# Solutions from Montgomery, D. C.

(2008) Design and Analysis of Experiments, Wiley, NY

Chapter 4
Randomized Blocks, Latin Squares, and Related Designs

Solutions
4.1.

The ANOVA from a randomized complete block experiment output is shown below.
Source

DF

SS

MS

F

P

Treatment

4

1010.56

?

29.84

?

Block

?

?

64.765

?

?

Error

20

169.33

?

Total

29

1503.71

(a) Fill in the blanks. You may give bounds on the P-value.
Completed table is:
Source

DF

SS

MS

F

P

Treatment

4

1010.56

252.640

29.84

< 0.00001

Block

5

323.82

64.765

Error

20

169.33

8.467

Total

29

1503.71

(b) How many blocks were used in this experiment?
Six blocks were used.
(c) What conclusions can you draw?
The treatment effect is significant; the means of the five treatments are not all equal.

4.2. Consider the single-factor completely randomized experiment shown in Problem 3.4. Suppose that
this experiment had been conducted in a randomized complete block design, and that the sum of squares for
blocks was 80.00. Modify the ANOVA for this experiment to show the correct analysis for the randomized
complete block experiment.
The modified ANOVA is shown below:
Source

DF

SS

MS

F

P

Treatment

4

987.71

246.93

46.3583

< 0.00001

Block

5

80.00

16.00

Error

20

106.53

5.33

Total

29

1174.24

4-1

Solutions from Montgomery, D. C. (2008) Design and Analysis of Experiments, Wiley, NY
4.3. A chemist wishes to test the effect of four chemical agents on the strength of a particular type of
cloth. Because there might be variability from one bolt to another, the chemist decides to use a randomized
block design, with the bolts of cloth considered as blocks. She selects five bolts and applies all four
chemicals in random order to each bolt. The resulting tensile strengths follow. Analyze the data from this
experiment (use α = 0.05) and draw appropriate conclusions.

Chemical
1
2
3
4

1
73
73
75
73

Design Expert Output
Response:
Strength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
157.00
4
39.25
Model
12.95
3
4.32
A
12.95
3
4.32
Residual
21.80
12
1.82
Cor Total
191.75
19

Bolt
3
74
75
78
75

2
68
67
68
71

4
71
72
73
75

F
Value

Prob > F

2.38
2.38

0.1211
0.1211

5
67
70
68
69

not significant

The "Model F-value" of 2.38 implies the model is not significant relative to the noise. There is a
12.11 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS

1.35
71.75
1.88
60.56

R-Squared
Pred R-Squared

0.3727
0.2158
-0.7426
10.558

Estimated
Standard
Mean
Error
1-1
70.60
0.60
2-2
71.40
0.60
3-3
72.40
0.60
4-4
72.60
0.60

Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4

Mean
Difference
-0.80
-1.80
-2.00
-1.00
-1.20
-0.20

DF
1
1
1
1
1
1

Standard
Error
0.85
0.85
0.85
0.85
0.85
0.85

t for H0
Coeff=0
-0.94
-2.11
-2.35
-1.17
-1.41
-0.23

Prob > |t|
0.3665
0.0564
0.0370
0.2635
0.1846
0.8185

There is no difference among the chemical types at α = 0.05 level.

4-2

Solutions from Montgomery, D. C. (2008) Design and Analysis of Experiments, Wiley, NY
4.4. Three different washing solutions are being compared to study their effectiveness in retarding
bacteria growth in five-gallon milk containers. The analysis is done in a laboratory, and only three trials
can be run on any day. Because days could represent a potential source of variability, the experimenter
decides to use a randomized block design. Observations are taken for four days, and the data are shown
here. Analyze the data from this experiment (use α = 0.05) and draw conclusions.

Solution
1
2
3
Design Expert Output
Response:
Growth
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
1106.92
3
368.97
Model
703.50
2
351.75
A
703.50
2
351.75
Residual
51.83
6
8.64
Cor Total
1862.25
11

1
13
16
5

2
22
24
4

Days
3
18
17
1

4
39
44
22

F
Value

Prob > F

40.72
40.72

0.0003
0.0003

significant

The Model F-value of 40.72 implies the model is significant. There is only
a 0.03% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS

2.94
18.75
15.68
207.33

R-Squared
Pred R-Squared

0.9314
0.9085
0.7255
19.687

Estimated
Standard
Mean
Error
1-1
23.00
1.47
2-2
25.25
1.47
3-3
8.00
1.47

Treatment
1 vs 2
1 vs 3
2 vs 3

Mean
Difference
-2.25
15.00
17.25

DF
1
1
1

Standard
Error
2.08
2.08
2.08

t for H0
Coeff=0
-1.08
7.22
8.30

Prob > |t|
0.3206
0.0004
0.0002

There is a difference between the means of the three solutions. The Fisher LSD procedure indicates that
solution 3 is significantly different than the other two.

4-3

Solutions from Montgomery, D. C. (2008) Design and Analysis of Experiments, Wiley, NY
4.5. Plot the mean tensile strengths observed for each chemical type in Problem 4.3 and compare them to
a scaled t distribution. What conclusions would you draw from the display?
S c a le d t D is tr ib u tio n

(1 )

(3 ,4 )

(2)

7 0 .0

7 1 .0

7 2 .0

7 3 .0

M e a n S tr e n g th

S yi . =

MS E
1.82
=
= 0.603
b
5

There is no obvious difference between the means. This is the same conclusion given by the analysis of
variance.

