections (all possess base conditions in common). This difference suggests that the original model is not capturing the variability in crashes in the Georgia data. Note that it was discovered subsequent to the analysis and draft report that some errors exist in the accident data due to changing mileposts although these errors are negligible and would have no effect on any conclusions drawn from the analysis. were insignificant when estimated with Georgia data. RT MAJ. HAU. All of the GOF statistics showed marginal differences between the validation data and the calibration (estimation) data. and HI1. The MSE per year squared of 0. with one exception. with RT MAJ becoming insignificant for Georgia.Predictive Ability Across Time To the extent possible. which represents a serious difference in fit.

If installed at intersections due to mostly capacity problems. This suggests that variables that quantify differences in facility design. Table 14 shows the GOF statistics of the Type II (TOTAL crashes model) on additional years of data. new data were collected from later time periods to validate the crash models. but most had large differences in magnitude. Table 13 shows a comparison of the parameter estimates for the original TOTAL crash model and one based on the additional years of data. Potential missing variables might include number of wet. The constant term and all variables had the same signs. the presence of a right-turn lane might be associated with increased numbers of crashes. and only the additional years of data for Minnesota data could be used for the validation test across time. or foggy days. The major assessments and conclusions derived from the analysis include the following: • The model performed better in the same jurisdiction (Minnesota) in a future time period than across a different jurisdiction (Georgia). The linear correlation coefficient shows consistent performance into the future.Model I Assessment and Conclusions The model for three-legged stop controlled intersections with two lanes both on minor and major roads has fairly mixed validation results. dark (nighttime) versus light (daylight) crashes. traffic. then the presence might indicate an improvement in crashes (due to reduction in conflicts and crashes). Presence of right-turn lanes (RT MAJ) can pose a problem in intersection models. or environmental conditions across States are missing from the model. The model based on additional years gains the HI1 variable as statistically significant. in particular those with low significance in both models. improvements to the models predictive power is desired. • • • 2. and a comparison of MSE per year squared (0. If installed due to high-crash conditions or in conjunction with another intersection improvement prompted by high-crash conditions. It was discovered subsequent to the analysis and draft report that some errors exist in the accident data due to changing mileposts.095) and MSPE per year squared 114 .7. This phenomenon might explain the switching of the sign associated with the RT MAJ variable observed in the validation effort associated with Model I and other models. icy.2 Model II Predictive Ability Across Time To the extent possible. Because the model will be used for predictions in jurisdictions other than Minnesota. although these errors are negligible and would have no effect on any conclusions drawn from the analysis. Note that Minnesota data were used in the original model calibration (estimation).

185) suggests that the variability in future crashes is not being captured as well as for the data on which the original model is based. indicating a strong linear trend between observed and predicted values. Data used to estimate Model II were obtained from Minnesota. are too small to detect the small effects some of the variables have on safety. Predictive Ability Across Space Georgia data were used to investigate the model’s performance across jurisdictional boundaries (see table 57). including increases in MAD (moderate) and a serious increase in MSPE per year squared (1. The validation statistics for TOTAL crashes (see table 59) show that the correlation coefficient between observed and predicted crashes is reduced by about 50 percent on Georgia data—a rather serious drop. the models again shared statistically significant AADT related variables. The major assessments and conclusions derived from the analysis include the following: • The model performed better in the same jurisdiction (Minnesota) in a future time period than across a different jurisdiction (Georgia). The correlation coefficient between predicted and observed crashes for the base condition model was 0. improvements to the models predictive power is desired.73) compared to MSE per year squared (0. Comparison of the original (Minnesota) and Georgia models for TOTAL crashes (table 58) shows that the two AADT variables had similar magnitudes and are the only variables the two models shared as statistically significant. The flip-flopping of variables that are statistically significant in original and validation models suggest that sample sizes. while the only other statistically significant variable was HAZRAT1 in the original (Minnesota) model (see table 60). This suggests that variables that quantify differences in facility design. icy.81 (see table 99). Potential missing variables might include number of wet. or environmental conditions across States are missing from the model. Other statistics show a moderate to serious increase in lack-of-fit. Crash Prediction Algorithms Nine sites in Georgia were available for validating the base condition for the crash prediction algorithm (see table 98). in general. traffic. The variable SPD1 became statistically significant for Georgia. 115 • • • . dark (nighttime) versus light (daylight) crashes.10) in the original model. Model II Assessment and Conclusions The model for four-legged stop controlled intersections with two lanes both on minor and major roads has fairly mixed validation results—although they are more favorable than for Model Type I. whereas the variable DRWY1 was statistically significant in the original (Minnesota) model. For INJURY crashes.(0. Since the model will be used for predictions in jurisdictions other than Minnesota. or foggy days.

new data were collected for additional time periods to validate the crash models (see table 21). MEDWDTH1 and DRWY1. as statistically significant.2. and table 28. Considering variant 2. and again.39) suggests a marginal increase in lack-of-fit. The linear correlation coefficient shows a moderate decline when calculated on the additional years of data.7. log of AADT 2. the coefficient for log of AADT2 is 0. MEDWIDTH1. The table shows that many variables have similar magnitudes across the calibration (California and Michigan) and validation (Georgia) data sets. and the coefficient for log of AADT2 was again twice as large for the additional years model. Table 23 shows the GOF statistics of Model III for TOTAL crashes on additional years of data. the variables HAU. The re-estimated models for INTERSECTION RELATED total and INJURY crashes (variant 1 and variant 2) are shown in table 24. the MAD per year shows a marginal increase in lack-of-fit. Recall that the original model was calibrated on data from Michigan and California from 1993 to 1995. Comparison of the original and Georgia models for TOTAL crashes (table 63) shows that the models are quite different. Predictive Ability Across Space Georgia data were used to investigate the model’s performance across jurisdictional boundaries. For the INTERSECTION RELATED total crash model. and ABSGRD1 became statistically significant for the additional years of data. In addition. Considering variant 1 of the INJURY models. The one notable exception is MEDWIDTH1. and DRWY1. while the original published model had all variables. which is considerably larger in the Georgia data than in the original data. Table 22 shows a comparison of the parameter estimates for the original published TOTAL crash model and one based on additional years of data. A comparison of MSE per year squared (1. that is unlike the case for the original calibrated model.8 percent of the original sites. In fact.26 in the original model and 0. the coefficient for log of AADT2 was twice as large for the additional years model. Similar results are seen for the TOTAL INTERSECTION 116 . Comparison of the original model and the model based on the later data reveals that two variables. More than half of the Georgia sites had medians on the major road compared to only 5. have become statistically insignificant in the additional years model. Table 25. table 26. log of AADT1. for INTERSECTION RELATED crashes. Similarly.52 in the model based on the additional years of data.22) and MSPE per year squared (1. DRWY1. the additional years of data did not produce statistically significant variables for MEDWDTH1 and DRWY1. A comparison of Georgia data for Model III is shown in table 62. shows similar performance assessments as the model for TOTAL crashes. Data from Michigan and California from the years 1996 and 1997 were used for the validation effort. only the variable log of AADT2 is statistically significant in the model calibrated using Georgia data. the variable HAU was not statistically significant for the additional years of data. while the original model did.3 Model III Predictive Ability Across Time To the extent possible.

This suggests that variables thought to be important may in fact not be important. However. The correlation between predicted and observed is seriously different on Georgia data. The plots of observed and predicted crashes for individual intersections in figures 5 and 6 shows that predictions are much less variable than are observations. Model III Assessment and Conclusions The model for three-legged stop controlled intersections with two lanes on minor and four lanes on major roads revealed some serious model concerns. All of the validation statistics (see table 64 and table 66) are consistent with the lack of agreement in model specification between original and Georgia data. and variant 1 (table 67) and variant 2 (table 69) of the INJURY crash models. despite an apparent similarity in the raw data. Model III is in need of improvement.RELATED crash model (table 65). based on the validation findings across time and space. These concerns are summarized below: • The TOTAL crash model performs moderately well across time. Thus. The MPB and MAD statistics show a significant worsening of the fit. or are highly correlated with truly important variables. even at the same sites the model specification changes significantly in a future time period—with the exclusion of two originally significant variables. The models for INTERSECTION RELATED TOTAL and INJURY crashes perform similarly. the MSE per year squared for the calibration data is considerably lower than the MSPE per year squared for Georgia data. It is clear that variables in the California and Michigan model are not specified appropriately for Georgia data. Thus it is likely that some of the explanatory variables are inadequate for predicting Georgia data. but both exhibit moderate to serious model specification difficulties. a serious lack of agreement between the originally calibrated models and the models estimated using Georgia data is observed. Crash Prediction Algorithms There are no AMFs to validate for Model III. The model was seriously deficient in predicting Georgia data. • • 117 . Finally.

2. shown in table 33.32 in the original published model to 0. The coefficient for log of AADT2 becomes insignificant in Georgia. All of the validation statistics (see table 75) are at least consistent with the lack of agreement in model specification between original and Georgia data.39) suggests a moderate improvement in the fit to later data. For the INJURY crash model (see table 38) and the INTERSECTION RELATED INJURY crash model (see table 40). The same minor differences exist for the INTERSECTION RELATED total crash model. meaning that in the calibration data the presence of a left-turn lane is associated with decreased TOTAL crashes. the variable log of AADT1 becomes statistically insignificant on additional years of data. with the exception of LTLNS. which changes from 0. All other coefficients are similar in magnitude and share the same sign. Table 31 shows a comparison of the parameter estimates for the original published TOTAL crash model and one based on the additional years of data. The linear correlation coefficient shows a marginal increase when calculated on the additional years of data. Recall that the original model was calibrated on data from Michigan and California from 1993 to 1995. and perhaps suggest 118 . and the sign of LTLN1S changes from negative to positive. A comparison of MSE per year squared (3. new data were collected from later time periods to validate the crash models (see table 30). Data from Michigan and California from 1996 and 1997 were used for this validation effort. Comparison of the original model and the model based on the later data reveals that the model performs very well on later data—all of the variables in the model are statistically significant in both models. The largest difference is the coefficient of log of AADT2. Table 32 shows the GOF statistics of Model IV for TOTAL crashes on the additional years of data. A comparison of TOTAL crashes compared to Georgia data for Model IV is shown in table 71. Table 34 shows similar performance assessments for the INTERSECTION RELATED total crash model. The MADs per year is similar.50 in the additional years model. which is moderately smaller (70 percent had left-turn lanes) in the calibration data (Michigan and California) than in the validation (Georgia) data. The table shows that many variables are similar across the calibration (California and Michigan) and validation (Georgia) data sets.7. Predictive Ability Across Space Georgia data were used to investigate the model’s performance across jurisdictional boundaries. where 83 percent had left-turn lanes.4 Model IV Predictive Ability Across Time To the extent possible.40) and MSPE per year squared (2. while in Georgia the presence of a left-turn lane is associated with increased TOTAL crashes. Some mathematical “work-arounds” had to be performed to circumvent a missing data problem in the validation data set. but the MPBs per year indicates a serious lack-of-fit to later years of data.

These concerns are summarized below: • • • The model performs well across time.7. The models for INTERSECTION RELATED and the two variants for INJURY crashes are more seriously deficient than the TOTAL crash models. The correlation between predicted and observed is significantly reduced. As a result. the accident data for 1996 and 1997 did not include the crossroad accidents where the crossroad was a State route. 2.05 and 0. Model IV needs to be improved. The MPB and MAD statistics show a moderate increase in lack-of-fit.a more substantial departure. two lanes on minor roads. Crash Prediction Algorithms There are no AMFs to validate for Model IV.5 Model V Predictive Ability Across Time To the extent possible. going from 0. and four lanes on major roads revealed model deficiencies ranging from moderate to serious. table 78. Finally. None of the models had statistically significant variables that corresponded with the original models. Data from Michigan and California from years 1996 and 1997 were used for the validation effort. The observations suggest that the most important differences are jurisdictional in nature—pointing towards the consideration of design. the later year crossroad accident numbers of the State routes should be systematically higher than those supposed to be.08 (Georgia). Table 76. the MSE per year squared (3. For signalized intersections in Michigan. The plot of observed and predicted crashes for individual intersections in figure 9 shows that predictions are much less variable than are observations. and environmental variables that are not included in the model. traffic.62) for the calibration data is moderately lower than the MSPE per year squared (4. Model IV Assessment and Conclusions The model for four-legged stop controlled intersections. All models were seriously deficient in predicting Georgia data. based on the validation findings across time and space. and table 80 show the serious differences between original published models and the models estimated using Georgia data. The model for TOTAL crashes performs better than the models for INTERSECTION RELATED and INJURY crashes. but all models show only marginal to moderate deficiencies.63) for Georgia data. because the crossroad milepost information was not available at the time of analysis. these differences would not have any effect on the conclusions drawn from the analysis. new data were collected from later time periods to validate the crash models (see table 39). Recall that the original model was calibrated on data from Michigan and California from 1993 to 1995. 119 .56 (original) to 0. as evidenced by the lack-of-fit and the inability of the Georgia data to fit the specified models. However. Thus it is likely that some of the explanatory variables are inadequate for predicting Georgia data.

Table 41 shows the GOF statistics of Model V for TOTAL crashes on the later data. The MAD per year shows a moderate increase in lack-of-fit.73 to 0. with a tendency to overpredict crashes. Table 42. going from 0. log of AADT2. PKLEFT2 and VEICOM are not statistically significant in the model based on additional years of data. Predictive Ability Across Space Georgia data (see table 82) were used to investigate the model’s performance across jurisdictional boundaries. 46. The correlation coefficient between 120 . The plot of observed and predicted crashes for individual intersections in figure 13 shows that predictions are much less variable than are observations. 44. the MSE per year squared for the calibration data is significantly lower than the MSPE per year squared for Georgia data. A comparison of TOTAL crashes compared to Georgia data for model V is shown in table 84. Finally. and VEICOM were not statistically significant in the Georgia data. The table shows moderate to serious departures between the original published models and the models based on Georgia data. All of the validation statistics (see table 85) are consistent with the serious lack of agreement in model specification between original and Georgia data. while the MPB per year indicates a serious lack-of-fit.Table 40 shows a comparison of the parameter estimates for the original published TOTAL crash model and one based on the later data. The correlation between predicted and observed is substantially reduced. Specifically. Crash Prediction Algorithms Fifty-one sites in Georgia were available for validating the base condition equation for the crash prediction algorithm (see table 100).18 (Georgia). The predictive ability of the models and their variants for INTERSECTION RELATED Total and INJURY crashes are difficult to assess.73 (original) to 0. A comparison of MSE per year squared (8. However. and also shows some systematic prediction error. since the original models could not be duplicated with similar results.40).56) and MSPE per year squared (21. The linear correlation coefficient between observed and predicted data shows a moderate to serious decrease when calculated on future year data (0. the models are deficient in different ways. at best. many or all of the statistically significant variables in the original published models could not be duplicated in the recalibration-so assessing the predictive ability is difficult. and 50 for INTERSECTION RELATED TOTAL and INJURY crash models and their variants also show serious differences between the original and models based on later data.19) suggests a poor fit to the additional years of data. 48. The MPB per year and MAD per year statistics show a poor fit. with different sets of variables becoming statistically insignificant. This comparison reveals that there are moderate to serious differences between the models. The variables PROT_LT. again suggesting a serious deficiency.

121 . indicating a weak linear trend between observed and predicted values.predicted and observed crashes for the base condition model was 0. however. Model V Assessment and Conclusions The model for signalized intersections of two-lane roads revealed model deficiencies ranging from moderate to serious. and a serious to moderate decline from the same statistics calculated on the original published data (0. part of the lack-of-fit is likely due to the inability to reproduce the original results. however.22. The models did not appear to perform adequately over time.36 for different base years).68 and 0. The models exhibited a serious lack-of-fit to the Georgia data. These concerns are summarized below: • • • The inability to recreate the original data hampered the ability to fully validate these models. Model V is likely in need of improvement. Other statistics show moderate to serious lack-of-fit. a better understanding of the model deficiencies can be gained only by recreating the original data.

.

This includes changes to parameter estimates. and perhaps minor changes to model functional forms. model functional form. It should be acknowledged that several anticipated end-uses of the crash prediction models guided all decisions made throughout this careful evaluation.1 RESEARCH APPROACH This model recalibration effort complemented the comprehensive model validation previously conducted as part of a larger technical evaluation of crash prediction models. there were multiple levels and types of models in the source documents—requiring a prioritization of models to be calibrated. which need to be prioritized. and decisions such as variable selection. with the sole intent to predict future crashes at intersections throughout the United States. Environmental effects on safety. This approach is based on the collective opinion that prior work. Considering the likely end uses of the crash prediction models within the IHSDM. 3. the treatment of explanatory variables is dependent upon the methodological approach taken. such as adverse weather and lighting conditions. and statistical model selection represent state-of-the-art knowledge with respect to intersection crash prediction models. the data and related issues are discussed. corrected for intersection type and State. Second. Sensitivity analysis results for the AMFs derived in this research then are given. The models need to be able to predict the change in safety as a result of changes in traffic and geometric features relative to nominal conditions. will be accounted for in State or regional correction factors. was done carefully by experts in the field of transportation safety. critical evaluation. a discussion and conclusions as a result of model recalibration are provided. First. Third. followed by fully parameterized model estimation results.or regional-specific effects. while important factors. The first section provides a discussion of the recalibration approach. including the estimation of statistical models. Finally. AADT model estimation results are presented. which resulted in some specific overriding considerations while conducting the model recalibration: • • • The most likely end-use of the crash prediction models is embedded code within the IHSDM. 123 . Past documentation. Finally. Before describing the research technical strategy. It was felt that the majority of effort in the recalibration should be devoted to refinements to existing models. considerable time was spent identifying a strategy for recalibrating statistical models. In the second section.3. there are numerous methodological approaches reflected in the source documents. RECALIBRATION This chapter presents recalibration results for the five types of rural intersections that were the subject of the validation exercise undertaken in the first part of the project. A strategy was needed for several reasons. some guiding philosophical principles used to guide the model recalibration effort are presented.

and it is not always a clear winner. AADT Models: One set of models represents intersection crash models that forecast crashes in frequency-per-year based on minor and major road AADT-only. roadside. The intended use of these models is to provide a baseline crash forecast.and discussion with other experts in the field confirm prior beliefs that the existing set of models represents a defensible and sound starting point. availability of explanatory variables. and operational features. It is through the assessment of many measures that a “best” model is chosen. and operational features of intersections that influence on crashes. Unfortunately. These models are meant to provide a fuller understanding of the geometric. There are no other independent variables in these models. and tstatistics and their associated p-values. Numerous measures are relied upon to avoid basing decisions on one single measure. there is no one single criterion that dominates to the point of rendering the remaining measures as invalid or unimportant. These provide a complement to the AADT models. Each of the strategies represents different possible end uses of the models. in a sense. Finally. roadside. It is believed that moderate to serious departures from existing models should be accompanied by detailed and defensible descriptions of the how. simply because of existing data limitations. Full Models: Another set of models represents statistical models with a full set of explanatory variables. When comparing and refining the three types models. Another use might be to develop or infer additional crash modification factors for the various types of intersections examined in this research. collection of explanatory variables. partly compensates for the loss of statistical precision resulting from the omission of variables other than AADT. why. and in what cases departures from previous methods and/or models were thought necessary and useful. capabilities with regard to model recalibration are limited. The technical strategy applied in this research effort is now described. 124 . The intended use of the AMFs is to provide percentage corrections to expected crash frequencies that result from the application of various crash countermeasures. including major and minor road AADT. and are handled rather simply in the IHSDM. When these limitations are thought to be critical they are identified and discussed. roadside. several GOF measures were used in addition to inspection of model coefficients. influenced by the stated guiding philosophical principles. AMFs represent a fairly intuitive approach to evaluating safety countermeasures. The sample available for calibrating these models was much larger than the sample available for calibrating full models that. and intersection representativeness across States. AMFs: A final set of “models” represents estimated effects of various geometric. which can then be modified with AMFs representing the effects of various geometric. and other relevant safety-related factors.

βin. Fully Parameterized Models The following model form and error distribution were assumed to represent the underlying phenomenon: n ⎞ ˆ i = AADT1β1 AADT2β 2 exp⎛ ⎜ α + ∑ β ij X ij ⎟ Y ⎟ ⎜ j= 3 ⎠ ⎝ ^ (13) where Y = the mean number of accidents to be expected at site i in a given time period.. Figure 19 illustrates the CURE plot for the covariate AADT1 for the total accidents for the selected AADT-only model for 125 . Var{m} = E{m} + K * E{m}2 ^ (14) where Var{m} = the estimated variance of the mean accident rate. E{m} = the estimated mean accident rate from the model. and β1 β2.3. estimated coefficients.1. A fifth approach to evaluating the GOF and in particular the suitability of alternate model forms was the Cumulative Residuals (CURE) method.).1 Model Functional Forms The negative binomial model form. Xi1. 3... α = the estimated intercept term.(8. = estimated coefficients.Xin.2) The following model form and error distribution were assumed to represent the underlying phenomenon: AADT Only Models ˆ = exp (α + β AADT + β AADT ) Y i 1 1 2 2 (12) where Y = the mean number of accidents to be expected at site i in a given time period.. in which the cumulative residuals (the difference between the actual and fitted values for each intersection) are plotted in increasing order for each covariate separately. = the values of the non-traffic highway variables at site i during that time period. α = the estimated intercept term.1.2 Goodness-of-Fit Evaluation Four GOF measures were used in the model selection process (refer to chapter 2 for a description of the GOF measures. proposed by Hauer and Hauer and Bamfo. and K = the estimated overdispersion constant. which is identical to that used in previous efforts. was used to provide the best fit to the data. Xi2.9) The graph shows how well the model fits the data with respect to each individual covariate.(1. and βi1 βi2….

CURE Plot for Type III TOTACC AADT Model 126 . Clearly.Type III intersections (presented in table 142). The indication is that the fit is very good for this covariate in that the cumulative residuals oscillate around the value of zero and lie between the two standard deviation boundaries. Figure 20 is a CURE plot for an alternate model. 25 20 15 10 Cumulative Residuals 5 Adj Cum Residuals 0 0 -5 10000 20000 30000 40000 50000 60000 70000 80000 Standard Deviation Standard Deviation -10 -15 -20 -25 Major AADT Figure 19. the alternate model cannot be judged to be an improvement over the selected model. Appendix D contains CURE plots for the TOTACC AADT models for all intersection types.

There were primarily minor differences in the summary statistics between those calculated on the available data and those stated in the reports. anomalies. such as traffic control or number of legs. and to identify any difficulties. The abbreviation employed in the modeling efforts and their descriptions are provided in the following section. CURE Plot for Type III TOTACC AADT Model Using the CURE Method: Alternative Model Now that the model’s end uses.1 Summary of Datasets The data used for recalibration were obtained from three sources. and V. and V. existing 127 .2.2) Additional years of accident and traffic data were obtained for those sites which did not experience a change in major variables.(1. 3. guiding research philosophies. 3.25 20 15 10 Cumulative Residuals 5 Adj Cum Residuals Standard Deviation Standard Deviation 0 0 -5 10000 20000 30000 40000 50000 60000 70000 80000 -10 -15 -20 -25 Major AADT Figure 20. and peculiar circumstances that needed to be remedied in the effort. However. IV. IV. The first set was the original calibration data used by Vogt and Bared from Minnesota and Vogt from California and Michigan. in general there were a larger number of variables available for estimation of model Types III. particularly for the vertical curvature variables for Type V sites. The first two sets were identical to the data used for the validation exercise described in chapter 2. and technical modeling strategy have been described.2 DESCRIPTION OF DATA Different variables were used in developing statistical models for Types I and II compared to Types III. the details of the technical modeling efforts are presented and discussed. It is useful to first describe the data that were used in the model recalibration efforts. Although average daily traffic variables are common to all models.

These statistics are also provided in this section of the report. their causes are unknown and these discrepancies were small enough that the data could confidently be used for recalibration. This data set was acquired to increase the size of the recalibration datasets with the aim of providing improved models with smaller standard errors of parameter estimates. and only the California HSIS sites were used to develop the full models 3 For Type V. the California HSIS sites were not included due to the limited availability of variables relevant to these models. Table 108 summarizes the sources of data used for recalibrating Models I to V. Again.e. The accident data obtained for the original sites included data for both the original and additional years. not available electronically in that database 2 Only the original sites were used to develop the base models for Types III. IV. However. The second source of data was for those sites selected in Georgia to provide and an independent set of validation data. IV. as site visits were not conducted.differences are sufficiently minor that further clarification was not necessary. 128 . Sources of Data No. It is also appropriate and useful to examine which variables strongly correlate positively or negatively with crashes and which potential independent variables are correlated to one another. of Sites Type Type Type II III IV In this section. and V. such as vertical curvature. Table 108. of Total (Injury Accidents) Type Type State V Type I II Type III Type IV Type V 2029 1892 Minnesota 1985-98 270 250 N/A4 N/A4 N/A4 (788) (878) N/A4 N/A4 N/A4 6494 6063 2136 1956 1159 California1 1991-98 1432 748 294 222 75 (2978) (3058) (847) (899) (370) 427 478 507 California2 1993-98 N/A4 N/A4 60 54 18 N/A4 N/A4 (196) (268) (200) 248 277 1262 Michigan3 1993-97 N/A4 N/A4 24 18 31 N/A4 N/A4 (63) (92) (159) 295 255 124 222 489 Georgia 1996-97 116 108 52 52 51 (110) (142) (56) (104) (118) 8818 8210 2935 2933 3417 Total 1818 1106 430 346 124 (3908) (4078) (1162) (1363) (847) 1 These data come from the California HSIS database and do not include variables. and V. For model Types III. summary statistics are provided for the data available for recalibrating the full models (i. Only 1996-97 injury accidents were available 4 N/A: not available Years of Data Type Available I No. Differences were found in the accident counts between the original data obtained and this new dataset for the original years. models with explanatory variables other than traffic volumes). although small differences exist. The third source of data was the California HSIS database. fewer variables were available for these sites. These data were collected with a minimum amount of effort with assistance from the HSIS staff.

the Georgia sites and the California HSIS sites.0 29.3750 0.38 2. with the additional years of accident and traffic data.2.2%) 547 (30.8%) 1818 1738 (95.56 -1. Table 109.13 35750 10001 N/A1 50.1%) 433 (23.6553 55 1 2 0 4 0 0 23 0 1 -90 0 0 0 55 8 7 85.477 1.1250 4475 270 Minimum 0 0 401 100 N/A1 N/A1 N/A1 N/A1 N/A1 Maximum 6. Some of the Minnesota sites experienced changes in some design feature or location information during the 1990–98 period and were not included in the analysis. Summary statistics by State are available in appendix C.0 Table 110 shows correlation statistics and p-values that indicate the association between crash counts and the independent variables for type I intersections.6%) 1818 1382 (76%) 436 (24%) 1818 1804 (99.2660 6011 492 Median 0.4%) 48 (2.2 Type I A summary of the full data for Type I intersections is shown in table 109. The frequency column indicates the number of sites for which the information was available. This dataset includes the original sites in Minnesota.6%) 80 (4. Table 111 shows 129 .6074 0. Note that some variables are not available in the data for the Minnesota sites and California sites.75 0. Summary Statistics for Type I Sites Variables TOTACC per year INJACC per year AADT1 AADT2 RT MAJ Total 0 1 RT MIN Total 0 1 LT MAJ Total 0 1 LT MIN Total 0 1 MEDIAN Total 0 1 TERRAIN Total Flat Rolling Mountainous SPD1 DRWY1 HAZRAT1 HAU SHOULDER1 VCI1 HI1 1 N/A: not available Frequency 1818 1818 1818 1818 1818 1563 (86%) 255 (14%) 1818 1770 (97.4%) 1548 568 (31.75 4.451 4.2%) 14 (0.3.1 16 14.8%) 381 386 386 386 1547 386 386 Mean 0.89 1.

000 0.000 0.427 0.619 0.002 0.052 0.127 0. terrain and posted speed are negatively correlated with crashes. major and minor road AADTs correlate positively with crashes.016 0.020 -0.089 p-value 0. although this correlation is much less than that of vehicle volumes and the correlation for right-turn lane on major roads is not significant.001 0.030 -0. Shoulder width and number of driveways were not significantly correlated with crashes. while HAU was negatively correlated with crashes although this correlation was not as strong.106 0.841 0.116 0.157 0.056 -0.000 0. 0. This counterintuitive result may arise because.000 0.312 0.402 0.110 0.101 0.694 0. As expected.030 0. Georgia sites have higher accident frequencies than California and Minnesota sites. With the presence of a median. as well as lower average posted speeds and a higher percentage of sites in rolling or mountainous terrain.033 0.020 0.202 0.080 Corr.012 0.426 0. as shown in Appendix C.013 0. Surprisingly.000 0.165 0.088 SHOULDER12 VCI1 1 1 HI11 0.000 0.059 -0.149 0. VCI1 and HI1 were positively correlated with crashes.085 0.327 0.000 0.000 0. Table 110.005 0. 0.001 0.correlations between the independent variables.428 0.065 0.072 -0.205 0. Only those correlations that are significant at the 90 percent level are shown. meaning that areas with rolling or mountainous terrain experience a lower crash risk than flatter terrains and that higher speeds are associated with fewer crashes.074 -0.558 0.076 -0.000 0.013 0.516 0.087 These variables are unknown for the California sites 2 These variables are unknown for the Minnesota sites 130 . Correlation Between Crashes and Independent Variables for Type I Sites TOTACC per YEAR Variables AADT1 AADT2 RT MAJ RT MIN LT MAJ LT MIN TERRAIN MEDIAN SPD11 DRWY1 HAU 1 1 INJACC per YEAR Corr. Turning lanes on the major and minor roads are also positively correlated with crashes.081 p-value 0.

LT MAJ. SHOULDER1 AADT1. MEDIAN. SPD1 MEDIAN. LT MAJ. LT MAJ. LT MIN. RT MAJ. RT MIN. HAZRAT1 Only those correlations are shown for which p-values are less than 0. RT MAJ. TERRAIN. HI1 Negative Correlates1 VCI1. AADT2. VCI1. DRWY1. SPD1. LT MIN. RT MAJ. the Georgia sites and the California HSIS sites. RT MAJ. SHOULDER1 HI1 AADT2. TERRAIN TERRAIN HAZRAT1. SPD1. HI1. MEDIAN. TERRAIN. SHOULDER1 AADT1. Note that some variables are not available in the data for the Minnesota sites and California sites. NODRWAY. SHOULDER1 AADT1. HI1 RT MAJ. MEDIAN. RT MIN. MEDIAN. AADT2. HAZRAT1 MEDIAN. TERRAIN AADT1. MEDIAN. RT MAJ.Table 111. RT MAJ. AADT2. RT MIN. VCI1 TERRAIN. HAZRAT1. HI1 SPD1 RT MAJ. VCI1.10. VCI1 RT MIN. Some of the Minnesota sites experienced changes in some design feature or location information during 1990–98 and were not included in the analysis. HI1 TERRAIN. LT MAJ. HAZRAT1. LT MAJ. TERRAIN.2. The frequency column indicates the number of sites for which the information was available. LT MIN. SPD1. LT MAJ. RT MIN. SHOULDER1 HI1 TERRAIN. VCI1. LT MIN. HI1 AADT2. RT MIN. HAZRAT1. RT MIN. LT MAJ. Summary of Correlations for Independent Variables for Type I Sites Variable AADT1 AADT2 RT MAJ RT MIN LT MAJ LT MIN MEDIAN TERRAIN SPD1 DRWY1 HAZRAT1 HAU SHOULDER1 VCI1 1 Positive Correlates1 AADT2. SPD1 AADT1. This dataset includes the original sites in Minnesota. LT MIN. LT MAJ. with additional years of accident and traffic data. SPD1. SHOULDER1 MEDIAN. SHOULDER1 AADT2. 2 Not all variables are available for Minnesota or California sites 3. MEDIAN AADT1. Summary statistics by State are available in appendix C.3 Type II A summary of the full data for Type II intersections is shown in table 112. LT MAJ. AADT2. 131 . LT MAJ. VCI1 AADT1.

4%) 1106 883 (79.2%) 1106 1105 (99.05 0 30 0 1 -120 0 0 0 55 6 6 150 16 8 14.00 0 6 0.7%) 37 (3.6%) 26 (2.Table 112.75 38126 7460 N/A1 52 0.5%) 98 (8. As expected.13 4. terrain and posted speed are 132 .4%) 195 (17.9%) 355 358 358 358 855 358 358 Mean 0. however left-turn lanes on the minor roads were not significantly correlated with crashes.553 HI1 N/A: not available Table 113 shows correlation statistics and p-values that indicate the association between crash counts and the independent variables for Type II intersections.45 0.9%) 1 (0.896 55 0 2.2500 4245 344 Minimum 0 0 407 100 N/A1 N/A1 N/A1 N/A1 N/A1 Maximum 7.1%) 1106 1069 (96. Summary Statistics for Type II Sites Variables TOTACC per year INJACC per year AADT1 AADT2 RT MAJ Total 0 1 RT MIN Total 0 1 LT MAJ Total 0 1 LT MIN Total 0 1 MEDIAN Total 0 1 TERRAIN Total Flat Rolling Mountainous SPD1 DRWY1 HAZRAT1 HAU SHOULDER1 VCI1 1 Frequency 1106 1106 1106 1106 1106 911 (82.83 2. while right-turn lanes on the minor roads were positively correlated with crashes.43 0. major and minor road AADTs correlate positively with crashes. Note that some variables are not included in the data for the Minnesota and California sites.9227 0. Again.8%) 223 (20. Right-turn lanes on the major roads were negatively correlated with crashes. Table 114 shows correlations between the independent variables.364 5. Left-turn lanes on the major roads were positively correlated with crashes.4665 5487 532 Median 0.3%) 856 520 (47%) 238 (21.426 0.6%) 1106 1080 (97.5357 0. Only those correlations that are significant at the 90 pecent level are shown.

000 0.000 0.003 0.029 p-value 0.258 0.251 0.246 0.086 These variables are unknown for the California sites 2 These variables are unknown for the Minnesota sites 133 .001 0.197 0.000 0.008 0.443 0. Intersection angle (HAU).133 0. meaning that areas with rolling or mountainous terrain experience a higher crash risk than flatter geographies and that higher speeds are associated with less crashes.004 0.106 HI11 0. Correlation Between Crashes and Independent Variables for Type II Sites TOTACC per YEAR Variables AADT1 AADT2 RT MAJ RT MIN LT MAJ LT MIN TERRAIN MEDIAN SPD11 DRWY11 HAZRAT11 HAU1 SHOULDER13 VCI11 1 INJACC per YEAR Corr.001 0.364 0.111 0.821 0. 0.000 0.046 0.384 0. and roadside hazard rating on the major roads were all positively correlated with crashes. 0. number of driveways.895 0.000 0.434 -0.027 -0.000 0.152 -0.123 p-value 0. Table 113.000 0.000 0.028 -0.020 Corr.000 0.001 0. shoulder width.003 0.000 0.126 0.000 0.007 -0.000 0.390 0. and VCI1 were not significantly correlated with crashes.046 0.101 0.115 0.103 0.000 0.444 0.184 0. HI1.041 0.580 0.negatively correlated with crashes.425 -0.057 0.105 0.265 0.060 -0.353 0.970 0. Presence of a median.088 -0.

DRWY1. HAZRAT1. HI1 RT MAJ. In total. HAZRAT1 RT MAJ. MEDIAN. As before the frequency column indicates the number of sites for which the information was available. RT MAJ. SPD1 LT MAJ. DRWY1. TERRAIN.2. LT MAJ. RT MIN. LT MAJ. Summary of Correlations for Independent Variables for Type II Sites Variable1 Positive Correlates2 AADT2. VCI1. HAU HAZRAT1 AADT1. VCI1. VCI1. HAZRAT1. LT MAJ. AADT2. median. HI1 AADT1. HAZRAT1. with the additional years of accident data. VCI1. etc) of relevance. DRWY1. HI1 SHOULDER1 SPD1. HAZRAT1. Some California sites experienced changes in some design features during 1996–98. AADT2. HI1 MEDIAN. LT MIN. SHOULDER1. LT MIN. TERRAIN. only 1993–95 data were used. This left the California and Michigan sites from the original study. TERRAIN AADT2. LT MIN AADT1. LT MAJ. LT MAJ. HAZRAT1. RT MIN. AADT2. TERRAIN. TERRAIN. 42 variables were available for model development. SPD1 VCI 1 Not all variables are available for Minnesota or California sites 2 Only those correlations are shown for which p-values are less than 0. SPD1. HAZRAT1. LT MAJ. terrain. DRWY1. VCI1. MEDIAN AADT1. for inclusion in the database. LT MIN. SHOULDER1 AADT1. MEDIAN. RT MAJ.4 Type III A summary of the full data for Type III intersections is shown in table 115. 134 .Table 114. DRWY1. MEDIAN. For these. SPD1 LT MAJ. HI1 AADT1. LT MAJ.10 3. SHOULDER1 AADT1. TERRAIN. TERRAIN. VCI1. AADT2. HAZRAT1. HI1 RT MAJ. LT MAJ. RT MIN. HI1 Negative Correlates2 AADT1 AADT2 RT MAJ RT MIN LT MAJ LT MIN MEDIAN TERRAIN SPD1 DRWY1 HAZRAT1 HAU SHOULDER1 VCI1 RT MAJ SPD1 AADT1. AADT2. SHOULDER1 AADT2. TERRAIN. TERRAIN. HAZRAT1 RT MIN. DRWY1. HAZRAT1. SHOULDER1 AADT1. SPD1 RT MAJ. DRWY1. The HSIS California data were excluded in developing Type III full models because this data set has only a few variables (turning lanes. HI1 RT MAJ. RT MIN. MEDIAN. RT MAJ. SPD1.