4.6. Plot the average bacteria counts for each solution in Problem 4.4 and compare them to an
appropriately scaled t distribution. What conclusions can you draw?
S c a le d t D is tr ib u t io n

(3 )

5

(1 )

10

15

20

(2 )

25

B a c t e r ia G r o w th

S yi . =

MS E
8.64
=
= 1.47
b
4

4-4

0935 0.27 Model 0.067 0.30 0.60 0. Consider the hardness testing experiment described in Section 4.047 2-2 9. Std.080 9 8. There is only a 0.5 9.067 0.047 4-4 9.8280 0. 2. Wiley.0009 0.7.067 0.0015 0.4.4563 15.0001 (b) Use the Fisher LSD method to make comparisons among the four tips to determine specifically which tips differ in mean hardness readings. Dev.8 9. Mean C. There is a difference between the means of the four tips.094 9.25 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0.0 9.63 0.6 9.27 -0.13 -0.44 implies the model is significant. The LSD method identifies a marginal difference between the means of tips 2 and 3.43 DF 1 1 1 1 1 1 Standard Error 0.13 14.44 Residual 0.38 3 0.0009 significant The Model F-value of 14.87 -4.047 Treatment 1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4 Mean Difference -0.1.0 4 10. the mean of tip 4 differs from the means of tips 1. C.635 Treatment Means (Adjusted. PRESS 0.98 0.6 10.V.889E-003 Cor Total 1.82 3 0. D. 4.38 3 0. If Necessary) Estimated Standard Mean Error 1-1 9.38 1.7 Coupon 2 3 9.57 0.0510 0.37 Prob > |t| 0. 4-5 .9 9.067 t for H0 Coeff=0 -0. NY There is no difference in mean bacteria growth between solutions 1 and 2.2 (a) Analyize the data from this experiment. Suppose that the experiment was conducted as described and the following Rockwell C-scale data (coded by subtracting 40 units) obtained: Tip 1 2 3 4 1 9.45 0.2 9.Solutions from Montgomery.13 14.7706 0.29 15 Prob > F 0.3 9.50 2. However. (2008) Design and Analysis of Experiments.15 -0.067 0.12 -6. solution 3 produces significantly lower mean bacteria growth.067 0.7 10. and 3.88 0.4 9. This is the same conclusion reached from the Fisher LSD procedure in Problem 4.44 A 0.09% chance that a "Model F-Value" this large could occur due to noise.047 3-3 9. Based on the LSD bars in the Design Expert plot below.25 -4.3 9. Design Expert Output Response: Hardness ANOVA for Selected Factorial Model Analysis of variance table [Terms added sequentially (first to last)] Sum of Mean F Source Squares DF Square Value Bock 0.4 9.4 9.025 0.7163 0.0026 0.

Wiley.025 0.0375 0.0375 5 1 2 -0.Solutions from Montgomery. Normal Plot of Residuals Residuals vs.22 0. D. NY One Factor Plot 10.1 -0.0875 90 70 50 30 0.15 9. The residual plots below do not identify any violations to the assumptions.96 10. (2008) Design and Analysis of Experiments.95 9.2 1 2 3 4 A: Tip (c) Analyze the residuals from this experiment.0875 9.025 2 20 10 -0.71 Predicted Residual 4-6 9.20 .45 9. Predicted 0. C.7 9.1 -0.47 9.2 Hardness 9.15 99 95 80 Residuals Normal % Probability 0.

A consumer products company relies on direct mail marketing pieces as a major component of its advertising campaigns.15 Residual 5428.Solutions from Montgomery.67 3 16345. as there are substantial differences in costs between the three designs. 4-7 Prob > F 0. (2008) Design and Analysis of Experiments. The number of responses to each mailing is shown below.83 6 904.0002 significant .22 Model 90755.452E+005 11 The Model F-value of 50. Design 1 2 3 NE 250 400 275 Region NW SE 350 219 525 390 340 200 SW 375 580 310 (a) Analyze the data from this experiment.025 -0. a square root transformation was applied as shown in the second ANOVA table.17 2 45377. The company has three different designs for a new brochure and want to evaluate their effectiveness. The company decides to test the three designs by mailing 5.0002 0. The residuals of the analsysis below identify concerns with the normality and equality of variance assumptions. Wiley.15 A 90755.02% chance that a "Model F-Value" this large could occur due to noise. Since there are known regional differences in the customer base.15 implies the model is significant. C. regions are considered as blocks. D. NY Residuals vs. There is only a 0.8.58 50. As a result.17 2 45377.58 50.15 Residuals 0. Design Expert Output Response: Number of responses ANOVA for Selected Factorial Model Analysis of variance table [Terms added sequentially (first to last)] Sum of Mean F Source Squares DF Square Value Block 49035.0375 -0.000 samples of each to potential customers in four different regions of the country. The residuals of both analysis are presented for comparison in part (c) of this problem. The analysis concludes that there is a difference between the mean number of responses for the three designs.0875 0.1 1 2 3 4 Tip 4.81 Cor Total 1. Tip 0.

197 Treatment Means (Adjusted.69 0.01 0.50 DF 1 1 1 Standard Error 21.81 9.37 60. however. (2008) Design and Analysis of Experiments.37 60.35 3-3 16.V.69 0. If Necessary) Estimated Standard Mean Error 1-1 17. Based on the LSD bars in the Design Expert plot below.48 4. Mean C.3769 < 0.99 DF 1 1 1 Standard Error 0.57 21715.47 A 60.24 0. There is only a 0.04 2-2 473.50 t for H0 Coeff=0 -9.8109 18.05 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0.V.27 21.89 3 11.0002 0.0001 (b) Use the Fisher LSD method to make comparisons among the three designs to determine specifically which designs differ in mean response rate.01% chance that a "Model F-Value" this large could occur due to noise. D. 4-8 .04 Treatment 1 vs 2 1 vs 3 2 vs 3 Mean Difference -175.75 15. C.9247 0. PRESS 30.04 3-3 281.52 0.Solutions from Montgomery.9527 0.50 Cor Total 99.0001 Design Expert Output for Transformed Data Response: Number of responses Transform: Square root ANOVA for Selected Factorial Model Analysis of variance table [Terms added sequentially (first to last)] Sum of Mean F Source Squares DF Square Value Block 35. Std.17 8. If Necessary) Estimated Standard Mean Error 1-1 298.25 192. Wiley.35 2-2 21.05 Prob > |t| 0.73 2 30.83 12.47 Residual 3.25 17.0001 0.96 Prob > |t| 0.25 15.27 21.96 Model 60.0001 0.50 15.71 18. Dev. designs 1 and 3 do not differ.9370 0. Dev.17 0.27 t for H0 Coeff=0 -8.50 0.64 11 Constant: 0 Prob > F 0. design 2 is different than designs 1 and 3. Mean C.9436 0.73 2 30.08 351.0001 significant The Model F-value of 60.47 implies the model is significant.52 3.01 6 0.50 0. NY Std. PRESS 0.191 Treatment Means (Adjusted.4483 0.33 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0.95 9.35 Treatment 1 vs 2 1 vs 3 2 vs 3 Mean Difference -4.7742 16.