0%) 16(32.7%) 52 0(0%) 2(4.0 30 15 14 7 15.0 0 0 1.0 2.80 0.Table 115.1%) 14(10.6%) 26(19.0 55 35 0 0 0.6 1.6 1.0 0 0 0.1%) 25(18.55 136 13011 136 709 136 69(50. Summary Statistics for Type III Sites Variables TOTACC per year INJACC per year AADT1 AADT2 MEDTYPE1 Total No Median Painted Curbed Other MEDWIDTH1 HAU HAZRAT1 Total 1 2 3 4 5 6 7 HAZRAT2 Total 1 2 3 4 5 6 7 COMDRWY1 RESDRWY1 DRWY1 NoCOMDRWY2 RESDRWY2 DRWY2 SPD1 SPD2 1 N/A: not available Frequency Mean 136 1.8%) 58(42.5 0.0%) 6(12.3%) 8(5.7 0 0 1.35 136 0.0%) 2(4.0 65 55 135 .5 33.0%) 6(12.3 6 0 0 -65 63 90 N/A1 N/A1 1.7%) 45(33.9%) 136 136 136 16(11.00 2360 15 Maximum 10.0%) 136 136 136 52 52 52 136 136 Median 0.0 3 6 6.5%) 1(0.00 0.4%) 8(5.60 4.0%) 20(40.0 52.5 1.4 0.00 33333 9490 N/A1 12.33 12100 430 Minimum 0.9%) 2(1.

16 53.3%) 5(3.29 26.6 2.0 4.0 7.13 14.5 0.7 24.0 0.0 1.01 1.28 2.0 0.0 0.7 1.0 1.4%) 28(20.6%) 136 48(35.0 1.09 25.15 6.7%) 136 117(86.26 2.0 1. Summary Statistics for Type III Sites (Continued) Variables LIGHT Total 0 1 TERRAIN1 Total Flat Rolling Mountainous TERRAIN2 Total Flat Rolling Mountainous RTLN1 Total 0 1 LTLN1 Total 0 1 RTLN2 Total 0 1 LTLN2 Total 0 1 HI1 HEI1 GRADE1 GRADE2 VEI1 VI2 LEGACC1 LEGACC2 SHOULDER1 PKTRUCK PKTURN PKLEFT 1 N/A: not available Frequency 136 97(71.0 0.16 0 0 0.1%) 136 108(79.6%) 21(15.68 3.79 4.7%) 136 136 136 52 136 52 52 52 52 84 84 84 Mean Median Minimum Maximum N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 1.0 0.9 4.3%) 39(28.00 0.1 4.Table 115.0 0.27 0.63 5.7 6.28 0.18 0.0 28.4%) 7(5.0 0.0 0.8 0.73 0.9 4.9%) 11(8.3%) 88(64.97 136 .0%) 136 131(96.0 9.0 0.0%) 19(14.0 10.2 0.7%) 136 83(61.1%) 52 24(17.0%) 42(30.

1211 0.1917 0. p-value 0.4201 0. shoulder width on major roads.0675 0.0254 0.47 55. although many of these correlations are insignificant.1190 0.2062 0.0007 INJACC per YEAR Corr.3703 0.29 100.0001 -0. There are several variables for which the correlation results are unexpected.0008 0.3606 0.9756 -0.00 0.0000 -0.4829 0.0697 0.1696 0.1677 0. PKLEFT1 and PKLEFT2 are also negatively correlated with each other.0005 0.0774 0. as well as other variables are correlated with crashes in the opposite direction to that expected. while PKTRUCK and PKLEFT2 correlate negatively with crashes.0051 0.5483 0.3168 -0.1184 0. According to table 114.00 2000 2000 2000 Table 116 shows correlation statistics and p-values that indicate the association between crash counts and the independent variables for Type III intersections.3959 0. be a consequence of the positive correlation of crashes with AADT variables.9534 0.69 60.1416 0. Table 116.0486 0. terrain on major roads.0947 -0.Table 115.” and the presence of left-and right-turn lane on minor roads.1307 0.7321 -0. Correlation Between Crashes and Independent Variables for Type III Sites Variables AADT1 AADT2 MEDWDTH1 HAU COMDRWY1 RESDRWY1 DRWY1 COMDRWY2 RESDRWY2 DRWY2 SPD1 SPD2 LIGHT TOTACC per YEAR Corr.4352 0.29 2000 1510 1555 Minimum 0. and PKLEFT1 correlate positively with crashes.0044 0.1603 0. Roadside hazard rating on major and minor roads. This suggests that the negative correlation of crashes with PKTRUCK may. in part.3330 0.00 500 40 80 Maximum 21. “LIGHT.0000 -0.2943 0. PKTRUCK correlates negatively with the AADT variables.3229 0.31 1515 1418 1428 Median 0.1295 137 . Major and minor road AADTs correlate positively with crashes as expected.2842 0. Table 114 shows correlations between the independent variables.1956 0.2342 0.0854 0. posted speed limits on major and minor roads. both positively and negatively. PKTURN.0519 0. Only those correlations that are significant at the 90 percent level are shown. Summary Statistics for Type III Sites (Continued) Variables PKLEFT1 PKLEFT2 SD1 SDL2 SDR2 Frequency 84 84 136 136 136 Mean 1.1425 -0. PKLEFT. Peak turning movement volumes also correlate with crashes.0001 0.1765 0. p-value 0. number of residential driveways on major and minor roads.0398 -0.2882 0.0000 -0.3299 0.1647 -0.

PKTRUCK.4986 INJACC per YEAR Corr. LIGHT.Table 116.1628 0.0052 0.3685 0.1943 0. RESDRWY1.9216 0. 0.2749 0.0791 0.0766 0.8043 Table 117.0003 -0. HAU.4771 0.0721 0. SDR2 MEDWDTH1. L1LT L1RT.0485 0.1247 -0.0752 -0. HAZRAT1.0162 0. VI2. DRWY2. Correlation Between Crashes and Independent Variables for Type III Sites (Continued) TOTACC per YEAR Variables L1RT L1LT L3RT L3LT HI1 HEI1 GRADE1 GRADE2 VEI1 VI2 LEGACC1 LEGACC2 SHOULDER1 PKTRUCK PKTURN PKLEFT PKLEFT1 PKLEFT2 SD1 SDL2 SDR2 Corr.2479 -0.0192 0. L1RT.5582 0.0633 -0.8243 0.2296 0.PKLEFT1.1481 0.2617 0. 0. L3LT. SPD2.1205 0. RESDRWY1.1534 -0.0583 0. PKTRUCK.0357 0.6116 0. L1LT.0230 0.0585 p-value 0. VEI1 COMDRWY1. L1LT. SHOULDER1. LIGHT.0344 0. GRADE1. RESDRWY1.7208 0. COMDRWY1. MEDTYPE2.2527 0. L1LT.0181 0.3249 0. SDL3.0140 -0. PKTURN. MEDTYPE2.2298 0. GRADE1.PKLEFT2.8915 0.5751 0.7284 0.2304 0.0309 0.0118 -0. PKTURN.1392 -0.2373 0. PKTRUCK. TERRAIN.2099 0.0831 -0.0129 -0.0204 0.3843 0.1353 0. PKTRUCK COMDRWY1. DRWY1.1977 0.0615 0. SDR2 MEDTYPE1. DRWY1.0115 0.6911 0. PKLEFT. L1RT. MEDTYPE3.4719 0.0071 0.0027 0. HI1.2744 -0.1610 -0.4633 0.2025 0.1020 -0. L1LT. PKTRUCK.0350 0.9970 0.1436 0. SDL2 DRWY1 SPD1.9748 0. HI1 Negative Correlates1 MEDTYPE2. LEFACC2.0300 -0. SPD1.9278 0.1717 0. Summary of Correlations for Independent Variables for Type III Sites Variable AADT1 AADT2 MEDWDTH1 HAU HAZRAT1 Positive Correlates1 L1RT.4642 0. SPD2.0456 0. DRWY2.0054 0. SDR3 138 .0994 -0.1039 -0.9520 0.1601 0. SPD1.1511 0. L3RT.4949 0. SHOULDER1. HAZRAT2. PKLEFT2. RESDRWY1. SPD1.0214 p-value 0. SPD2. LIGHT.0746 0. TERRAIN1. COMDRWY2.0968 0.

TERRAIN1. PKLEFT1 AADT2. L1LT. COMDRWY1. VEI1. MEDWDTH1. DRWY1. L1RT. SPD2. GRADE1. PKTURN. HAZRAT1. SPD1. L1RT.2. HI. LEGACC2. DRWY1. LIGHT. LIGHT. SDL2. L1LT. PKTURN. SPD2. TERRAIN2 HAZRAT2. LEGACC2 HAZRAT1.Table 117. SDR2 AADT2. L3RT. SDL2. L1RT. In total. RESDRWY1. SDR2 AADT1. TERRAIN1. LIGHT. SPD1. MEDWDTH1. RESDRWY1. HAZRAT2. only 1993–95 139 . MEDTYPE1. SD1. etc. PKLEFT1 MEDTYPE1. SD1. SD1. Some California sites experienced changes in some design features during 1996–98. TERRAIN2 HAZRAT2 AADT1. SPD1. GRADE1. L1RT. yes=1) TERRAIN1 L1RT L1LT L3RT L3LT PKTRUCK PKTURN VEI1 HEI1 GRADE1 MEDWDTH1. SDL2. SPD1 SPD2 LIGHT (no=0. TERRAIN1. TERRAIN1. HI1. 53 variables were available for model development. For these. terrain. VEI1. AADT2. HAZRAT1. SPD2. MEDTYPE1. GRDE2. SPD2. SPD2. PKLEFT1 MEDTYPE1. Instead. COMDRWY1. SDR2 3. RESDRWY1. TERRAIN1. SD1. GRADE2. PKTRUCK. LEFACC1. SDR2 GRADE1. PKLEFT. L1RT. L3RT. L1LT. GRADE2. MEDWDTH1. HI1.10 COMDRWY1. PKLEFT. DRWY1. RESDRWY1. PKLEFT.5 Type IV A summary of the full data for type IV intersections is shown in table 118. PKTRUCK. with the additional years of accident data were included in the database. GRADE2. HAU. GRADE1. SD1. SDL3 VEI1. DRWY1. SPD1. L1LT. DRWY1. COMDRWY2. HI1. PKLEFT1 AADT1. SPD1. MEDTYPE1. HAZRAT2. SDL2. LIGHT MEDTYPE2. SPD1. VEI1 SPD1. DRWY1. L1RT. L3LT. DRWY1. SDR2 HAZRAT2. VI2 AADT1. COMDRWY1. SPD2. SD1. L1LT. MEDTYPE3. SD1. LEFACC2. COMDRWY2. DRWY1 SDR2 1 Only those correlations are shown for which p-values are less than 0. VEI1. TERRAIN1. MEDTYPE3.) of relevance. SD1. VI2 MEDTYPE1. HAZRAT1. SDR2 MEDTYPE1. LIGHT. Negative Correlates1 AADT2. HAZRAT1. SDR2 SD1. L3LT. L1LT. HI1. L3LT. The HSIS California data were again excluded because of a lack of sufficient variables (turning lanes. PKTRUCK. MEDTYPE2. TERRAIN2. Summary of Correlations for Independent Variables for Type III Sites (Continued) Variable Positive Correlates1 MEDTYPE1. median. HI1. PKTURN. L3RT. MEDWDTH1. L3LT. GRADE1. MEDWDTH1. PKLEFT. PKTRUCK. HAZRAT1. TERRAIN2 HAZRAT2. L3LT. AADT2. HAZRAT1. SDL2. L1RT. MEDWDTH1.MEDTYPE3. GRADE2. MEDWDTH1. PKTURN. SDL2 SPD1. SPD2. the California and Michigan sites from the original study. SHOULDER1 AADT1. SHOULDER1. PKTRUCK. HI1. COMDRWY1. RESDRWY1. SDR3 AADT2. VEI1 COMDRWY1. HI1. DRWY2. TERRAIN1. SPD1. LIGHT.

0 4.7 3150 73799 21 2990 N/A1 N/A1 16.6%) 2(1.0 5.6%) 20(16.8%) 124 24(19.data were used.4 0.6%) 0(0%) Median 1.1%) 13(10.1 0.0 124 0. frequency indicates the number of sites for which the information was available.5%) 27(21.7%) 5(4.4%) 43(34.2%) 3(5.5 11496 430 Minimum Maximum 0.0%) 52 52(100%) 124 52 52 52 124 69(55. As before.5 0 4 0 0 0 2 0 60 1 6 2 N/A1 N/A1 N/A1 N/A1 N/A1 N/A1 140 .6%) 52 49(94. Summary Statistics for Type IV Sites Variables TOTACC per YEAR INJACC per YEAR AADT1 AADT2 MEDTYPE on major Total 0: No Median 1: Painted 2: Curbed 3: Other MEDTYPE on minor Total 0: No Median MEDWDTH1 MEDWDTH2 SHOULDER1 SHOULDER2 L1RT Total 0 1 2 L3RT Total 0 1 2 L3LT Total 0 1 LEGACC1 Total 0 1 LEGACC2 Total 0 1 HAZRAT1 1 2 3 4 5 6 7 1 N/A: not available Frequency Mean 124 2.9%) 2(1.7%) 32(25.9 124 12881 124 621 124 70(56. Table 118.8 0.0 10.8%) 22(17.2%) 3(5.3 6.5%) 124 122(98.1%) 35(28.2 0.8%) 52 49(94.2%) 124 72(58.4%) 2(1.5%) 39(31.8%) 21(16.

8 0 0 0 0 0 0 0 0 0 0 0 0 12 7 15 4 3 6 N/A1 Total N/A1 Total N/A1 0.3%) 52 19(36.71 7.00 0.28 1.71 1.00 0.6%) 25(20.18 11.00 0.66 12.50 12.5%) 124 124 124 124 52 52 52 52 124 51 124 124 124 124 Mean Median Minimum Maximum N/A1 0.5%) 27(51.2%) 9(7.5 55.6 0.62 0.00 0.2%) 37(29.00 0.3 0.33 55 65 141 .00 2.8%) 124 90(72.87 0.00 0.5%) 15(28.Table 118.50 10.00 0.92 3.35 0.7 1.00 -50 25 12.00 0.08 1.43 3.00 0.00 0.50 12.32 0.60 0 55 0.50 12.8%) 16(30.33 233.63 0.80 3. Summary Statistics for Type IV Sites (Continued) Variables HAZRAT2 1 2 3 4 5 6 7 COMDRWY1 RESDRWY1 DRWY1 COMDRWY2 RESDRWY2 DRWY2 LIGHT 0 1 TERRAIN1 Flat Rolling Mountainous TERRAIN2 Flat Rolling Mountainous VEI1 VCEI1 VI1 VCI1 VEI2 VCEI2 VI2 VCI2 GRADE1 GRADE2 HI HEI HAU SPD1 1 N/A: not available Total Frequency 52 0(0%) 7(13.31 2.94 1.05 2.50 5.36 9.00 0.8%) 0(0%) 124 124 124 52 52 52 124 87(70.4 0.60 0.48 0.9%) 6(11.8%) 12(23.02 0.65 0.08 0.6 0.84 2.4 0.62 2.00 0.1%) 2(3.97 2.

89 124 1399 124 1314 124 1329 Median 35 8. as expected. as table 120 shows.00 0. Only those correlations that are significant at the 90 pecent level are shown.7 72 10. horizontal curves on major roads.Table 118. both positively and negatively.00 400 324 215 Maximum 55 37. Peak turning movements also correlate with crashes. Shoulder width on the road. light.00 0.09 100. absolute grades on major and minor roads. intersection angle.77 0. median width on major road is insignificant with a counterintuitive sign.51 10. there is a negative correlation between median width on major roads and median types.36 96.56 3.75 67. There are several variables for which the correlation results are contrary to expectations.25 100. For example. posted speed limit on major roads.08 1.95 6.00 0.52 25. Table 120 shows correlations between the independent variables. acceleration lane on major roads.69 72 38. However.66 1332 1262 1354 Minimum 25 1. and others are correlated with crashes in the opposite direction than expected.00 0.00 48.96 68. although many of these correlations are insignificant.41 72 9. right-and left-turn lane on minor roads. terrain on major and minor roads.00 2000 2000 2000 Table 119 shows correlation statistics and p-values that indicate the association between crash counts and the independent variables for Type IV intersections.80 72 2.82 36.26 13. vertical curves on major and minor roads. Major and minor road AADTs correlate positively with crashes.95 72 94. Summary Statistics for Type IV Sites (Continued) Variables SPD2 PKTRUCK PKTHRU1 PKTURN PKLEFT PKLEFT1 PKTHRU2 PKLEFT2 SD1 SDL2 SDR2 1 N/A: not available Frequency Mean 124 34. 142 . the result of which is that median type is skewing the effect of median width at Type IV intersections.78 72 15. residential driveway and total driveway on minor roads.47 72 4.

6999 0.3243 -0.0770 -0.3369 0.0038 -0.6815 -0.0081 0.0692 0.1220 0.5137 0.7397 -0.9133 0.0490 0.0494 0.0097 0.4633 0.7215 0.0765 0.0698 0.1547 0.0350 0.0608 0.0311 0.0592 0.5887 -0.0657 0.2826 0.1876 0.0672 0.0506 0.2156 0.8245 -0.0099 0.0015 INJACC per YEAR Corr.1184 0.2626 0.0942 0.8676 -0.1825 0. p-value 0.1770 -0.0596 0.1594 0.1929 0.2809 0.6360 -0.5109 0.1900 0.0409 0.1137 0.8459 0.0211 0.1187 -0.0288 -0.0318 0.1905 -0.3503 -0.3099 0.1040 0.0022 0.0084 0.7994 -0.7171 -0.3242 0.9418 -0.7963 0.2979 0.2195 -0.1248 -0.0373 0.1145 -0.0818 0.4408 0.0863 0.8161 -0.3058 0.1631 0.2258 0.0055 0.2544 0.0104 0.0314 0.0174 0.6806 0.0323 0.2526 0.7289 -0.9709 -0.9519 -0.Table 119.1263 -0.1772 -0.2714 0.0738 0.0319 0.2089 0.0892 0.3550 0.4411 0.1999 -0.1500 0.0151 0.1372 -0.3984 -0.2613 0.0770 0.0051 -0.0015 0.2600 0.0849 0.0607 0.0000 0.5026 0.2665 0.3486 -0.2324 0.1408 0.0090 0.6006 -0.4432 0.2323 0.0562 -0.9267 -0.0695 0.0995 0.2481 0.2147 0.1839 0.8332 0.0627 0.5277 -0.1331 0.1569 0.0033 0.1407 -0.0367 0.1428 0.0581 0.0583 143 .1705 0.0361 0.8476 0.2285 0.0117 0.0000 1.0846 0.1732 0.1155 0.0492 0.4152 0.0191 0. Correlation Between Crashes and Independent Variables for Type IV Sites Variables AADT1 AADT2 MEDWDTH1 MEDWDTH2 SHOULDER1 SHOULDER2 L1RT L1LT L3RT L3LT LEGACC1 LEGACC2 COMDRWY1 RESDRWY1 DRWY1 COMDRWY2 RESDRWY2 DRWY2 LIGHT VEI1 VCEI1 VI1 VCI1 VEI2 VCEI2 VI2 VCI2 GRADE1 GRADE2 HI1 HEI1 HAU SPD1 SPD2 PKTRUCK PKTHRU1 PKTURN PKLEFT PKLEFT1 PKTHRU2 PKLEFT2 SD1 SDL2 SDR2 TOTACC per YEAR Corr.2474 0.0283 0.1092 0.0176 0.0329 0.1964 0.2086 -0.9867 0.0975 0.5033 0.5212 -0.0107 0.0035 0.8423 -0.2209 0.6257 -0.0437 -0. p-value 0.0301 0.0600 -0.3268 0.0055 0.0572 0.6434 -0.3028 0.1017 0.1633 0.2474 -0.

PKLEFT1. HEI. PKLEFT1. LIGHT. L1RT. PKLEFT. DRWY2. L1RT. L3RT. PKLEFT. PKTURN. PKLEFT. RESDRWY1. VCI2. HEI1. SDR2 MEDWDTH1. VI2. LEGACC1. MEDWDTH1. SPD2. DRWY1. MEDWDTH2. HAZRAT2. MEDWDTH1. HI1. SDR2. PKLEFT. PKTRUCK. PKLEFT. DRWY1. PKTURN. DRWY2. PKTHRU2 MEDTYPE2. TERRAIN1. MEDTYPE2. L1LR. SPD1. DRWY1. L1LT. RESDRWY2. TERRAIN1. SPD2. COMDRWY1. PKTHRU1. VI2. DRWY1. L1LT. PKLEFT1 MEDWDTH1 HAU HAZRAT1 DRWY1 SPD1 SPD2 LIGHT (no=0. COMDRWY1. GRADE2. PKTURN. PKTHRU1 AADT2. PKLEFT1 MEDTYPE1. PKTURN. VI2. GRADE1. SD1. L1LT. SDR2 COMDRWY1. LEGACC1. RESDRWY2. LIGHT. PKTHRU1. MEDWDTH1. RESDRWY1. PKTHRU2 Negative Correlates1 VCEI2. PKTURN PKLEFT. PKLEFT2. L1RT. COMDRWY1. PKTHRU1. SPD1 COMDRWY1. MEDWDTH1. PKTHRU2 144 . PKTRUCK. L1LT. PKTRUCK L1RT L1LT L3RT L3LT MEDTYPE2. RESDRWY1. PKLEFT. SD1. PKLEFT1 MEDTYPE1. SPD2. PKTRUCK. L2RT. HAU. PKLEFT1 LIGHT MEDTYPE2. LIGHT. PKTURCK. SPD1. PKTURN. VEI2. VEI2. MEDWDTH1. PKTRUCK. VI2. MEDTYPE1. PKTURN. SPD2. VCI2. DRWY2. RESDRWY1. SPD2. HAZRAT1. COMDRWY2. HI1. SPD1. GRADE2. DRWY2. HAU. VCEI2. RESDRWY2. MEDTYPE3. VEI2. TERRAIN2. VCEI2. PKLEFT1 AADT1. RESDRWY2. LIGHT. SDL2. L1LT. HAZRAT2. L1RT. HAU. L1LT. MEDTYPE2. COMDRWY2. DRWY2. GRADE1. L1LT. DRWY1. PKLEFT1. SPD1. HI1. DRWY1. COMDRWY2. PKLEFT1 COMDRWY1. SPD1. RESDRWY1. RESDRWY2. LIGHT. PKLEFT1 HAZRAT2. VCEI2. RESDRWY1. VEI2. Summary of Correlations for Independent Variables for Type IV Sites Variable AADT1 AADT2 Positive Correlates1 MEDTYPE1. SPD2. GRADE2. PKTURN. PKLEFT AADT1. COMDRWY2.Table 120. SPD2. PKTRUCK. COMDRWY2. HEI1. GRADE1. HEI1. SDR2 HAZRAT1. HAZRAT1. HAZRAT1. TERRAIN1. PKLEFT2 MEDTYPE1. PKTRUCK. L1RT. L1RT. TERRAIN2. PKTHRU1 MEDWDTH1. PKTRUCK. MEDTYPE2. TERRAIN2. L3RT. RESDRWY2. SHOULDER2. HI1. GRADE1. PKTHRU1. GRADE1. DRWY2. PKTHRU2 MEDTYPE1. PKLEFT. PKLEFT. MEDWDTH1. PKTURN. LIGHT. SHOULDER2. HEI1 MEDTYPE2. PKTHRU2. L3RT. SDL2. DRWY2. SDL2. COMDRWY1. L1RT.yes=1) TERRAIN1 AADT2. HEI1. PKTRUCK. RESDRWY1. DRWY1. MEDWDTH2. SD1. MEDWDTH1. SPD1. SPD1. SD1. SHOULDER2. TERRAIN2 MEDTYPE1. MEDTYPE3. L3RT. VCI2. LIGHT. PKLEFT2 MEDWDTH1. SDL2. MEDTYPE2. L3RT. MEDWDTH1. L1RT. PKTHRU1 MEDTYPE3. L1LT. L3RT. L3RT. VCI2. HAZRAT1.

RESDRWY1. VCI2. L1LT. SDL2.0 0. SPD1. SD1. L1LT. HEI1 SDL2 MEDTYPE2. TERRAIN1. SPD1. MEDTYPE2. AADT2. LIGHT. RESDRWY1. PKTURN PKLEFT. PKTURN. AADT2. Summary Statistics for Type V Sites Variables TOTACC per YEAR INJACC per YEAR AADT1 AADT2 Frequency 100 100 100 100 Mean 5. Variable Positive Correlates1 3. TERRAIN1.6 Type V A summary of the full data for Type V intersections is shown in table 121. L1RT. The HSIS California data were again excluded because only five Type V sites were available. PKTRUCK. DRWY2. LEGACC2. Summary of Correlations for Independent Variables for Type IV Sites (Continued) Negative Correlates1 AADT1. RESDRWY1. HAZRAT1. SPD1. HI1 PKTHRU2. 53 variables were available for model development.5 25132 12478 145 . VCEI1. only 1993–95 data were used. SD1. TERRAIN1. PKTHRU2. HAZRAT2.8 9126 3544 Median 5. For these. LIGHT. SDL2 VEI1. L3RT. TERRAIN1. L1RT. As before. SD1. DRWY1. HAZRAT1. PKTRUCK PKTHRU2. SD1. GRADE1 MEDTYPE2. PKTRUCK. SDR2 AADT1. COMDRWY1.0 430 420 Maximum 26.10. RESDRWY1. HAZRAT1.9 1. Table 121.Table 120.5 8700 3100 Minimum 0. SPD2. PKTHRU1 VEI1 VI1. VCI1. PKTRUCK. SHOUDLER2. PKLEFT. SDL2 HEI1 MEDTYPE1. DRWY1. RESDRWY2. SDL2. L1LT. DRWY1. the frequency column indicates the number of sites for which the information was available. MEDTYPE1. COMDRWY1. In total. MEDTYPE2. HI1 SDR2 1 Only those correlations are shown for which p-values are less than 0.5 6. with the additional years of accident data for inclusion in the database. AADT2. HAZRAT2. SPD1. PKLEFT1. LIGHT. DRWY1. DRWY1.VCI2. SDR2 MEDTYPE2. HI1. SD1.2.3 1. PKLEFT2 SPD1. GRADE1.VI2. VCEI1. L3RT. PKLEFT1. L1LT. Some California sites experienced changes in some design features during 1996–98 period. GRADE1. SDR2 VEI1. HAZRAT2. This left the California and Michigan sites from the original study. SDR2 AADT2. MEDTYPE2. L1RT. SHOULDER1. MEDWDTH1. SD1. HAZRAT1. GRADE1 VEI1. SDL2.

3 1.9 1.2%) 5(9.Table 121.1%) 3(5.5 0 0 2 2 0 0 0 0 13 12 10 10 N/A1 N/A1 N/A1 N/A1 N/A1 146 .9%) 0(0%) 100 100 51 51 100 51(51%) 21(21%) 28(28%) 100 23(23%) 2(2%) 75(75%) 100 59(59%) 20(20%) 21(21%) 100 45(45%) 5(5%) 50(50%) 51 46(90.8%) Mean Median Minimum Maximum N/A1 N/A1 N/A1 1. Summary Statistics for Type V Sites (Continued) Variables SIGTYPE Total 0:Pre-timed 1:Actuated 2:Semi-actuated MEDTYPE on major Total 0:No Median 1:Painted 2:Other MEDTYPE on minor Total 0:No Median 1:Painted 2:Other MEDWDTH1 MEDWDTH2 SHOULDER1 SHOULDER2 L1RT Total 0 1 2 L1LT Total 0 1 2 L3RT Total 0 1 2 L3LT Total 0 1 2 LEGACC1 Total 0 1 1 N/A: not available Frequency 100 33(33%) 45(45%) 22(22%) 100 87(87%) 12(12%) 1(1%) 51 48(94.3 0.

5%) 8(15.7%) 17(33.8%) 100 12(12%) 29(29%) 27(27%) 16(16%) 13(13%) 3(3%) 0(0%) 51 1(2%) 8(15.7%) 3(5.16 2.3%) 14(27.9%) 0(0%) 100 100 100 100 100 100 100 29(29%) 71(715) Mean Median Minimum Maximum N/A1 N/A1 N/A1 N/A1 2.13 2 0 3 2 0 3 N/A1 0 0 0 0 0 0 11 6 15 10 8 11 N/A1 147 .44 0.2%) 4(7.64 0. Summary Statistics for Type V Sites (Continued) Variables LEGACC2 Total 0 1 PROTLT1 Total 0 1 PROTLT2 Total 0 1 HAZRAT1 Total 1 2 3 4 5 6 7 HAZRAT2 Total 1 2 3 4 5 6 7 COMDRWY1 RESDRWY1 DRWY1 COMDRWY2 RESDRWY2 DRWY2 LIGHT Total 0 1 1 N/A: not available Frequency 51 50(98%) 1(2%) 100 70(70%) 30(30%) 51 47(92.52 3.Table 121.69 3.

91 1.79 14.07 43.00 0.9 8.97 13.36 0.79 94.77 41.52 2.99 28.59 0.90 18.90 1.00 -45.15 24.54 42.00 65 55 45.00 0. Table 123 shows correlations between the independent variables.73 84.00 0.00 0.00 0.71 34.97 13.00 45 40 7.00 25 20 2.13 10.3%) 31(60.81 1.66 96.09 37.00 0. 148 .96 35.07 45.00 0.03 1.61 0.00 0.00 0.43 72.19 43.78 2.45 4.00 0.73 Table 122 shows correlation statistics and p-values that indicate the association between crash counts and the independent variables for type V intersections.48 73.10 1.01 8.28 0.00 0.00 0.39 1.05 40.50 8.56 2.23 75.45 0.9%) 100 100 100 100 51 51 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 49 49 49 49 49 49 49 Mean Median Minimum Maximum N/A1 1314 1213 774 910 822 1042 1.8%) 2(3.00 7.17 14.00 36.58 2.Table 121.50 3.2 40.69 7.15 2.58 0.87 60.00 0.99 17.00 4.00 0.88 N/A1 235 224 122 142 103 224 0.20 1.00 0.45 1.00 0.41 47.19 1.20 1.64 71.00 0.98 7.44 32.95 2.21 1246 1091 673 750 798 934 1.07 18.00 1.59 2000 2000 2000 2000 2000 2000 11.54 1.32 1. Summary Statistics for Type V Sites (Continued) Variables TERRAIN1 Total Flat Rolling Mountainous TERRAIN2 Total Flat Rolling Mountainous SD1 SD2 SDL1 SDL2 SDR1 SDR2 VEI1 VEI2 VEICOM VCEI1 VCEI2 VCEICOM GRADE1 GRADE2 HEI HI HEI2 HI2 HEICOM HICOM HAU SPD1 SPD2 PKTRUK PKTURN PKTHRU1 PKTHRU2 PKLEFT PKLEFT1 PKLEFT2 1 N/A: not available Frequency 100 59(59%) 38(38%) 3(3%) 51 18(35. Only those correlations that are significant at the 90 percent level are shown.

1377 0.1633 0.0925 0. as expected.Again.0474 0.0552 -0. both positively and negatively.1924 0.0130 0.0850 0.9035 -0.0020 0. protected left lane on major and minor roads.9883 -0. acceleration lane on major and minor roads.1064 0.0942 0.3064 0.0084 0.1837 0. absolute grades. residential driveway on major and minor roads.0445 0.4458 0.0231 0. Table 122.2886 0.1012 0.2113 0.1044 -0.0015 0.1138 0.7036 0. Peak turning movement volume also correlates with crashes. although many of these correlations are insignificant.2033 0.5684 0.6212 -0.0500 0.0036 0.1649 0.1325 0.0095 0.0106 0.0692 0.2545 0.1323 0.0385 0.1079 0.2324 0.3163 -0.0348 0.1622 0.3513 0.1885 0.1692 0.0424 0.0911 0.2391 0.2452 0.1461 0.1408 0. p-value 0.9251 -0.4510 0. vertical curves.1340 0.1321 0.2057 0.2801 0.1888 0.4004 0.3602 0.2826 0.1008 0.1731 0. intersection angle.0828 -0. Shoulder width on major and minor roads.0000 -0.3652 0. left-and right-lane on major and minor roads.2552 -0.2437 0. major and minor road AADTs correlate positively with crashes.6399 -0.2581 0.4313 0.1718 -0.6296 INJACC per YEAR Corr.0048 0. terrains.3556 149 .1400 0.2964 0.0771 0.2919 0. sight distance.1072 0.0095 0.1938 0.8976 0.0146 0.1475 0.1667 0.2883 0.1122 -0.2271 0. and other variables are correlated with crashes in the opposite direction than expected.1922 -0.0975 0.1315 0.3056 0.0027 0.0818 0.0557 0.1591 0.1598 0.1149 0. horizontal curves.0094 0.0603 0.1516 0.2178 0.0295 0.6979 0. Correlation Between Crashes and Independent Variables for Type V Sites Variables AADT1 AADT2 MEDWDTH1 MEDWDTH2 SHOULDER1 SHOULDER2 L1RT L1LT L3RT L3LT LEGACC1 LEGACC2 PROTLT1 PROTLT2 COMDRWY1 RESDRWY1 DRWY1 COMDRWY2 RESDRWY2 DRWY2 LIGHT SD1 SD2 SDL1 SDL2 SDR1 SDR2 TOTACC per YEAR Corr. p-value 0.0123 0.

8268 -0.0376 0.1541 0.1988 0.0222 0. Summary of Correlations for Independent Variables for Type V Sites Variable AADT1 Positive Correlates1 Negative Correlates1 AADT2.1208 0. HI2.0275 0. RESDRWY2. PKTHRU2.0467 0.0904 0.4604 0. HAZRAT1. L1RT. HICOM GRADE1.3070 0.0510 0.1009 0.2103 0. PKTURN.2097 0.0073 Table 123.1794 -0.1739 0. LEGACC1. PKLEFT SIGTYPE3.3471 0.2234 0.2396 0.VEICOM. PKTURN.6302 -0. p-value 0.3829 -0. HEI.2059 0.0378 0. L1RT. COMDRWY2. L1LT. PROTLT2.1468 0.5722 -0. PROTLT1.8208 0. L1LT.4105 0.0374 INJACC per YEAR Corr.0702 0. SIGTYPE2. SDR1. HI2.6317 0.1464 0.0822 0.1950 0.2116 0.1023 0.1203 0.Table 122. HAZRAT2.0487 0. LIGHT.9450 -0.2312 -0.1228 0.0572 0.0070 0. RESDRWY2.3179 -0.1276 0. TERRAIN2.1326 0.3141 -0.7734 -0.0367 0.2983 0. HAU.0292 0. PKLEFT1 AADT1.7857 0.0312 0. MEDTYPE2. SIGTYPE2. L1LT.0145 -0.0357 0. HEICOM. VCEI1. p-value 0.0369 0.1079 0.0973 0.5368 0.6443 0. PKLEFT.7103 0. LEGACC1.1922 -0. VCEICOM.1032 0.0001 0.3819 0. 150 .8578 -0. GRADE2.1445 -0. SIGTYPE1.1639 -0.1837 0.4325 0.7155 0.0836 -0.7169 -0.1692 0. Correlation Between Crashes and Independent Variables for Type V Sites (Continued) Variables VEI1 VEI2 VEICOM VCEI1 VCEI2 VCEICOM GRADE1 GRADE2 HEI HI HEI2 HI2 HEICOM HICOM HAU SPD1 SPD2 PKTRUK PKTURN PKTHRU1 PKTHRU2 PKLEFT PKLEFT1 PKLEFT2 TOTACC per YEAR Corr.1167 0.1403 0. DRWY1.6144 0. HEI1.1482 0.2106 0. L3RT. HEI2.7090 0. PKTHRU1 AADT1.3784 0.1886 0.0000 0. VEI2. L3LT.0882 0.7580 -0.1895 0.4162 -0.0229 0.0474 0.1258 -0. PKTHRU1 AADT2 PROTLT1 SIGTYPE1. LEGACC2.0924 -0.2474 0.0674 0.3112 -0.0181 0.

SDR1. HAU SIGTYPE1 AADT2. SPD1. HI1. DRWY2. SDR2. SIGTYPE1. VEI1. SPD1. L1LT. DRWY1. HAZRAT1. HAZRAT2. VCEICOM. GRADE1. L3RT. PKTURN. HEI1. SD2. HAZRAT2. METYPE1minor. SDL1.Table 123. HAZRAT1. SPD2 L1LT 151 . SPD1. MEDTYPE2. HEICOM. SDR1. LEGACC1. SPD2. SDL1. PKTHRU2 AADT1. L1LT. MEDTYPE1. L3LT AADT2. GRADE2. SDL1. SDL1. DRWY1. Summary of Correlations for Independent Variables for Type V Sites (Continued) Variable Positive Correlates1 MEDTYPE1. PROTLT1. VEICOM. SDR2 HAZRAT1. SD1. HEI1. SD2. L1LT. HEICOM COMDRWY1. GRADE2. GRADE1. COMDRWY1. VCEI1. L1LT. SIGTYPE1. SD1. SD1. SD1. VCEICOM. L1LT. RESDRWY2. SPD2. LIGHT. DRWY1. SD1. COMDRWY1. SDL2. COMDRWY1. VCEI1. L1RT. HEI2. COMDRWY1. PKLEFT1 TERRAIN1. SDR2. VEI2. TERRAIN1. LIGHT. SDR2. SPD2 L1RT. PKLEFT2 LIGHT (no=0. SHOULDER2. PKTRUCK Negative Correlates1 MEDWDTH1 HAU HAZRAT1 DRWY1 SPD1 SPD2 L1RT. HICOM SIGTYPE1. COMDRWY2. L3LT. SDL2. L3LT. L3RT. HEI1. PROTLT2. DRWY2. DRWY1. SHOULDER1. TERRAIN1. SDL2. TERRAIN1. SDL1. MEDTYPE1minor. SPD1. SDL2. VCEICOM. SDR1. L1RT. VCEI1. VEI1. MEDWDTH2. PKLEFT2 MEDTYPE2. HI2. SDL1. PKLEFT1 TERRAIN1 L1RT. HAZRAT2. HICOM. SDL2. L1LT. MEDWDTH1. GRADE1 L1RT L1LT. GRADE2. DRWY2. VCEICOM. MEDWDTH2. DRWY1. RESDRWY1. GRADE2. SD2. L3LT. SPD1. PKLEFT1 SIGTYPE2. SDR2. SDL2. PKTRUCK. DRWY2. L3RT. SPD2 AADT1. SDL1. COMDRWY1. PROTLT1. L3LT. SD2. SD2. HICOM. DRWY2. TERRAIN2. PKTURN. LIGHT. PKLEFT HAZRAT1. LEGACC1. COMDRWY2. L4LT. L3RT. SDR1. L1RT. COMDRWY2. PKTHRU1 HAZRAT1. COMDRWY2. PKTHRU1 SIGTYPE3. SPD1. VEICOM. SDL2. L1RT. LIGHT. PROTLT1.yes=1) SIGTYPE3. VEI2. VEI1. SPD2. L3RT. SD2. PKLEFT. SDD1. GRADE1. HAZRAT2. COMDRWY2. SIGTYPE2. PRTLT1.

RESDRWY2. PROTLT1. SDL2. SPD1. SPD1. SDR2 SDL1. VCEICOM. SD2. SD1. SPD2. HI1. SDL2. SDL1. VCEI1. GRADE2 GRADE1 SDL2 SHOULDER1. HAZRAT1. PKLEFT2 HAZRAT1. VCEI1. SD2. TERRAIN1. HEI2. COMDRWY1. PKTURN. PKLEFT1 PROTLT1.1 were selected 152 . HICOM. HICOM. VEICOM. SDR2. SPD1. SPD2 AADT1. PKTHRU1. LEGACC1. PKTHRU1 HAZRAT1. SIGTYPE2. HEICOM. SDL1. HEI1. VCEICOM. SDR1. SHOULDER1. SPD2 AADT2. HI2. TERRAIN2. SDL1. VEI1. VCEICOM. SPD1. GRADE2. VEI2. PKTHRU1 SD1. SDR1. L3RT. SD1. SD2. GRADE1. L1RT. HI2.Table 123. COMDRWY2. SDL2. TERRAIN1. L1RT. LIGHT. HEICOM. PKTHRU2 VEICOM HEICOM L3RT. GRADE1. GRADE1. PKTHRU1 L3LT PKTRUCK PROTLT1. SDR1. DRWY1. COMDRWY1. PKLEFT2 HAZRAT1. GRADE2. SDR1. L1RT. SDR2. RESDRWY2. VCEI2. VEICOM. GRADE1. HAZRAT2. HICOM HAZRAT1. L3LT. SDR1. PKLEFT PKTURN AADT1. PKLEFT1 MEDWDTH1. L3RT. HI1. SDR2. SPD2. PKLEFT. HI2. VEI1. VCEI1. TERRAIN1. HEI2. SDL1. DRWY2. SD1. SDL2. GRADE1. SD2. VEI2. HEI1. VEI1. HI1. SD2. SD1. L1LT. VEI1. VEI1. VCEI1. L1RT. HI2. SDR2. DRWY2. HAZRAT2. LIGHT. SDL2. HEI1. SPD1. GRADE2. COMDRWY1. HAZRAT2. TERRAIN2. VEICOM. HEI2. L1LT. SPD1. GRADE2. HAZRAT2. HAU. COMDRWY2. VEICOM. L3LT. Summary of Correlations for Independent Variables for Type V Sites (Continued) Variable Positive Correlates1 AADT2. PKTURN. PKTHRU2 AADT2. VCEICOM. SDP2 1 Variables only significant with p-value of 0. L3RT. SPD2 SHOULDER1. DRWY1. VCEI2. VEI2. DRWY2. PROTLT1. VEI1. PKTHRU2 Negative Correlates1 L3RT HAZRAT1. LEGACC2.

the original sites from Minnesota and Michigan. including Minnesota Michigan Michigan California HSIS1 Georgia Georgia Georgia Subset of Group B Sites: AADT Model for sties meeting base conditions for California HSIS California HSIS California HSIS project data plus Minnesota Michigan Michigan California HSIS1 Georgia Georgia Georgia Subset of Group A Sites: Model for sites meeting base California project California project conditions for Michigan Michigan project data Georgia Georgia Not calibrated 1 The California project data for Types III.1 Description of Base Conditions The accident prediction algorithm outlined in the “Red Book” provides AMFs for four variables: intersection skew angle. IV. These AADT-only models used the data for the original sites from California and Michigan along with the Georgia validation data. Table 124. For Types III. 3.3. Second.(3) It was sought to develop AADT-only base models using sites which met the base condition criteria for these AMFs and for any other variables that logically and practically could provide additional AMFs. and three types of models are given for Type III.3. Table 124 shows the summary of the data used for AADT-only models. left-turn lanes on major approach.3 AADT MODEL ESTIMATION RESULTS This section discusses the development of AADT-only models. and V were not used Type V California HSIS Michigan Georgia Not calibrated California project Michigan Georgia Models were developed for total accidents (TOTACC) and injury (fatal + nonfatal injury) accidents (INJACC) using accidents within 76. Types I to IV models were developed for a subset of these sites that met specified conditions for possible use as base models in the IHSDM accident prediction algorithm. This group of data is referred to as Group A. Summary of AADT Models Recalibrated and Data Used Model Site Types and Data Used Description Types I and II Type III Type IV Group B Sites: AADT Model for California HSIS California HSIS California HSIS all sites. models were calibrated using all available data from the HSIS California database. V sites. First. This group of data is referred to as Group B. right-turn lanes on major approach. IV. Two types of models are presented for Types I and II sites. and the number of intersection quadrants with inadequate sight distance. additional AADT models were calibrated from a dataset that met the base conditions of the significant variables in the full models. IV. 153 . and the Georgia validation data.25 m (250 ft) of the intersection center. and V sites in this report.

original California. while the Group B dataset is more useful for the AADT model development. the combination of which is referred to as Group B. and that inadequate sight distance could be taken into account with the use of an AMF. Georgia. sight distance information is not available. For the Minnesota sites and the California HSIS sites. have deficient site distance. it is reasonable to believe that the majority of sites have adequate sight distance since roads are constructed to design standards and exceptions are made only where necessary. no turning lanes on the major or minor road and no medians on the major road were selected as base conditions. Selecting a skew angle of zero as a base condition would have left few sites for calibrating a base condition model. The Group B nominal or base conditions and the number of sites meeting these conditions for model types I to IV are presented in table 125. Group A and B Data For type I and II intersections. The Group B dataset has more sample sites but fewer variables than the Group A dataset for Type I to IV sites. 154 . and California HSIS data. Type V base models were calibrated only for the Group A sites because only five sites could have been added from the Group B California HSIS data.Base conditions were defined by examining the distribution of variables and selecting the most common condition. For Types I and II. However. approximately 65 percent of all the sites met all of the specified base conditions. the base models developed could be applied assuming they represent sites with adequate sight distance. Group B Base Conditions Intersection skew angle was not included as a base condition because the California HSIS data does not contain this variable and the skew angle at those sites where it is known is in fact highly variable. The second included the Michigan. the percentage of sites that met all of the base conditions was approximately 20 percent and 15 percent. the base models calibrated with a contrary assumption would produce artificially high predictions if crashes actually increased with deficient sight distance. For Types I and II. For Type III and IV intersections. the Group A datasets benefit the full model development because of the large number of variables. keeping in mind that enough sites must remain to calibrate reliable models. The first included the Michigan. Therefore. base models were developed from two datasets. Whether a variable exhibited an impact on safety was also considered in defining base conditions. and was therefore not included as a base condition. Type III and IV considered both Group A and B sites for the full model and AADT model development. respectively. In the event that many of the sites did. Georgia. and California HSIS sites (Group B). the base models were developed using all of the Minnesota. Thus. in fact. while for Types III and IV. and Georgia sites (Group A).

78) 62 (21. an intersection angle between plus or minus 5 degrees was defined as the base condition representing “no skew.65) Yes N/A 1 No No No Flat 718 (64.00) 1213 (66.72) 1818 (100.16) 174 (59.60) No 1804 (99.” No commercial driveways within 76.87) 176 (79.87) 164 (55. Table 126 shows the base conditions for these variables. only a few sites met the “no vertical curve” condition and. Separate base conditions were defined for TOTACC and INJACC models.5 m (100 ft) (which is relatively flat) was used for the base condition. commercial driveways on the major road. and unlike the cases for Types I and II. terrain on major road was included as a base condition for Types III and IV because it showed a significant impact on safety.25 m (250 ft) of the intersection center is the final base condition.00) Total sites N/A: not available 294 (100.05) 268 (91.97) 1382 (76.65) 148 (66.18) 292 (99. Also for Types III and IV.02) Type II Frequency (Percent) Type III and IV Base Condition Type III Frequency (Percent) 253 (86. Similarly.36) 1080 (97.32) 219 (98. the base conditions included the presence of a left-turn lane on the major road and the presence of a median on the major road.84) Yes 1770 (97.00) Unlike Types I and II.00) 222 (100.32) Variable Right turn on major No Right turn on minor No Left turn on major Left turn on minor Median on major Terrain on major Total sites meeting all base conditions 1 911 (82.67) 148 (66. 155 . Group A Base Conditions The Group A nominal or base conditions and the number of sites meeting these conditions for model Types III to V are presented in tables 126 to 131. a VEI less than 1 degree per 30. therefore. Type III For total accidents. and intersection angle were selected as significant variables from the full models.92) 1106 (100. vertical curves on the major road.Table 125. For vertical curves. Group B Base Conditions for Type I to IV AADT Models Type I and II Base Condition Type I Frequency (Percent) 1563 (85.67) 34 (15.40) No 883 (79.91) No 1738 (96.28) 145 (65.32) 212 (73. The percentage of sites meeting all of the base conditions ranged from approximately 26 percent to 54 percent.09) Type IV Frequency (Percent) 164 (73.23) 1105 (99.60) 1069 (96.

0) Type IV For total accidents. values of 1 and 2 were taken as the base condition.(3) Table 128.68) 136 (100.4) 39 (28. Group A Base Conditions for Type III TOTACC AADT Models Variable Base Conditions Flat (≤1° per 100 ft) VEI1 0 COMDRWY1 -5° ~ 5° HAU Total sites meeting all base conditions Total sites Frequency (Percent) 88 (64.0) For injury accidents. Table 129 shows the base conditions for these variables.25 m (250 ft) of the intersection center were the other base conditions. report.47) 136 (100. report. Table 128 shows the base conditions for these variables. not enough sites with a single posted 156 .Table 126.7) 83 (61. The same rule was applied to judge adequate sight distance as the rule described on page 48 of the Harwood et al. Adequate sight distance.0) 89 (65. Table 127. right sight distance from minor road. “right” sight distance from minor road and median type on major road were selected as significant variables from the full models. As for total accidents.0) 89 (65. the same rule was applied to estimate adequate sight distance as was described on page 48 of the Harwood et al.03) 124 (100. Adequate sight distance and no median are the base conditions.0) 70 (56.(3) For posted speed limit. Table 127 shows base conditions for these variables. and posted speed limit on minor road were selected as significant variables from the full models.4) 36 (26. Group A Base Conditions for Type IV TOTACC AADT Models Base Conditions SDR2 Adequate right sight distance MEDTYPE No median type Total sites meeting all base conditions Total sites Variable Frequency (Percent) 116 (96. no median type.0) For injury accidents. median type on major roads.4) 74 (54. commercial driveways on the major road. and speed limit between 48 and 56 kilometers per hour (km/h) (30 and 35 miles per hour (mi/h)) are the base conditions. and intersection angle were selected as significant variables from the full models.5) 67 (54. Group A Base Conditions for Type III INJACC AADT Models Variable Base Conditions 0 COMDRWY1 -5° ~ 5° HAU 1 or 2 HAZRAT1 Total sites meeting all base conditions Total sites Frequency (Percent) 83 (61. For hazard rating. hazard rating on the major road. An intersection angle between plus and minus 5 degrees and no commercial driveways within 76.

Presence of lighting and speed range between 72 and 89 km/h (45 and 55 mi/h) were considered as the base conditions. Group A Base Conditions for Type V TOTACC AADT Models Base Conditions HEICOM ≤ 3.0) 49 (49.0) 70 (56. posted speed limit on major road and presence of lighting at intersection were selected as significant variables from the full models.0) 66 (66. Table 130.00) 100 (100. Table 131. Table 130 shows the base conditions for these variables. For HEICOM. For posted speed limit.3.0) 39 (39.5) 84 (67. and posted speed limit on major roads were estimated as significant in the full models and used for base conditions. Table 129. the data for and models calibrated using all available sites are referred to in the table headings as main models. median type on major roads.speed limit could represent a base condition.0) 87 (87. Table 131 shows the base conditions for these variables.65) 124 (100.82 (radius ≥ 1500 ft) MEDTYPE No median type SPD1 72–88 km/h (45 ~ 55 mi/h) Total sites meeting all base conditions Total sites Variable Frequency (Percent) 80 (80. radium larger than 458 m (1500 ft) was considered as the base condition.0) Type V Total accidents HEICOM for major and minor roads.7) 38 (30.0) 3. Group A Base Conditions for Type IV INJACC AADT Models Base Conditions SDR2 Adequate right sight distance MEDTYPE1 No median type SPD2 30 ~ 35 mi/h Total sites meeting all base conditions Total sites Variable Frequency (Percent) 116 (96.0) For injury accidents.0) 71 (71.2 Model Results The AADT-only modeling results are discussed next. 157 . a range with the highest frequency (48 to 56 km/h (30 to 35 mi/h)) was considered as the base condition.00) 100 (100. the speed range between 72–88 km/h (45–55 mi/h) was used as the base condition since this range had the highest frequency. In the tables below. Group A Base Conditions for Type V INJACC AADT Models Variable Base Conditions SPD1 72–88 km/h (45 ~ 55 mi/h) LIGHT Yes Total sites meeting all base conditions Total sites Frequency (Percent) 66 (66. Therefore.

Table 132. The results indicated that the recalibrated 158 . with the additional years of accident data. appear to be of suspect quality. sometimes as low as one vehicle per day. but AADTonly base models subsequently were derived by Harwood et al. The total number of sites and the number of Group B base condition sites are given in table 132. Number of Sites Used for Type I Main and Group B Base AADT Models Dataset Minnesota 1985–98 Georgia 1996–97 California 1991–98 All Sites 270 116 1432 Group B Base Condition Sites 133 107 973 Percent (Base/All Sites) 49. the overdispersion parameter.13 Group B Base Condition Sites Mean 0. With a large number of available sites.Type I Models The datasets used to develop Type I AADT-only models included the Minnesota sites from the original study.1250 3800 201 0 0 100 6.2 67.1250 4475 270 Minimum Maximum 0 0 6. intersections with a major road AADT below 400 or a minor road AADT below 100 were removed from the data. have been estimated for each location using a specially written maximum likelihood program.9 Table 133. Summary statistics on these datasets are presented in table 133.75 4.3750 0. The CURE method described earlier was used to suggest any alternate model forms that could provide an improved fit to the data.(1.3) A comparison between this model applied to the recalibration data and the recalibrated model for TOTACC is given. Separate values of K. A similar model for INJACC from the original calibration is not available for comparison.2368 4885 374 Median Minimum Maximum 0. A unique constant term was estimated for each location to account for the differences in accident reporting and other characteristics between jurisdictions.5326 0. by setting default values of two for HAZRAT1 and no right-turn lane on the major road in a model for predicting total intersection related accidents.3750 0. Summary Statistics for Type I Sites: Main and Group B Base AADT Models All Sites Variable TOTACC/yr INJACC/yr AADT1 AADT2 Mean 0.3 92. Vogt and Bared did not report AADT-only models for Type I intersections.00 4. These low volume counts. the Georgia sites.13 6480 401 35750 100 10001 401 20190 Tables 134 and 135 report the parameter estimates for the Type I TOTACC and INJACC models calibrated using all sites as well as for those sites meeting the Group B base conditions. The comparison indicates the recalibrated model for total accidents is improved as measured by the GOF using the K value. and sites extracted from the California HSIS database.2660 6011 492 Median 0.6074 0. Sites with very low volume typically are not representative of sites in the general population.

001) 0.394 1 The base model published in Harwood et al.584 0. For the TOTACC model.model using the original exponential model form adequately fit the data and that the various adjustments to this model form did not improve the AADT models.900 (0.e.246.206 (0.148.9 (0. Parameter Estimates for Recalibrated TOTACC Type I: Main and Group B AADT Models All Sites Estimate (s. <0.6180 (0. <0. <0.0215.368 0.7900 (0.0680.0310. the base conditions model was estimated with a lower overall K than the model using all sites.4900 (0..001) 0.001) N/A2 0.001) -8.540 0.658 MPB/yr -0. Table 134.280 MAD/yr 0.303. This would be expected.7900 (0. <0. <0. and MAD per year were negligibly different.400 0.600 -0.140 (0. <0. (p.001) -7.055 (0.229 (0. which seems to be a reasonable expectation on the basis of other models reported in the literature.315.004 0.001) -7.e.0249.161 0.4900 (0.001) 0. <0.192 0.367 159 . Appendix D shows the CURE plot for the Type I AADT TOTACC model.370 0. <0.. For the INJACC model.0630.256 0. <0. MPB per year. <0.232.668 0. the Pearson product-moment correlation coefficient.030 (0.294. <0.001) 0.293 0. Therefore it was decided to retain the original model form. <0.001) Group B Base Condition Sites Estimate (s.540 Pearson productmoment correlation coefficient 0.417 0.3872 (0. the overall K was the same for the two AADT models.001) 0.476 Original AADT Model1 Applied to All Sites -10. <0.132 (0.225.0296.6445 (0.3540 (0. <0.0630. 2000.001) -8.321 Original AADT Model1 Applied to Base Condition Sites -10.001) 0. because the base condition sites should be more homogeneous in their design characteristics. Specifically. 21) 2 N/A: not available N/A2 0.001) 0.148.001 0. <0.001) Variable Minnesota constant Georgia constant California constant Log of AADT1 Log of AADT2 K–all locations K–Minnesota K–Georgia K–California 0.001) 0.0680.001) 0. p-value) -8. Tables 134 and 135 indicate that the β coefficient of the log of major road AADT is about two times that for minor road AADT.244 0.. <0. p-value) -8.

0306.834 (0.283.001) -8.376. <0. <0.e. The total number of sites and the number of Group B base condition sites are given in table 136. the Georgia sites. <0.303 0.Table 135. <0.942 (0. <0.275. <0. Summary statistics on these datasets are presented in table 137. Number of Sites Used for Type II Main and Group B Base AADT Models Dataset Minnesota 1985–98 Georgia 1996–97 California 1991–98 All Sites 250 108 748 Group B Base Condition Sites 120 100 498 Percent (Base/All Sites) 48.001) 0.8524 (0. and sites extracted from the California HSIS database.001) 0.0001 Group B Base Condition Sites Estimate (s.001) 0.6 66.227 0. <0.304.405.435 0.500 0.263 0. the large number of sites available allowed for the removal of those sites with a major road AADT less than 400 or a minor road AADT less than 100.001) 0.0364. <0.0253.182 No previously calibrated AADT-only model for injury accidents exists for comparison for Type I sites Type II Models The datasets used to develop the Type II main and Group B base models included the Minnesota sites from the original study.778 (0..435 0.0 92.0001 Minnesota constant Georgia constant California constant Log of AADT1 Log of AADT2 K–all locations K–Minnesota K–Georgia K–California Pearson product-moment correlation coefficient MPB/yr 1 MAD/yr 0.586 -0. <0.195 0. with the additional years of accident data. the rationale again being that the omitted sites either have errors in AADTs or are unrepresentative of those in the general population..476 0.001) -8.294 0. p-value) -8. Table 136. As for Type I.788 (0.e.6 160 .001) 0. Parameter Estimates for Recalibrated INJACC Type I: Main and Group B AADT Models1 Variable All Sites Estimate (s.035 (0.001) -7.3551 (0.3382 (0.0397.001) -8.001) 0.520 0.6426 (0.6159 (0.388. <0. p-value) -8.

Table 139 indicates that the β coefficient of the log of major road AADT is about 10 to 20 percent higher than that for minor road AADT. A unique constant term was estimated for each location to account for the differences in accident reporting and other characteristics between jurisdictions. which is consistent with previous modeling efforts. this would be expected.4665 0. Table 138 indicates that the β coefficient of the log of major road AADT is about 35 percent higher than that for minor road AADT value.5357 0 7.50 AADT1 5487 4245 407 38126 4855 3907 407 28860 AADT2 532 344 100 7460 448 310 100 5801 Tables 138 and 139 report the parameter estimates for the Type II TOTACC and INJACC models.(3) A comparison between this model applied to the recalibration data and the recalibrated model for TOTACC is given. The CURE method was again used to suggest any alternate model forms that could provide an improved fit the data.75 0. Vogt and Bared did not report AADT-only models for Type II intersections. there was a negligible difference in Pearson product-moment correlation coefficient. and the various adjustments to this model form did not improve the AADT models. Therefore it was decided to retain the original model form. Appendix D shows the CURE plot for the Type II AADT TOTACC model. A similar model for INJACC for the original calibration is not available for comparison. the model does a poorer job for the Georgia and California sites.07 INJACC/yr 0. Summary Statistics for Type II Sites: Main and Group B Base AADT Models All Sites Variable Group B Base Condition Sites Mean Median Minimum Maximum Mean Median Minimum Maximum TOTACC/yr 0. The comparison indicates the recalibrated model for total accidents has an inferior GOF compared to the Vogt and Bared models using the K value for all States. Again. For both the TOTACC and INJACC models. and MAD per year. MPB per year. In particular. Separate values of K.4249 0. because the base condition sites should be more homogeneous in their design characteristics.5000 0 7.(1) However.8440 0. both combined and individually.2500 0 3. Specifically. which were calibrated using all sites and those meeting the base conditions.13 0.2500 0 4.Table 137. AADT-only base models were subsequently derived by setting default values of “no intersection skew angle” and “no driveways within 76. The results indicated that the recalibrated model using the original exponential model form adequately fit the data. 161 . have been estimated for each location using the specially developed maximum likelihood program.9227 0.25 m (250 ft) of the intersection” on the major road for a model for predicting total intersection related accidents. the overdispersion parameter. the base condition models were estimated with a lower overall K than the model using all sites.

001) -8.435 0.0267.7128 (0. p-value) -8.4532 (0.686 -0.400 0.733 0. <0.415.0361.385 0.e. <0.301.494 (0.001) 0.e.001) -8.669 0. <0.305 0.001) -9..001) 0.299 (0. <0.001) 0. <0.424. <0.370 0.001) 0.0690.240 0.720. <0.001) 0.312. <0. <0.001) 0.556 0.335 0.407.294.720. <0.001) 0.0423.747 -0.001) N/A1 0. Parameter Estimates for Recalibrated TOTACC Type II: Main and Group B AADT Models All Sites Estimate (s.6723 (0.6010 (0. <0.0780.4873 (0. p-value) -9.6010 (0. <0.747 (0.256 0.542 The base model published in Harwood et al. 2000 (p.240 0.001) 0.001) 0. <0.6100 (0.001) Variable Minnesota constant Georgia constant California constant Log of AADT1 Log of AADT2 K–all locations K–Minnesota K–Georgia K–California Pearson productmoment correlation coefficient MPB/yr N/A1 0.514 1 MAD/yr 0.480 Original AADT Model1 Applied to Base Condition Sites -9. <0.832 (0. <0.010 Original AADT Model1 Applied to All Sites -9.532 0.340 (0.970 (0. 22) 2 N/A: not available 162 ..001) Group B Base Condition Sites Estimate (s.001) -8.340 (0.0780.004 0.256 0.5445 (0.0331. <0..0690.6100 (0.588 0.Table 138.

<0. Summary statistics on these datasets are presented in table 141.161 0. p-value) -9.679 -0.001) 0. p-value) -9.5118 (0.001) -8.476 0.Table 139. <0. because as many as 87 out of the 294 available sites had an AADT on the minor road of less than 100. the researchers did not have the luxury of removing sites with a minor road AADT less than 100. <0.001) 0.385 0. <0. The total number of sites and the number of Group B base condition sites are given in table 140.0411.636 (0.217 0.435 0.492) -9.472. 163 . the Georgia sites. and those extracted from the California HSIS database.071 (0. Parameter Estimates for Recalibrated INJACC Type II: Main and Group B AADT Models 1 Variable Minnesota constant Georgia constant California constant Log of AADT1 Log of AADT2 K–all locations K–Minnesota K–Georgia K–California Pearson product-moment correlation coefficient MPB/yr 1 All Sites Estimate (s.417 0.707 (0. <0. <0.213 (0.6210 (0.5611 (0.302 No previously calibrated AADT-only model for injury accidents exists for comparison for type II sites Type III Models The dataset used to develop the main Type III model included the Michigan sites from the original study.6282 (0.0380.001) -9.0490.e.347.001) 0.266 (0. <0.303 0.279 MAD/yr 0. Unlike the case for Types I and II models.001 0.e.626 -0.001) 0.361.001) 0. Base models were calibrated using both the Group A and Group B datasets.339.001) -9.0301.5692 (0..005 Base Condition Sites Estimate (s. <0.481) 0.294 0.. with the additional years of accident data.

the differences in Pearson product-moment correlation coefficient.13 0. and 145.00 0. MPB per year.13 5. a unique constant term was not included for each State. Specifically.13 74500 9490 Mean 1. For reference. Number of Sites Used for Type III Main and Group B Base Condition AADT Models Dataset Michigan 1993–97 Georgia 1996–97 California 1991–98 All Sites 24 52 218 Group B Base Condition Sites 0 14 48 Percent (Base/All Sites) 0.45 17002 449 Median Minimum Maximum 0. Table 142 indicates that the β coefficient of the log of major road AADT is about two to three times that for minor road AADT. The comparisons indicate the recalibrated models for TOTACC and INJACC have a better GOF measures compared to the Vogt models.0 22.05 0.38 15433 325 Minimum Maximum 0 0 6500 10 6. comparisons between the Vogt AADT models applied to the recalibration data and the recalibrated models for TOTACC and INJACC are also given in tables 142. Therefore the original model form was retained.65.88 4.50 0. Summary Statistics for Type III Sites: Main and Group B Base AADT Models All Sites Variable TOTACC/yr INJACC/yr AADT1 AADT2 Mean 1. Pearson product-moment correlation coefficients for TOTACC and INJACC models were approximately 0. 143. Table 143 shows that β coefficient of the log of major road AADT is about three to four times that for minor road AADT.13 57731 2500 Tables 142 and 143 report parameter estimates for the main Type III TOTACC and INJACC models.63 0. The CURE plot method proposed by Hauer was also used to suggest any alternate model forms that could provide an improved fit for the data.0 Table 141.Table 140. because the State indicator variables were insignificant. Unlike models for Type I and II intersections.00 1902 1 15. 164 . and MAD per year were negligible.25 12909 206 0.57 18933 466 Group B Base Condition Sites Median 0.0 27. which is also consistent with previous efforts. Appendix D shows the CURE plot for the Type III AADT TOTACC model. calibrated using all sites and as well as those meeting the Group B base conditions. a reasonable expectation. The results indicated that the recalibrated models using the original exponential model form adequately fit the data and that the various adjustments to this model form did not improve the AADT models.

0001) 0. 1999.0001) 1.0.5162 0.e.3082.0931.0001) 1.2028 (0..1925 (0.03 MAD/yr 0.0317.8435.0.3074 0. p-value) -12.96 (0.5649 0. 0.0319. Parameter Estimates for Recalibrated INJACC Type III: Main and Group B Base AADT Models All Sites Estimate (s.3682.0.49 (0.0892..0.0366.0.0001) 1.0.66 0.0.20 0.67 0.2648 (0.5849 0.0001) 0.70 0.9357.3027 (0.08 0.0002) 0. 0.0613.3316 (0.0001) 0.0.0814.0000) 0.02 0.0000) 0.3027 (0. Parameter Estimates for Recalibrated TOTACC Type III: Main and Group B Base AADT Models All Sites Estimate (s.32 The AADT model published in Vogt.2629. 0. 1999.60 Original AADT Model1 Applied to Base Condition Sites -12.Table 142. 0.58 -0.0636.0388) 0.9980 (0.1925 (0.52 1. 0.6835 Constant Log of AADT1 Log of AADT2 K Pearson productmoment correlation coefficient MPB/yr 1 Variable Original AADT Model1 Applied to All Sites -12.e.0000) 1. p-value) -15.1762.e.68 -0.5256 0.1989 (0.0000) 0.0000) 0.02 Group B Base Condition Sites Estimate (s.2081.2028 (0. 0. 0. 0.01 MAD/yr 0.1685 (3. (p.2477.0.0000) 1.0007) 0. (p.3125 0.0000) 0.0000) 0.e..0000) 0.0001) 0.2817 (2.2477.0001) 0.9243 (2.0001) 1..59 Group B Base Condition Sites Estimate (s. 0.1685 (3.2720 (0.3682.71 The AADT model published in Vogt.0.6109.0007) 0.66 0. p-value) -12. p-value) -9.2544 (0.0892.0717. 165 .62 -0.5256 0.67 0. 111) Table 143.3206 (0.0. 0. 0.01 0.8238 (0.3082.113) Summary statistics on the Group A base condition datasets are presented in table 144.0.0000) 0.0931.5649 0.9243 (2.63 0.0590 Constant Log of AADT1 Log of AADT2 K Pearson productmoment correlation coefficient MPB/yr 1 Variable Original AADT Model1 Applied to All Sites -13.0319.32 Original AADT Model1 Applied to Base Condition Sites -13.0388) 0. 0.1332 (1.1989 (0.0941 (0.

5448.58 The TOTACC AADT model published in Vogt.01 0. (p.1989 (0.3280 (0.0421) 0.65 -0.65 and 0.76 12087 324 Median 0. 1999. which is reasonable.2874.0.55 0.35 Original INJACC AADT Model2 Applied to Group A Sites -13.0931. Again. p-value) -13.1326..5649 0. (p.3693 (0. 0.58 0.47 0.3027 (0. 1999.3924 (2.0.23 0.23 11610 359 Median 0.1925 (0.0086) N/A3 N/A3 0. Parameter Estimates for Type III Group A TOTACC and INJACC Base Models TOTACC Estimate (s.e. Pearson productmoment correlation coefficients of TOTACC and INJACC AADT models were 0. 1999 Pearson productmoment correlation coefficient MPB/yr 1 MAD/yr 0. 0. 111) 2 The INJACC AADT model published in Vogt.2194 (0.1685 (3.0.3082.0000) 0.0001) 0.e. the coefficient of the log of AADT2 was quite insignificant because of large standard error..0388) N/A3 0. Table 145. the Georgia sites.0054) N/A3 0.0001) 1.50 10977 180 Minimum Maximum 0 4.5256 0. The INJACC models differed from the TOTACC models in the use of LOG (AADT1 * AADT2) rather than the individual logs of the major and minor road AADTs. of Sites = 36 Mean 0.5227 0.3682.2028 (0.05 Original TOTACC AADT Model1 Applied to Group A Sites -12. 0. Summary Statistics for Type III Sites: Group A Base AADT Models Group A Base Condition Sites TOTACC No. of Sites = 39 Variable ACC/yr AADT1 AADT2 Mean 0. respectively.60 28000 1128 2360 33333 20 2500 Table 145 reports the parameter estimates for the Group A Type III TOTACC and INJACC base models.1614.20 Group A Base Condition Sites INJACC No.5256 INJACC Estimate (s. The β coefficient of the log of major road AADT is about three to four times that for minor road AADT.0.0001) 0.0.00 10217 292 Minimum 0 4794 15 Maximum 1. 0. p-value) -6.0.2477.0000) 1.5649 Variable Constant Log of AADT1 Log of AADT2 Log of (AADT1*AADT2) K K–Vogt.25 0.64 -0.Table 144.0892. 113) 3 N/A: not available Type IV Models The datasets used to develop the main Type IV model included the Michigan sites from the original study.0319.0001) 1.9243 (2. and those extracted from the California HSIS database.0007) N/A3 0.7691 0. 0. with the additional years of accident data. For INJACC.4337.00 0. a large number of sites (52 out of 166 .47.8087 (2.

Therefore. Pearson product-moment correlation coefficients 167 . Appendix D shows the CURE plot for the Type IV AADT TOTACC model.8 4.6 0.5 69521 7400 Group B Base Condition Sites Mean 1. the original model form was retrained. Summary statistics on these datasets are presented in table 147.0 2192 10 10.4 12950 420 Minimum Maximum 0. the β coefficient of the log of major road AADT for the INJACC model is also significantly larger than that for minor road AADT. the estimated overdispersion parameter for the original full models is also given. Summary Statistics for Type IV Sites: Main and Group B Base AADT Models All Sites Variable TOTACC/yr INJACC/yr AADT1 AADT2 Mean 1. so. For reference.4 3. Specifically.3 0. the Pearson productmoment correlation coefficient.0 1. The results indicated that the various adjustments to this model form did not improve the AADT models. calibrated using all sites and those sites meeting the Group B base conditions. Base models were calibrated using both the Group A and Group B datasets. as expected.0 0. The total number of sites and the number of Group B base condition sites are given in table 146. Table 148 indicates that. The CURE method was again used to suggest any alternate model forms that could provide an improved fit the data.3 43167 2625 Tables 148 and 149 report the parameter estimates for the main Type IV TOTACC and INJACC models.9 21. MPB per year.7 15477 552 Median 1. Unlike models for Type I and II intersections. Table 149 shows that. the β coefficient of the log of major road AADT is significantly larger than that for minor road AADT. again. Number of Sites Used for Type IV Main and Group B Base Condition AADT Models Dataset Michigan 1993–97 Georgia 1996–97 California 1991–98 All Sites 18 52 152 Group B Base Condition Sites 0 1 33 Percent (Base/All Sites) 0. and MAD per year were negligibly different. Table 146.7 Table 147.222) had a minor road AADT of less than 100. the researchers could not remove sites with a minor road AADT less than 100.7 18385 345 Median Minimum Maximum 0.4 13865 186 0 0 2367 11 5.0 0. unlike the case for Types I and II.6 0. a State indicator variable was insignificant.

Table 148.e.5082.2826 (0.4683 (0. 0. 0.0779.90 0. 116) 168 .6136 Log of AADT1 (0.78 -0. 0.0000) 0.0444) 0.89 for the base sites.0035) 0. 149.0444) 0.6677 0.0.27 1.4671 (0.10 0. and 0. p-value) -14. and 151.0001) 0.75 and 0.. 0.1398.0.0001) 0.9352 (2.0035) 0.4683 (0.73 -0. 1999. p-value) -8.75 0. (p. The comparisons indicate the recalibrated models for TOTACC and INJACC have better GOF measures compared to the Vogt models.2070 Original AADT Model1 Applied to Base Condition Sites -6.5135 (0.0360.0000) 1.63 using all sites.00 0.90 and 0.76 MAD/yr 0.0289 (0.23 0.0.5135 (0.0959.0000) 0.6144 Constant (0.2330.0000) 0.8553.0896.0896. Parameter Estimates for Recalibrated TOTACC Type IV: Main and Group B Base AADT Models All Sites Estimate (s.0000) 0.0.0.for TOTACC and INJACC models were 0..98 The AADT model published in Vogt.6144 Group B Base Condition Sites Estimate (s.3767.48 0.9469 (1.2330.9352 (2.0.3767.e. Comparisons between the Vogt AADT models applied to the recalibration data and the recalibrated models for TOTACC and INJACC are also given in tables 148.03 0. 0. 0.4359 Log of AADT2 K Pearson productmoment correlation coefficient MPB/yr 1 Variable Original AADT Model1 Applied to All Sites -6.0000) 0.