58333 20 10 -22. design plot indicates a slight inequality of variance. Wiley.75 -22. D.5833 199.113 16.Solutions from Montgomery.1667 5 1 -41. The second set of residual plots represent transformed data and do not identify significant violations of the assumptions.58333 17 36.88 372.75 -41.13 544. however. C. Concerns with normality as well as inequality of variance are presented. Normal Plot of Residuals Residuals vs.142 1 2 3 A: Design (c) Analyze the residuals from this experiment.083 21. The residuals vs. (2008) Design and Analysis of Experiments.627 14.00 Predicted 4-9 458.1667 -2. not a strong violation and an improvement over the non-transformed data. NY One Factor Plot Sqrt(Number of responses) 24.75 Residual 285.25 . The first set of residual plots presented below represent the untransformed data.5833 99 17 90 80 70 Residuals Normal % Probability 95 50 30 -2.598 19. Predicted 36.

Design 36.Solutions from Montgomery.476292 90 70 50 30 0. D. C. Normal Plot of Residuals Residuals vs. NY Residuals vs.476292 14.921041 -0.24 23.52 .58333 -22.942069 16.1667 -41. Predicted 0.455263 5 1 -0.921041 -0. (2008) Design and Analysis of Experiments.68 18.5833 Residuals 17 -2.0105142 0.96 Predicted Residual 4-10 21.75 1 2 3 Design The following are the square root transformed data residual plots.41 0.942069 99 95 80 Residuals Normal % Probability 0. Wiley.0105142 20 10 -0.455263 0.

From the analysis below.35 A 6.023 Model 6. The effect of three different lubricating oils on fuel economy in diesel truck engines is being studied.222E-003 8 5.520 0.814 4-11 Prob > F 0.023 0.942069 Residuals 0.10 14 The Model F-value of 6. Design 0.092 4 0.540 0.35 Residual 4.0105142 -0. there is a significant difference between lubricating oils with regards to fuel economy.V.353E-003 6.015 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0. Five different truck engines are available for the study. PRESS 0. There is only a 2.51 4.23% chance that a "Model F-Value" this large could occur due to noise. and the experimenters conduct the following randomized complete block design.5170 -0.353E-003 6. NY Residuals vs.706E-003 2 3.329 0. Dev.0223 0. Std.487 0.35 implies the model is significant.510 (a) Analyize the data from this experiment.49 0.455263 -0.595 4 0.634 0. Mean C.706E-003 2 3.921041 1 2 3 Design 4. C.512 0. Fuel economy is measured using brake-specific fuel consumption after the engine has been running for 15 minutes.0223 significant .3583 18.278E-004 Cor Total 0. Wiley.476292 0. (2008) Design and Analysis of Experiments.488 2 0.400 5 0.Solutions from Montgomery.500 0.513 Truck 3 0.535 0.9.6136 0. D.435 0.675 0. Design Expert Output Response: Fuel consumption ANOVA for Selected Factorial Model Analysis of variance table [Terms added sequentially (first to last)] Sum of Mean F Source Squares DF Square Value Block 0. Oil 1 2 3 1 0.

49 0. 4-12 . (2008) Design and Analysis of Experiments. however.049 1 vs 3 -8.0102 0.010 2-2 0.0255 (b) Use the Fisher LSD method to make comparisons among the three lubricating oils to determine specifically which oils differ in break-specific fuel consumption.Solutions from Montgomery.50 0.74 Prob > |t| 0.61 2.015 0. NY Treatment Means (Adjusted. C. oil 2 is different than oils 1 and 3.54 0.800E-003 2 vs 3 0.502 0.5615 0. Wiley.040 DF 1 1 1 Standard Error 0. Based on the LSD bars in the Design Expert plot below.5885 0. the means for break-specific fuel consumption for oils 1 and 3 do not differ. The residual plots below do not identify any violations to the assumptions. One Factor Plot 0.015 0.675 Fuel consumption 0. D.010 Mean Treatment Difference 1 vs 2 -0.34 -0.015 t for H0 Coeff=0 -3. If Necessary) Estimated Standard Mean Error 1-1 0.4155 0.329 1 2 3 A: Oil (c) Analyze the residuals from this experiment.010 3-3 0.

78 0.0243167 -0. August 1981) describes an experiment in which a shape factor was determined for several different nozzle designs at six levels of jet efflux velocity.59 0.83 0.0223333 99 95 80 Residuals Normal % Probability 0.37 0. (2008) Design and Analysis of Experiments. C. Predicted 0.85 0.0398667 -0.0223333 0.37 0.81 0.86 16.89 0.95 0. The data are shown below: Jet Efflux Velocity (m/s) Nozzle Design 1 2 3 4 5 11.0398667 1 2 3 Oil 4. 4.00678333 -0.00876667 30 20 10 -0.43 0.0398667 -0.59 0.” Vol.86 0.14 0.77 0. Interest focused on potential differences between nozzle designs. NY Normal Plot of Residuals Residuals vs.86 0.78 0.00678333 90 70 50 -0.74 0.66 Predicted Residual Residuals vs.52 0.0223333 Residuals 0.78 4-13 20. Oil 0.0243167 5 1 -0. An article in the Fire Safety Journal (“The Effect of Nozzle Design on the Stability and Performance of Turbulent Water Jets.83 0.73 0.44 0.75 0. Wiley.75 .93 1. with velocity considered as a nuisance variable.92 0. D.81 0.00876667 -0.88 0.46 0.97 0.98 0.Solutions from Montgomery.00876667 0.10.80 0.92 0.00678333 0.97 14.0243167 -0.76 23.85 0.89 0.83 0.76 28.