1128.0.0007) 0.0811.e.63 The AADT model published in the Vogt’s report. conforming to expectations.80 4210 25799 21 2990 Group A Base Condition Sites INJACC No.Table 149.0.01 0.5741 0. 118) Summary statistics on the Group A base condition datasets are presented in table 150. 0. Parameter Estimates for Recalibrated INJACC Type IV: Main and Group B Base AADT Models All Sites Estimate (s.39 (0. It appears that the sample size of 38.00 12601 592 Minimum Maximum 0 10.4535 (0.53 0. 1999.63 0.0000) 0.0053) 0.0.15 0.5741 0. 0. the INJACC model used LOG (AADT1 * AADT2) because log of AADT2 was quite insignificant. 0.7490 0.0053) 0.29 Original AADT Model1 Applied to Base Condition Sites -9.1486 0. Summary Statistics for Type IV Sites: Group A Base AADT Models Group A Base Condition Sites TOTACC No. of Sites = 67 Variable ACC/year AADT1 AADT2 Mean 2. Table 150. with a relatively low number of injury accidents.0000) 1.86 -0.. (p.0445. even with LOG (AADT1 * AADT2).1401.9947. 0.1688. However.82 13740 498 Minimum Maximum 0 4210 43 3.39 12444 764 Median 2. 0.2591. p-value) -8.5844 Log of AADT1 Log of AADT2 K Pearson productmoment correlation coefficient MPB/yr 1 Variable Constant Original AADT Model1 Applied to All Sites -9.4778 (0.3961 (0.0.2849 (0.89 -0. 169 .1858 (1.2591. This table indicates that the β coefficient of the log of major road AADT is about one and one half times that for minor road AADT.22 Group B Base Condition Sites Estimate (s..7442.5675.93 12417 717 Median 0.2513 (0. 0.0001) 0.0.7224 (0.8454 (2. was not sufficient to provide reasonable coefficients and p-values for the INJACC model.02 MAD/yr 0.33 25799 2990 Table 151 reports the parameter estimates for the Group A Type IV TOTACC and INJACC base models.1401.7224 (0.0.61 -0.4778 (0.8454 (2.0007) 0.0000) 0.e.5675. of Sites = 38 Mean 0.0001) 0.0000) 0.0000) 0. the p-value was not statistically significant. As was the case for Type III. p-value) -15.0000) 0.

9225.0.0001) 0.0001) 0.0187) N/A3 0.64 -0. were calibrated.2330.5741 0. 116) 2 The INJACC AADT model published in the Vogt’s report. 0.0.7224 (0. (p.0.63 The TOTACC AADT model published in the Vogt’s report. Number of Sites Used for Type V Main AADT Model Dataset Michigan 1993–97 (only 1996–97 INJACC used) Georgia 1996–97 All Sites 18 51 170 . (p. the additional years of accident data.0283) 0.75 Variable Constant Log of AADT1 Log of AADT2 Log of (AADT1*AADT2) K Pearson productmoment correlation coefficient MPB/yr 1 MAD/yr 1.0444) 0. 118) 3 N/A: not available Type V Models The datasets used to develop the Type V main models included the Michigan sites from the original study.5675.0. so it was decided not to calibrate Group B base models that would have included these data.6144 0.62 0. the Georgia sites.0007) N/A3 0.0053) 0.5135 (0. for Michigan. Table 152. p-value) -3. 1999. 0.0.2677 (2.5947 (0.1401.e.Table 151.0093) 0.1685.10 0. and those extracted from the California HSIS database.03 0.2407 (0. Only five sites were available from the California HSIS database.1952) 0.63 -0.0896. 0.5675.16 INJACC Estimate (s.72 -0.4778 (0.04 Original TOTACC AADT Model1 Applied to Group A Sites -6..4683 (0.6768 0. Parameter Estimates for Type IV Group A TOTACC and INJACC Base Models TOTACC Estimate (s. Summary statistics are presented in table 153. Only Group A base models.. because injury data for the original years were not available.5771 0.1910) N/A3 N/A3 0.63 1. 0.8454 (2.73 Original INJACC AADT Model2 Applied to Group A Sites -9.1858. using the Georgia data and the original sites for Michigan and California. The numbers of sites comprising the dataset used for the main model are given in table152.0001) N/A3 0.2591. 0.e. 1999. As seen.9352 (2. the INJACC model used only the additional years of accident data.3964 (0.8217 (2.7953.0.2712. p-value) -7.

0. 1999 All Sites Estimate (s.1970 (0. the differences in Pearson product-moment correlation coefficient.5776 0. Parameter Estimates for Recalibrated TOTACC Type V: Main AADT Model Variable Constant Log of AADT1 Log of AADT2 K Pearson product-moment correlation coefficient MPB/yr MAD/yr * There are no Type V AADT models for TOTACC in the Vogt report.0 430 420 Maximum 26.0908. 0. The results indicated that the recalibrated models using the original exponential model form adequately fit the data and that the various adjustments to this model form did not improve the AADT models.e. Unlike models for Type I and II intersections. MPB per year. the Vogt report does not provide AADT-only models for Type V. p-value) -4.66 and 0.Table 153. Therefore. The CURE plot method proposed by Hauer was used to explore alternate model forms that could provide an improved fit the data.01 2. Summary Statistics for Type V Sites: Main AADT Model All Sites Variable TOTACC/yr INJACC/yr AADT1 AADT2 Mean 4.2267.0368) 0.95 171 .2285.0 26000 10600 Tables 154 and 155 report the parameter estimates for the main Type V TOTACC and INJACC models calibrated using all sites.2 9643 2995 Median 4. Unlike the case for Types III and IV. Appendix D shows the CURE plot for the Type V AADT TOTACC model. a State indicator variable was insignificant. Table 154. in accord with expectations. Therefore.0 0.0300) 0.5 12. Specifically.1 1. a comparison between the Vogt models and the newly calibrated AADT models for TOTACC and INJACC could not be done. Both tables show that the β coefficient of the log of major road AADT is.2638 (2.9 2.44. Pearson product-moment correlation coefficients for TOTACC and INJACC were 0.4771 (0.66 -0. respectively. about two to three times that for minor road AADT. and MAD per year were negligible. the original model form was retained.5 9000 2700 Minimum 0..0555) 0. 0.

0292) 0.02 1.Table 155.57.50 1700 430 Maximum 26. 0. 0.94 10059 3772 Group A Base Condition Sites INJACC No. 0.00 8700 3459 0. Table 156. Pearson product-moment correlation coefficients of TOTACC and INJACC AADT models were 0..8599 (0. As was the case for Type III and IV models.00 8750 3452 Minimum 0 700 430 Maximum 5.7110 (2.23 9238 3890 Median Minimum 6.0000) 0. 1999 All Sites Estimate (s. 172 . respectively.64 Summary statistics on the Group A base condition datasets are presented in table 156.3372 (0.e.44 0. the INJACC model used LOG (AADT1 * AADT2) because the individual logs for separate AADT terms were insignificant. p-value) -9. This table indicates that the β coefficient of the log of major road AADT is about two to three times greater than that for minor road AADT.50 25132 10280 Table 157 reports the parameter estimates for the Group A Type V TOTACC and INJACC base models.3118. which is reasonable. Parameter Estimates for Recalibrated INJACC Type V: Main AADT Model Variable Constant Log of AADT1 Log of AADT2 K Pearson product-moment correlation coefficient MPB/yr MAD/yr * There are no Type V AADT models for INJACC in the Vogt report.70 and 0.0001) 0.7267 0. Summary Statistics for Type V Sites: Group A Base AADT Models Group A Base Condition Sites TOTACC No. of Sites = 49 Variable ACC/yr AADT1 AADT2 Mean 7.2210. of Sites = 39 Median 2.1547.50 20067 10280 Mean 1.

08 Constant Log of AADT1 Log of AADT2 Log of (AADT1*AADT2) K Pearson product-moment correlation coefficient MPB/yr MAD/yr * No AADT models previously calibrated 1 N/A: not available 3.1815 (0.2891 0. Summary of Models Recalibrated and Data Used Model Site Types and Data Used Description Types I and II Type III Type IV Group A Sites: Full Model with California project California project variables in Minnesota Michigan Michigan project data Georgia Georgia Georgia Group B Sites: Full Model with California HSIS variables from Minnesota California HSIS Georgia Not calibrated Not calibrated * The California project data for Types II.e.08 1.57 -0.Table 157.e.1063. Unlike the AADT models. models were developed using two different datasets.0759) 0. and V used only one dataset.25 m (250 ft) of the intersection were developed.. these models include many variables with the intent to explain as much of variation in crash occurrence as possible.05 3. 0. 0. p-value) -4. and V were not used Type V California project Michigan Georgia Not calibrated As before.4199 (2. Table 158 summarizes what data was used for each model. 173 .0182) 0.7087. given the available set of potential explanatory variables.2396 (0.1966.2429 0.70 -0.2752.0109) 0. IV.. Parameter Estimates for Type V Group A TOTACC and INJACC Base Models Variable TOTACC Estimate (s.0878) N/A1 0.4 FULLY PARAMETERIZED STATISTICAL MODEL ESTIMATION RESULTS This section discusses the development of the fully parameterized statistical models. For Types I and II. IV. p-value) -3. 0.5005 (0.10 INJACC Estimate (s. Table 158.1328) N/A1 N/A1 0.0357 (1. 0.1350. models for total accidents (TOTACC) and injury (fatal + nonfatal injury) accidents (INJACC) within 76. Types III. 0.

The Group B data set contained an additional 1432 sites from the California HSIS database. Group A.(1) Group A Tables 159 and 160 report the parameter estimates for the Type I TOTACC and INJACC models for the Group A data. was comprised of the sites from Minnesota and Georgia and consisted of many variables. including horizontal and vertical curvature.1 Type I Models Full models were developed using two groups of data. For reference the recalibrated models are compared to the models calibrated using data from Minnesota and recommended by Vogt and Bared. The California HSIS sites were not in this group because many of the variables were not available. Group B. included the California HSIS sites. 174 . The second.3. sites with a major road AADT below 400 or a minor road AADT below 100 were not used.4. As was the case for the AADT-only models. Two model variants are reported. the first. Summary statistics on these datasets by State are available in appendix C. As indicated earlier. but fewer variables were available for modeling. 270 sites from Minnesota and 116 sites from Georgia were used. The first includes a State indicator term and the second does not.

3211 (0.5037 (0.0470.005) -0.244.025) 0. 0.0460. <0.3004) 0.0327.001) 0.010 0.3785 (0.. <0. (p.0561) N/A2 0.416 0.825 (0.0045 (0.0708.001) 0.001) N/A2 Original Main Model1 Applied to Full Model Data Coeff.014) Variant 2 Coeff.0314 (0. p-value) -12.726 0.2901 (0.1398.0177) 0. p-value) -8.726 0.233 0. <0. 0.972 (0.0639.086 (0.0677. <0. 0.086) -0.Table 159.2671 (0.e.001) 0.2935.0032.481 N/A2 Variables Intercept for Minnesota Intercept for Georgia LOG of AADT1 LOG of AADT2 HI1 RT MAJ RT MIN LT MAJ HAZRAT1 VCI1 SPD1 HAU K–overall K–Minnesota K–Georgia Pearson product-moment correlation coefficients MPB/yr 1 N/A2 0.6529 (0.0362.400 0.0439. 0. <0.0263 (0.0841.0001) 0.0107.001) -7.1887 (0.001) 0. 0.0108) 0.001) 0. 0.407. 1998.992 (1.048 0. 0. <0.110.419 (0. Parameter Estimates for Recalibrated TOTACC Full Models: Type I Group A Variant 1 Coeff.0339 (0.742 -0. 0. <0. 115) 2 N/A: not available 175 .0285 (0. <0.001) 0.263 N/A2 0.0362.030 MAD/yr Vogt and Bared.157) 0.155 (0.3229) 0.0001) 0.438. 0.1578) 0. (s. 0.151.e. (s..7001 (0. <0.0113. 0.1726 (0.026 0.250 0. 0. (s..0416. 0.1433 (0.141 0.0001)) 0.398. p-value) -7.e.8052 (0.001) 0.1204 (0.

0337 (0. right-turn on major roads.035) Variant 2 Coeff. <0. 176 .690 0.3657.001) 0.0001) 0.1678 (0.4551 (0. 0. For total accidents. (p.120 (0. <0.0552.494 N/A2 Variables Intercept for Minnesota Intercept for Georgia LOG of AADT1 LOG of AADT2 HI1 RT MAJ LT MAJ HAZRAT1 VCI1 SPD1 DRWY1 HAU K–overall K–Minnesota K–Georgia Pearson product-moment correlation coefficients MPB/yr 1 N/A2 N/A2 0.3047) 0.003 0.001) 0.8122 (0.0286 (0. 0.258 (0. and the angle variable HAU were not included. 0.002 0.0460) N/A2 0. p-value) -8.0001) 0.0263) 0.0930.6092) 0. <0. <0.051) 0.0269. (s.0973.8671) 0.3620 (0. 0.0457.691 -0.022 0.129) Original Main Model1 Applied to Full Model Data Coeff.001) 0.212 0.0001) 0.660 0. 0.5618) -0. 0. 0. 116) 2 N/A: not available There were similarities and differences between the recalibrated models and the Vogt and Bared models.0846 (0. number of driveways on major roads.2870 (0. 0.0136.0557.0977.0596.537.523.566.0139. 0. while right-turn lane on minor roads and left-turn lanes on major roads were included. while left-turn lane on major roads were included. Parameter Estimates for Recalibrated INJACC Full Models: Type I Group A Variant 1 Coeff.6809 (0.e.0335 (0.037 (1. posted speed on major roads.0120 (0. (s.2065 (0.010 MAD/yr Vogt and Bared. 0.791.934 (0. (s.0045.3251 (0. p-value) -9.0569. <0. 0.2594) 0..001) 0.1869 (0.727 (0.001 0.001) -8.001) 0.. 1998.0556. <0.0327.256 0.132. 0.Table 160. For injury accidents.256 0. 0.1814. <0. 0.0714.016) N/A2 -0..001) 0. <0.7174 (0.001) 0.0156 (0.0051 (0.256 0.e.e. posted speed on major roads and the angle variable HAU were not included in the recalibrated model.278 N/A2 0. p-value) -13.

including the State location variable. and sociodemographics. on average. weather. and HAZRAT1. RT MIN. The data summary in appendix C shows that the Georgia sites experience. The following analysis further supports this conclusion. HAZRAT1. including the State location term.000 -150. Although the cumulative residuals go outside the 95 percent plots for few sites at low volumes and AADTs above 15000. HI1. The cumulative residuals on occasion go outside the 95 percent confidence intervals for a random walk around 0. have lower overdispersion parameters.000 Major AADT Figure 21. and VCI1 were all estimated to be significant at the 10 percent level or better in one or both of the TOTACC and INJACC models.000 -100. 150. 177 . The variables RT MAJ. more accidents than sites in Minnesota and also are higher in values for HI1.When a State indicator variable was used in the models in addition to the logs of AADTs. only the variable HI1 proved to be significant. Without the State indicator. The question arises then.000 100. indicating that the model is performing poorly. more geometric variables were statistically significant. However. the overdispersion parameter K is higher for the models without the State indicator variable and with more variables. LT MAJ.000 50. VCI1. the results are a slight improvement over the model without the State location variable in that the oscillations tend to be closer to the x-axis.000 Cumulative Residuals Adj Cum Residual 0. Because the models. it can be concluded that this model should be used. while being less likely to have a turning lane on the major road. CURE Plot for Alternate TOTACC Type I Group A Model Figure 22 plots the cumulative residuals (Y-axis) versus major AADT (X-axis) for the TOTACC model. whether the significantly higher accident risk in Georgia is due to differences in these geometric features or other fundamental reasons such as reporting levels. Figure 21 plots the cumulative residuals (Y-axis) versus major AADT (X-axis) for the TOTACC model without the State location variable.000 0 5000 10000 15000 20000 25000 Standard Deviation Standard Deviation -50.

150 Cumulative Residuals 100 50 Adj Cum Residual 0 0 5000 10000 15000 20000 25000 Standard Deviation Standard Deviation -50 -100 -150 Major AADT Figure 22. CURE Plot for TOTACC Type I Group A Model The GOF as measured by K was improved over the Vogt and Bared models. 178 . although the measures of MPB per year and MAD per year were better for the original model with the exception of the MPB per year for the TOTACC and INJACC Variant 1 models. Group B Tables 161 and 162 report the parameter estimates for the Type I TOTACC and INJACC models for the Group B data.

0561) N/A3 0.6112 (0. <0. 0.024) -0.2671 (0..3010 (0.4128 (0.118.669 MPB/yr -0.265.0001)) 0.0574. (s. 0. 0. <0.001) -8. <.001) 0.0339 (0.2042 (0.657 0.3229) 0.0032.0800 (0.001) 0.0264.e.035) 0.001) 0.2935.0045 (0.0224.8052 (0.992 (1.379 N/A3 N/A3 N/A3 179 .001) Variant 2 Coeff.413 (0.189 K–California 0.0509.0001) 0.0108) 0.001) -7.641 (0.455 Pearson product-moment correlation coefficients 0. 0.0630.251.001) N/A3 -0. p-value) -8.2838 (0. <0.250 K–Georgia 0.0677. (p.0285 (0. 115) 2 These variables area not available in the Group B data 3 N/A: not available 0.400 K–Minnesota 0. 0.009) -0.0639.001) N/A3 -0.0222.603 (0. p-value) -12.0001) 0.1140. <0.0177) 0.e.164) 0. 0.1326 (0.003 MAD/yr 0. p-value) -8..250. 0.1726 (0.001) 0.4607 (0.0327.6557 (0.001) Original Main Model1 Coeff.5037 (0.0263. <0. 0.0500..265 (0. 0. 0.Table 161.1578) 0.3004) 0.021 0. <0.244.0708. 0.151.435 (0. <0. 1998. (s. <0. <. Parameter Estimates for Recalibrated TOTACC Full Models: Type I Group B Variant 1 Coeff.481 N/A3 Variables Intercept for Minnesota Intercept for Georgia Intercept for California LOG of AADT1 LOG of AADT2 HI12 RT MAJ RT MIN LT MAJ HAZRAT12 VCI12 SPD12 N/A3 N/A3 HAU2 K–overall 0.367 1 Vogt and Bared. (s.435 N/A3 0.1398. 0.e.2901 (0.

116) 2 These variables area not available in the Group B data 3 N/A: not available The GOF as measured by K was improved over the Vogt and Bared models. <0.3820 (0. (s. 0.0596.1330. In the recalibrated models. (s.0973.0004 0.0051 (0.001) -9.4551 (0.009 0.. right-turn lanes on minor roads and left-turn lanes on major roads are significant in addition to right-turn lanes on major roads.077) 0.6590 (0.2253 (0.1346 (0.1814. 0.022) -0. <0.001) -8.3657. <0.300.0260.001)) 0. 0.8671) 0. p-value) -13.001) 0..248 (0.0261. <0.303.329.0001) 0.0930.0320. 180 ..037 (1. 0.345 0.1340.001) 0.0325.2065 (0.309.593 -0. <0.1869 (0.e.3000 (0.0045.3620 (0. (s.182 (0. 0.417 0. <0.588 0.e. <0. 0.0120 (0.435 N/A3 0.0597.001) N/A3 -0. <0.0977.256 0.1360 (0.182 (0. 1998.791. 0. Parameter Estimates for Recalibrated INJACC Full Models: Type I Group B Variant 1 Coeff. <0.0677. 0. 0. p-value) -9.494 N/A3 Variables Intercept for Minnesota Intercept for Georgia Intercept for California LOG of AADT1 LOG of AADT2 HI12 RT MAJ RT MIN LT MAJ HAZRAT12 VCI12 SPD12 DRWY12 HAU2 K – overall K – Minnesota K – Georgia K – California Pearson product-moment correlation coefficients MPB/yr 1 N/A3 N/A3 0.001) Variant 2 Coeff.3060 (0. (p.194 Vogt and Bared.2655 (0.0001) 0.e. <0.0335 (0.0269. 0.Table 162.001) 0. 0.001) N/A3 -0.0460) N/A3 0.0156 (0.0760.0001) 0.195 N/A3 N/A3 N/A3 MAD/yr 0.6402 (0. p-value) -9.4069 (0.2594) 0.001) Original Main Model1 Coeff.6092) 0.476 0.5618) -0.463 (0.0263) 0.025) -0.3047) 0.045) 0.0327.0714. 0.8122 (0. 0. 0.

The Group B data contained an additional 748 sites from the California HSIS database. sites with a major road AADT below 400 or a minor road AADT below 100 were not included. The California HSIS sites were not included in this group because many of the variables were not available. Group A. Summary statistics on these datasets are available in chapter 3. full models were developed using two groups of data. Two variants of models were calibrated. As for Type I. Group B.3.2 Type II Models As for Type I. As was the case for the AADT-only models. included the California HSIS sites. such as horizontal and vertical curvature. 181 .(1) Group A Tables 163 and 164 report the parameter estimates for the Type II TOTACC and INJACC models for the Group A data. The second. 250 sites from Minnesota and 108 sites from Georgia were used. and listed by State in appendix C. but fewer variables were available for modeling.4. the recalibrated models are compared to the models calibrated by Vogt and Bared. The first. For reference.2. the first variant included a State indicator term and the second did not. consisted only of the sites from Minnesota and Georgia and included many variables.

7079 (0.0033. (s.0529. 0.001) 0.2875) 0.e.1235 (0.001) 0.001) -8.e.475.278 N/A2 0. 0.0531.053 0. 0.0694.1735 (0. <0.630 (0..080 (0.256 0.001) Variant 2 Coeff.167) N/A2 0. <.001) N/A2 N/A2 0. p-value) -9.2576.111 Variables Intercept for Minnesota Intercept for Georgia LOG of AADT1 LOG of AADT2 RTMAJ DRWY1 HI1 SPD1 VCI1 HAU K–overall K–Minnesota K–Georgia Pearson product-moment correlation coefficients MPB/yr MAD/yr 1 Vogt and Bared.488.0187 (0.206 N/A2 0. 0. (s.0529.440 Original Main Model1 Applied to Full Model Data Coeff.1341) 0. p-value) -9.001) 0.0832.1375 (0.227 0.0176.777 0.317.0766 (0.0554.4774 (0.426 (1.0001) 0.777 -0.0519. p-value) -10.2885 (0..6990 (0.6091 (0. <..6026 (0.3431) 0.033 0.0049 (0.003 0.0836. 1998.0001) 0.001) 0.0173) 0. <0.001) 0.499.e.0277. 0.1219 (0. (s. <0.037) 0. 0.0001) N/A2 0.0282. (p.001) N/A2 0. <.0473.248 (0.0537.500 0. 115) 2 N/A: not available 0. <0.431 182 . 0. 0. <0.Table 163. 0. Parameter Estimates for Recalibrated TOTACC Full Models: Type II Group A Variant 1 Coeff.753 0.5153 (0. 0.0449 (0.2628) -0. <.

005) 0.254 -0. right-turn lanes on major roads. 0.557. (s.021) Variables Intercept for Minnesota Intercept for Georgia LOG of AADT1 LOG of AADT2 DRWY1 HI1 SPD1 VCI1 HAZRAT1 RT MAJ HAU K–overall K–Minnesota K–Georgia Pearson product-moment correlation coefficients MPB/yr MAD/yr 1 Vogt and Bared.0635.6229 (0.0944 (0.0001) 0. The Vogt and Bared model also included the angle variable HAU. <0.0639.0638. and the number of driveways for the variant that included a State indicator variable. 0. p-value) -9.001 0.001) -9.6196 (0. The GOF as measured by K was not as good as for the Vogt and Bared model. 0.5893 (0.0044.2789 (0. 0.4623.263 0..565.0626 (0.154 0. and number of driveways and the vertical curvature variable VCI1 for the variant without the State indicator variable. 0. and the horizontal curvature variable HI1.181 N/A2 0.724 0.5464) -0.570.0614. 0. <.783 (1.7865) -0.25 m (250 ft) of the intersection center were significant at the 10 percent level or better for both the State indicator and non-State indicator variants. 0.001) 0.0612.0857 (0. significant variables at approximately the 85 percent level or better for TOTACC included major and minor road AADTs.e.0729 (0.1055.0001) 0.e. 1998.0619.6392 (0. (p.1799) 0. 0.033 0. The Vogt model included roadside hazard rating as a significant variable and others that were not significant. 0.001) 0.125) Variant 2 Coeff.e.0870. For the recalibrated models.0043 (0.063 N/A2 N/A2 0. <0.6339 (0.0266.249 There were some similarities and differences between the recalibrated models and the Vogt and Bared models. 0. <.278 N/A2 0. p-value) -10. For the INJACC models. Parameter Estimates for Recalibrated INJACC Full Models: Type II Group A Variant 1 Coeff. the major road posted speed.001) 0. p-value) -10. 117) 2 N/A: not available Original Main Model1 Applied to Full Model Data Coeff.6567) 0.3258) 0..0824 (0.800 (0.250 0. number of driveways and horizontal curvature within 76. 0. 0.304 (0. (s.0451 (0.0408 (0.747 0. although the last two were of low significance. <.0889) 0. 183 . <0. (s.Table 164.2513) 0.6584 (0.1225 (0.303 (0.001) 0.033 0. 0.0337.0251. <0.0001) 0.727 0..001) 0.001) 0.766.1665.0720.0272.012) 0. 0.0112 (0.0328.

<0.0001) 0.015) Original Main Model1 Coeff. (s.0686.296. <0. 0.e. 0. 0.. <0. Table 165.6026 (0. 0.316.0274.4905 (0.0176.6091 (0.Group B Tables 165 and 166 report the parameter estimates for the Type II TOTACC and INJACC models using the Group B data.321. 0.718 (0. p-value) -8.0033. 0.001) -8.648 (0.1070 (0. 115) 2 These variables area not available in the Group B data 3 N/A: not available 0.091) -0.0633.6674 (0.275 (0. <.0173) 0.033) Variant 2 Coeff.3431) 0.400 N/A3 0.2875) -0. 1998.0331.001) -0.. p-value) -8.0473.2885 (0.e.0519.0694. 0.0187 (0. 0.2576. <..0633.799 (0.2628) 0.0449 (0.0049 (0.0836.5200 (0.1462 (0. (s..435 Pearson product-moment correlation coefficients 0.426 (1. <0.008 MAD/yr 0.1235 (0.304. 0.206 N/A3 Variables Intercept for Minnesota Intercept for Georgia Intercept for California LOG of AADT1 LOG of AADT2 RTMAJ LTMAJ DRWY12 HI12 VCI12 SPD12 N/A3 N/A3 HAU2 K–overall 0.001) 0. p-value) -10. <. Parameter Estimates for Recalibrated TOTACC Full Models: Type II Group B Variant 1 Coeff.6483 (0.001) 0.1341) 0.001) 0.317.528 N/A3 N/A3 N/A3 184 .e.001) -0.0001) 0.400 K–Minnesota 0.0268.1535 (0.001) -8.0327.695 0. <0.694 MPB/yr -0.009 0. 0.278 K–Georgia 0. 0.001) 0.528 1 Vogt and Bared. (s.0001) N/A3 0.555 K–California 0. (p.

6339 (0. significant variables in addition to major and minor road AADTs.1055. The GOF as measured by K was not as good as that for the Vogt and Bared model.0346. median and right-turn lane on major road were significant for both the State indicator and non-State indicator variants.130 (0.001) -0.5464) -0.143.0639.001) -8.6567) 0. <0.023) -0.285 (0.1799) 0.006) Original Main Model1 Coeff.2789 (0. 0.0001) N/A3 0.0729 (0. <.4623.645 MPB/yr -0. p-value) -9.5182 (0. (p.0001 N/A3 N/A3 N/A3 Variables Intercept for Minnesota Intercept for Georgia Intercept for California LOG of AADT1 LOG of AADT2 MEDIAN RT MAJ DRWY12 HI12 VCI12 HAZRAT12 HAU2 N/A3 N/A3 SPD12 K–overall 0.019) -0. 0.303 K–Georgia 0.001) 0.e. p-value) -10. For INJACC.0251.0857 (0. (s.0001) 0. <0.0889) -0.001) -9.001) 0.343.299 1 Vogt and Bared.6295 (0.3258) 0. 0.. (s.6006 (0.0299.0044.766.0727.400 K–Minnesota 0.. 0.6229 (0. 0.001) 0.334.204 K–California 0.0382.035) Variant 2 Coeff.298 (0.. <.0043 (0. p-value) -9.0870. Parameter Estimates for Recalibrated INJACC Full Models: Type II Group B Variant 1 Coeff. 0. include right-turn lane on major road for the variant with the State indicator variable and rightturn lane on major road and left-turn lane on major road for the variant without the State indicator variable. <.2014 (0. (s.352. <0.1646 (0.0451 (0.010 0.1225 (0. 117) 2 These variables area not available in the Group B data 3 N/A: not available 0.001) 0. 0.0635. 0.Table 166. 1998. 0.3320 (0.0001) 0. 0. 185 .181 N/A3 N/A1 0.0720.417 For TOTACC.2513) 0. 0. 0.455 Pearson product-moment correlation coefficients 0. <0.001) -0.e.642 0. 0.0299.1410.366.325 (0.7865) 0.5376 (0.0780.003 MAD/yr 0.e. <0.0112 (0.1665.729 (0. 0.783 (1.

The other model was the one judged to be next best on in terms of these measures. State indicator variables were statistically insignificant and therefore not included in the models. For TOTACC. MPB per year. commercial driveways on major roads. major and minor AADTs. Some of the California sites experienced changes in their design features during the 1996–98 period. the recalibrated main model had the additional terms COMDRWY1. accidents during the 1996–98 period were not included. peak turning percentage.3 Type III Models The data used to calibrate full models consisted of the California and Michigan sites from the original study. commercial driveways on major road. Tables 167 and 168 report the model results for TOTACC and INJACC. PKTRUCK. roadside hazard rating on major road. HAZRAT1. For INJACC. there were some differences between the recalibrated models and the Vogt models. VEI1. The main model was selected based on the highest Pearson product-moment correlation coefficient. The GOF as measured by Pearson product-moment correlation coefficients. and PKTURN compared to the Vogt model. For the INJACC. Summary statistics on the data are provided in section 3.2. major and minor AADTs. which was in the Vogt model. Two models each are reported for TOTACC and INJACC. and painted medians on major roads were found to be significant in the main models (15 percent level or better). MPB per year. For such sites. and the Georgia sites. State indicator variables were again statistically insignificant.4. and MEDTYPE1 compared to the Vogt model but did not include DRWY1. median width on major roads. vertical curve rate on major roads.3. As expected. and peak truck percentage were found to be significant in the main model at the 15 percent level or better. HAU. lowest overdispersion. For TOTACC. the GOF as measured by K was not as good as those for the Vogt models. For reference a comparison between Vogt’s full models applied to the recalibration data and the newly recalibrated models for TOTACC and INJACC also is given. intersection angle. However. and only 1993–95 accident data were used. additional years of accident data from the California and Michigan sites. 186 . intersection angle. and MAD per hear indicates that the recalibrated main models for TOTACC and INJACC provide better GOF than the Vogt models. the recalibrated main model had the additional terms COMDRWY1. and MAD per year.

0059.0681 (0.1014) -0.0001) 0.0.2624 (0.1064.2972 (0.0700) N/A3 0.0276.0.0060.0000) 0.70 -0.0088 (0.0546 (0.0239.e.0000) 0.0106 (0.0523.3209 (0.0024) Variables Intercept LOG of AADT1 LOG of AADT2 COMDRWY1 VEI1 HAU MEDWDTH1 MEDTYPE12 DRWY1 K Pearson product-moment correlation coefficients MPB/yr MAD/yr 0.90 N/A3 Original Main Model1 Applied to Full Model Data Coeff.4229 (0.1160.0249.0.0.9214 (1.2527.1665.0.3228 (0.0585.0154) 0.0. 111) 2 Median Type 1 (painted) on major roads 3 N/A: not available 1 187 .1081 (0.0000) 0.0.0.1914 (1.0912 (0.0001) 1.0.0866.0000) 0.0281.4552 (0.88 N/A3 Variant 1 Coeff. (s.0000) 0.0760) -0.0000) 0. Parameter Estimates for Recalibrated TOTACC Full Models: Type III Main Model Coeff.2196 (2..1479 (0.8509 (0.0.5100.0. 1999.0008) 0.8877 (0.0.0.0.0519) 0.0.3893 (0. (s. (p.0556.37 0.0.0590.1771.1109. (s.68 0.0101 (0.0000) 0.0285) N/A3 0...0054.0. p-value) -12. p-value) -9.0861) -0.Table 167.09 0.02 0.0010) 0.3575.70 0.1023) 0.0.1666.0.0.1044 (0.5232.0.0461) 0.84 The main model published in Vogt. p-value) -10.0391 (0.e.e.0001) 0.0.

For TOTACC. Summary statistics on the data are provided in section 3. and only 1993– 95 accident data were used.0605) 0.0592) N/A2 0.0. Two models were developed for both TOTACC and INJACC.0901.0.0001) 1.Table 168.1889 (0.0135.4.0001) 0.0001) 0.43 INJACC Variant 1 Model.5178 (0..0.0.1426.. peak left-turn percentages.0.0. and the Georgia sites.0797) 0.0053.0412) 0.0021) -0.2763.0063) 0.14 0. 0.0923. accidents during the 1996–98 period were ignored.56 -0.0003) 0.0131.0212) 0.47 MAD/yr 0. and right-side sight distance on minor roads were found to be significant in the main model at the 10 percent level or better.0003) 0.1357 (0. there is no main INJACC model in Vogt.0163 (0. (s.4 Type IV Models The data used to calibrate full models include the California and Michigan sites from the original study. with the additional years of accident data.1437.0.0.0.0186.0.0. MPB per year.0845.0.1897 (0. Parameter Estimates for Recalibrated INJACC Full Models: Type III Original Main Model1 Applied to Full Model Data Coeff.0001) 0.0407) 0.8498 (0.1872) N/A2 0.0949.4453 (2. p-value) -10. Again. Some of the California sites experienced changes in their design features during the 1996–98 period.2146. peak through percentages on minor roads.0.1436 (0. the main model was selected based on the highest Pearson product-moment correlation coefficient.0756) 0.0.64 -0.1029.0.e. p-value) -10. and MAD per year.0.66 -0.0019) -0.e.2.0000) 0.0346) 0. peak truck percentages.1792.0230 (0. major and minor road AADTs.0331 (0.0135.05 Variant 1 Coeff.0346.0627 (0.0762) N/A2 0.0.0. 1999 2 N/A: not available 3.0.0474.0790) Variables Intercept LOG of AADT1 LOG of AADT2 COMDRWY1 HAZRAT1 HAU PKTRUCK PKTURN PKLEFT K Pearson product-moment correlation coefficients MPB/yr 1 Main Model Coeff.2188 (0.0333 (0.0758) 0. (s.0253 (0.8076.0. For such sites.0930. State indicator variables were statistically insignificant in the main model and 188 . (s.0168 (0.6443 (2.05 0.3787 (0.2460 (0.e.0188.0353.0.0254 (0.0054.5102 (0. p-value) -12..3246 (2.0607 (0.0000) 0.8260 (0. and lowest overdispersion.0.2097.47 N/A2 0.

includes a Michigan indicator variable. and PKLEFT compared to the Vogt model. For TOTACC. which means there was more influence of the Michigan data on the model. 189 . the recalibrated main model had additional terms SDR2. the GOF as measured by K was not as good as those for the Vogt models. For reference. a comparison between Vogt’s full models applied to the recalibration data and the newly recalibrated models for TOTACC and INJACC is given. The GOF as measured by Pearson product-moment correlation coefficients. there were some differences between the recalibrated models and the Vogt’s models. MPB per year. and MAD per year indicates that the recalibrated models for TOTACC and the main model for INJACC provide better GOF than the Vogt models. Variant 1. However. However. the recalibrated main model included the additional term PKTRUCK compared to the Vogt model. respectively. PKTRUCK. PKTHRU2. but did not include PKLEFT1 and LTLN1S. As expected. For INJACC. peak truck percentages. For the INJACC. major and minor road AADTs. Tables 169 and 170 report the model results for TOTACC and INJACC. peak left-turn percentages on major roads. which were in the Vogt model. State indicator variables were statistically insignificant.were therefore not included. and speed limits on minor roads were selected as significant in the main model at the 10 percent level or better. the second best model.

(s.0.0.0.0000) 0.2311.0683) N/A4 0.0005 (0.20 N/A4 0. (s. 1999.2390 (0.8930.0229 (0.1849.0099) -0.0110.4631 (2.0034) 0.0001.0.2779.0479 (0..0.0975.0525) Variant 1 Coeff. p-value) -7.0001.0.0000) 0.0403) -0..0.0.0362) 0.0154 (0.e.4350 (1.4713 (1.0003 (0.0022) 0.0118.0018) N/A4 0.0001) 0.0000) 0. (s.48 1. p-value) -9.0003) 0.0.4841 (0.0.12 MAD/yr 1.31 N/A4 0.1255.5991.0.0158 (0. 116) 2 Median Type 1 (painted) on major roads 3 Indicator variable for Michigan 4 N/A: not available 190 .77 0.0.0249 (0.2084. Parameter Estimates for Recalibrated TOTACC Full Models: Type IV Original Main Model1 Applied to Full Model Data Coeff.6933.0000) 0.0591) N/A4 -0.4001 (0.28 1.0083.4027 (0.75 0.0.0.Table 169.0087) Variables Intercept LOG of AADT1 LOG of AADT2 SDR2 PKTRUCK PKTHRU2 PKLEFT PKLEFT1 MEDTYPE12 MICHIGAN3 LTLN1S (0 or 1) K Pearson product-moment correlation coefficients MPB/yr 1 Main Model Coeff.0082.4578 (0.7193 (0.64 0..1100 (0.0926.0.0.2586 (0.0565) -0.0.8503 (0.0.0958.2645.0. p-value) -7.1722.0076) N/A4 -0.0080) -0.0000) 0. (p.1307.0.4382 (0.0.4823 (0.0.7350 (0.0965.0085.0.16 The main model published in Vogt.0005) 0.0001) 0.0412.e.3294 (0.e.0533) 0.

Two models were developed for both TOTACC and INJACC.2186. speed limits on major roads.2145. Summary statistics on the data are provided in section 3.0003 (0.0149.67 INJACC Variant 1 Model.9505 (0. and only 1993–95 accident data were used.9801 (2. but not in other GOF measures.0523 (0.0465) -0.0. Compared to the main model.1279.65 0.0182) 0. there is no main INJACC model in Vogt.0.5400 (0.3237 (0. and horizontal curvature variables were found to be significant in the main model.0001) 0. (s.0..05 0.5296 (2.2 of this report.Table 170.60 0.0.0216) N/A3 0. For TOTACC.0577) N/A3 0.14 0.0. 1999 2 Median Type 1 (painted) on major roads 3 N/A: not available 3.0038) 0.70 -0.0.e.0000) N/A 3 Variant 1 Coeff.0240) 0.3284. the main model was selected by examining the Pearson product-moment correlation coefficient. p-value) -12.0. p-value) -7.0.0. p-value) -7.3027 (0.0.1345. overdispersion.0.0.71 0. commercial driveways on major roads.4.4308 (0.0491) 0. and the Georgia sites. (s. Again.1341.1296. Parameter Estimates for Recalibrated INJACC Type IV: Full Models Original Main Model1 Applied to Full Model Data Coeff.5299 (0.0145.5008 (0.. MPB per year.0. Some of California sites experienced changes in their design features during the 1996–98 period..3927 (2.0220) 0.0420) -0.0870.e.9908.68 Variables Intercept LOG of AADT1 LOG of AADT2 SPD2 PKTRUCK PKLEFT1 SDR2 MEDTYPE12 K Pearson product-moment correlation coefficients MPB/yr 1 Main Model Coeff. Variant 1 showed a 191 .1824.5 Type V Models The data used to calibrate full models include the California and Michigan sites from the original study.0003) 0.0044) 0.0262 (0.0520 (0.0000) 0.0179.0.0994 (0.0.0433.0127.3452 (0. Variant 1 in table 171 yields improvement in overdispersion.05 MAD/yr 0. (s.e.0128.4671 (0.0001) 0.0.0082) 0.0385) 0.0289 (0. and MAD per year.0002.0. For such sites.0.0339 (0. accidents during the 1996–98 period were ignored in the database.5670 (0.1645. additional years of accident data from the California and Michigan sites.0001) 0.0795) N/A3 -0.2560.1213.0.0. light.0005) 0. major and minor AADTs.

lower Pearson product-moment correlation coefficient than the main model, but MPB per year and MAD per year were higher. For INJACC, major and minor AADTs, presence of lighting, speed limits on major roads, and horizontal curve on minor roads within 244 m (800 ft) of intersection were significant variables in the main model. Variant 1 in table 172 was superior to the main model in terms of lower overdispersion and other GOF measures, with the exception of MPB per year. However, the model includes a Michigan indicator variable, which means more influence of the Michigan data on the model. Because the IHSDM requires the main model to be recalibrated to work in any State, the model with the State indicator was selected as a variant, not as the main model. For reference, a comparison between the Vogt full models applied to the recalibration data and the newly recalibrated models for TOTACC and INJACC also is given. As expected, there were some differences between the recalibrated models and the Vogt models. For TOTACC, the recalibrated main model included additional terms COMDRWY1, SPD1, LIGHT, and HEICOM compared to the Vogt model but did not include PKTRUCK, PKLEFT2, PROT_LT, and VEICOM, all of which were in the Vogt model. For the INJACC model, the recalibrated main model separated the safety effect of major and minor road AADTs. The main model included LIGHT, SPD1, and HEI2 but excluded PKLEFT2, PKTRUCK, PROT_LT, and VEICOM when compared to the Vogt model. The GOF as measured by MPB per year and MAD per year indicates that the recalibrated models for TOTACC and INJACC provide better GOF than the Vogt models. Pearson product-moment correlation coefficients were also higher for INJACC but were a little lower for TOTACC. The GOF as measured by K was not as good as that for the Vogt models. Tables 171 and 172 show the model results for TOTACC and INJACC, respectively.

192

Table 171. Parameter Estimates for Recalibrated TOTACC Full Models: Type V
Original Main Model1 Applied to Full Model Data Coeff. (s.e., p-value) -6.9636 (2.7911,0.0132) 0.6199 (0.2504,0.0133) 0.3948 (0.1737,0.0230)

Variables Intercept LOG of AADT1 LOG of AADT2 COMDRWY1 SPD1 LIGHT HEICOM HI MEDTYPE12 PKTRUCK PKLEFT2 PROT_LT (0=no, 1=yes) VEICOM K Pearson product-moment correlation coefficients MPB/yr

Main Model Coeff. (s.e., p-value) -5.1527 (1.8653,0.0057) 0.4499 (0.1968,0.0223) 0.2699 (0.0767,0.0004) 0.0539 (0.0304,0.0757) 0.0177 (0.0090,0.0482) -0.2938 (0.1837,0.1098) -0.0288 (0.0153,0.0597)

Variant 1 Coeff. (s.e., p-value) -5.4718 (1.8686,0.0034) 0.4220 (0.1976,0.0327) 0.2913 (0.0755,0.0001) 0.0494 (0.0296,0.0948) 0.0229 (0.0088,0.0092) N/A3 -0.0221 (0.0116,0.0571) -0.4941 (0.2349,0.0354)

N/A3

N/A

3

N/A3

0.4019 (0.0765,0.0000) 0.77 -0.02 2.90

0.3954 (0.0781,0.0000) 0.73 -0.03 3.03

0.0315 (0.0143,0.0275) -0.0142 (0.0047,0.0023) -0.6754 (0.1824,0.0002) 0.1299 (0.0450,0.0039) 0.1161 (0.0323,0.0003) 0.78 -0.79 3.31

MAD/yr The main model published in Vogt, 1999, (p. 122) 2 Median Type 1 (painted) on major roads 3 N/A: not available
1

193

Table 172. Parameter Estimates for Recalibrated INJACC Full Models: Type V
Original Main Model1 Applied to Full Model Data Coeff. (s.e., p-value) -3.2562 (2.9932,0.2767) N/A3 0.2358 (0.1722,0.1707)

Variables Intercept LOG of AADT1 LOG of AADT2 Log of (AADT1*AADT2) LIGHT SPD1 HEI2 PKLEFT2 PKTRUCK MICHIGAN2 PROT_LT (0=no, 1=yes) VEICOM K Pearson product-moment correlation coefficients MPB/yr

Main Model Coeff. (s.e., p-value) -9.0707 (1.9064,0.0000) 0.6697 (0.1899,0.0004) 0.2509 (0.0929,0.0069) N/A3 -0.3985 (0.1702,0.0192) 0.0397 (0.0093,0.0000) -0.0284 (0.0126,0.0244)

Variant 1 Coeff. (s.e., p-value) -8.6196 (1.7838,0.0000) 0.6875 (0.1756,0.0001) 0.1731 (0.0933,0.0636) N/A3 -0.4054 (0.1641,0.0135) 0.0370 (0.0091,0.0001) -0.0230 (0.0124,0.0642) N/A
3

N/A3

-0.0113 (0.0062,0.0678) 0.0323 (0.0146,0.0267) N/A3 -0.2943 (0.1864,0.1144) 0.0822 (0.0551,0.1358) 0.1630 (0.0662,0.0138) 0.43 -1.02 1.40

N/A3

0.3499 (0.1653,0.0343) N/A3

0.2360 (0.0958,0.0138) 0.68 0.00 0.98

0.2065 (0.0918,0.0244) 0.70 -0.01 0.94

MAD/yr The main model published in Vogt, 1999, (p. 124) 2 Indicator variable for Michigan 3 N/A: not available
1

3.5 ESTIMATION OF ACCIDENT MODIFICATION FACTORS
This section discusses the derivation of AMFs for both total and injury (fatal and nonfatal injury) accidents. FHWA and its contractors have developed a new approach that combines historical accident data, regression analysis, before-and-after studies, and expert judgment to make safety performance predictions that are better than those obtained by any of the individual approaches. A recent report documents an accident prediction algorithm for implementing the new approach for two-lane rural highway sections that include road 194

segments and intersections.(3) Ongoing efforts aim to produce similar documents for other types of facilities. The accident prediction algorithm has been developed for incorporation in the IHSDM as the Crash Prediction Module, but is suitable for stand-alone applications. As indicated earlier, the structure of the accident prediction algorithm for the five types of rural atgrade intersections is as follows:

Nint = Nb (AMF1 AMF2 … AMFn)

(15)

where Nint = predicted number of total intersection-related accidents per year after application of AMFs; Nb = predicted number of total intersection-related accidents per year for base conditions; and AMF1 AMF2 … AMFn = AMFs for various intersection features The accident algorithm report, referred to as the “Red Book,” provides AMFs for twolane rural roads and intersections.(3) An additional report provided AMFs for turning lanes at intersections.(5) As part of this project, the available data have been used to attempt to derive AMFs for these and any other variables at rural intersections to compare and refine the current AMFs. The AMFs provided in the “Red Book” are shown below in table 173. These AMFs apply to total intersection accidents (TOTACC). Also shown are the AMFs relevant to intersection types I, II, and V from Harwood et al.(5)

195

Table 173. AMFs from Previous Studies
AMF SKEW Type I exp(0.004SKEW) 1.00 if none exist 0.78 if at least one exists (0.56 if at least one exists) Type II exp(0.0054)SKEW 1.00 if none exist 0.76 on one approach (0.72 on one approach) 0.58 on both approaches (0.52 on both approaches) 1 Type V

LT MAJ

RT MAJ

SIGHT DISTANCE

1.00 if none exist 0.95 on one approach (0.86 on one approach) 0.90 on both approaches (0.74 on both approaches) 1.00 if sight distance is not limited 1.05 if limited in 1 quadrants 1.10 if limited in 2 quadrants 1.15 if limited in 3 quadrants 1.20 if limited in 4 quadrants

1.00 if none exist 0.82 on one approach 0.67 on both approaches 1.00 if none exist 0.975 on one approach (0.96 on one approach) 0.95 on both approaches (0.92 on both approaches)

1

Derivation of AMFs in this project was attempted in two ways. First, AMFs were inferred from the parameter estimates of the full models presented in section 3.4. Second, a relatively new and untested regression procedure that relates the difference between the observed number of crashes at a site and is predicted by a base model to the nonbase condition factors was applied. This procedure is explained in detail later in this section. 3.5.1 AMFs Derived from Recalibration Full Models One approach to deriving AMFs is to apply a model using the estimated parameter values from only statistically significant variables in accident prediction models. This approach suffers from correlation between geometric variables and traffic, and the difference in accident experience between sites is possibly due to the substantial unexplained variation resulting from omitted factors. Nevertheless, AMFs derived in this manner from the full models of section 3.4 are presented in tables 174 and 175.

196

Table 174. AMFs Derived from Type I and II Full Models
Type I AMF Group A Data HI11 RT MAJ DRWY1 Group B Data RT MAJ RT MIN LT MAJ
1

Type II Injury Total Not calibrated 1.19 1.13 0.86 Injury

Total

exp(0.0263HI1) Not calibrated 0.88 1.35 0.82

exp(0.0286HI1) Not calibrated 0.87 1.36 0.80

exp(0.0408HI1) Not calibrated 1.09 0.85 Not calibrated 0.72

MEDIAN Not calibrated Not calibrated As a continuous variable, the AMF for HI1 is a function and not a factor

Not calibrated

197

0163HAU) exp MEDWIDTH1 (-0.Table 175.0249PKTHRU2) Not calibrated Not calibrated 198 Not calibrated exp(0.0284HEI2) exp(0.0101HAU) exp(0.0627COMDRWY1) Not calibrated exp(0. and V Full Models Type III AMF COMDRWY1 VEI1 HAU Total exp (0.0520PKTRUCK) exp(0.1889HAZRAT1) exp(-0.0523PKLEFT1) HEI2 1 Medtype1: Painted .0479PKTRUCK) exp(-0.0681COMDRWY1) exp(0. AMFs Derived from Type III.73 Not calibrated exp(0.0003SDR2) 0.0539 COMDRWY1) Type V Injury exp (0.0106MEDWDTH1) Not calibrated Not calibrated 2 MEDTYPE11 HAZRAT1 PKTRUCK PKTURN PKTHRU2 PKLEFT PKLEFT1 SDR2 LIGHT HEICOM 0.67 Not calibrated exp(-0.75 Not calibrated Not calibrated exp(-0.0254PKTURN) Not calibrated exp(0.0253PKTRUCK) exp(-0. IV.0288HEICOM) 0.1081VEI1) Type IV Injury Total Injury Total exp(0.0229PKLEFT) Not calibrated Not calibrated Not calibrated exp(-0.

3. it was assumed that these factors have no effect on the safety of the intersection. Y in the accident prediction algorithm should be the observed Y.2 AMFs Derived from Regression Models Recalibration of AMFs was also undertaken using a relatively untested regression analysis procedure in which the following steps were taken: 1. The AADT base models that meet all of the base conditions were estimated. The difference between a site’s accident count per year and the expected value per year was used as an estimate of the dependent variable. Y = Y * AMFs Equation (16) can be rewritten as: ∧ (16) AMFs = From steps 3 and 5: Y Y ∧ (17) Y =Y −β Equations (17) and (18) can be combined to estimate an AMF: ∧ (18) AMF = Y Y −β (19) 7. they were set to a value of 1. The β coefficient of a factor was estimated from a simple linear regression model. for a turning lane AMF. In the case that β coefficients in equation (19) are insignificant (at the 90 percent level in this research). The β coefficients of the insignificant AMFs were re-estimated by repeating step 2 and subsequent steps. AMFs were estimated from the following equations: If perfect knowledge for base models and AMFs is available. Y. 4.. 9. If the β coefficients became significant. 6.5. If not. Suitable base conditions for qualifying the current AMFs were selected.e. This offered a larger sample size to re-estimate the insignificant AMFs. the estimated coefficients were adopted as the relevant AMFs. left-turn lane on major road) that differs from the base conditions was specified as an independent variable. i. 5. 2. Any factor (for example. 8. 3. the base condition is no turning lane. 199 .

71 1 0. The base conditions include no turning lanes on major or minor road and no median on the major road.3 and which were developed using the Group B dataset were used.42 0. for which the Group B dataset was applied.(3) The base condition for intersection skew angle was plus or minus 5 degrees of skew to have more base sites and to accommodate any possible measurement errors. Table 177 shows the base conditions for Type III AMFs. Table 176. where AMFs were not estimated to be significantly different from 1.77 Type II INJACC 1 1 1 0. 200 . a value of 1 has been assigned. For Type I. report was used to define adequate sight distance.and left-turn lanes on major and median on major are expected to reduce the expected number of crashes from the base model. The base condition for intersection sight distance is the availability of adequate intersection sight distance along the major road in all quadrants of the intersection. The nominal or base condition for “intersection left-turn lane” is the presence of left-lane on the major-road approach because as many as 79. Type III Unlike the case for Type I and II. the Group A dataset was used to calibrate AMFs for Type III intersections because the Group B dataset does not provide sight distance and angle data. Table 178 displays the base model estimated using the sites meeting all of the base conditions.” the absence of right-turn lane is the base condition.56 1 1 1 1 TOTACC 1 0. The same definition regarding sight distance in a quadrant as described on page 48 of the Harwood et al.4 percent of the available sites have left-turn lanes. new base models had to be calibrated for the AMFs. As mentioned. the AMF for right-turn lane on the minor road increases the expected number of crashes from the base model. For Type II. AMFs derived from this procedure are shown in table 176.52 Few AMFs could be derived with statistically significant results. As a result. right. log of (AADT1 * AADT2) was used as the traffic variable.48 1 1 1 1 INJACC 1. AMFs Estimated from Regression Procedure Type I AMF RT MIN RT MAJ LT MIN LT MAJ MEDIAN TOTACC 1. For “right-turn lane.71 0. the base conditions and base models described in section 3. Because the β coefficient of log of AADT1 was quite insignificant.Types I and II For types I and II.

0.0.016*SKEW1)/ 1+(0.0000) Log (AADT1* AADT2)1 K 0.6) Total1 All Total means sites meeting all of the base conditions Table 178.0000) 0.98 = mean of the observed TOTACC accidents per year of the sites meeting no angle..0. no right lane. and presence of left lane AMFs Left lane on major road Right lane on major road Sight distance 201 . Table 177.4) 88 (64.523+0.0. (s.3172 1 Logs of AADT1 and AADT2 were combined because the β coefficient of log of AADT1 was insignificant Table 179.1251.9821..AMFs derived from this procedure are given in table 179.e.4) 28 (20.982+0. Base Model for Type III Sites TOTACC Variables INJACC Coeff.5) 89 (65.0000) (0.7) 104 (76. (s.0740.1313. and presence of left lane 3 0.5342 (0.017*SKEW1)/ (0.52 = mean of the observed INJACC accidents per year of the sites meeting no angle.5583 -9. p-value) -7. AMFs for Type III Sites Recalibrated (TOTACC) Recalibrated (INJACC) 0.016*SKEW) (0.017*SKEW) Intersection angle 1 SKEW = intersection skew angle (degrees). p-value) Coeff.0000) (1. no right lane.e.71 (One approach) 1 (One approach) 1 (One approach) 1 (One approach) 1 1 1+(0.2045 0.0573 Intercept (1. expressed as the absolute value of the difference between 90 degrees and the actual intersection angle 2 0.4909 0. These AMFs apply to total accidents and total injury accidents. Base Conditions for Type III Sites AMF Left lane on major road Right lane on major road Sight distance Intersection angle 1 Base Condition Presence of left lane Absence of right lane Adequate sight distance -5° ~ 5° Frequency Meeting Base Conditions (Percent) 108 (79.

0.6284. The AMFs derived are shown in table 182.9261 Log (AADT1* AADT2)1 (0.6 percent of the available sites have leftturn lane on both approaches.8220 0.3871 1 Logs of AADT1 and AADT2 were combined because the β coefficient of log of AADT1 was insignificant 202 . For “intersection right-turn lane. These AMFs apply to total accidents and total injury accidents.1609.” the absence of right-turn lane is the base condition. p-value) Coeff. Table 181 represents the base model estimated using the sites meeting all of the base conditions..0662) (7.4677. Base Conditions for Type IV Sites AMFs Left-turn lane on major road Right-turn lane on major road Sight distance Intersection angle 1 Base Condition Presence of left lane (both approach) Absence of right lane Adequate sight distance -5° ~ 5° Frequency Meeting Base Conditions (Percent) 90 (72. (s. New base models were calibrated for the AMFs.6) 81 (65. the base condition for intersection skew angle was plus or minus 5 degrees of skew to have more base sites and to accommodate any possible measurement errors. The base conditions for Type IV AMFs are shown in table 180.Type IV As was the case for Type III. Base Model for Type IV Sites TOTACC Variables INJACC Coeff. log of (AADT1 * AADT2) was used as the traffic volume variable.9) Total1 All Total means sites meeting all of the base conditions Table 181.0477) K 0.0. Table 180. As for Type III.0470) 0.3) 75 (60.e. the Group A dataset was used to calibrate AMFs for Type IV intersections because the Group B dataset does not provide sight distance and angle data.2207 Intercept (6. The base condition for intersection sight distance is the availability of adequate intersection sight distance along the major road in all quadrants of the intersection. Since the β coefficient of log of AADT1 was quite insignificant.6) 69 (55.0. (s. p-value) -12. The base condition for “intersection left-turn lane” is the presence of left-lane on the major road approaches because as many as 72.e.5159 0..0576) (0.0.5) 11 (8.4330.1778 -14.

048*SKEW Intersection Angle 1 SKEW = intersection skew angle (degrees). which indicates that the variable was not found to have a significant impact on the safety of the intersection. the Group A dataset was used to recalibrate the current AMFs for Type V intersections.” the absence of right-turn lane is the base condition.74 1 1 1+(0. and presence of left lane 3 0. 203 . Because the β coefficient of log of AADT1 was insignificant.72 = mean of the observed INJACC accidents per year of the sites meeting no angle.053*SKEW1)/ 1+(0. As a result. As many as 75 percent of the available sites have left-turn lane on both approaches.86 1 Both Approaches 0.053*SKEW) (0. Table 184 represents the base model estimated using the sites meeting all of the base conditions. Therefore. and new base models were calibrated for applying the AMFs. the base condition for “intersection left-turn lane” is the presence of left-lane on the major road approaches. The base condition for intersection skew angle was plus or minus 5 degrees of skew for the same reasons as those for Type III and IV.723+0. expressed as the absolute value of the difference between 90 degrees and the actual intersection angle 2 0. there would be insufficient sites for a base model. In addition. no right lane. no right lane. Sight distance was ignored for the base conditions to develop a base model because the number of sites with adequate sight distance was almost equal to the ones with sight distance limited in one or two quadrants. and presence of left lane Type V Because the Group B dataset does not provide sight distance and angle. if one of the two sight distance levels was considered as a base condition.43 = mean of the observed TOTACC accidents per year of the sites meeting no angle.048*SKEW1)/ 2 (1.43 +0.Table 182. No AMFs were statistically significant using this procedure for Type V sites. sight distance is believed to have no effect on safety because conflicting traffic movements are controlled by signals. log of (AADT1 * AADT2) was used to represent the traffic volume variable. AMFs for Type IV Sites Recalibrated (TOTACC) AMFs Left-turn lane on major road Right-turn lane on major road Sight Distance One Approach 1 1 Both Approaches 1 1 1 Recalibrated (INJACC) One Approach 0. Therefore. a value of 1 was assigned. The base conditions for Type V AMFs are shown in table 183. For “intersection right-turn lane.

0.0. no AMF was estimated for SKEW.2780.3907 (0.4147 1 Log of AADT1 was insignificant. SKEW did not provide a significant AMF different from 1 in the new data. turning lanes at most sits in this dataset were provided on both approaches and separate effects could not be detected for one versus two approaches. For right-turn lane on major roads. Tables 185 through 187 compare the AMFs from the “Red Book” and those derived here.1969. Compared to the previous works. are also shown in brackets.0422) (0.0. AMFs greater than one and less than one were estimated for Type II sites using the Group A dataset for the former and the Group B for the latter.Table 183. The AMFs from Harwood et al.1265 Intercept (3.0991) 0.3 Summary of AMFs This research introduced two ways of deriving AMFs. so logs of AADT1 and AADT2 were combined 3. none of the variables investigated showed any significant impact on safety for Type V sites. (s. for Types I and II. This difference will explain a real effect of the presence of individual design features on safety. 204 .(5) As mentioned. p-value) -5..4001 0.2164.0.0710) Log (AADT1* AADT2)1 K 0. resulting in AMFs of one for these variables. The other approach uses a relatively untested regression procedure. And very few sites had deficient sight distance so an AMF could not be estimated. Base Model for Type V Sites TOTACC Variables Coeff.e.5.6499 0.1136) (3. Base Conditions for Type V Sites AMFs Base Condition Left lane on major Presence of left lane (both road approach) Right lane on major road Absence of right lane Angle -5° ~ 5° Total1 All 1 Total means sites meeting all of the base conditions Frequency Meeting Base Conditions (Percent) 75 51 60 22 Table 184.. Whereas the “Red Book” provides separate AMFs for right-and left-turn lanes on major roads at Type II intersections.e. This approach aims to attribute the difference between the observed number of crashes at a site and that predicted by a base model to the presence of non-base condition factors. (s. the second approach is recommended. p-value) INJACC Coeff.1864 -6.7145. One approach is to adopt the estimated parameter values from the statistically significant variables in the accident prediction models. Although in general the first approach is easier to apply.

71 Not Not Not Not calibrated calibrated calibrated calibrated 1.19.48 Not exp(0.72 on one 1.86 on one approach) 0.56 if at least one (0.0. (2002) Type I Type II Full Models Regression Models Type II 1 exp(0.00 if none exist 0.13 Not calibrated Not calibrated 1 1.10 if limited in 2 quadrants SIGHT DISTANCE RT MIN HI1 DRWY1 MEDIAN 1 Group A = 1.19.78 if at least one 0.0263) calibrated Not calibrated 1.77 205 .0054SKEW) Type I Type II Type I Not Not calibrated calibrated 1 1.05 if limited in 1 quadrants 1.71 LT MAJ 1 0. Right-turn lanes on major roads provided significant AMFs for Type IV intersections. no AMFs were found to be significantly different from 1.00 if none exist 0.004SKEW) exp(0. For Type V intersections.74 on both approaches) RT MAJ 0.82 1. Comparison of Type I–II AMFs for TOTACC AMF SKEW “Red Book” and Harwood et al. Group B = 0.15 if limited in 3 quadrants 1.58 on both exists approaches (0.00 if none exist approach) 0.861 1 0.20 if limited in 4 quadrants 1.88 1. intersection angle (SKEW) was estimated as significant in the regression models.35 1 Not calibrated 0.76 on one approach (0.For Type III and IV intersections. Table 185.90 on both approaches (0.86 Not calibrated Not calibrated 1.52 on both exists) approaches) 0.95 on one approach (0.

016*SKEW) (1. Comparison of Type III–V AMFs for TOTACC “Red Book” and Harwood et al.1081VEI1) exp (-0.053*SKEW)/ (0.0106MEDWDTH1) 0.82 on one approach 0.Table 186. (2002) AMF SKEW Type III-IV Type V Type III AMFs Derived From Full Models Type IV Type V AMFs Derived From Regression Models Type III Type IV Type V 1+(0.053*SKEW) 1 RT MAJ LT MAJ SIGHT DISTANCE COMDRWY1 VEI1 MEDWIDTH1 MEDTYPE1 PKTRUCK PKTHRU2 PKLEFT SDR2 LIGHT HEICOM None provided 1 exp(0.71 1 exp (0.75 exp (-0.67 on both approaches 1 exp (0.0249PKTHRU2) exp (0.010SKEW) 1.96 on one approach) 0.00 if none exist 0.95 on both approaches (0.0479PKTRUCK) Not calibrated exp (0.0288HEICOM) Not calibrated Not calibrated Not calibrated .016*SKEW)/ 1+(0.0229PKLEFT) exp(-0.0539 COMDRWY1) 1 1 1 1 206 Not calibrated exp (-0.73 Not calibrated Not calibrated Not calibrated 1 1 1 0.43+0.98+0.975 on one approach (0.0681COMDRWY1) exp(0.92 on Not calibrated both approaches) 1.0003SDR2) 0.00 if none exist 0.

87 0.0254PKTURN) Not calibrated Not exp calibrated (0.0627COMDRWY1) exp (0.52+0.36 Not calibrated Not calibrated Not calibrated Not Not calibrated calibrated 0.1889HAZRAT1) exp exp (-0.52 Not calibrated Not Not calibrated calibrated Not calibrated Not calibrated exp(0.0520PKTRUCK) 0.8 1.09 None provided 0.017SKEW)/ 1+(0.0408) 1.V AMFs Derived From Full Models Type I Type II Type III Not calibrated Not calibrated exp(0.74 both approaches 1 1 1 1 1 1 1 Not calibrated exp exp (0.0253PKTRUCK) (-0.67 exp (-0.0286) (0.72+0.Table 187.0163SKEW) 0.017SKEW) (0.048SKEW)/ (0.42 1 AMFs Derived From Regression Models Type II Type III Type IV 1+(0.85 Type IV Type V 1 1 1 1.0523PKLEF) Not calibrated Not calibrated 0.56 Type I 1 1 0.72 207 MEDIAN COMDRWY1 HAZRAT1 PKTRUCK PKTURN PKLEFT1 SDR2 LIGHT HEI2 1 Not calibrated exp (0. 0.048SKEW) 0.0284HEI2) .86 one approach. Comparison of AMFs for INJACC “Red Book” AMF SKEW RT MAJ LT MAJ RT MIN SIGHT DISTANCE HI DRWY1 Type I .

772 1.421 4.539 0.722 2.200 3.701 1. The sensitivity test results for these AMFs are presented in tables 188 through 191.3.000 100 500 1.483 2.675 5.000 5.590 2.081 7.170 2.000 100 500 1.000 5.411 0.796 0. 3.115 0.318 3.000 5.240 0. Table 188.000 3.692 2.6 SENSITIVITY ANALYSIS RESULTS A sensitivity analysis was performed to examine the estimated effect on safety of the AMFs derived from the full models and regression base models.835 1.299 0.017 1.919 2.069 2.472 0. Type I TOTAL Accidents The AMFs derived for TOTAL accidents for Type I intersections are HI1.999 3.033 2.732 0.890 1.217 1.522 2.000 10.113 0.526 0.6.336 0.067 1.461 0. right-turn lane on minor.975 20 0. right-turn lanes on major.075 6.312 0.1 Type I Intersections The predicted TOTAL and INJURY accident frequencies per year for each AMF derived from the full models and regression base models are presented in tables 188 through 195.380 0.583 2.952 1.265 0.414 0.233 0.324 1.147 0.469 0.610 0. The sensitivity analysis applies each AMF to the base model estimate for various levels of AADT.594 4.496 2.691 3.440 5. The AMF for HI1 was derived from full models.000 3.101 0.160 1.749 1.150 0.099 5.936 1.814 2.361 0.644 3. 208 .000 Note: Group B AADT base model was used.191 0.026 1.780 0.238 10 0.000 100 500 1.622 1.000 10.900 1.000 HI1 0 0.901 4.211 0.600 0.204 0.656 1.300 1.185 0.239 1. and left-turn lanes on major roads. Sensitivity of Safety to HI1 for Type I TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 50 100 400 100 500 1.000 3.593 5 0.990 1.574 0.247 3.188 2.345 0.000 3.088 0.129 0.974 1.576 1.

000 Note: Group B AADT base model was used.380 0.162 4.792 1.000 3.078 0.919 2.000 5.936 1.688 2. The AMF for RT MAJ was derived from full models.185 0.900 1.406 0.789 2.000 500 1.000 3.300 1.639 3.042 50 100 400 100 1.299 0.000 10.000 100 10.088 0.000 500 1.000 100 3.215 0.017 1.033 2.162 0.576 1.594 4.023 0.144 1.000 5.000 3.204 0.000 500 1.461 0.999 3.000 100 5.180 0.414 0.400 1.361 0.506 0.823 1.593 Presence of Right-Turn Lane on Major Road 0.113 0.590 2.732 0.Table 189.364 0.099 0.000 500 1.895 1. 209 .317 0. Sensitivity of Safety to RT MAJ for Type I TOTAL Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 RT MAJ No Right-Turn Lane on Major Road 0.644 0.

925 2.083 1.204 0.Table 190.017 1.506 1.380 0.000 Note: Group B AADT base model was used.131 0.732 0.593 RT MIN Presence of Right-Turn Lane on Minor Road A B 0.852 1.000 5.049 4.185 0.300 1.000 500 1.088 0.103 3.798 50 100 400 100 1.461 0.000 500 1. B: AMF derived from regression models.000 500 1.851 5.000 100 5.919 2.263 1.153 0.043 0.000 100 10. Sensitivity of Safety to RT MIN for Type I TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) No Right-Turn Lane on Minor Road 0.590 2.119 0. A: AMF derived from full models.000 3.319 6.590 2.000 5.000 100 3.999 3.215 1.385 1. 210 .374 1.439 4.744 3.988 1.354 2.099 4.682 0.000 3.361 0.900 1.302 0.559 0.576 1.777 0.622 0.756 1.936 1.000 500 1.487 0.000 10.402 1.275 0.000 3.167 0.534 0.839 3.147 2.613 0.594 4.200 6.033 2.113 0.299 0.273 0.864 2.414 0.332 2.249 0.

594 4.000 500 1.340 0.593 Presence of Left-Turn Lane on Major Road 0.378 0.093 0.414 0.000 3.361 0.000 3. Sensitivity of Safety to LT MAJ for Type I TOTAL Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 LT MAJ No Left-Turn Lane on Major Road 0.Table 191. The AMF for LT MAJ was derived from full models.885 0.000 100 3.113 0.667 2.738 1.017 1.834 1.000 500 1.204 0.919 2.947 3.999 3.296 0.000 5.590 2.600 0.000 5.766 50 100 400 100 1.299 0.472 0.767 1.300 1.088 0. 211 .000 500 1.167 0.573 1.000 500 1.066 1.000 3.461 0.576 1.000 10.000 100 10.732 0.459 2.132 0.936 1.000 100 5.000 Note: Group B AADT base model was used.151 0.185 0.900 1.380 0.304 1.033 2.073 0.

129 0.000 3. The AMF for HI1 was derived from full models.606 0.046 1.317 0.382 0.179 0. and left-turn lanes on major roads.530 0.154 0.331 0.073 0.000 100 500 1.562 1.409 0. right-turn lanes on minor.042 0.730 0.000 10.000 3.786 1.000 5.238 0.867 0.398 0.842 1.712 5 0.000 Note: Group B AADT base model was used.482 0.642 0.033 1.971 1. Table 192.581 0.101 1.700 0.399 3.142 0.516 1.000 100 500 1.000 5.975 10 0.000 10.314 1.827 1.621 0.293 1.480 0.266 0.418 0.000 5.231 0.295 0.084 0.802 2.097 0. right-turn lane on major. 212 .892 1.526 0.061 0. Sensitivity of Safety to HI1 for Type I INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 50 100 400 100 500 1.307 0.053 0.705 0.166 0.508 0.109 0.036 0. The sensitivity test results for these AMFs are presented in tables 192 through 195.287 0.163 0.000 100 500 1.000 3.081 0.537 0.416 0.392 2.251 0.139 1.906 1.670 0.717 0.362 0.046 0.064 0.361 0.018 2.459 0.082 0.221 0.095 0.192 0.206 0.000 HI1 0 0.001 0.000 3.145 0.188 0.931 0.503 0.048 0.639 1.Type I INJURY Accidents The AMFs derived for INJURY accidents for Type I intersections are HI1.354 1.278 20 0.

000 500 1.031 0.000 5.201 0.000 5.231 0.000 3.000 100 5.362 0.178 1.730 0.000 500 1.139 1.000 3. 213 .786 1.249 0.073 0.071 0.000 500 1.287 0.036 0.179 0.Table 193.000 100 3.635 0.755 0.000 3.000 Note: Group B AADT base model was used.082 0.166 0.123 0.040 0.315 0.063 0.712 Presence of Right-Turn Lane on Major Road 0.438 0.346 0.145 0. The AMF for RT MAJ was derived from full models.354 1.991 1.000 10.314 0.398 0.046 0. Sensitivity of Safety to RT MAJ for Type I INJURY Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 RT MAJ No Right-Turn Lane on Major Road 0.361 0.000 500 1.142 0.457 0.503 0.156 0.621 0.684 0.000 100 10.489 50 100 400 100 1.526 0.867 0.541 0.

000 3.000 3.777 1.867 0.526 0.565 0.353 0.243 0.279 0. A: AMF derived from full models.221 0.712 RT MIN Presence of right-turn lane on minor road A B 0.845 0.114 0.226 0.036 0.398 0.000 10.503 0.621 0.139 1.670 50 100 400 100 1.260 0. 214 .684 0.000 500 1.142 0.138 1.785 0.226 1.328 2.179 0.046 0.000 100 5. B: AMF derived from regression models.112 0.082 0.000 3.166 0.231 0.362 0.354 1.071 0.549 1.490 0.541 0.820 0.000 5.049 0.841 2.000 500 1.786 1.000 5.073 0.000 Note: Group B AADT base model was used.287 0.563 0.390 0.361 0.314 0.180 1.062 0.192 0.112 2.493 0.360 0.000 100 3.000 500 1.099 0.068 1.Table 194.992 1.970 1.730 0.621 0.000 500 1.128 0.715 0.447 0.056 0. Sensitivity of Safety to RT MIN for Type I INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) No right-turn lane on minor road 0.000 100 10.