031 0. (2008) Design and Analysis of Experiments.0003 The Model F-value of 8. Mean C.85 0.92 8.865E-003 Cor Total 0. NY (a) Does nozzle design affect the shape factor? Compare nozzles with a scatter plot and with an analysis of variance.0097 0.88 -5.022 Treatment 1 vs 2 1 vs 3 1 vs 4 1 vs 5 2 vs 3 2 vs 4 2 vs 5 3 vs 4 3 vs 5 4 vs 5 Mean Difference -0.031 0. If Necessary) Estimated Standard Mean Error 1-1 0.2103 0.81 0.91 1.057 20 2.V.048 -0.031 0.10 4 0.088 0.05.031 0.3177 0.022 4-4 0.022 5-5 0.78 0.0009 < 0.1926 0.86 4.031 0. Design Expert Output Response: Shape ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 0.86 6.23 -1.013 Model 0.054 0.0001 0.022 3-3 0.13 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0. C.23 0.10 4 0.0311 0.94 0.026 A 0.35 2.5688 0. 4-14 Prob > |t| 0. There is only a 0.0086 0.022 2-2 0.040 -0.29 -1.0004 significant .21 Nozzle design has a significant effect on shape factor. using α = 0. Wiley.031 0.56 -2.031 0.22 29 F Value Prob > F 8.031 0.438 Treatment Means (Adjusted.90 0.03% chance that a "Model F-Value" this large could occur due to noise.026 Residual 0.1335 0.072 -0.1916 9.6407 0.32 -3.063 5 0.031 0.042 0. PRESS 0.12 -0.090 0.16 -0.13 DF 1 1 1 1 1 1 1 1 1 1 Standard Error 0.0003 0.02 -1. D. Dev. Std.Solutions from Montgomery.92 implies the model is significant.032 -0.92 0.031 t for H0 Coeff=0 -2.

Solutions from Montgomery. C.0286667 5 1 -0.944718 2 2 0.847076 2 2 0.14 Shape 1.121333 0. The plots shown below do not give any indication of serious problems. D.0213333 50 30 20 10 -0.0213333 0.0786667 -0.02 .95 1. Residuals vs.121333 99 0.749435 1 2 3 4 5 Nozzle Design (b) Analyze the residual from this experiment. Wiley.87 Predicted Res idual 4-15 0. (2008) Design and Analysis of Experiments.73 0.04236 0. NY One Factor Plot 1.0713333 90 80 70 Res iduals Norm al % probability 95 0. Predicted Normal plot of residuals 0.0786667 -0.0713333 0.0286667 0.80 0. Thre is some indication of a mild outlier on the normal probability plot and on the plot of residuals versus the predicted velocity.

i.021852)= 0.04833 0.05(5.03167 0.08833 0.18)(0.05(4.Solutions from Montgomery.16167 0.121333 Residuals 0.021852)= 0.0786667 1 2 3 4 5 Nozzle Design (c) Which nozzle designs are different with respect to shape factor? Draw a graph of average shape factor for each nozzle type and compare this to a scaled t distribution.20) S y = (2.06774 R4= r0.10)(0.06446 0.13000 0.07102 1 vs 4 1 vs 3 1 vs 2 1 vs 5 5 vs 4 5 vs 3 5 vs 2 2 vs 4 2 vs 3 3 vs 4 i. NY Residuals vs. Wiley.04000 0.25)(0.06446 R3= r0. Compare the conclusions that you draw from this plot to those from Duncan’s multiple range test.06949 0.07102 0. D. (2008) Design and Analysis of Experiments.20) S y = (3.12000 0.95)(0.04167 > > > < > > < > < < 4-16 R 0. Nozzle Design 0.0213333 2 -0.05(3.021852 6 R2= r0.06446 0.002865 = 0.20) S y = (3.05(2.06446 0. i.07167 0.06446 different different different different different different . Mean Difference 0.20) S yi . = (3.06949 0.06774 0. = MS E = b 0.06774 0.06949 R5= r0.06774 0.021852)= 0. C. S yi .0713333 2 0.09000 0.021852)= 0.0286667 -0.

and all four ratio control algorithms were tested in each time period.19 0.Solutions from Montgomery.11.86 (0.12) 4 4.15) 4.028 4-4 4.06) 4.069 4.83 (0. The experiment was actually conducted as a randomized block design.13) 4.04) 4.75 (0.812E-003 Cor Total 0.86 (0.95 (0.153E-004 A 2. (2008) Design and Analysis of Experiments.02) 4. C.88 (0. Mean C.05) 2 4.43 0.09) 4.03) 4.94 (0.05.072 15 4.746E-003 3 9.19 0.8 (2 (3 0.84 1.9 Shape 4. D.08) 4.79 (0. where six time periods were selected as the blocks.90 (0.05) 4.05) 4.153E-004 Residual 0. (Use α = 0.02) (a) Analyze the average cell voltage data.93 (0.05) 4. If Necessary) Estimated Standard Mean Error 1-1 4.14 % chance that a "Model F-value" this large could occur due to noise.86 0.05) 4.85 (0.75 (0.04) 4.017 5 3.V.85 0.77 (0.9.9014 0.092 23 F Value Prob > F 0.9014 not significant The "Model F-value" of 0.03) 4.84 0.19 implies the model is not significant relative to the noise.028 4-17 .03) 4. NY Scaled t (1) 0.04) 4. Std.91 (0.028 2-2 4.487E-003 Model 2.89 (0.02) 3 4.7 (5) 0.9 (4 0.688 Treatment Means (Adjusted.79 (0.85 (0.85 (0.03) 4.8 0.) Does the choice of ratio control algorithm affect the cell voltage? Design Expert Output Response: Average ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 0.746E-003 3 9.88 (0.82 (0. PRESS 0.05) 4. There is a 90. The average cell voltage and the standard deviation of voltage (shown in parentheses) for each cell are as follows: Ratio Control Time Period Algorithms 1 2 3 4 5 6 1 4.83 0.03) 4.1560 -1. Consider the ratio control algorithm experiment described in Section 3.76 (0.4662 2.0366 -0.11) 4.90 (0.75 (0. Dev.79 (0.18 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0. Wiley.028 3-3 4.