786 1.318 0.Table 195.185 0.621 0.362 0.029 0.000 Note: Group B AADT base model was used.000 3.000 500 1.000 500 1.354 1.000 100 10.526 0.584 0.073 0.000 500 1.179 0.036 0.712 Presence of Left-Turn Lane on Major Road 0. Sensitivity of Safety to LT MAJ for Type I INJURY Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 LT MAJ No Left-Turn Lane on Major Road 0.000 3.403 0.000 5.287 0.867 0.143 0.166 0. The AMF for LT MAJ was derived from full models.000 100 3.000 5.000 3. 215 .503 0.142 0.113 0.290 0.082 0.730 0.369 50 100 400 100 1.046 0.000 100 5.288 0.361 0.139 1.694 0.066 0.231 0.911 1.229 0.000 10.629 0.058 0.420 0.133 0.398 0.083 1.497 0.036 0.000 500 1.

204 100 0.527 2.769 12.016 Note: Group B AADT base model was used.346 3. 216 .503 3.491 0. Table 196.369 2.391 0.000 4. The sensitivity test results for these AMFs are presented in tables 196 through 199.636 1.960 1.190 100 0.158 1.350 1.3. and median on major roads.064 0. Type II TOTAL Accidents The AMFs derived for TOTAL accidents for Type II intersections are ND. right-turn lanes on major.831 1.179 0.867 10.000 10.2 Type II Intersections The predicted TOTAL and INJURY accident frequencies per year for each AMF derived from the full models and regression base models are presented in tables 196 through 204. The AMF for DRWY1 was derived from full models.072 0.760 11.000 1.485 0.876 7.198 0.380 1.051 5.650 6.200 5.816 10.223 0. Sensitivity of Safety to DRWY1 for Type II TOTAL Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 DRWY1 0 3 0.498 5.971 3.000 500 0.212 3.735 0.585 4.000 6.639 8. left-turn lanes on major.429 0.173 5.000 500 0.626 0.442 1.230 5.202 0.707 0.215 3.815 0.351 1.000 1.093 0.789 50 0.859 3.744 3.548 2.000 0.000 2.000 500 1.554 100 0.318 5 0.061 1.922 2.175 100 0.000 3.6.082 400 0.939 1.056 100 0.227 4.105 0.000 500 1.196 1.000 2.000 5.042 2.563 1.

867 8.049 0.000 2.190 4.208 0.000 5.728 2.173 3.423 1.692 5.082 0.200 6.919 8.590 1.614 7.172 5.459 2.076 1. A: AMF derived from full models.744 2.402 2.000 500 0.136 0.960 2.715 0. Sensitivity of Safety to RT MAJ for Type II TOTAL Accidents Per Year RT MAJ Presence of Right-Turn Lane on Major Road No Right-Turn Lane on Major Road A B C 50 0.346 0. Group A data.831 0.876 10.040 400 100 0.896 1.204 2.252 5.111 Note: Group B AADT base model was used.332 1.775 2. Group B data.986 3.151 0.816 0.603 2.554 0. B: AMF derived from full models.000 4.042 0.188 4.592 0.212 1.000 1.392 1.196 1.Table 197.702 0.056 0.476 0.579 10.067 0.906 4.028 0.058 400 0.124 100 0.188 0.071 0.686 1.971 0.412 0.000 2.270 1.380 0.327 0.030 3.000 500 1.000 500 1.175 0.472 3.500 1.246 3.000 6.000 10.000 500 0.238 3.112 1.989 0.442 1.000 3.860 3.000 0.849 1.859 3.659 0.393 100 0.158 0.016 11. C: AMF derived from regression models Major Road Minor Road AADT AADT (veh/day) (veh/day) 217 .297 0.353 5.975 100 0.000 1.623 1.428 0.498 0.452 0.565 100 0.098 0.

860 1.000 2.867 10.000 500 0.353 0.196 1.158 1.816 10.380 1.111 50 0. Sensitivity of Safety to LT MAJ for Type II TOTAL Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 LT MAJ No Left-Turn Lane on Major Road Presence of Left-Turn Lane on Major Road 0.565 0.000 10.000 500 1.252 2.000 4.000 6.270 0.190 100 0.346 3.859 3.016 Note: Group B AADT base model was used.831 1.000 1.8449 1.030 3.590 0.056 100 0.058 0.000 500 0.000 1.000 2.173 5.204 100 0.000 500 1.238 2.554 100 0.579 1.212 3.000 3.200 5.124 0.246 0.082 400 0. The AMF for LT MAJ was derived from regression models.392 2.112 0.876 7.393 0.000 0.040 0.000 5.692 4.960 1.175 100 0.975 0.498 5. 218 .744 3.Table 198.

628 1.000 Note: Group B AADT base model was used.697 0.867 10.204 0.380 0.000 100 5.000 10.082 0.056 0.859 5.921 1.000 3.443 3.383 0.201 4.016 Presence of Median on Major Road 0.266 0.000 3.292 0.000 500 1.000 100 3.000 500 1.004 5.744 3.346 0.043 0.000 100 10.190 0.000 500 1.288 7.933 1.343 2.426 0.554 0.135 0.498 1.712 50 100 400 100 1.063 0.175 0.173 4. The AMF for MEDIAN was derived from regression models.000 5.200 6. Sensitivity of Safety to MEDIAN for Type II TOTAL Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 MEDIAN No Median on Major Road 0.196 1.000 3.000 500 1.212 2.960 2.640 0.Table 199.122 0.831 1. 219 .509 2.816 1.226 0.158 0.000 5.

049 0.259 0.447 2.000 3.529 0.000 5.050 3.124 3.831 2. Sensitivity of Safety to HI1 for Type II INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 50 100 400 100 500 1.831 4.196 0.640 1.479 2.292 0.074 0. and median on major roads.326 2.394 0.082 1. Table 200.078 3.648 2.161 0.382 1.242 0.704 0.962 1.199 7.672 2.798 0.106 0.055 HI1 5 10 0.894 1.060 0. rightturn lanes on major.407 5.553 1.727 4.204 0.087 0.321 0.602 20 0.212 0.178 5.483 0.000 100 500 1.073 0.703 3.433 1.238 0.882 1.358 0.196 0.995 3.705 11.538 1.344 1.131 0. 220 .000 10.919 1.000 0 0.021 2.124 6.000 Note: Group B AADT base model was used.000 100 500 1.000 10.000 100 500 1.431 0. DRWY1.975 1. The sensitivity test results for these AMFs are presented in tables 200 through 204.985 0.000 5.040 0. left-turn lanes on major.649 0.000 3.110 0.726 0.548 3.127 1.000 3.173 0.368 0.390 0.000 3.785 0.000 5.Type II INJURY Accidents The AMFs derived for INJURY accidents for Type II intersections are HI1. The AMF for HI1 was derived from full models.315 0.451 0.466 1.083 5.325 0.729 0.107 0.049 0.265 0.761 7.594 0.130 0.363 2.216 0.489 0.033 0.

000 0.303 0.000 3.092 1.548 5.363 3.055 Note: Group B AADT base model was used. Sensitivity of Safety to DRWY1 for Type II INJURY Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 DRWY1 0 3 0.640 3.204 100 0.216 1.000 500 0.403 0.000 500 0.512 0.619 0.196 100 0.000 500 0.000 500 0.797 2.401 1.053 0.470 0.087 1.961 1.648 0.173 3.919 1.000 1.350 0.127 0.000 1.283 0.238 5.048 6.236 0.039 0.033 100 0.510 5 0.714 5.421 0.958 2.407 10.000 0.000 5.117 0.094 0.697 1.188 0. The AMF for DRWY1 was derived from full models.648 5.027 4.437 1.882 3.381 0.259 0.777 3.368 10.000 1.594 1.431 1.000 2.000 0.049 400 0.107 100 0.205 0.257 0.760 1.001 1.000 2.Table 201.036 0.006 50 0.706 1.321 100 0. 221 .486 2.048 1.058 0.103 0.620 3.

000 5.000 3.049 0.273 0. The AMF for RT MAJ was derived from full models.919 1.544 1.640 1.033 0.107 0.363 2.000 500 1.297 50 100 400 100 1.321 0.000 500 1.087 0.368 0.000 100 3.000 5.196 0. 222 .055 Presence of Right-Turn Lane on Major Road 0.431 0.000 3.000 3.401 1.594 0.028 0.216 0.202 0.000 100 10.016 0.238 0.Table 202.000 Note: Group B AADT base model was used. Sensitivity of Safety to RT MAJ for Type II INJURY Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 RT MAJ No Right-Turn Lane on Major Road 0.505 0.367 0.882 1.000 100 5.184 0.041 0.896 4.750 1.648 2.781 1.000 10.000 500 1.874 0.407 5.548 3.312 0.074 0.204 0.165 2.000 500 1.173 0.159 2.091 0.147 0.

926 0.173 0.502 0.548 3.Table 203.000 10.000 3.000 Note: Group B AADT base model was used.407 5.648 2.368 0.000 500 1.107 0.000 5.123 50 100 400 100 1.135 0.196 0.049 0.036 0.386 0.014 0.033 0.370 0.000 5.091 0.072 0.431 0.238 0.692 0.919 1.321 0.573 1.000 3.000 3. The AMF for LT MAJ was derived from regression models.000 100 3.055 Presence of Left-Turn Lane on Major Road 0.154 0.000 500 1.250 0.640 1.000 100 10.269 0.045 0.216 0.363 2.594 0.020 0.000 100 5.100 0.087 0.070 1.000 500 1. 223 .882 1.181 0.204 0.000 500 1.431 2. Sensitivity of Safety to LT MAJ for Type II INJURY Accidents Per Year Major Road Minor Road AADT AADT (veh/day) (veh/day) 400 LT MAJ No Left-Turn Lane on Major Road 0.

Sensitivity of Safety to MEDIAN for Type II INJURY Accidents Per Year MEDIAN Presence of Median on Major Road No Median on Major Road A B 50 0.981 0.629 Note: Group B AADT base model was used.428 0.231 0.407 2.124 5.478 1.311 0.648 1.017 400 100 0.265 0.062 0.124 0.000 5.224 1.191 10.000 1.156 0.000 500 0.321 0.000 0.834 1.363 0.000 0.772 10.035 0.622 100 0.662 0.107 0.173 0.077 0.333 3.087 0.325 5.594 0. A: AMF derived from full models.167 100 0.000 500 0.238 0.635 0.045 1.640 2.187 0.056 100 0.861 0.000 2.000 0.112 1.548 1.000 500 0.709 3.461 0.000 1.453 1.055 3.171 0.431 0.049 0.000 500 0. B: AMF derived from regression models.000 1. Major Road Minor Road AADT AADT (veh/day) (veh/day) 224 .025 400 0.000 3.216 0.919 0.090 3.000 2.024 0.204 1.196 0.309 1.882 0.459 3.587 1.146 100 0.033 0.368 0.Table 204.857 5.640 0.

19 0.02 0.15 0.01 0.03 0.01 0.52 0.60 0.62 0.09 1.83 0. 400 Major Road Minor Road AADT AADT (veh/day) (veh/day) A B A SKEW Angle (degrees) B A B A 10 15 30 0.06 100 0. Sensitivity of Safety to Skew Angles for Type III TOTAL Accidents Per Year 0 50 0.12 0.49 0.07 0.01 0.58 0.01 0.23 0.33 1.52 1.08 0.01 100 0.04 0.17 0.000 0.82 0.56 0.13 0.12 1.71 0.20 3.02 0. painted median type on major road.52 100 0.19 0.000 500 0.01 0.30 0.31 0.51 1.01 0. B: AMF derived from regression models.30 1.06 0.26 0.01 0.28 0.98 1.11 3.07 0.23 0.85 0.46 0.59 1.42 0.74 3.19 0.59 1.15 0.22 0.000 500 0.02 0.20 0.07 0.000 1.90 225 .74 0.65 0.01 0.86 0. Table 205.98 0.30 1. The sensitivity test results for these AMFs are presented in tables 205 through 210.23 0.48 0.80 B 0.12 1.40 1.02 0.16 0.06 1.21 0.22 0.14 1.01 400 0.59 0.55 1.000 0.63 0.89 1.49 0.12 10.41 0.52 0.06 0.29 0.55 0.69 0.74 0.30 0.05 0.18 0.48 1.35 0.000 500 0.09 0.62 0.02 0.72 0.24 1.04 0.11 0.26 0.04 0.20 0.31 0.35 0.05 0. commercial driveways on major road.09 0.3.34 0.000 0.72 0.99 1.29 1.26 0.07 0.06 0.01 0.05 1.13 0. and major road left-turn lane.38 0.40 0. vertical curves on major road.01 0.06 0.01 0.05 0.34 1.53 0.47 0.000 0.36 0.3 Type III Intersections The predicted TOTAL and INJURY accident frequencies per year for each AMF derived from the full models and regression base models are presented in tables 205 through 215.61 0.28 0.07 0.41 0.82 0.70 0.02 0.17 1.84 0.000 0.07 0.46 5.61 0.29 0.33 1.40 0.01 0.000 0.74 0.10 B 0.02 0.33 0.47 1.47 0.000 0.39 0.02 0.35 0.32 0.17 1.65 0. median width on major road.51 0.08 0.13 0.46 0.000 500 0.06 0.62 1.02 0.46 0.05 0.26 0.23 0.55 0. A: AMF derived from full models.24 0.33 Note: Group B AADT base model was used.55 0.17 0.02 0.04 0.27 1.39 0.00 1.000 0.41 0.41 10.98 5.31 0.22 0.6.000 1.55 1.07 0.02 0.04 0.02 100 0.27 0.01 0.08 0.46 0.26 0.19 5.76 2.37 0.35 3.03 0.23 0. Type III TOTAL Accidents The AMFs derived for TOTAL accidents for Type III intersections are intersection SKEW angle.24 0.71 0.89 1.17 1.77 A 45 0.02 0.02 0.03 1.26 100 0.

28 0.22 0.46 0.03 0.08 0.96 0.07 0.000 Note: Group B AADT base model was used.01 0.33 0.10 1.05 0.000 100 500 1.000 3.Table 206.20 0.70 50 100 400 1.41 0.000 10.17 0. Sensitivity of Safety to Commercial Driveways on Major Road forType III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) Commercial Driveways on Major Road (density) 0 5 10 15 0.81 0.73 0.06 0.27 0.74 1.28 0.000 3.12 1.35 0.02 0.87 1.58 0.09 0.04 1.63 3.000 5.21 3.55 0.000 3.52 0.87 2.19 0.02 0.16 0.000 100 500 1.58 0.12 0.82 1.07 0.02 0. 226 .29 0.38 0.02 0.65 0.03 0.47 2.37 0.57 2.72 1.000 3.01 0.000 10.52 0.15 0.02 0.03 1.54 0.46 0.05 0.49 0.000 5.41 0.000 5.38 1.06 0.17 0.91 1.94 2.01 0.14 0.05 0.23 1.03 0.26 0.03 0.10 0.000 100 500 1. The AMF for CD on major road was derived from full models.31 0.98 1.39 0.33 1.23 0.45 0.62 0.69 0.000 100 500 1.74 1.11 0.73 0.

05 0. Sensitivity of Safety to Vertical Curves on Major Road (VEI1) for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) VEI1 0 0.27 0.54 1.57 0.36 1.10 0.79 0.54 1.000 100 500 1.29 0.93 50 100 400 1.41 0.000 100 500 1.26 0.27 1.84 5 0.000 10.18 0.71 1.27 0.000 5.29 0.19 0.000 10.01 0.86 1.000 Note: Group B AADT base model was used.000 3.10 0.08 0.02 0.29 10 0.86 1.35 0.77 0.000 100 500 1.000 3.35 1.05 0.000 100 500 1.17 0.23 0.000 5.46 0.02 0.04 0.02 0.06 0.02 1.000 3.33 0.15 0. The AMF for VEI1 was derived from full models.09 0.60 0.06 0.02 0.000 3.03 0.03 0.Table 207.33 3 0.57 0.83 2.19 2.15 0.07 0.92 2.07 1.49 0.89 3.90 0.45 0.01 0.19 0.36 0. 227 .48 0.72 0.68 1.29 3.62 0.20 0.98 1.01 0.52 0.03 1.03 0.34 0.000 5.50 0.11 0.33 0.05 0.40 0.02 0.22 1.59 0.12 1.74 0.64 0.

03 500 0.46 0.01 400 0.33 0.02 0.09 0.70 0.26 5.25 20 0.62 1.74 3.20 3.36 0.02 0.01 0.17 0.79 0.000 0.52 10.000 100 0.86 0.03 0.98 1.06 3.25 0.01 0.24 0.01 0.50 0.40 0.08 50 0.17 0.58 0.26 0.000 0.42 0.65 0. The AMF for MEDWIDTH1 was derived from full models.16 0.41 500 0.18 0.000 0.37 0.Table 208.16 0.02 1.35 3.01 100 0.11 500 0.90 1.000 0.06 0.19 0.05 1.05 1.10 0.29 1.05 0.04 0.000 0.33 0.05 0.33 Note: Group B AADT base model was used.15 0.28 0.43 0.16 0.05 0.000 100 0.01 0.10 0. Sensitivity of Safety to Median Width on Major Road (MEDWIDTH1) for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 0 MEDWIDHT1 6 12 0.19 500 0.98 5.04 0.31 0.01 0. 228 .21 0.17 1.000 0.000 1.000 1.60 0.17 0.23 0.03 0.39 0.92 1.12 10.000 100 0.13 0.55 0.27 0.000 0.03 0.01 0.000 100 0.46 5.000 0.01 0.49 0.

34 0.35 0.29 0.000 5.97 229 .19 0. Sensitivity of Safety to Painted Median Type on Major Road (MEDTYPE1) for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 50 100 400 1.98 1.01 0.20 0.000 100 500 1.06 0.74 0.01 0.41 0.03 0.02 0.000 3.000 3.15 0.21 0.01 0.46 0.000 100 500 1.01 0.12 0.000 100 500 1.04 0.52 0.04 0.30 0.Table 209.000 5.14 0. Median Type on Major Road No Median Painted Median 0.000 3.000 Note: Group B AADT base model was used.000 5.000 10.54 0.26 0.25 0.11 0.38 0.17 0.01 0.72 0.02 0.62 0.82 0.19 0.45 0.000 3.12 1.000 10.000 100 500 1. The AMF for MEDTYPE1 was derived from full models.05 0.33 0.08 0.

25 0. Sensitivity of Safety to Major Road Left-Turn Lane for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 50 100 400 1.46 0.01 0.52 0.000 100 500 1.000 5.19 0.01 0.000 100 500 1.Table 210.95 230 .21 0.44 0.14 0.35 0.000 5.03 0.12 0.20 0.19 0.000 3.62 0.04 0. The AMF for Major Road Left-Turn Lane was derived from regression models.02 0.06 0.000 100 500 1.000 5.33 0.000 10.08 0. Major Road Left-Turn Lane No Left-Turn Lane One Left-Turn Lane 0.29 0.29 0.37 0.000 100 500 1.17 0.05 0.12 0.01 0.33 0.04 0.02 0.000 3.000 3.000 Note: Group B AADT base model was used.53 0.000 3.98 0.01 0.41 0.26 0.01 0.14 0.000 10.70 1.11 0.74 0.79 1.

467 10.004 0.077 0.066 5.218 0.761 0.000 0.033 0.164 0.061 3.528 0. peak truck percentage on major road.212 0.121 0.414 0.181 0.002 400 100 0.134 0.000 500 0.076 0.097 0.128 0. and peak turning percentage.231 0.078 0.109 0.064 0.008 1.000 0.018 0.137 0.010 0.131 0.000 500 0.065 0.542 0.100 0.002 0.121 3.000 0.015 0.000 0.107 0.023 0. Sensitivity of Safety to Skew Angles for Type III INJURY Accidents Per Year 0 50 0.000 0.082 100 0.379 0.151 0.021 0.162 0.456 0.915 B 0.018 0.Type III INJURY Accidents The AMFs derived for INJURY accidents for Type III intersections are intersection SKEW angle.105 0.101 1.359 0.003 0.151 0.092 0.405 0.745 0.129 0.084 0.004 0.195 0.003 0.698 0.337 0.610 0.003 0.000 500 0.101 0.044 0.549 0.264 0.002 0.098 0.561 0.068 0.665 0.033 3.237 0.005 0.405 0.039 0.002 0.005 0.849 0.018 0.270 0.005 0.043 0.699 0.008 0.082 0.166 0.191 0.002 400 0.054 0.002 0.003 0.126 0.012 1.620 0.277 0.000 0.210 0.486 0.004 0.003 0.202 0.650 0.019 0.521 0.076 0.004 0.408 0.000 0.171 0.162 5. commercial driveways on major road.716 0.003 0.337 0.299 0.119 0.302 0.386 0.017 0.102 0.023 0.316 0.014 100 0.345 0.069 0.895 231 .082 0.193 0. Table 211.003 0.185 0.003 0.106 0.083 0.215 0.143 0.207 0.305 0.014 0.220 0.242 0.508 0. B: AMF derived from regression models.258 0.497 0.839 A 45 0.051 1.000 500 0.480 0.582 0.060 0. hazard rating on major road.134 0.042 0.324 0.246 0.002 0.061 0.000 0.053 0.264 0.009 0.015 0.012 0.389 0.168 B 0.635 0.467 0. Major Road Minor Road AADT AADT (veh/day) (veh/day) A B A 10 0.012 0.051 0.161 0.305 3.013 0.066 0.003 100 0.014 0. The sensitivity test results for these AMFs are presented below.000 0.004 0.007 0.016 0.005 0.000 0.206 0.003 0.248 0.072 0.252 0.081 0.030 0.596 0.003 0.254 0.408 5.012 0.019 0.745 0.019 0.155 0.004 0.296 0.087 0.016 0.197 0.098 0.105 0.123 0.185 100 0.660 SKEW Angle (degrees) B A B A 15 30 0.025 0.010 0.050 0.380 0.166 10.972 1.004 0.082 0.161 0.254 1.561 Note: Group B AADT base model was used A: AMF derived from full models.010 0.

062 0.154 0.000 100 500 1.000 3.781 0.051 0.166 0.000 3.002 0.070 0.003 0.112 0.000 3.196 0.131 0.066 0.000 5.467 0.227 0.019 0. Sensitivity of Safety to Commercial Driveways on Major Road for Type III INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) Commercial Driveways on Major Road (density) 0 5 10 15 0.016 0.189 0.000 5.011 0.258 0.006 0.036 0.157 0.082 0.305 0.123 0.310 0.166 0.763 1.000 10.012 0.084 0.096 0.044 0.014 0.424 0.475 0.254 0.571 0.415 0.138 0.185 0.033 0.061 0.210 0.000 100 500 1.000 100 500 1.000 10.008 0. 232 .436 50 100 400 1.085 0.004 0.004 0.Table 212.003 0.030 0.046 0.050 1.417 0.000 Note: Group B AADT base model was used The AMF for Commercial driveways on major road was derived from full models.101 0.020 0.639 0.121 0.005 0.022 0.162 0.115 0.169 0.767 1.008 0.254 0.014 0.347 0.475 0.090 0.558 0.006 0.222 0.874 1.000 3.005 0.000 5.002 0.227 0.303 0.310 0.000 100 500 1.408 0.347 0.027 0.561 0.003 0.650 0.

000 5.000 100 500 1.719 0.988 5 0.108 0.823 0.200 1.169 0.408 0.082 0.467 0.125 0.000 3.784 1.292 0.006 0.000 5.Table 213.537 0.327 0.000 3.311 0.608 0.185 0.286 0.000 3.308 0.002 0. 233 .158 0.049 1.131 0.003 0.000 10.014 0.000 100 500 1.561 3 0.033 0.622 0.066 0.005 0.530 1.009 0.751 2.002 0.454 0.000 100 500 1.442 7 0.000 10.000 Note: Group B AADT base model was used The AMF for HAZRAT1 was derived from full models.053 0.447 0.116 0.378 0.247 0.417 0.090 0.254 0.192 0.305 0.004 0.051 0.211 0.166 0.037 0.213 0.086 0.030 0.014 0. Sensitivity of Safety to Hazard Rating on Major Road (HAZRAT1) for Type III INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) HAZRAT 1 0.104 50 100 400 1.101 0.061 0.021 0.162 0.652 0.012 0.008 0.029 0.020 0.000 5.230 0.044 0.006 0.144 1.000 3.059 0.259 0.477 0.696 0.012 0.025 0.952 1.008 0.178 0.000 100 500 1.145 0.121 0.007 0.426 0.003 0.

045 0.279 0..000 5.000 10.000 5.003 0.029 0.001 0.127 0.001 0.000 3.237 0.002 0.000 Note: Group B AADT base model was used.129 0.Table 214.048 0.163 0.058 0.166 0.069 0.012 0.408 0.064 0.061 0.002 0.051 0.209 0.003 0.056 0.014 0.162 0.384 0.362 0.002 0.002 0.113 0.143 0.174 0.011 0.000 100 500 1.035 0.000 3.008 0.254 0.089 0.023 0.007 0.051 0.054 0.010 0.033 0.000 100 500 1.008 0.101 0.082 0.107 0.078 0.003 0. Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type III TOTAL Accidents Per Year 0 50 100 400 1.197 0.040 0.013 0.005 0.083 0.000 10.042 0.319 0.066 0.111 0.094 0.411 0.305 0.000 100 500 1. 400 Major Road AADT (veh/day) Minor Road AADT (veh/day) PKTURCK (percent) 5 10 0.002 0.006 0.269 0.002 0.494 0.126 0.045 0.359 0.317 0.000 3.000 100 500 1.185 0.002 0.000 5.561 234 .467 0.121 0.010 0.224 0.026 0.144 0. The AMF for PKTRUCK was derived from full models.000 3.435 15 0.146 0.072 0.009 0.

025 0.932 30 0.000 5.653 0.004 0.544 0.071 0.526 0.Table 215.033 0.000 100 500 1.043 0.008 0.408 0.141 0.209 0.106 0.723 0.015 0.678 0.347 0.002 0.030 0.000 100 500 1.275 0.109 0.507 0.393 0.018 0.003 0.061 0.000 10.010 0.467 0.020 0.136 0.162 0.024 0.002 0.000 3.121 0.131 0.874 1.101 0.327 0.397 0. The AMF for PKTURN was derived from full models.002 0.051 0.005 0.004 0.259 0.007 0.166 0.066 0.014 0.000 10.422 0.216 0.305 0.269 0.130 0.602 0.082 0.003 0.000 5.000 3.355 0.000 3.079 0.004 0.239 0.005 0.012 0.308 0. Sensitivity of Safety to Peak Turning Percentage (PKTURN) for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 0 0.561 PKTURN (percent) 10 20 0.202 50 100 400 1.109 0.185 0.000 1.254 0.776 0.102 0.003 0.017 0.156 0.201 0.000 100 500 1.214 0.085 0.000 3.085 0. 235 .055 0.168 0.066 0.000 5.000 Note: Group B AADT base model was used.176 0.000 100 500 1.013 0.

46 0.000 100 0.000 2.01 0.09 1.27 6.35 1.02 0.000 0.04 0. The sensitivity test results for these AMFs are presented in tables 216 through 220.02 0.01 0.92 0.06 5. Sensitivity of Safety to Skew Angles for Type IV TOTAL Accidents PerYear Major Road Minor Road SKEW Angle (degrees) AADT AADT (veh/day) (veh/day) 0 10 15 400 50 0.000 4.01 400 0.80 1.06 0. peak through percentage on minor road.44 1.000 1.26 500 0.07 1.49 0.13 500 0.62 0.46 4. Type IV TOTAL Accidents The AMFs derived for TOTAL accidents for Type IV intersections are intersection SKEW angle.95 1.01 0.71 1.000 1.71 1.44 0.01 0.09 0.03 0.59 0.27 7.13 0.11 3.6.37 0.01 0.07 0.06 1.000 100 0.52 0.53 45 0.79 0.000 1.32 5.57 0.17 0.01 0.15 0.02 0.000 100 0.10 0.63 500 1.02 236 .23 0.69 Note: Group B AADT base model was used.12 0.01 1.77 5.01 0.3.66 0. The AMF for Intersection SKEW angle was derived from regression base models.38 3.10 0.42 1.31 4.93 6.48 5.000 100 0.000 3.46 4.000 0.66 3.32 0.13 0. peak truck percentage.31 1.29 0.01 100 0.72 2.31 0.4 Type IV Intersections The predicted TOTAL and INJURY accident frequencies per year for each AMF derived from the full models and regression base models are presented in tables 216 through 224.69 10.84 3.61 5.67 2.92 2.75 1.35 0.24 0.000 0.03 500 0. Table 216.05 3.08 0.19 0.000 0.51 1.02 0.59 1.41 0.01 0.02 0.31 0.62 2.44 3.04 0.87 0.16 0. 30 0.72 0.39 4.93 10.55 0.39 1. and peak left-turn percentage. right sight distance from minor road.08 0.78 2.29 0.02 1.62 8.

032 0.000 3.099 0.615 1.085 0.577 0.015 0. 237 .000 Note: Group B AADT base model was used.092 0.121 0.131 0.375 0.420 0.396 0.940 1.010 0.063 0.399 0.000 100 500 1.964 1.226 0.231 0.195 0.000 5.932 500 0.164 0.444 0.007 0.000 100 500 1.068 0.002 0. Sensitivity of Safety to Right Sight Distance from Minor Road (SDR2) for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) SDR2 0 0.323 0.Table 217.203 3.301 2.021 0.000 100 500 1.278 0.000 5.405 2.003 0.278 0.431 0.145 1500 0.000 3.695 2.435 0.005 0.059 1.245 1. The AMF for SDR2 was derived from full models.206 0.000 0.040 0.697 0.005 0.362 0.015 0.675 0.024 0.000 3.296 0.000 10.005 50 100 400 1.455 4.006 0.093 1.512 2.622 1.077 0.535 0.658 3.006 0.079 0.177 0.003 0.912 1.005 0.050 0.288 2.000 10.035 0.000 3.144 0.000 100 500 1.000 5.567 0.188 0.055 0.013 0.081 1.463 1.222 0.048 0.974 4.906 0.191 0.253 0.022 0.488 0.010 0.

093 1.277 0.157 0.006 0.055 15 0.455 4.000 10.118 0.003 0.015 0.512 2.034 0.000 3.055 0.000 100 500 1.140 3.004 0.677 0.447 0.226 0.049 0. The AMF for PKTRUCK was derived from full models.226 0.000 100 500 1.489 0.190 2.062 0.079 0.000 5.404 238 .093 0.303 0.533 0.254 0.004 0.200 0.463 1.343 0.737 1.287 0.000 3.882 10 0.005 0.323 0.000 100 500 1.719 3.099 0. 400 Major Road AADT (veh/day) Minor Road AADT (veh/day) PKTRUCK 5 0.296 1.015 0.516 0.567 0.365 0.150 0.078 0.000 3.212 0.435 0.061 0.937 1.684 2.140 0.191 0.007 0.000 5.270 0.880 0.118 0.043 0.003 0.048 0.038 0.420 0.692 0.860 1.932 0.351 0.178 0.834 1.000 100 500 1.656 0.024 0.110 0.Table 218.012 0.385 0.658 3.059 1.009 0. Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type III TOTAL Accidents Per Year 0 50 100 400 1.000 3.007 0.092 2.000 5.646 2.012 0.019 0.000 Note: Group B AADT base model was used.000 10.005 0.027 0.622 1.

051 0.011 0.410 50 100 400 1.622 1.306 2.191 5.024 0.010 0.455 4.005 0.373 4.763 1.567 0.797 1.000 100 500 1.822 2.079 0.000 5.312 2.009 0.226 0.000 10.025 0.000 3.939 2.032 0.290 0.127 0.166 0.016 0.000 10.414 0.000 100 500 1.313 0.000 3.059 1.402 0.610 7.372 0.000 100 500 1.093 1.091 0.023 1.116 30 0.488 3.512 2.071 0.000 5.293 10.401 1.432 5.327 8.934 0. Sensitivity of Safety to Peak Through Percentage on Minor Road (PKTHRU2) for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) 0 0.209 0.000 3.919 1.236 3.420 0.099 0.012 0.031 0. The AMF for PKTHRU2 was derived from full models.463 1. 359 1.117 0.015 0.244 0.337 0.000 5.477 0.743 1.055 0.323 0.000 3.007 0.000 100 500 1. 239 .020 0.000 Note: Group B AADT base model was used.685 6.163 0.130 0.998 0.007 0.435 0.681 1.198 0.101 0.978 2.Table 219.191 0.040 0.409 4.531 0.798 1.716 0.932 PKTHRU2 10 20 0.658 3.558 0.728 0.595 0.

688 0.000 3. 240 .099 0. Sensitivity of Safety to Peak Left-Turn Percentage (PKLEFT) for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) PKLEFT 0 0.283 6.005 5.000 100 500 1.124 0.Table 220.015 0.000 3.512 2.344 6.010 0.547 0.727 2.000 10.019 0.781 1.202 20 0.462 7.048 0.055 0.197 0.079 0.323 0.000 5.921 2.005 0.191 0.897 0.000 100 500 1.804 50 100 400 1.983 1.342 4.901 3.128 0.008 0.069 0.284 0.420 0.000 5.007 0.332 1.007 0.245 0.642 1.099 0.797 30 0.455 4.510 0.240 0.157 0.125 0.449 0.390 4.009 0.000 10.024 0.202 5.786 0.868 9.567 0.105 3.000 Note: Group B AADT base model was used.030 0.583 1.932 10 0.374 1.030 0.024 0.000 5.093 1.015 0.406 0.823 0.000 100 500 1.226 0.435 0.379 0.235 2.713 0.012 0.301 0.622 1.674 2.087 0.156 0.357 0.172 2.000 3. The AMF for PKLEFT was derived from full models.865 1.000 3.463 1.038 0.733 1.000 100 500 1.658 3.059 1.110 0.

317 0.019 0.456 5.617 2. The sensitivity test results for these AMFs are presented in tables 221 through 224.271 1.004 0. 30 0.000 0.095 0.208 0.020 0.229 0.064 0.012 0.005 0.004 0. Table 221.311 500 0.000 0.196 3.000 1.016 0.698 2.034 0.376 0.077 0.017 500 0.006 0.713 1.134 0.301 0.181 0.000 0.359 1.323 5.003 0.153 0.081 0.039 2.143 1.055 0.010 0.036 1.000 0.835 10.Type IV INJURY Accidents The AMFs derived for INJURY accidents for Type IV intersections are intersection SKEW angle.131 500 0. Sensitivity of Safety to Skew Angles for Type IV INJURY Accidents Per Year Major Road Minor Road SKEW Angle (degrees) AADT AADT (veh/day) (veh/day) 0 10 15 400 50 0.010 1.131 0.932 241 .183 0.223 1.032 1.359 0.000 0.792 45 0.347 0.042 0.046 0.675 2. major right-turn lane.215 0.713 0.006 0.004 100 0.000 100 0.000 0.983 1.291 0.046 0. peak truck percentage.571 0.899 0.513 Note: Group B AADT base model was used.413 0.006 400 0.000 1.024 0.771 10.159 0. and peak left-turn percentage.679 0.603 0.058 0.000 0.167 0.007 0.050 3.884 3.970 1.000 100 0.145 0.718 0.611 5.033 0.590 0.141 2.345 2.000 100 0.087 0.253 0.218 0.069 500 0.005 0.012 0.646 1.514 0.408 0.372 3.122 0.346 0. The AMF for Intersection SKEW angle was derived from regression base models.433 0.826 0.004 0.248 0.301 0.000 0.719 0.012 0.040 0.754 1.000 100 0.857 0.431 0.363 0.

835 0.082 0.002 0. Sensitivity of Safety to Major Right-Turn Lane for Type IV INJURY Accidents Per Year Right-Turn Lane on Major Road Major Road Minor Road Right-Turn Lane on Right-Turn Lane on Two AADT (veh/day) AADT (veh/day) No Right Turn One Approach Approaches 400 50 0.223 1.028 0.675 1.018 1.370 0.071 1.000 0.000 0.442 0.046 0.040 0.208 0.000 100 0.905 10.159 5.003 0.213 0.021 0.009 500 0.302 5.000 100 0.034 500 0.065 500 0.514 0.052 0.436 3.000 100 0.319 1.431 0.718 5.000 1. 242 .087 0.185 0.006 0.000 0.183 3.024 0.178 0.240 Note: Group B AADT base model was used.441 1.005 1.590 0.000 0.033 0.003 0.000 0.Table 222.134 1.248 0.156 0.970 0.095 0.097 3.007 0.000 0.012 0.131 0.112 0.000 0.000 1.181 0.380 10.215 0. The AMF for Major Right-Turn Lane was derived from full models.408 0.003 400 0.000 100 0.004 0.002 100 0.024 3.154 500 0.351 0.010 0.075 0.000 0.507 0.