01% chance that a "Model F-Value" this large could occur due to noise.12 -1.446 Treatment Means (Adjusted.04 23 Constant: 0.29 Prob > |t| 0.93 -6.040 0. NY Mean Treatment Difference 1 vs 2 0.25 -3.0001 0.26 < 0. Wiley.0001 0.013 2 vs 4 -1.37 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0. PRESS 0.14 0.9674 0.14 0.040 0. Std.14 t for H0 Coeff=0 2.”) Does the choice of ratio control algorithm affect the pot noise? Design Expert Output Response: StDev Transform: Natural log ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 0.89 0.33 0.0103 < 0.18 2.20 0.10 Treatment 1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4 Mean Difference 0.7438 0.15 1.3042 < 0.667E-003 3 vs 4 0.17 3 2. (b) Perform an appropriate analysis of the standard deviation of voltage.062 Cor Total 8.26 implies the model is significant.040 0.040 t for H0 Coeff=0 0.42 -0.06 Residual 0.042 0.19 1.17 3 2.6654 12. D.31 -0.8693 0. 4-18 .013 1 vs 4 0.012 DF 1 1 1 1 1 1 Standard Error 0.06 Prob > |t| 0.87 -9.040 0.36 0.5156 0.Solutions from Montgomery.040 0.0001 significant The Model F-value of 33.16 DF 1 1 1 1 1 1 Standard Error 0.67 0.33 -0.5419 0.000 F Value Prob > F 33.27 -1.06 8.93 15 0.10 3-3 -2.04 -8.26 33.0001 A natural log transformation was applied to the pot noise data. There is only a 0.8432 0. (2008) Design and Analysis of Experiments.027 1 vs 3 0.14 0. If Necessary) Estimated Standard Mean Error 1-1 -3. C.09 0.0001 < 0.7438 0.10 2-2 -3.14 0.10 4-4 -3.62 -0. The ratio control algorithm does affect the pot noise.19 Model 6.14 0.94 5 0.7748 The ratio control algorithm does not affect the mean cell voltage. Mean C.51 0.06 A 6.V. (Recall that this is called “pot noise.025 2 vs 3 -0. Dev.0813 < 0.

D.84 Predicted Residuals vs.19708 5 1 -0.78 -2.19708 -0. Predicted 0.31 -1.288958 -3.73 Res idual -3.26 -2.126945 -0. that is algorithm #2.0350673 30 20 10 -0. NY (c) Conduct any residual analyses that seem appropriate. Algorithm 0.19708 -0.Solutions from Montgomery.126945 90 80 70 Res iduals Norm al % probability 95 50 -0. Normal plot of residuals Residuals vs.0350673 -0. (2008) Design and Analysis of Experiments. however.359093 -0. Wiley.359093 1 2 3 4 Algorithm The normal probability plot shows slight deviations from normality.126945 0. select the algorithm that minimizes pot noise.288958 Res iduals 0. C. 4-19 . (d) Which ratio control algorithm would you select if your objective is to reduce both the average cell voltage and the pot noise? Since the ratio control algorithm has little effect on average cell voltage.0350673 0. still acceptable.288958 99 0.359093 -0.

9071 0.85 0.85 0.75 1.85 implies the model is not significant relative to the noise. The company produces the product in four furnaces. PRESS 2.96 -1. Mean C.75 1.3620 0.12. A randomized block design is run for a particular refiner and the resulting grain size data is as follows. There is a 49.2213 -0.44 15 F Value Prob > F 0.06 Model 22. 4-20 . Wiley.47 Treatment 1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4 Mean Difference -2.75 -0.69 38.19 3 55.12 -0. C.67 Cor Total 265.44 0.47 2-10 8.Solutions from Montgomery.08 2.0382 -1.1836 0. so any experiment run in the foundry that involves more than one furnace will consider furnaces as a nuisance variable.31 246.7270 0. D.72 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0.48 Prob > |t| 0.95 % chance that a "Model F-value" this large could occur due to noise.40 A 22.50 1.08 2. An aluminum master alloy manufacturer produces grain refiners in ingot form.00 -3.47 4-20 8.40 Residual 78.75 1. Each furnace can be run at four different stirring rates.08 2.08 2.06 9 8.4995 not significant The "Model F-value" of 0. The process engineers suspect that stirring rate impacts the grain size of the product.4995 0.19 3 7.19 3 7.25 -1. (2008) Design and Analysis of Experiments.390 Treatment Means (Adjusted.36 -0.4610 5. Dev.08 2.2193 0.75 -2.95 7.08 t for H0 Coeff=0 -1.V.00 0.6425 The analysis of variance shown above indicates that there is no difference in mean grain size due to the different stirring rates. NY 4.00 DF 1 1 1 1 1 1 Standard Error 2. Stirring Rate 5 10 15 20 Furnace 2 3 4 5 5 6 6 9 9 3 1 8 14 14 17 4 6 9 2 6 (a) Is there any evidence that stirring rate impacts grain size? Design Expert Output Response: Grain Size ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 165.47 3-15 7. If Necessary) Estimated Standard Mean Error 1-5 5.32 -0. Std. Each furnace is known to have its own unique operating characteristics.

8125 1 2 3 4 1 Stirring Rate 2 3 4 Furnace The variance is consistent at different stirring rates. Stirring Rate 3. (d) What should the process engineers recommend concerning the choice of stirring rate and furnace for this particular grain refiner if small grain size is desirable? There really is no effect due to the stirring rate. Furnace Residuals vs. NY (b) Graph the residuals from this experiment on a normal probability plot. C.Solutions from Montgomery.0625 -3.4375 1.1875 -0.1875 1.3125 1. Not only does this validate the assumption of uniform variance. (2008) Design and Analysis of Experiments. Normal plot of residuals 99 Norm al % probability 95 90 80 70 50 30 20 10 5 1 -3. it also identifies that the different stirring rates do not affect variance. Wiley. Does this plot convey any useful information? Residuals vs.0625 -0. D.8125 -2. 4-21 .1875 Res idual The plot indicates that normality assumption is valid.3125 2 -0.3125 -2.8125 -3.4375 Residuals Res iduals 3. Interpret this plot. (c) Plot the residuals versus furnace and stirring rate.0625 -2.4375 3.