114 0.002 0.943 1.000 Note: Group B AADT base model was used.727 0.057 0.306 0.044 0.000 10.007 0. 400 Major Road AADT (veh/day) Minor Road AADT (veh/day) PKTRUCK 5 0.208 0. The AMF for PKTRUCK was derived from full models.099 0.996 15 0.005 0.256 0.166 0.014 0.292 10 0.078 0.011 0.748 0.015 0.408 0.160 0.007 0.577 0.123 0.000 3.004 0.060 0.101 0.002 0.128 0.027 0.087 0.675 0.270 0.083 0.131 0.215 0.Table 223.314 0.514 0.001 0.000 3.052 0.236 0.000 3.024 0.021 0.108 0.431 0.035 0.181 0.000 5.003 0.000 5.025 0.004 0.000 3.970 1.191 0.396 0.455 0.248 0.003 0.000 100 500 1.223 1.445 0.332 0.033 0.067 0.012 0.095 0.590 0.197 0.074 0.000 100 500 1.242 0.000 100 500 1.009 0.003 0.040 0.046 0.002 0.147 0.000 100 500 1.768 243 .139 0.187 0.000 10.561 0. Sensitivity of Safety to Peak Truck Percentage (PKTRUCK) for Type IV INJURY Accidents Per Year 0 50 100 400 1.005 0.000 5.020 0.351 0.002 0.095 0.019 0.

087 0.012 0.000 10.007 0.147 0.000 100 500 1.957 2.408 0.705 1.Table 224.069 0.970 1.768 30 0.590 0.094 0.000 3. The AMF for PKLEFT was derived from full model 244 .223 1.131 0.011 0.867 0.159 0.660 5.726 0.997 2.305 0.272 0.004 0.006 0.221 0.012 0.995 1.020 0.000 100 500 1.003 0.431 0.418 0.160 1.000 5.482 4.000 3.161 0.675 10 0.000 5.033 0.869 1.419 0.000 Note: Group B AADT base model was used.372 0.221 0.056 0.000 100 500 1.181 0.068 2.591 1.248 0.020 0.637 2.041 0.189 1.044 50 100 400 1.248 0.226 1.628 1.013 0.463 0.831 4.064 2.612 0.762 3.688 0.000 5.678 2.000 3.515 0.000 10.033 0.116 0.046 0.024 0.033 0.875 8.468 0.131 0.033 0.005 0.000 100 500 1.008 0.363 0.458 0.826 20 0. Sensitivity of Safety to Peak Left-Turn Percentage (PKLEFT) for Type IV INJURY Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) PKLEFT 0 0.078 0.095 0.215 0.056 0.208 0.018 0.514 0.000 3.350 0.

37 4. Type V TOTAL Accidents The AMFs derived for TOTAL accidents for Type IV intersections are commercial driveways on major road.89 7.37 16.49 7.44 7.03 9.79 4.76 7.59 9.27 2.33 2.40 12. The AMF for Commercial driveways on major road was derived from full models.07 1.05 1.000 10.20 12.45 12.40 1.21 50 100 400 1.70 4.16 5.28 18.13 9.90 1.15 6.74 3.73 2.14 10.40 5.91 14. Table 225.71 10.41 2.000 10.02 17.000 100 500 1.40 4.05 5. Sensitivity of Safety to Commercial Driveways on Major Road for Type V TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) Commercial Driveways on Major Road (density) 0 5 10 15 0.62 0.000 3.36 1. horizontal curve combinations on major and minor roads.80 2.65 8.97 3.38 1.37 7.000 5.000 5.87 5.46 5.89 1.00 3.04 8.96 7.09 13.000 3.64 4.84 7.000 3.70 9.90 3.66 13.82 1.02 9.24 2.84 5.000 5.03 3.000 3.94 1.22 2. and light.88 5.000 100 500 1.3.000 Note: Group A AADT base model was used.10 5.94 13.57 3.84 1.22 8.33 10.72 0.24 1.31 6.6.29 1. 245 .96 2.19 5.94 3.50 3.20 21.36 7.18 9.69 2.5 Type V Intersections The predicted TOTAL and INJURY accident frequencies per year for each AMF derived from the full models and regression base models are presented in tables 225 through 229.93 5.22 4.000 100 500 1.000 100 500 1.54 9.96 6.08 6. The sensitivity test results for these AMFs are presented in tables 225 through 227.

75 0.95 3.92 5.Table 226.81 4.60 2.69 6.000 Note: Group A AADT base model was used.17 2.82 1.59 1.67 5 0.16 2.66 5.000 100 500 1.39 6.70 1. Sensitivity of Safety to Horizontal Curve Combinations on Major and Minor Roads (HEICOM) for Type III TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) HEICOM 0 0.54 0.65 5.33 9.94 2.91 3.59 8.91 1.40 5.45 3 0.75 5.96 1.79 0.71 0.03 4.000 3.30 1.000 100 500 1.61 0.000 3.03 5.000 3.11 4.75 3.02 4.66 0.10 5.71 6.24 3.66 3.12 2.49 6. 246 .000 5.64 8.05 1.00 3.000 10.19 1.73 1.90 3.30 4.96 2.55 3.60 2.40 4.88 4.25 2.000 5.10 3.000 3.62 0.57 7.06 2.37 5.42 3. The AMF for HEICOM was derived from full models.56 4.68 2.72 0.08 50 100 400 1.21 8.36 3.89 4.97 1.000 5.76 5.000 10.50 1.25 7.96 7.18 10 0.12 1.12 3.40 3.47 1.000 100 500 1.22 7.07 4.81 2.80 2.29 1.51 3.000 100 500 1.55 4.

96 2. Sensitivity of Safety to Light for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) Presence of Light No 0.42 3.000 3.55 3.12 2.33 9.25 2.73 1.Table 227.30 1.69 6.00 3.45 Yes 0.47 1. The AMF for LIGHT was derived from full models.68 2.24 3.16 2.000 3.25 7.67 5.000 10.10 5.000 Note: Group A AADT base model was used.000 5.88 4. 247 .40 4.17 2.97 1.22 7.40 5.000 100 500 1.49 6.59 8.79 0.000 100 500 1.09 50 100 400 1.03 4.000 3.000 10.000 5.90 3.30 4.000 100 500 1.11 4.000 3.72 0.91 3.000 100 500 1.29 1.89 4.37 5.61 0.07 4.000 5.54 0.05 1.82 1.

76 0.51 0.41 0.000 5.74 0.53 0.52 1.14 0.67 0.000 3.66 0.000 5.76 1.000 100 500 1. The AMF for HEI2 was derived from full models.47 0.21 1.86 1.97 1.94 0.76 1.57 0.90 1.90 0.000 3. 248 .35 0.36 0.86 2.12 1.52 0.50 0.39 0.68 0.58 0.02 2.99 1.01 1.48 5 0.02 1.58 0.57 1.32 1.67 0.30 0.000 10.91 1.82 1.38 0.35 1.03 50 100 400 1.62 0.32 1.31 0.82 0.66 0.000 3.52 0.70 1.56 2.49 1.000 100 500 1.000 Note: Group A AADT base model was used.99 1.70 0. The sensitivity test results for these AMFs are presented in tables 228 to 229.14 1.84 0.000 100 500 1.88 1.32 0.34 10 0.000 3.57 0.32 0.14 1. Sensitivity of Safety to Horizontal Curves on Minor Road (HEI2) for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) HEI2 0 0.45 0.26 0.000 5.29 1.000 100 500 1.43 0.99 1.67 0.39 0.17 1. Table 228.70 3 0.78 0.000 10.78 0.Type V INJURY Accidents The AMFs derived for INJURY accidents for Type V intersections are horizontal curves on minor roads and light on intersections.07 1.71 1.78 1.98 2.46 0.21 1.72 2.43 1.10 2.17 1.29 2.

000 10.36 1. the recalibration efforts were focused on this application. In this sense. the algorithm was not developed and the calibration philosophy was somewhat different.66 0.67 0.000 100 500 1.56 2.51 0.000 3.90 1.7 SUMMARY.94 0.000 100 500 1.88 1. Sensitivity of Safety to Light for Type IV TOTAL Accidents Per Year Major Road AADT (veh/day) 400 Minor Road AADT (veh/day) Presence of Light No 0.28 0.51 0.29 2. DISCUSSION. the research reflected here represents a different perspective from the original calibration by Vogt since.71 1.000 5.32 1.000 3.02 2.17 1. Because the models are required for use in the IHSDM accident prediction algorithm. The AMF for LIGHT was derived from full models.12 1. To remind the reader. Type II: Four-legged stop controlled intersections of two-lane roads.81 50 100 400 1.000 Note: Group A AADT base model was used.78 1.60 0.02 0.000 3. the five intersection types examined are: • • • • • Type I: Three-legged stop controlled intersections of two-lane roads.30 0.15 1.75 0.52 0. 249 .45 0.76 0.35 0.60 0.000 10. AND CONCLUSIONS In summarizing and discussing the model recalibration exercise. Type III: Three-legged stop controlled intersections with two lanes on the minor and four lanes on major roads.35 0.000 100 500 1. 3.90 0.53 1. Models with a comprehensive set of regression parameters were to be used directly for predicting the expected number of accidents at intersections and for deriving AMFs for the five types of intersection crash models examined in this research.24 0. at that time.88 1.32 1.000 5.52 0.76 1.41 0.000 3.04 1.Table 229.39 0.99 1. Type V: Signalized intersections of two-lane roads.58 0. it is important not to lose sight of the fundamental purpose of the statistical models that are the focus of this research. Type IV: Four-legged stop controlled intersections with two lanes on the minor and four lanes on major roads.000 100 500 1.000 5.70 Yes 0.

The full models.The accident prediction algorithm enables the number of total intersection-related accidents per year to be estimated by multiplying the predicted number of such accidents for base conditions by AMFs for various features specific to an intersection. The Harwood et al. two fundamental classes of statistical models were developed—models using AADT as the sole predictor of crashes (referred as base or AADT models) and models containing a fairly comprehensive set of predictor variables (referred as full models). full models were developed in keeping with the original intent of the project and with the expectation that it may be possible to derive AMFs similar to what were accomplished in the earlier FHWA work. For each intersection type. such as sources of published results for the earlier models and comparison tables.2) Understanding what was required for the recalibration effort is an improvement of the base models and AMFs to be used in the IHSDM accident prediction algorithm—including possible enhancements to model functional forms.and four-legged stop controlled with four lanes on the major and two on the minor.and four-legged intersections of two-lane roads. At the same time.25 m (250 ft) of the intersection center. and updated parameter estimates.and four-legged intersections of two-lane rural roads with STOP control. and translating these results into meaningful observations and conclusions. These details of the models calibrated and data used are summarized in table 230. while Vogt documents models for three other types of rural intersections: three.1 Model Recalibration For the five intersection types previously described. are presented in two FHWA reports. Vogt and Bared present models for three.(1. Those projects also developed full models with additional variables with the intention of using the variable coefficients to estimate AMFs for use in IHSDM. These base models were the best of available accident prediction models developed in the earlier Vogt FHWA projects and included only variables that were statistically significant at the 15 percent level. statistical models were developed for total accidents (TOTACC) and injury (fatal + nonfatal injury) accidents (INJACC) within 76. and signalized intersections of two-lane roads. The discussion provided here focuses primarily on summarizing and briefly discussing the results of the recalibration detailed in the body of this report. Not all levels were calibrated for all model Types (I to V). The reader interested in additional details. This requirement guided the approach. “Red Book” presented base models and AMFs for three. addition or exclusion of variables. and four-legged signalized intersections of two-lane roads. There are two levels for each class. along with several variants. should refer to the earlier sections of this report. Descriptions of all variable abbreviations and definitions used in this report can be found at the beginning of this document. 250 .7. 3.

This research strongly supported their use. and Georgia validation data. models were calibrated using all available data from the HSIS California database. Second. original sites from Minnesota and Michigan. Michigan. the likely application is to apply them for forecasting crashes across the entire country. The reasoning behind this apparently simplistic approach is that models with AADT as the only predictor are more likely to be transferable across jurisdictions than models that include other variables. models were developed for a subset of these sites that met specified conditions for possible use as base models in the IHSDM accident prediction algorithm. however. This is an appealing feature. Type V California project Michigan Georgia California HSIS Michigan Georgia Not calibrated Not calibrated California project Michigan Georgia AADT Models Overview The AADT statistical models include AADT as the sole predictor variable and are proposed for consideration as base models to be used in IHSDM. and V were not used. 251 . and three types for Type III. including Minnesota Michigan Michigan California HSIS* Georgia Georgia Georgia Group B Sites: Full Model with California HSIS variables from Minnesota California HSIS Georgia Not calibrated Not calibrated Subset of Group B Sites: AADT Model for sties meeting base conditions for California HSIS California HSIS California HSIS project data plus Minnesota Michigan Michigan California HSIS* Georgia Georgia Georgia Subset of Group A Sites: AADT Model for sites meeting base California project California project conditions for Michigan Michigan project data Georgia Georgia Not calibrated * The California project data for Types II. IV. additional AADT models were calibrated from a dataset that met the base conditions of the significant variables in the full models. and V sites. Two types of AADT models are presented for Type I and II sites. and V sites. and the sites used for the Georgia validation data. IV. These AADT-only models were calibrated on data from the original sites (intersections) from California. IV. For Types III.Table 230. considering that the models are calibrated on data from three States within the United States. First. Summary of Models Recalibrated and Data Used Model Site Types and Data Used Description Types I and II Type III Type IV Group A Sites: Full Model with California project California project variables in project Minnesota Michigan Michigan data Georgia Georgia Georgia Group B Sites: AADT Model for California HSIS California HSIS California HSIS all sites.

For the INJACC model the overdispersion was similar for the two AADT models. full models were developed using two groups of data. with certain States possessing intersections that systematically share geometric traits. Group A. was comprised of the sites from Minnesota and Georgia and consisted of many variables. more geometric variables were statistically significant. with the intent to explain as much of variation in crash occurrence as possible. The second. For the Group A models. posted speed limit on major roads and the angle variable HAU were not included in the recalibrated model. There was no equivalent to Group B of the Types I and II models because there were very few HSIS sites. was comprised of the sites from Minnesota and Georgia and consisted of many variables. but fewer variables were available for modeling. For 252 . and the Georgia sites. For the TOTACC model. Type I Full Models (see table 160. including the additional years of accident data. table 161. The first. For Types III. For total accidents. when a State indicator variable was used in the models. The California sites were not in this group because many of the variables were not available. given the available set of potential explanatory variables.Full Models Overview Statistical models with relatively comprehensive sets of predictor variables were also developed. For Types I and II. Without the State indicator. One includes a State indicator term and the other does not. while right-turn lanes on minor roads and left-turn lanes on major roads were included. This effect suggests that geometric variables are correlated with State of origin. Unlike their AADT counterparts. Georgia. Group B consisted of the Minnesota. because the base condition sites should be more homogeneous in their design characteristics. which resulted in fewer variables being available for modeling. these models include many variables. including horizontal and vertical curvature. Group B. including horizontal and vertical curvature. (such a model was not presented in the Vogt reports). Group A. The CURE plots confirmed the superiority of the chosen model form. The first. There were some similarities and differences between the recalibrated models and the Vogt and Bared models. added California HSIS sites. Type I Model Results Type I AADT Models (see table 134 and table 135) The recalibrated models for total accidents represent improvements to the one reported in Harwood et al. the data used to calibrate full models consisted of the California and Michigan sites from the original study. the only geometric variable that proved to be significant was HI1. and California HSIS sites. This was expected. Two model variants are reported. IV.(3) The β coefficient of the log of major road AADT is about two times that for minor road AADT. which seems to be a reasonable expectation on the basis of other models reported in the literature. the base condition model (calibrated from data that met specified base conditions) was estimated with a lower overall overdispersion parameter than the model using all sites. and these had almost no variables of interest. which testifies to the reasonableness of the calibrated models. and table 162) Full models were developed using two groups of data. and V.

significant variables (in addition to major and minor road AADTs) include right-turn lanes on major roads for the variant with the State indicator variable. medians and right-turn lanes on major roads were significant for both the State indicator and non-State indicator variants. table 165.25 m (250 ft) of the intersection center were significant at the 10 percent level or better for both the State indicator and non-State indicator variants. In the case of the Group B models. there were some similarities and differences between the recalibrated models and the Vogt and Bared models. which appears to be reasonable. Group A. Type II Model Results Type II AADT Models (see table 135 and table 139) Unlike the case for Type I models. added California HSIS sites. The first.injury accidents. significant geometric variables at approximately the 10 percent level of significance or better for TOTACC included right-turn lane on major roads and the number of driveways for the model l variant. right-turns on major roads. For the recalibrated models. the models do a poorer job for the Georgia and California sites. For the INJACC models. As for Type I. For the Group A models. and right. The GOF as measured by the overdispersion parameter was improved over the Vogt and Bared models. the β coefficient of the log of major road AADT is about 20– 30 percent higher than that for minor road AADT. just one of numerous GOF measures. the GOF for 253 . Group B. For INJACC. number of driveways on major roads. The second. Nevertheless. including a State indicator variable. The GOF as measured by the overdispersion parameter was not as good as that for the Vogt and Bared model. with and without the State indicator variable. The Vogt and Bared model also included the angle variable HAU. The Vogt model included roadside hazard rating as a significant variable and others that were not significant. was comprised of sites from Minnesota and Georgia and included variables such as horizontal and vertical curvature. and table 166) As was the case for Type I. Two model variants were reported. full models were developed using two groups of data. the recalibrated Type II models for total accidents have more overdispersion than models reported in Harwood et al. number of driveways and horizontal curvature within 76. and the horizontal curvature variable HI1. again not surprising. and number of driveways and the vertical curvature variable VCI1 for the variant without the State indicator variable. for TOTACC. the base condition models were estimated with a lower overdispersion than the model using all sites. table 164. (such a model was not presented in the Vogt reports). For both of the Group B models. Type II Full Models (see table 163. posted speed on major. and the angle variable HAU were not included while left-turn lanes on major roads was included. For both the TOTACC and INJACC models. which resulted in fewer variables being available for modeling. the major road posted speed. one included a State indicator term and the other did not.(3) In particular. because the base condition sites should be more homogeneous in their design characteristics.and left-turn lanes on major roads for the variant without the State indicator variable. Again. right-turn lanes on minor roads and left-turn lanes on major roads are significant in addition to rightturn lanes on major roads. although these last two were of low significance.

which had an angle variable HAU as the only geometric variable. Type III Model Results Type III AADT Models (see table 142. 254 . for both TOTACC and INJACC. State indicator variables were again statistically insignificant in the recalibration. For TOTACC. The other model was the one judged to be next best in terms of these measures. which is in accord with reasonable expectations. lowest overdispersion.all of the Group B models. roadside hazard ratings on major roads. once again. the β coefficient of the log of major road AADT is about three to four times that for minor road AADT. The recalibrated models for both TOTACC and INJACC produce better GOF measures than the Vogt’s models. median width on major roads. the GOF statistics were better for the models using base condition sites than for models using all sites. State indicator variables were statistically insignificant in the recalibration. For the Group A models. major and minor AADTs. crest curve rates on major roads. only median widths on major roads and the AAADT variables were included in the Vogt model. and MAD per year. In the case of the Group B models. is reasonable. commercial driveways on major roads. None of these were included in the Vogt model. For INJACC. the β coefficient of the log of major road AADT is significantly larger than that for minor road AADT. intersection angles. for TOTACC. major and minor AADTs. and table 145) Unlike models for Type I and II intersections. which specified these variables as separate terms. Type III Full Models (see table 167 and table 168) Two models each are reported for TOTACC and INJACC. except for the overdispersion parameter. MPB per year. as measured by the overdispersion parameter. intersection angle. which. Of these. As expected. table 143. Like TOTACC. For INJACC. the AADT variable effect was captured as the product of the major and minor AADTs as opposed to the TOTACC model. and painted medians on major roads were found to be significant in the main model. peak turning percentages. commercial driveways on major roads. The recalibrated models for TOTACC and INJACC generally have better GOF measures than comparable models from the earlier FHWA research. but that model did have a driveway variable DRWY1 instead of COMDRWY1. and peak truck percentages were found to be significant in the main model. was not as good as that for the Vogt and Bared model. The main model was selected based on the highest Pearson product-moment correlation coefficient. a State indicator variable was insignificant for all models.

a Michigan indicator variable was found to be significant. table 149. and speed limits on minor roads were significant in the main model. conforming to expectations. as seems reasonable. peak truck percentages. indicating more influence of the Michigan data on the model. the β coefficient of the log of major road AADT is about one and one half times that for minor road AADT. peak through percentages on minor roads. By contrast. The main model showed the lowest overdispersion among candidate models and indicated the best GOF results. Again. which specified these variables as separate terms. major and minor road AADTs. For the TOTACC model using all sites. MPB per year. The Vogt model contained a speed limit on minor road variable SPD2 and a peak left-turn percentage on major road variable PKLEFT1 in addition to the AADT variables. the GOF statistics were improved for the models using base condition sites compared to models using all sites. the β coefficient of the log of major road AADT is significantly larger than that for minor road AADT for both TOTACC and INJACC. the Vogt main model included only a peak left-turn percentage on major road variable PKLEFT1 and a left-turn lane on major road variable LTN1S in addition to the AADT variables. 255 . lowest overdispersion. For TOTACC. as opposed to the TOTACC model. The recalibrated models for TOTACC and INJACC resulted in generally improved GOF measures than comparable models from the previous FHWA research. As expected. For INJACC. a State indicator variable was not statistically significant for all models. The AADT variable effect for the INJACC model was captured as the product of the major and minor AADTs. and table 151) Unlike models for Type I and II intersections. peak left-turn percentages. The variant model yields an improvement in overdispersion and Pearson product-moment correlation coefficient. Type IV Full Models (see table 170) Two models were presented for both TOTACC and INJACC. but for the variant model. but not in MPB per year and MAD per year. The overdispersion values of the recalibrated models were lower than Vogt’s for TOTACC but slightly higher for INJACC. the main model was selected based on the highest Pearson product-moment correlation coefficient. the small sample size was insufficient to provide good coefficients and p-values for a model with specified base conditions. The State indicator variable was statistically insignificant in the main model. In the case of the Group B models. major and minor road AADTs. State indicator variables were statistically insignificant. peak left-turn percentages on major roads. and right-side sight distances on minor roads were found to be significant in the main model. For the Group A models.Type IV Model Results Type IV AADT Models (see table 148. The recalibrated models for TOTACC and INJACC generally provide better GOF measures than Vogt’s models. peak truck percentages. and MAD per year.

In the case of the Group B models. in accord with expectations. For INJACC. the AADT variable effect was captured as the product of the major and minor AADTs as opposed to the TOTACC model. The variant model provides an improvement over the main model in overdispersion and Pearson product-moment correlation. for the INJACC model. Again. Because the IHSDM requires the main model to be recalibrated to work in any State. Therefore. peak left-turn percentages on minor roads. the β coefficient of the log of major road AADT is about two to three times greater than that for minor road AADT. peak truck percentage PKTRUCK. and MAD per year. it does include a Michigan indicator variable. but not in MPB per year and MAD per year. The recalibrated models for both TOTACC and INJACC provide a better GOF measures than Vogt’s models. the Vogt main model had a completely different set of non-AADT variables: peak left-turn percentage on minor road PKLEFT2. 256 . the main model was selected based on the highest Pearson product-moment correlation coefficient.Type V Model Results Type V AADT Models (see table 154. peak left-turn percentage on minor road PKLEFT2. MPB per year. Although the recalibrated model variant was superior to the main model in terms of lower overdispersion and better fit to the data. the Vogt report does not provide AADT-only models for Type V. and speed limits on major roads were significant variables in the main model. a comparison between the Vogt models and the newly calibrated AADT models for TOTACC and INJACC could not be done. except for the overdispersion parameter. presence of lighting. protected left lane PROT_LT. lowest overdispersion. In addition. table 155. and vertical curvature VEICOM. and horizontal curvature variables were found to be significant in the main model. By contrast. and table 156) Statistical models could not be calibrated for specified base conditions due to lack of data. Unlike the case for the other model types. for TOTACC. speed limits on major roads. For TOTACC. peak truck percentages. and vertical curvature VEICOM. For the Group A models. As was the case for the Type III and IV models. major and minor AADTs. protected left lane PROT_LT. which specified these variables as separate terms. commercial driveways on major roads. major and minor AADTs. the AADT variable effect was captured as the product of the major minor AADTs. as opposed to specifying these variables as separate terms in the recalibrated model. Type V Full Models (see table 171 and table 172) Two models were developed for both TOTACC and INJACC. Again. similar to what was done for Types I and II. the main Type V TOTACC and INJACC models calibrated using all sites have a β coefficient of the log of major road AADT. the model with the State indicator was selected as a variant and not as the main model. about two to three times that for minor road AADT. which means more influence of the Michigan data on the model. which is reasonable. the Vogt main model included a completely different set of non-AADT variables: peak truck percentage PKTRUCK. presence of lighting.

SKEW was estimated as statistically significant in the regression models.” those from Harwood et al.’s 2002 report.7. to infer AMFs from full models. 2. in 2000 and 2002. the AMF estimates developed were of the same direction of effect and reasonably close in magnitude to those provided by Harwood et al. no variables showed any statistically significant impacts on safety in the regression model. 257 . so an AMF could not be estimated for the effect of this variable. For Type V intersections. None of the variables used showed any significant impacts on safety for Type V sites. Tables 231. similar to the technique used by Vogt. There were only few sites with deficient sight distance. sites in this dataset had turning lanes on both approaches. and 234 compare the AMFs from the “Red Book. An attempt. Implementation of a relatively untested regression analysis procedure.2 Summary of AMFs Several strategies for assessing and recalibrating the AMFs corresponding with the five intersection models were explored. and those derived during the course of this research. 233. 232.3. For Type III and IV intersections. Independent variables were factors at the site that differed from conditions assumed in base or AADT model. The dependent variable was the difference between a site’s observed accident count and its AADTbased model prediction. Whereas the “Red Book” provides separate AMFs for major road right-and left-turn lanes at Type II intersections. In general. Right-turn lanes on major roads provided statistically significant AMFs for Type IV intersections. It is believed that a SKEW AMF significantly different than 1 was not supported by the data. and separate effects could not be detected for one versus two approaches. including: 1.

0101SKEW) 1.00 if none exist 0.016*SKEW)/ 1+(0.43+0.1081VEI1) Not calibrated exp (0.053*SKEW)/ (0.0539 COMDRWY1) 1 1 1 1 258 VEI1 MEDWIDTH1 MEDTYPE1 PKTRUCK PKTHRU2 PKLEFT SDR2 LIGHT HEICOM Not calibrated Not calibrated 0.975 on one approach (0.96 on one approach) 0.016*SKEW) (1.0249PKTHRU2) Not calibrated exp (0.0681COMDRW Y1) exp(0.71 1 exp (0.92 on both approaches) Not calibrated Not calibrated 1.053*SKEW) 1 RT MAJ LT MAJ SIGHT DISTANCE COMDRWY1 1 exp(0.0288HEICOM) .0003SDR2) Not calibrated Not calibrated 1 1 1 0.0229PKLEFT) exp(-0.82 on one approach 0.98+0.0479PKTRUCK) exp (0.00 if none exist 0.67 on both approaches 1 exp (0.95 on both approaches (0.73 exp (-0.0106MEDWDT H1) None provided Not calibrated 0. Comparison of Type III–V AMFs for TOTACC AMF “Red Book” Type III-IV SKEW Type V AMFs Derived From Full Models Type III Type IV Type V AMFs Derived From Regression Models Type III Type IV Type V 1+(0.Table 231.75 exp (-0.

19.86 on one approach) 0.77 259 .004SKEW) exp(0.Table 232.10 if limited in 2 quadrants 1.20 if limited in 4 quadrants Not calibrated Not calibrated Not calibrated Not calibrated Not calibrated Not calibrated Not calibrated Not Not calibrated calibrated Not calibrated Not calibrated MEDIAN Not calibrated 1 Group A = 1.0.90 on both approaches (0.71 LT MAJ SIGHT DISTANCE RT MIN HI1 DRWY1 1. Group B = 0.74 on both approaches) RT MAJ 0.861 1 0.76 on one approach (0.19.58 on both exists approaches (0. Comparison of Type I–II AMFs for TOTACC AMF Type I SKEW “Red Book” Type II Full Models Type I Type II Not Not calibrated calibrated Regression Models Type I 1 1 Type II exp(0.15 if limited in 3 quadrants 1.71 Not calibrated 1 0.56 if at least one (0.88 1.95 on one approach (0.05 if limited in 1 quadrants 1.52 on both exists) approaches) 0.86 Not calibrated Not exp(0.82 1.0054SKEW) 1.48 Not calibrated Not calibrated 1 1 Not calibrated Not calibrated 0.0263) calibrated Not calibrated 1.00 if none exist approach) 0.35 1.00 if none exist 0.00 if none exist 0.13 Not Not calibrated calibrated 1.78 if at least one 0.72 on one 1.

82 on one approach 0.1081VEI1) Not calibrated exp (0.0106MEDWDTH1) 0.0288HEICOM) Not calibrated Not calibrated .0101SKEW) 1.92 on both approaches) Not calibrated 1.0539 COMDRWY1) 1 1 1 1 260 VEI1 MEDWIDTH1 MEDTYPE1 PKTRUCK PKTHRU2 PKLEFT SDR2 LIGHT HEICOM exp(0.95 on both approaches (0.0681COMDRWY1) None provided Not calibrated Not calibrated 1 1 1 LT MAJ SIGHT DISTANCE COMDRWY1 0.053*SKEW) RT MAJ 1 exp(0.053*SKEW)/ (0.98+0.67 on both approaches 1 exp (0.016*SKEW)/ 1+(0.016*SKEW) (1.75 exp (-0.Table 233.975 on one approach (0.96 on one approach) 0. Comparison of Type III–V AMFs for TOTACC “Red Book” AMF SKEW Type III-IV Type V Type III AMFs Derived From Full Models Type IV Type V AMFs Derived From Regression Models Type III Type IV Type V 1 1+(0.0479PKTRUCK) exp (0.0229PKLEFT) exp(-0.73 Not calibrated exp (-0.0003SDR2) Not calibrated 0.43+0.00 if none exist 0.71 1 exp (0.0249PKTHRU2) Not calibrated exp (0.00 if none exist 0.

87 0.0286) 0.017SKEW) (0. Comparison of AMFs for INJACC “Red Book” AMF SKEW RT MAJ LT MAJ RT MIN SIGHT DISTANCE HI1 DRWY1 Type I-V Type I Type II AMFs Derived From Full Models Type III Type IV Type V 1 1 1 Not calibrated Not calibrated exp (0.0520PKTRUCK) 1 0.56 Type I 1 1 0.09 Not calibrated Not calibrated 1.0284HEI2) .0408) 1.86 one approach.Table 234.0163SKEW) 0.048SKEW) 0.048SKEW)/ (0.8 1.0523PKLEFT1) Not calibrated Not calibrated 0.36 Not calibrated exp (0.42 1 AMFs Derived From Regression Models Type II Type III Type IV 1 1 1 1 1 Type V Not Not calibrated calibrated exp(0.0627COMDRWY1) exp (0.67 exp (-0.52+0.52 Not calibrated Not Not calibrated calibrated Not calibrated Not calibrated Not calibrated exp(0.017SKEW)/ 1+(0. 1 0.85 1+(0.74 both approaches 1 1 1 1 1 1 Not Not calibrated calibrated 261 MEDIAN COMDRWY1 HAZRAT1 PKTRUCK PKTURN PKLEFT1 SDR2 LIGHT HEI2 None provided Not calibrated 0.72+0.0254PKTURN) Not calibrated exp (0.1889HAZRAT1) exp (-0.0253PKTRUCK) exp (-0.72 exp (0.

the models calibrated using all sites should be used. and AMFs. Based on these extensive analyses. Only Group B data models available. 262 . the models shown in table 235 are recommended for use in the IHSDM: Table 235. or the appropriate AMFs not available. those calibrated with AADT as the only explanatory variable are recommended. 157 Further Notes on Selection Only Group B data models available. AADT models. For AMFs.3. 155. several conclusions and recommendations can be drawn with respect to full models. The first were calibrated using all available data and the second for a subset of sites meeting base condition criteria. AADT Models Recommended for Use in IHSDM Model Type I Type II Type III Type IV Type V Table Numbers 134. Group B selected due to larger sample size. Only Group A data models available.7. the main consideration is logical appeal. Expert opinion derived AMFs should be used in the IHSDM in the short term. GOF. it is recommended that there be a continuation of the current approach whereby AMFs are applied to base model accident predictions to account for factors for intersections under consideration that are different from the base condition. 139 142. and practical issues of concern. and instead were created by substituting constants for the non-AADT variables in the calibrated models. It is important to recognize that AMFs derived from expert opinion lack the conventional statistical variability measures associated with statistically derived AMFs. 135 138. models estimated in previous efforts should be replaced with improved versions described in this report. Numerous GOF indices were used to assess the models. Two sets of AADT models are recommended for both total and injury accidents. the following recommendations regarding AMFs are made: 1. Combined with the fact that the research team had difficulty producing sufficient sample sizes for testing and/or validating AMFs. and the most defensible and representative data set from which models were estimated. Specifically. These are made in the context of IHSDM and general applications. For base models. 143 148. Also important is that expert derived AMFs were borne out of a perceived need among respected safety professionals that statistical information on empirically derived AMFs is unreliable. the main considerations for recommendation surround the issues of variable selection. as described in previous sections. If these base condition criterion are not known. 149 154. IHSDM Application For IHSDM model development. Because previous AADT models were not directly calibrated as AADT-only. For these models.3 Conclusions and Recommendations Extensive work was conducted as part of this effort to examine the appropriateness and defensibility of various models in the IHSDM. Group B selected due to larger sample size.

Only Group A available. Full Models Recommended for Use in Crash Prediction Model Type I Type II Type III Type IV Type V Table Numbers 159.g. the intuitive appeal and agreement with engineering expectations. 164 167 main model. crash prediction models are most relevant to ongoing and planned Highway Safety Manual projects and to the FHWA initiative known as SafetyAnalyst that consists of software tools to help manage site-specific safety improvements. These models are appropriate for crash prediction applications such as network screening and safety treatment evaluation. California) were the target of crash forecasts. If crashes in States other than those identified are to be forecast. In these applications. 172 main model Further Notes on Selection Group B not selected due to few available variables. Some of the main considerations in this regard include whether or not to include indicator variables for individual State effects (as opposed to a single constant term) and the appeal of individual model variables. Only Group A available. useful byproducts are the full crash prediction models that have been calibrated. and b) accrue data to help validate and/or recalibrate and refine AMFs in support of the IHSDM. General Applications of Full Models Although this research was focused on IHSDM applications.. and the GOF criterion. then models without State effects should be applied. 160 163. the full models shown in table 236 are recommended as “best available” statistical models: Table 236. 263 . It is generally believed that individual State effect models should be used if the individual States (e. The overriding concerns for full models surround the selection of variables in the models. Group B not selected due to few available variables. Considering these factors.2. 170 main model 171 main model. 168 main model 169 main model. Only Group A available. Continued studies should be undertaken to: a) develop analytical methods for obtaining AMFs under observational study conditions.

.

p. p. 113 Table 30. p. p. 112 Table 29. 9 15 16 17 18 19 20 21 22 23 24 25 26 27 28 Model Reference 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 Reference 1 1 1 1 3 3 3 3 1 1 1 1 1 3 3 3 3 2 2 2 2 2 2 2 2 2 2 2 Equation/Table Table 35. p. p. 111 Table 28. 116 Table 31. 5 2 3 4 5. p. 111 Table 28. p. p. 112 Dependent Variable1 A (TOT) A (INJ) A (TOT) A (TOT) A (TOT) A (TOT) B (TOT) B (TOT) A (TOT) A (INJ) A (TOT) A (TOT) A (TOT) A (TOT) A (TOT) B (TOT) B (TOT) C (TOT) C (TOT) C (TOT) C (INJ) C (INJ) C (INJ) B (TOT) B (TOT) B (TOT) B (TOT) B (TOT) Remarks Final Model Final Model Interim Model Interim Model Full Model Recommended Base Model Full Model Alternative Base Model Final Model Final Model Interim Model Interim Model Interim Model Full Model Alternative Base Model Full Model Base Model Recommended ADT ONLY Main Model Variant ADT ONLY Variant 1 Variant 2 ADT ONLY Main Model Variant 1 Variant 2 Variant 3 265 . 61. p. p. All Models in the Reports of Relevance Validation Number (V) 1. 141 Eq. 115 Table 36. p. 55. p. p. p. p. 113 Table 29. 140 Eq. 136 Eq. 111 Table 34. 137 Eq. p. 59. 115 Table 37. p. 114 Table 34. p. 143 Table 28. 112 Table 29. 58. p. 112 Table 29. 114 Eq. p. p. 113 Table 30. 112 Table 29. 53. 52. 54. 142 Eq. 14 10 11 12 13 14. p. 117 Table 32. 60. p. p. 138 Table 35.APPENDIX A: ALL MODELS IN THE REPORTS OF RELEVANCE Table 237. 111 Table 30. p. p. 114 Eq. 112 Table 34. 134 Eq. 1 6 7 8 9. p.