βˆ 2 = .50 = SS Treatments μˆ = ( ) Model Restricted to β j = 0 : 4-22 .4 using the general regression significance test. τˆ 3 = . β4 = 12 12 12 12 12 12 12 12 78 − 129 − 89 − 25 225 51 ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ R(μ .τ . β ) − R(μ .17 12 ⎝ ⎠ ⎝ 12 ⎠ 2 ij − R(μ .83 Model Restricted to τ i = 0 : μ : 12μˆ +3βˆ1 +3βˆ2 β1 : 3μˆ β2 : 3μˆ β3 : 3μˆ β4 : Applying the constraint ∑ βˆ j +3βˆ3 +3βˆ4 +3βˆ1 = 225 = 34 +3βˆ2 = 50 +3βˆ3 3μˆ = 36 +3βˆ4 = 105 = 0 .τ . β2 = . we obtain: 225 ˆ −25 ˆ −81 ˆ 195 . τˆ 1 = . Wiley.67 12 12 12 12 ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ 12 ⎠ R τ μ .Solutions from Montgomery. SS E = ∑∑ y ⎛ − 81 ⎞ ⎛ 195 ⎞ ⎜ ⎟(36 ) + ⎜ ⎟(105) = 6029. Now: 12 12 12 12 ⎛ 225 ⎞ ⎛ − 89 ⎞ ⎛ − 25 ⎞ ⎛ − 81 ⎞ ⎛ 195 ⎞ R (μ . (2008) Design and Analysis of Experiments. β ) = ⎜ ⎟(225) + ⎜ ⎟(34 ) + ⎜ ⎟(50 ) + ⎜ ⎟(36 ) + ⎜ ⎟(105) = 5325. β = R(μ . β3 = . Analyze the data in Problem 4. NY 4.13. β ) = ⎜ ⎟(225) + ⎜ ⎟(92) + ⎜ ⎟(101) + ⎜ ⎟(32 ) + ⎜ ⎟(34 ) + ⎜ ⎟(50 ) + ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ μˆ = ∑∑ y 2 ij = 6081 . D. β1 = . β4 = . τˆ 2 = . μ : 12μˆ +4τˆ1 +4τˆ2 +4τˆ3 +3βˆ1 + βˆ +3βˆ2 + βˆ +3βˆ3 + βˆ +3βˆ4 + βˆ = 225 + βˆ2 + βˆ + βˆ3 + βˆ + βˆ4 + βˆ = 101 +4τˆ3 + βˆ1 + βˆ +3βˆ1 τ1 : 4 μˆ τ2 : 4 μˆ τ3 : 4 μˆ β1 : 3μˆ +τˆ1 +τˆ2 +τˆ3 β1 : 3μˆ +τˆ1 +τˆ2 +τˆ3 β1 : 3μˆ +τˆ1 +τˆ2 +τˆ3 β : 3μˆ +τˆ1 +τˆ2 +τˆ3 Applying the constraints +4τˆ1 1 +4τˆ2 ∑τˆ = ∑ βˆ i j 1 2 2 3 3 4 4 = 92 = 32 = 34 +3βˆ2 = 50 +3βˆ3 = 36 +3βˆ4 = 105 = 0 . β ) = 6029. we obtain: −129 ˆ 225 51 78 −89 ˆ −25 ˆ −81 ˆ 195 .17 = 51. C. β ) = 6081 − 6029.67 = 703.τ .17 − 5325. β1 = −89 / 12 . β3 = .

we obtain: μˆ = 35 13 17 ˆ −23 −7 35 ˆ 75 ˆ 20 ˆ −65 ˆ −65 . β4 = . τˆ 3 = 20 20 20 20 20 1 20 20 20 20 20 4.63 and β ≅ 0.25 = 1106.55 .70 3(1) = 2σ 2 = 2MS E .τ ) = 6029. etc. Wiley. We have: υ 2 = (a − 1)(b − 1) = (2 )(3) = 6 .92 = SS Blocks 4. we obtain: −129 225 51 78 . β3 = .Solutions from Montgomery. then: Φ= 4 ( 2) 3 (1) = 1.4. υ1 = a − 1 = 2 If ∑ τˆ 2 i = σ 2 = MS E .3. τˆ 1 = . we use the OC curve in appendix V. β2 = .14. Assuming that chemical types and bolts are fixed. applying the constraints. Does this test seem to be sensitive to small differences in treatment effects? Assuming that solution type is a fixed factor. This test is not very sensitive to small differences. τˆ 2 = . β = .τ ) = ⎜ ⎟(32 ) = 4922. Calculate Φ = 2 b∑τ i2 aσ 2 = 4∑τ i2 3 ( 8.17 − 4922. 4-23 . NY μ : 12μˆ +4τˆ1 +4τˆ2 τ 1 : 4μˆ +4τˆ1 τ 2 : 4μˆ +4τˆ2 τ 3 : 4μˆ Applying the constraint ∑ τˆ i +4τˆ3 = 225 +4τˆ3 = 92 = 101 = 32 = 0 . D. β ) − R(μ . β5 = . τˆ 1 = . C. (2008) Design and Analysis of Experiments. τˆ 2 = .64 ) using MSE to estimate σ2. Draw an operating characteristic curve for the design in Problem 4. τˆ 3 = 12 12 12 12 ⎛ − 129 ⎞ ⎛ 78 ⎞ ⎛ 51 ⎞ ⎛ 225 ⎞ R(μ . estimate the model parameters τi and βj in Problem 4. Using Equations 4.15 and β ≅ 0.τ .18.τ ) = R(μ . then: Φ= If ∑ τˆ i 4 = 1.25 ⎟(225) + ⎜ ⎟(92 ) + ⎜ ⎟(101) + ⎜ ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ ⎝ 12 ⎠ μˆ = R (β μ . τˆ 4 = .15.

y 23 0 y23 equal to the average of the observations available for 2 = 0.. C. Chemicals are not significant. and y.' + by.154 1.25 Source Chemicals Bolts Error Total SS 12.133333 F0 19.. y 23 is missing.04 = = 5.4 − y'. 4 y'2.80 0. Two missing values in a randomized block. and ˆy44 = 4. Wiley. ˆy23 = 4 y' + 5 y'.3.7.Solutions from Montgomery.3=38. Suppose that the observation for tip 2 in coupon 3 is missing.3=302.82 Source Tip Coupon Error Total SS 0.3' − y.18.3.3' − y.3 − y'.5 . Thus.4344 DF 3 4 11 18 MS 4. Then .72. Consider the hardness testing experiment in Problem 4. (2008) Design and Analysis of Experiments.' 4 ( 28.8875 21. 4. Analyze the problem by estimating the missing value.' 4 ( 282 ) + 5 ( 227 ) − 1360 = = 75.6 ) + 4 ( 29.29 0.17.2622 DF 3 3 9 15 MS 0.25 ( 3)( 4 ) ( a − 1)( b − 1) Therefore.. y2.40 0..22. This is the same result as found in Problem 4. As an initial guess. Tips are significant.7844 158.=1435. yˆ 23 = ay2. D.2615 F0 2. y.25. This is the same result as found in Problem 4.05.13.3.=38. and y.59.=357.3 the observations for chemical type 2 and bolt 3 and chemical type 4 and bolt 4 are missing. yˆ 23 = ay2.006914 F0. + 5 y'.' + by.04 12 4(2 ) + 5(17 ) − 28.0622 1.25. y 23 is missing.3. 4.7. NY 4.=153..9784 F0.5 = 3.1) − 144.86. y2.2 = = 9.11=3.16. Perform the exact analysis and compare the results.1.9=3.7625 193.. (a) Analyze the design by iteratively estimating the missing values as described in Section 4.62 ( 3)( 3) ( a − 1)( b − 1) Therefore. 12 12 Data is coded y-70. Analyze the problem by estimating the missing value. set 0 = chemical 2. Suppose that in Problem 4. 4 4(8) + 5(6 ) − 25. Suppose that the observation for chemical type 2 and bolt 3 is missing in Problem 4.05.41 12 0 ˆy 44 = ˆy 123 4-24 . y.

+by'..6 y44 − 6..59 3 3.44 12 4(8) + 5(6 ) − 30.21 Model 9.63 2 ˆy 44 = = 5. NY 4(8) + 5(6 ) − 30.08 2.20 Residual 18. (2008) Design and Analysis of Experiments.1ˆy44 = 6. b + ykv ) − ( y′ 2 .1560 0. we obtain: ∂y23 ∂y44 (a − 1)(b − 1)⎤ = ay' i . (c) Derive general formulas for estimating two missing values when the observations are in different blocks. + yiu ) + ( y′ 2 k. ˆy 44 = 2.63 12 4(2 ) + 5(17 ) − 27.1y23 y44 + R From ∂SS E ∂SS E = = 0 . D.63 12 ∴ ˆy23 = 5.20 A 9. ˆyiu ˆykv ⎢ − ⎥= ab ab ab ⎣ ⎦ 4-25 ) + ( y′ 2 .2 ˆy23 + 0.. − 14 ∑y 2 . C.v + ykv a ∂SS E ∂SS E = = 0 .v − y'.44 ˆy 44 = 2. Wiley.1560 not significant (b) Differentiate SSE with respect to the two missing values. +by' .2 ˆy44 = 3. j − y' . we obtain: ∂y23 ∂y44 1. The analysis of variance using these new data does not differ substantially from part (a).7 ⇒ ˆy23 = 5. SS E = y + y 2 iu From 2 kv ( y′ − i. and solve for estimates of the missing values.u + yiu ) + ( y′ 2 .44 2 ˆy 44 = = 2.59 3 3. SS E = ∑∑ y 2 ij − 15 ∑y 2 i. − ˆy kv ˆy iu ⎡⎢ ⎥ ab ab ab ⎦ ⎣ ⎡ ( a − 1 )( b − 1 ) ⎤ ay'k .j 1 + 20 ∑y 2 . + yiu + ykv ab ) 2 .Solutions from Montgomery.83 19 F Value Prob > F 2.83 4 39. 2 2 SS E = 0. Analyze the design using these two estimates of the missing values.8 y23 − 3.41 = 2.8 0.1ˆy23 + 1.6 y23 + 0.7 y44 + 0.53 Cor Total 184.63 ˆy 144 = Design Expert Output ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 156. equate the results to zero.63 These quantities are almost identical to those found in part (a).41 12 1..45 .08 0.

] ⎡1 − ( a − 1)2 ( b − 1)2 ⎤ ⎣ ⎦ (d) Derive general formulas for estimating two missing values when the observations are in the same block.Solutions from Montgomery. He has five subjects available for the experiment. − ( b − 1)( a − 1) [ ay 'k .′ ( a − 1)( b − 1) ( b − 1) ⎡⎣ ayk′ . + by. + by.05) and draw appropriate conclusions. ⎡1 − ab ( a − 1) ( b − 1) ⎤ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦+ yˆ iu = 2 2 ⎡ ⎤ 1 1 1 1 1 a − b − − a − b − ( )( ) ⎣ ( ) ( ) ⎦ ab [ ay 'k .′ + ˆykj (a − 1)(b − 1)2 ayk′ . Four different distances are of interest... ] yˆ kv = ⎡1 − ( a − 1)2 ( b − 1)2 ⎤ ⎣ ⎦ ay 'i..′j − y.′j − y.u − y '.′j − y. The data obtained follow. Distance (ft) 4 6 8 10 1 10 7 5 6 2 6 6 3 4 4-26 Subject 3 6 6 3 4 4 6 1 2 2 5 6 6 5 3 .. + by.. + by. + by '.. i≠k (same block j).. NY whose simultaneous solution is: 2 2 2 2 2 2 y 'i . ) + (y ′ 2 k.′ + ( a − 1)( b − 1) ( ayi′. (2008) Design and Analysis of Experiments.v − y '. Suppose that two observations yij and ykj are missing..u b ⎡1 − ( a − 1) ( b − 1) − ab ⎤ − y '. + by. + by '.j b + yij + ykj a ) + (y ′ 2 .′ + ˆyij (a − 1)(b − 1)2 (a − 1)(b − 1) (a − 1)(b − 1) whose simultaneous solution is: yˆij = ayi′. He is interested in the effect of the distance of the object from the eye on the focus time.′ − ( b − 1) ( a − 1) ⎡⎣ ayi′. Analyze the data from this experiment (use α = 0. + by. he decides to conduct the experiment in a randomized block design..′ )⎤⎦ 2 + ⎡1 − ( a − 1)2 ( b − 1)4 ⎤ ⎣ ⎦ ayk′ . we obtain ∂ yij ∂ ykj ˆyij = ˆy kj = ayi′. An industrial engineer is conducting an experiment on eye focus time.′j − y.′j − y. a ⎡1 − ( a − 1) ( b − 1) − ab ⎤ + y '.′j − y. D.v − y '. ) − (y ′ 2 + ykj .. + yij + ykj ) 2 ab ∂ SS E ∂ SS E = = 0 .′j − y. + by. Because there may be differences among individuals. SS E = yij2 + y kj2 − From (y ′ + yij i. Wiley.19. C.′ ⎤⎦ 2 yˆ kj = ( a − 1)( b − 1) ⎡⎣1 − ( a − 1) ( b − 1) 2 4 ⎤ ⎦ 4... + by '.