117 Table 34. 66. p. p. 117 Table 33. 64. 58 52 53. 124 Table 36. 118 Table 34. 55 50. 146 Dependent Variable1 C (TOT) C (TOT) C (TOT) C (TOT) C (TOT) C (INJ) C (INJ) B (TOT) B (TOT) B (TOT) B (TOT) B (TOT) B (INJ) B (INJ) C (TOT) C (TOT) C (TOT) C (TOT) C (INJ) B (TOT) B (TOT) B (TOT) B (TOT) B (INJ) C(TOT) C(TOT) B (TOT) B (TOT) B (TOT) Remarks ADT ONLY Main Model Variant 1 Variant 2 Variant 3 ADT ONLY Variant 1 ADT ONLY Main Model Variant 1 Variant 2 Variant 3 ADT only Variant 1 Main Variant 1 Variant 2 Variant 3 Main Main Variant 1 Variant 2 Variant 3 Main Full Full Full Full Full Full 58. p. 122 Table 37. C = Total number occurring at intersection 266 . 116 Table 32. 123 Table 36. 146 Eq. 43 55. 67. p. 117 Table 33. 122 Table 35. p. p. 148 B (TOT) Recommended 1 A = Police definition of intersection related. 123 Eq. p. 123 Table 36. 123 Table 36. 49 56. p. p. 51 5 3 Eq. 116 Table 32. 146 Eq. 116 Table 32. 116 Table 34. 116 Table 32. p. All Models in the Reports of Relevance (Continued) Validation Number (V) 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43. p. 48 57. p. p. p. B = Accidents generally considered to be intersectionrelated (BMI). 63. p. 123 Table 36. 117 Table 33. 68. p. p. 118 Table 34. 50 Model Reference 4 4 4 4 4 4 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 Reference 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 Equation/Table Table 32. p. p. 57 51. 118 Table 35. 44 54. p. p. 54 44. p. 122 Table 35. 145 Eq. p. p. 56 49. 53 45 46 47 48. p. 117 Table 33.Table 237. p. 118 Table 33. 122 Table 35. 65. p. p. p. p. 146 Eq.

p. 123 Eq. 116 Table 34. 118 Table 35. 61. 113 Table 30. p. 134 Eq. p. 52. 58. 148 Dependent Variable A (TOT) A (INJ) A (TOT) A (TOT) A (TOT) A (INJ) A (TOT) B (TOT) C (TOT) C (INJ) C (INJ) B (TOT) C (TOT) C (INJ) B (TOT) B (INJ) C (TOT) C (TOT) C (INJ) B (TOT) B (TOT) B (INJ) B (TOT) Validation Tasks All All See V1 Algo. 136 Table 35. 117 Eq. 111 Table 30. p. 118 Table 33. 123 Table 36. 68. 56 51. 123 Table 36. p. p. 117 Table 34. p. 54 44. 9 17 19 22 23 25 30 35 37 42 43. 143 Table 28. 53 47 48. p. 116 Eq. 112 Table 32. p. 53. Models Validated Validation Number Model Reference (V) 1. 122 Table 37. p. p. p. 140 Eq. p. p. p. 115 Table 37. 122 Table 35. All All See V9 Algo. 14 10 14. 1 6 9. p. All All All All All All All All All All All All See V58 All Algo. p. p. p. p. 124 Table 36.APPENDIX B: MODELS VALIDATED Table 238. 58 52 58. 5 2 5. p. 51 1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 5 5 5 Reference 1 1 3 3 1 1 3 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 Equation/Table Table 35. 115 Table 36. Remarks Final Model Final Model Full Model Recommended Base Model Final Model Final Model Full Model Base Model Recommended Main Model Variant 1 Variant 2 Main Model Main Model Variant 1 Main Model Variant 1 Main Variant 1 Main Main Variant 3 Main Full Recommended 267 . p. 113 Table 29. p.

.

5669 0.3750 0.3%) rolling 489 (34.1%) mountainous 423 (29. Summary Statistics for Type I Sites by State Variables TOTACC per Minnesota year Georgia California INJACC per Minnesota year Georgia California AADT1 Minnesota Georgia California AADT2 Minnesota Georgia California RT MAJ Minnesota Georgia California RT MIN Minnesota Georgia California LT MAJ Minnesota Georgia California LT MIN Minnesota Georgia California MEDIAN Minnesota Georgia California TERRAIN Minnesota Georgia California Number of Sites 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 Mean 0.5368 1.0000 0.1429 0.6%) 121 (8.APPENDIX C: DESCRIPTIVE STATISTICS FOR FULL DATASETS BY STATE TYPE I Table 239.9%) 0 (0%) 3 (2.5%) 269 .2600 3981 3593 6590 536 639 472 Median 0.9%) 13 (0.2716 0.57 2.6%) 77 (5.4%) rolling 58 (50%) mountainous 10 (8.4%) unknown flat 48 (41.1%) 0 (0%) 1 (0.5000 0.86 6.4%) 40 (2.1250 2611 2950 5097 339 430 210 Minimum 0 0 0 0 0 0 401 420 450 101 300 100 Maximum 3.3571 1.2085 0.6%) 388 (27.6%) flat 520 (36.4%) 4 (1.75 1.4741 0.8%) 45 (16.13 22067 16900 35750 4608 6480 10001 139 (51.00 6.7%) 3 (2.5%) 3 (2.50 4.5%) 4 (3.

05 0.0000 29.Table 239.0000 23.1088 1.00 1.00 85 60 7.89 unknown 1.2900 270 .00 0.50 -90 -65 0.2689 2.00 1.00 2.50 0.6435 Minimum 25 25 0 0 1.0200 14.00 7.35 4.0000 0.00 3.50 0.0000 0.00 0.13 3.5545 unknown Median 55.0000 0.29 1.00 16.00 0.3337 unknown 1.0000 Maximum 55 55 8 7 4.0000 0.97 unknown unknown 1.56 unknown -0.54 46.80 -2.00 0.0000 0.00 45. Summary Statistics for Type I Sites by State (Continued) Variables SPD1 Minnesota Georgia California DRWY1 Minnesota Georgia California HAZRAT1 Minnesota Georgia California HAU Minnesota Georgia California SHOULDER1 Minnesota Georgia California VCI1 Minnesota Georgia California HI1 Minnesota Georgia California Number of Sites 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 270 116 1432 Mean 52.00 1.02 5.00 1.8855 0.57 unknown 2.

5000 0.50 4.75 14141 12300 38126 3209 7460 6700 130 (52%) 3 (2.07 6.2509 0.5406 1.3750 1983 2100 5620 258 430 310 Minimum 0 0 0 0 0 0 419 420 407 100 100 100 Maximum 7.00 7.TYPE II Table 240.9%) 23 (3.1429 0.6%) 217 (29%) 0 (0%) 0 (0%) 1 (0.4%) 2 (1.1%) 0 (0%) 6 (5.6574 0.14 3.3%) 1 (0.1806 1.5110 2605 3181 6784 395 644 562 Median 0.0132 0.9%) 36 (4.3571 0.1%) 0 (0%) 1 (0.13 3.8%) 62 (8.5000 0.6250 0. Summary Statistics for Type II Sites by State Variable and Abbreviation TOTACC per year Minnesota Georgia California INJACC per year Minnesota Georgia California AADT1 Minnesota Georgia California AADT2 Minnesota Georgia California RT MAJ Minnesota Georgia California RT MIN Minnesota Georgia California LT MAJ Minnesota Georgia California LT MIN Minnesota Georgia California MEDIAN Minnesota Georgia California Number of Sites 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 Mean 0.8%) 271 .

Summary Statistics for Type II Sites by State (Continued) Variable and Abbreviation TERRAIN Minnesota Georgia California SPD1 DRWY1 HAZRAT1 HAU SHOULDER1 VCI1 HI1 Minnesota Georgia California Minnesota Georgia California Minnesota Georgia California Minnesota Georgia California Minnesota Georgia California Minnesota Georgia California Minnesota Georgia California Number of Sites 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 250 108 748 Mean Median Minimum Maximum unknown flat 52 (48.66 0.00 3.00 2.00 unknown 2.1000 7.69 0.1%) mountainous (3.00 -0.04 2.41 3.00 unknown 0.55 0.0300 1.7%) flat 468 (62.9%) mountainous (12.00 14.5000 9.0000 0.30 16.00 0.00 0.0000 0.50 6.1089 0.9045 unknown 0.58 unknown 30 30 0 0 1 1 -120 -58 0.00 1.10 6.Table 240.52 -0.00 55 55 6 6 6 6 150 50 2.75 55.17 55.00 unknown 0.1350 0.6%) rolling 186 (24.21 1.25 unknown unknown 0.60 272 .75 0.00 1.00 0.00 0.00 49.1%) rolling 52 (48.6%) 53.73 0.

2 2.6 0.0 -65.0 0.0 0.0 63. of Sites 60 24 52 136 60 24 52 Mean 1.5%) 8(5.6%) 24(17.0 90.0 0.3 2.0 0.6%) 0(0%) 5(3.5 0.4 0.6 3.1 1.0 0.6 60 13484 24 11635 52 13100 136 13011 60 500 24 835 52 892 136 136 21(15.6%) 69(50.4 0.0 10.0 0.0 3. Michigan (MI).0 4.0 12.4%) 24(17.0 90.5 10.5 20.2 0.4%) 0(0%) 2(1.0 2360 6817 6500 2360 20 15 80 15 Maximum 4.0 0.0 36.0 33333 23716 28601 33333 3001 2957 9490 9490 INJACC per Year AADT1 136 0.6 -2.0 63.2 0.7 2.0 273 .0 27. and Georgia (GA) Variables TOTACC per Year State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total Total No Median MEDTYPE CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.0 0.0 50.0 0.0 4.3%) 6(4.0 0.5 Median 0.1 1.0 0. Summary Statistics for Type III Sites by State: California (CA).0 0.2 1.7 1.TYPE III Table 241.1%) 9(6.8 4.0 6.0 0.0 0.8 0.7%) 14(10.0 -65.6 6.0 0.9%) 60 24 52 136 60 24 52 136 709 AADT2 Painted Curbed Other MEDWDTH1 HAU 5.0 -45.0 0.6%) 0(0%) 21(15.8 0.3 0.2 0.7%) 24(17.3 12082 11958 12200 12100 190 779 430 430 Minimum 0.0 55.4%) 45(33.5 0.

0 0.5%) 14(10.7 1.0 274 .7 0.0 11.0 0.0 9.0 1.1%) 9(6.0 1. and Georgia (GA) (Continued) Variables Total 1 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total MI GA CA MI GA Total CA MI GA Total CA MI GA Total No.0%) 16(11.0 9.7%) unknown 2(4.0 0.0 15.4%) 6(4.0 14.0 0.0 5. Summary Statistics for Type III Sites by State: California (CA).4%) 58(42.7%) 2(1.6%) 12 (8.0 2.5 1.0 2. of Sites 136 2(1.5 2.0 5.7%) 13(9.5 2.0%) 1(0.0 15.6%) 26(19.0 0.0%) 2(1. Michigan (MI).5 0.4%) 0(0.0 0.5 4.0 0.0%) 1(0.0 0.0 6.8%) 29(21.0 0.0 1.0 0.0 7.0 0.9 0.0 7.7%) 15(11.0%) 60 24 52 136 60 24 52 136 60 24 52 136 Mean Median Minimum Maximum 2 HAZRAT1 3 4 5 6 7 COMDRWY1 RESDRWY1 DRWY1 1.0 0.0 0.7%) 0(0.3%) 8(5.5%) 1(0.0 0.9%) 21(15.7%) 0(0.0 0.0%) 0(0.0 0.0 11.3%) 0(0.0 0.0 1.0 14.Table 241.8%) 1(0.6%) 1(0.6 1.5%) 8(5.8 1.9%) 1(0.0%) 25(18.0 0.

0 55.0 55.0 35.0 35.6%) 35(25.3%) 17(12.1%) 39(28. and Georgia (GA) (Continued) Variables SPD1 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.0 55.0 35.0 SPD2 Total LIGHT 0 1 Total TERRAIN1 Flat CA MI GA Total CA Rolling MI GA Total CA Mountainous MI GA Total Total Flat TERRAIN2 Rolling CA MI GA CA MI GA CA MI GA Mountainous 275 .0 25.0 65.6%) 45(33.6%) 9(6.0 45.5 34.7%) 83(61.0%) 4(2.0 65.1%) 97(71.0 15. of Sites 60 24 52 136 60 24 52 136 136 43(31.5%) 15(11. Summary Statistics for Type III Sites by State: California (CA).0 30.0 15.8 52.0 65.6%) unknown unknown 21(15.3 43.1 33.0 55.0%) 7(5.9%) 11(8.0%) 29(21.1%) 52 unknown unknown 24(17.1 55.0 30.9%) 7(5.7 Median Minimum Maximum 55.0 40.0 55.6%) 24(17.3%) 0(0.0 37.0 15.0 35.4%) unknown unknown 7(5.2 25.6%) 42(30.7%) 136 24(17.Table 241.0 35.1%) Mean 53.0 25.0 30. Michigan (MI).0%) 13(9.1%) 0(0.

00 0.00 0.29 14.6%) 88(64.00 0.01 0.7%) 60 24 52 136 60 24 52 136 Mean Median Minimum Maximum 1 Total LTLN1 0 1 Total RTLN2 0 1 Total LTLN2 0 1 HI1 1.00 0.30 14.30 26.2%) 0(0.00 0.0%) 12(8.00 0.4%) 19(14.00 0.0%) 43(31.6%) 9(6.6%) 136 15(11.4%) 24(17.00 0.00 0.63 HEI1 276 .8%) 131(96.00 0.0%) 24(17.6%) 19(14.7%) 136 55(40.15 6.1%) 0(0.Table 241.29 9.0%) 136 57(41.90 1.0%) 5(3.7%) 5(3.6%) 50(36.3%) 3(2.00 0.9%) 24(17. Summary Statistics for Type III Sites by State: California (CA).8%) 28(20.00 0.4%) 108(79.0%) 43(31.3%) 45(33.73 0.28 2.29 26.00 0.6%) 40(29.63 6. Michigan (MI).52 1.26 2.5%) 5(3.6%) 48(35. of Sites 136 44(32.88 0.00 14.4%) 16(11. and Georgia (GA) (Continued) Variables Total RTLN1 0 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.6%) 117(86.8%) 0(0.36 0.21 1.17 3.7%) 9(6.0%) 2(1.

46 0.00 1.59 7.71 5.27 0.76 1.00 0.00 100.16 0.97 21.00 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 VEI1 PKTRUCK PKTURN PKLEFT PKLEFT1 PKLEFT2 SD1 SDL2 SDR2 277 .00 0.35 0.16 16.75 5.54 1.59 3. of Sites 60 24 52 136 60 24 52 136 60 24 unknown 84 60 24 unknown 84 60 24 unknown 84 60 24 unknown 84 60 24 unknown 84 60 24 52 136 60 24 52 136 60 24 52 136 Mean 1. Michigan (MI).00 0.37 2.62 0.97 11. Summary Statistics for Type III Sites by State: California (CA).00 0.00 0.23 0.85 0.18 0.29 2000 2000 1715 2000 1365 1800 1533 1510 1424 1350 1865 1555 Minimum 0.13 0.28 1.20 2.83 6.78 1.27 0.90 6.83 2.12 0.93 1.55 6.13 0.09 18.90 9.44 1.21 9.29 100. and Georgia (GA) (Continued) Variables GRADE1 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.79 3.09 25.70 0.00 0.69 60.53 8.97 55.48 0.50 5.95 21.85 1.51 0.Table 241.18 2.00 0.15 3.00 500 600 560 500 40 400 170 40 80 500 510 80 Maximum 5.85 0.70 60.99 6.00 1.00 0.28 1.97 0.00 0.16 53.60 7.87 53.17 25.15 6.00 100.70 28.68 3.63 28.00 0.29 5.44 54.91 7.77 58.87 4.00 0.31 1553 1519 1470 1515 1364 1486 1448 1418 1380 1408 1493 1428 Median 0.88 0.00 0.00 0.47 55.14 0.73 3.03 0.

Summary Statistics for Type IV Sites by State: California (CA).5 5.4 3.0 0.0 0. Michigan (MI).0 1.0 0.0%) 10(8.0 1.0%) 5(4.9 1.5%) 39(31.7%) 0(0.7%) 2(1.4 0.7%) 0(0.8 5.0 0.0%) Mean 1.0 0.0 0.8 2.TYPE IV Table 242.8%) 22(17.2 10.0 0.0 0.5%) 17(13.1 2. of Sites 54 18 52 124 54 18 52 124 54 18 52 124 54 18 52 124 124 13(10.1 2. and Georgia (GA) Variables TOTACC per year State CA MI GA Total INJACC per year CA MI GA Total AADT1 CA MI GA Total AADT2 CA MI GA Total Total CA No Median MI GA Total MEDTYPE Painted CA MI GA Total CA Curbed MI GA Total CA Other MI GA Total No.0 0.9 13847 10707 12631 12881 441 913 706 621 Median Minimum Maximum 0.7 3.5 0.7 73799 19384 25799 73799 1850 2018 2990 2990 278 .5%) 70(56.0 3150 5967 5300 3150 21 254 300 21 7.8 6.1%) 27(21.0%) 0(0.5 10.8 1.5%) 18(14.0 0.4%) 22(17.2 2.6%) 0(0.5 11199 10550 12850 11496 301 733 463 430 0.0%) 3(2.2 4.

0%) 13(10.0%) 0(0.0 HAZRAT1 COMDRWY1 0.8%) 20(16.0 0.6%) 1(0.0 36 0 60 60 22.1 16.5 30.6%) 0(0.0%) 54 18 52 124 Mean 5.1 0.8 1.0 7. of Sites 54 18 52 124 54 18 52 124 124 13(10.0 -20.3 0.5 0.8%) 0(0.7%) 11(8.8%) 8(6.5%) 21(16.0 0.0 6.5 Median Minimum Maximum 4.6 0 40.Table 242.7 1.0%) 0(0.0 0 12.0 -50.8%) 0(0.5%) 8(6.0 1. Summary Statistics for Type IV Sites by State: California (CA).0%) 1(0.0 0 33.40 1.0 0 0.1 2.0%) 0(0. Michigan (MI).0 -50.0 12 279 .0 0 -20.9%) 1(0.25 0.0%) 1(0.0 2.5 0.6 0.0 55.4%) 24(19.3%) 43(34.4%) 20(16.9%) 1(0.0 0.1%) 9(7.5%) 0(0.0 55.3%) 14(11.0 0 0. and Georgia (GA) (Continued) Variables MEDWDTH1 State CA MI GA Total HAU CA MI GA Total Total 1 CA MI GA Total 2 CA MI GA Total 3 CA MI GA Total 4 CA MI GA Total 5 CA MI GA Total 6 CA MI GA Total 7 CA MI GA Total CA MI GA Total No.1%) 32(25.8%) 2(1.0 0.8%) 2(1.5%) 3(2.0 0.3 0.

4 55.0%) 0(0.6 35.6 34.2 36.0 2 0 7 0.0%) 4(3.0 0.0 3. Summary Statistics for Type IV Sites by State: California (CA).0 28.0 0.0 0.8 52 0.0 2.5%) 0(0.Table 242.7 Median Minimum Maximum 0.3 58.0 35. of Sites Mean 54 0.0 0 55.6 18 2.4 45.4 124 54 18 52 124 54 18 52 124 124 36(29.7 54 1.0%) 12(9.0 6.3%) 90(72.0%) 9(7.9%) 87(70.3 56.2%) 9(7.0 0.8%) 18(14.0 35.2%) 47(37.0 9.0 25.7%) 25(20.5%) 14(11.3%) 0(0.0 42.5%) 40(32.5 55.8%) 124 32(25.2 124 0.5 52 0.0%) 37(29.6%) 13(10.3%) 5(4.0 0 25 35 45 25 25 25 25 25 15 65 55 65 65 50 45 55 55 DRWY1 SPD1 SPD2 Rolling Mountainous 280 .0 0 0 7 0.0 4.0 15.0 0.3%) 1. Michigan (MI). and Georgia (GA) (Continued) Variables RESDRWY1 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total Total LIGHT 0 CA MI GA Total 1 CA MI GA Total Total TERRAIN1 Flat CA MI GA Total CA MI GA Total CA MI GA Total No.0 55 35.2%) 18(14.0 0.3 18 3.

2%) 18(14.0%) 46(37.5%) 14(11. Michigan (MI).1%) 0(0.1%) 20(16.6%) 0(0. Summary Statistics for Type IV Sites by State: California (CA).6%) 124 33(26.5%) 24(19.4%) 4(3.4%) 3(2.3%) 3(2.4%) 22(17.5%) 7(5.Table 242.0%) 1(0.8%) 5(4.0%) 13(10.8%) 18(14.1%) 7(5.6%) 3(2.4%) 13(10.1%) 35(28. and Georgia (GA) (Continued) Variables Total RTLN1 0 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No. of Sites 124 27(21.6%) 7(5.1%) 0(0.5%) 90(72.6%) 29(23.6%) 12(9.0%) 44(35.0%) 15(12.7%) 39(31.2%) 124 4(3.4%) 69(55.7%) 27(21.8%) 72(58.2%) 0(0.5%) 20(16.5%) Mean Median Minimum Maximum 1 2 Total LTLN1 0 1 2 Total RTLN2 0 1 2 281 .

49 233.00 0.71 0.0 0.0 0.0 0.00 0.5 HEI1 GRADE1 VEI1 VI1 VCI1 282 .19 0.00 0 0.35 0.24 0.07 7.80 3.00 0.0%) 0(0.0 0.00 0.89 0.4%) 2(1.00 0.3 1.0%) 0(0.92 1.33 2.0 0.00 0.00 0 0.5 3.83 0.5 3.1 1.25 2.85 0.0%) 0(0.1 12. of Sites 124 52(41.00 0.9%) 122(98.Table 242.80 2.4 1.43 0.50 0.0 0.47 15.28 1.0%) 54 18 52 124 54 18 52 124 54 18 52 124 54 18 52 124 54 18 52 124 54 18 52 124 Mean Median Minimum Maximum HI1 1.33 5.30 5.00 0.09 0.65 0.0 1.00 0.6 0.32 0.71 0.00 0.5 1. and Georgia (GA) (Continued) Variables Total LTLN2 0 State CA MI GA Total CA 1 MI GA Total CA 2 MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.00 0.33 7.6%) 0(0.79 12.45 0 0.5 3.00 0.87 0.23 12.5 2.05 0.0%) 2(1.04 0.1 12.00 0.0 0.00 0 7.00 0 0.00 0 0.7 12.07 233.00 0.0 0.29 0 0.72 0.67 0.62 0. Summary Statistics for Type IV Sites by State: California (CA). Michigan (MI).00 0 0.62 0.6%) 0(0.0 0.88 3.5%) 52(41.33 3.1 0.9%) 18(14.94 0.57 12.50 7.50 2.0%) 0(0.00 0.

95 8.87 9.12 34.08 1.85 94.83 0 0.96 11.93 14. and Georgia (GA) (Continued) Variables VCEI1 State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.00 4.54 4.75 13.00 58.22 37.26 100 68.22 38.52 25.00 0.20 67.05 7.96 0 400 580 639 400 37.19 10.45 15.14 0 0.00 2.10 9.41 15.39 68.04 4.10 1.75 0. of Sites 54 18 52 124 54 18 unknown 72 54 18 unknown 72 54 18 unknown 72 54 18 unknown 72 54 18 unknown 72 54 18 unknown 72 54 18 unknown 72 54 18 52 124 Mean 0.77 0.26 13.8 2.25 48.66 97. Michigan (MI).26 18.89 95.99 7.12 0.26 16.63 11.78 36.00 13.25 4.94 1.54 0 0.69 1431 1430 1356 1399 Median Minimum Maximum 0.76 25.51 37.52 32.78 39.95 7.56 2.92 2.80 0.77 79.82 1519 1310 1301 1332 1. Summary Statistics for Type IV Sites by State: California (CA).47 4.09 2000 2000 2000 2000 PKTRUCK PKTURN PKLEFT PKLEFT1 PKLEFT2 PKTHRU1 PKTHRU2 SD1 283 .10 36.56 13.71 94.0 3.53 6.96 100.5 8.Table 242.3 1.65 100 100.00 0.44 16.09 39.2 0.57 0 0 12.00 12.0 0.00 2.36 4.00 98.36 2.13 8.56 0.61 0 67.00 0.50 0.25 7.55 48.02 3.83 10.75 3.51 96.27 91.

Michigan (MI). and Georgia (GA) (Continued) Variables SDL2 State CA MI GA Total SDR2 CA MI GA Total No. of Sites 54 18 52 124 54 18 52 124 Mean 1382 1286 1254 1314 1384 1356 1263 1329 Median Minimum Maximum 1470 1203 1182 1262 1470 1255 1206 1354 324 394 579 324 215 479 540 215 2000 2000 2000 2000 2000 2000 2000 2000 284 . Summary Statistics for Type IV Sites by State: California (CA).Table 242.

8 9007 7798 9126 3642 4796 2749 3544 Median 5.0 0 0.5 5.0 3.5 6.2 8.2 1.5 12650 8435 7400 8700 3026 4434 2200 3100 Minimum 1.8 5.7 14.6 1.5 26. Michigan (MI). of Sites 18 31 51 100 18 31 51 100 31 51 100 18 31 51 100 100 0 22 11 33 17 4 24 45 1 5 16 22 Mean 5.8 26.0 6.3 0.5 1.0 0.0 2.TYPE V Table 243.5 4.0 0 7340 4917 430 430 940 1961 420 420 Maximum 10.5 0.8 2.1 4.0 0. and Georgia (GA) Variables TOTACC per year State CA MI GA Total INJACC per year CA MI GA Total AADT1 CA MI GA Total AADT2 CA MI GA Total Total CA Pretimed MI GA SIGTYPE Actuated Total CA MI GA Total CA Semiactuated MI GA Total No.3 8.5 25132 17483 15200 25132 10280 12478 10400 12478 18 13095 285 .3 1.3 2.5 6.9 2. Summary Statistics for Type V Sites by State: California (CA).

07 0 0 0 0 0 0 0 0 0 0 0 0 -10 -45 -45 -45 13 0 13 13 30 40 35 40 Total 3 4 5 286 . of Sites 100 12 31 44 87 5 0 7 12 1 0 0 1 18 31 51 100 18 31 51 100 100 3 7 2 12 5 15 9 29 6 8 13 27 2 1 13 16 1 0 12 13 Mean Median Minimum Maximum Painted Other MEDWIDTH1 HAU 3 0 1.82 0.33 -3.3 3. Summary Statistics for Type V Sites by State: California (CA).06 0.5 1. and Georgia (GA) (Continued) Variables Total MEDTYPE No Median State CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total 1 HAZRAT1 2 CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total CA MI GA Total No.Table 243. Michigan (MI).

Table 243. Summary Statistics for Type V Sites by State: California (CA).9 3.4 0.8 47.9 2.2 45 43.5 1.6 2.8 45.9 1 3 2 2 0 0 0 0 1 3 3 3 55 50 45 45 45 45 35 40 0 0 0 0 0 0 0 0 0 0 0 0 35 30 25 25 30 25 20 20 6 11 9 11 5 6 4 6 7 15 9 15 65 55 55 65 55 55 55 55 Mean Median Minimum Maximum 287 .4 41.9 2.6 0. of Sites 1 0 2 3 0 0 0 0 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 100 1 9 19 29 17 22 32 71 1.3 0.1 40.1 38. and Georgia (GA) (Continued) Variables HAZRAT1 State 6 CA MI GA Total 7 CA MI GA Total COMDRWY1 CA MI GA Total CA RESDRWY1 MI GA Total CA DRWY1 MI GA Total CA SPD1 MI GA Total CA SPD2 MI GA Total LIGHT Total 0 CA MI GA Total 1 CA MI GA Total No. Michigan (MI).2 50.9 0.8 3.2 3.

of Sites 100 11 25 23 59 5 6 27 38 2 0 1 3 100 1 27 42 70 17 4 9 30 100 7 14 30 51 3 5 13 21 8 12 8 28 Mean Median Minimum Maximum 288 . Michigan (MI).Table 243. and Georgia (GA) (Continued) Variables Total Flat TERRAIN1 CA MI GA Total Rolling CA MI GA Total CA Mountainous MI GA Total Total PROTLT1 CA 0 MI GA Total 1 CA MI GA Total Total CA 0 MI RTLN1 GA Total CA 1 MI GA Total CA 2 MI GA Total State No. Summary Statistics for Type V Sites by State: California (CA).

of Sites 100 0 7 16 23 0 1 1 2 18 23 34 75 100 12 14 33 59 3 7 10 20 3 10 8 21 100 8 9 28 45 0 3 2 5 10 19 21 50 Mean Median Minimum Maximum 289 . and Georgia (GA) (Continued) Variables Total LTLN1 0 State CA MI GA Total CA 1 MI GA Total CA 2 MI GA Total Total RTLN2 CA MI GA Total CA 1 MI GA Total CA 2 MI GA Total 0 Total LTLN2 CA MI GA Total CA 1 MI GA Total CA 2 MI GA Total 0 No. Michigan (MI). Summary Statistics for Type V Sites by State: California (CA).Table 243.

45 1.72 42.89 2.90 94.00 0.95 4.00 0. Michigan (MI).45 3.00 0.85 2.00 0.00 0.47 1.22 19.82 94.00 0.00 0.54 42.00 0.31 1.82 60.45 1.98 4.00 0.41 12. and Georgia (GA) (Continued) Variables HEI State CA MI GA Total CA MI HEI2 HI1 GA Total CA MI GA Total HI2 CA MI GA Total HEICOM CA MI GA Total HICOM CA MI GA Total GRADE1 CA MI GA Total GRADE2 CA MI GA Total VEI1 CA MI GA Total No.00 0.65 0.00 0.38 3.00 1.15 3.00 0.88 0.52 1.44 27.00 1.48 1.00 0.00 36.46 0.54 10.36 0.00 0.65 1.00 0.00 0.79 11.00 0.93 2.00 0.71 1.10 6.00 0.99 1.00 0.09 0.28 0.00 0.99 2.97 290 .58 0.20 1.94 1.58 2.05 47.00 0. of Sites 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 Mean 2.00 0.83 4.59 0.00 0 0.00 19.00 0.16 47.87 49.31 4.70 2.00 0.00 0.00 0.72 30.00 0.00 0.83 1.50 1.79 0.19 Minimum 0.56 2.40 1.Table 243.00 0.97 6.48 0.00 0.80 1.46 1.25 3.43 1.78 2.00 32.00 7.00 1.98 5.23 3.98 11.00 49.22 3.58 0.63 1.61 0.00 0.00 0.00 0.00 0.90 60.74 2.44 18.87 36.00 0.00 0. Summary Statistics for Type V Sites by State: California (CA).05 32.30 4.00 0.00 Maximum 9.01 1.00 0.00 0.79 7.00 0.00 0.00 0.54 32.00 0.00 0.56 1.00 0.40 0.00 0.45 Median 1.00 0 0.65 0.41 9.05 3.13 3.

43 291 .50 7.75 4.36 9.90 1.Table 243.45 0.00 5.65 1.89 Median 0.00 0.32 4.66 1.79 4.15 1.69 2.78 4.60 37.51 19.00 0.59 0.97 Maximum 11.00 2.17 0.08 7.23 36.07 7.87 1.78 43.00 0. and Georgia (GA) (Continued) Variables VEI2 State CA MI GA Total VEICOM CA MI GA Total CA MI VCEI1 VCEI2 GA Total CA MI GA Total VCEICOM CA MI GA Total PKTRUCK CA MI GA Total PKTURN CA MI GA Total PKLEFT CA MI GA Total PKLEFT1 CA MI GA Total unknown 49 14.69 1.43 72.03 No.38 0. Michigan (MI).00 0.54 1.47 1.07 26.65 1.00 8.60 1.31 1.00 0.72 1.00 0.13 10.17 13.46 1.66 61.97 11.79 13.90 35.48 14.66 1.83 15.07 4.13 14.79 12.00 0.09 7.66 37.13 6.43 8.45 45.90 0.26 1.00 0.91 2.54 34.00 1.33 1.23 unknown 49 18 31 18.58 unknown 49 18 31 8.24 1.62 1.00 6.99 13.74 38.18 10.41 1.00 15.65 1.41 0.00 0.03 1.00 0.00 0.40 14.64 15.96 30.37 Minimum 0.71 29.36 16.15 0.79 8.88 72.77 5.53 13.55 0.69 7.00 0.00 0.32 7.30 1.82 0.00 45.95 1.00 0.67 unknown 49 18 31 35.71 0.80 19.00 0.10 1.2 1.81 1.00 0.00 0.07 43.50 8.48 7.00 0. of Sites 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 18 31 Mean 2.00 0.03 6.29 2. Summary Statistics for Type V Sites by State: California (CA).20 11.39 1.66 17.07 31.00 0.

45 267 522 235 235 390 509 224 224 188 200 122 122 269 253 142 142 84.59 9.59 18.91 Maximum 68.78 18.68 43.44 73.34 8.19 44.21 74.17 45.88 Minimum 2.21 24.01 9.37 2.83 70.01 44.99 2000 1500 1086 1246 1750 1212 961 1091 454 615 709 673 787 850 739 750 8.77 39.88 80.90 27.80 Median 25. and Georgia (GA) (Continued) Variables PKLEFT2 State CA MI GA Total PKTHRU1 CA MI GA Total PKTHRU2 CA MI GA Total SD1 CA MI GA Total SD2 CA MI GA Total SDL1 CA MI GA Total SDL2 CA MI GA Total unknown 49 18 31 51 100 18 31 51 100 18 31 51 100 18 31 51 100 43.09 73.45 96. Summary Statistics for Type V Sites by State: California (CA).73 292 .57 75.09 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 unknown 49 18 31 71.40 75.73 84.73 96.49 69.73 85.9 1482 1447 1175 1314 1420 1389 1033 1213 815 814 735 774 1011 1041 794 910 41.14 unknown 49 18 31 28.33 No. Michigan (MI). of Sites 18 31 Mean 28.Table 243.27 24.

IV.APPENDIX D: CURE PLOTS FOR TYPE I. CURE Plot for Type I TOTACC AADT Model 293 . III. II. AND V AADT MODELS 250 200 150 100 Cumulative Residuals 50 Adj Cum Resi Standard Deviation Standard Deviation 0 0 -50 5000 10000 15000 20000 25000 30000 35000 40000 -100 -150 -200 -250 Major AADT Figure 23.

CURE Plot for Type III TOTACC AADT Model 294 . CURE Plot for Type II TOTACC AADT Model 25 20 15 10 Cumulative Residuals 5 Adj Cum Residuals 0 0 -5 10000 20000 30000 40000 50000 60000 70000 80000 Standard Deviation Standard Deviation -10 -15 -20 -25 Major AADT Figure 25.250 200 150 100 Cumulative Residuals 50 Adj Cum Resi Standard Deviation Standard Deviation 0 0 -50 5000 10000 15000 20000 25000 30000 35000 40000 45000 -100 -150 -200 -250 Major AADT Figure 24.

30 20 10 Cumulative Residuals 0 0 10000 20000 30000 40000 50000 60000 70000 80000 Adj Cum Residuals Standard Deviation Standard Deviation -10 -20 -30 Major AADT Figure 26. CURE Plot for Type IV TOTACC AADT Model 40 30 20 Cumulative Residuals 10 0 0 5000 10000 15000 20000 25000 30000 Adj Cum Residuals Standard Deviation Standard Deviation -10 -20 -30 -40 Major AADT Figure 27. CURE Plot for Type V TOTACC AADT Model 295 .

.

.. Vogt A.. 2003.. Hauer. Herf.. E. Crash Models For Rural Intersections: Four-Lane by Two-Lane Stop-Controlled and Two-Lane by Two-Lane Signalized.J. 2001. D. Washington D. Hughes. L. Validation of the FHWA Crash Models for Rural Intersections: Lessons Learned. No 1840. FHWA-RD-99-128. Lund.M.. W. and Bamfo. American Association of State Highway and Transportation Officials. December 2000..C. Journal of the Transportation Research Board. Vogt A. and J.E. Jutaek. Hauer. J. Hauer.. 8. L.. Oh. Sweden.. 2. 7. Hummer. (draft)..R. Safety Effectiveness of Intersection Left. and Bared.. Reestimation of Models for Two-Lane Rural Road Segments. Lyon. D. K.. Bauer. and Vogt A. FHWA-RD-99-207. A Policy on Geometric Design of Highways and Streets. October 1999. Reinfurt. J. Fourth Edition. Transportation Research Board. Torbic.M. Richard.” In Proceedings of the ICTCT 1997 Conference. Potts. 1988. 6. Harwood. F. 5. Safety Effects of Cross-Section Design for Two-Lane Roads. Washington. October 1998. FHWA-RD-02-089. J. 9. National Academy Press. Harwood. and Hunter.B. and Elefteriadou. Bared. K. Council.W. D.V. Prediction of the Expected Safety Performance of Rural Two-Lane Highways. Accident Prediction Models for Two-Lane Rural Roads: Segments and Intersections. FHWA-RD-98-133.REFERENCES 1. Persaud.R. “Two Tools for Finding What Function Links the Dependent Variable to the Explanatory Variables. 1997. Kohlman Rabbani. J. K..W. S. January 2001. pp 41-49. C. 3. Hauer. B. July 2002. Transportation Research Record 1195. 4. E.W. E. Zegeer.and RightTurn Lanes.. C. D. W. E.. 297 .

Recycled Recyclable HRDS-05/05-05(250)E .