71 0. Each batch of new material is only large enough to permit five runs to be made.96 -0. D. E) on reaction time of a chemical process is being studied.27 Cor Total 84.95 3 10. NY Design Expert Output Response: Focus Time ANOVA for Selected Factorial Model Analysis of variance table [Partial sum of squares] Sum of Mean Source Squares DF Square Block 36.0448 0.0448 0.20 0.0025 0.0008 0.28 Prob > |t| 0.13 4.80 0.20 2.50 3-8 3.80 0. Wiley.6829 0. 4-27 . The effect of five different ingredients (A. B. 4.55 19 F Value Prob > F 8. Dev. D.61 8.28 42.50 Treatment 1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4 Mean Difference 1.61 implies the model is significant. C.30 4 9.50 R-Squared Adj R-Squared Pred R-Squared Adeq Precision 0. PRESS 1. each run requires approximately 1 1/2 hours.0025 significant The Model F-value of 8.71 0.432 Treatment Means (Adjusted.95 3 10.61 0.7842 Distance has a statistically significant effect on mean focus time. The experimenter decides to run the experiment as a Latin square so that day and batch effects can be systematically controlled.48 4. Batch 1 2 3 4 5 1 A=8 C=11 B=4 D=6 E=4 2 B=7 E=2 A=9 C=8 D=2 Day 3 D=1 A=7 C=10 E=6 B=3 4 C=7 D=3 E=1 B=6 A=8 5 E=3 B=8 D=5 A=10 C=8 The Minitab output below identifies the ingredients as having a significant effect on reaction time.71 0.20 DF 1 1 1 1 1 1 Standard Error 0. Std.98 A 32. C.71 0.60 0.50 2-6 5.00 1. She obtains the data that follow.25% chance that a "Model F-Value" this large could occur due to noise. Furthermore. Mean C.24 1.0736 0.24 4.20. Analyze the data from this experiment (use α = 0.Solutions from Montgomery.98 Residual 15.0012 0.07 Model 32.05) and draw conclusions.71 t for H0 Coeff=0 2.71 0. so only five runs can be made in one day.40 -0.6036 0.20 3. (2008) Design and Analysis of Experiments.85 23.V.50 4-10 3.60 1. There is only a 0.30 12 1.60 3. If Necessary) Estimated Standard Mean Error 1-4 6.1192 10.

Suppose that in Problem 4.500 Adj MS 24.500 153.21.440 15. y 354 is missing.′ k − 2 y .455 4. Minitab Output General Linear Model Factor Type Levels Values Order random 4 1 2 3 4 Operator random 4 1 2 3 4 Method fixed 4 A B C D Analysis of Variance for Time.127 F 11. ( p − 2)( p − 1) = 5[28 + 15 + 24] − 2(146) = 3.750 F 13.22. An industrial engineer is investigating the effect of four assembly methods (A.500 10.Solutions from Montgomery. Analyze the data from this experiment (α = 0.52 9. and perform the analysis using this value. using Adjusted SS for Tests Source Catalyst Batch Day Error Total DF 4 4 4 12 24 Seq SS 141.240 37. using Adjusted SS for Tests Source Method Order Operator Error Total DF 3 3 3 6 15 Seq SS 72..500 10.348 0. Furthermore.31 1. C. + y . the engineer knows that each assembly method produces such fatigue that the time required for the last assembly may be greater than the time required for the first. + y .500 18. NY Minitab Output General Linear Model Factor Type Levels Values Batch random 5 1 2 3 4 5 Day random 5 1 2 3 4 5 Catalyst fixed 5 A B C D E Analysis of Variance for Time.167 6. That is. B.24.. the engineer uses the Latin square design shown below..440 15.440 12.000 0.500 51.167 1.520 206. regardless of the method.010 4.440 12.520 Adj MS 35. a trend develops in the required assembly time. Order of Assembly 1 2 3 4 1 C=10 B=7 A=5 D=10 2 D=14 C=18 B=10 A=10 Operator 3 A=7 D=11 C=11 B=12 4 B=8 A=8 D=9 C=14 The Minitab output below identifies assembly method as having a significant effect on assembly time.20 the observation from batch 3 on day 4 is missing.167 17.500 18.240 37.640 Adj SS 141.05) draw appropriate conclusions.81 3.089 0.060 3.58 (3)(4) 4-28 . Estimate the missing value from Equation 4. D) on the assembly time for a color television component.360 3.860 3.000 Adj SS 72. To account for this source of variability.23 0.. (2008) Design and Analysis of Experiments. Four operators are selected for the study.98 P 0. C. ˆy 354 = [ ] ′ p y i′. D.500 51.′j . Wiley.81 P 0.004 0.

.23. τj.. 2.. p j =1 There are 3p+1 equations in 3p+1 unknowns. j ... p τˆ j = y.092 8. C..860 F 11.191 2. = y. p k =1 p β k : pμˆ + p k k =1 ∑ βˆ p ∑ ∑ βˆ p τˆ j + p j =1 τ j : pμˆ + p p τˆ j + p ∑ p αˆ i + p i =1 ∑τˆ j + pβˆ k = y. and treatments (τj) fixed. . 2.023 2.. p μ : p μˆ + p 2 ∑ p αˆ i + p i =1 ∑ j =1 p α i : pμˆ + pαˆ i + p ∑ = y. − y.2k + 2⎜ . SS E = ∑∑∑ 2 yijk − ∑ yi2......676 16. The usual conditions imposed are: ∑ i =1 p αˆ i = ∑ p τˆ j = j =1 ∑ βˆ k = 0 .. i = 1. Obtain least squares estimates of the model parameters αi.2... .290 0. k = 1.317 Adj MS 32..2 ⎟ ⎜p ⎟ p ⎠ ⎝ 4-29 . Derive the missing value formula (Equation 4.24) for the Latin square design.. j = 1.092 8.41 0.2j . − p ∑ y.169 4.... NY Minitab Output General Linear Model Factor Type Levels Values Batch random 5 1 2 3 4 5 Day random 5 1 2 3 4 5 Catalyst fixed 5 A B C D E Analysis of Variance for Time. using Adjusted SS for Tests Source Catalyst Batch Day Error Total DF 4 4 4 12 24 Seq SS 128.. Three side conditions are p necessary. The solution is then: k =1 y.. − y.. p − ∑ ⎛ y2 ⎞ y. The rank of the system is 3p-2..77 P 0.2.... D. k = yi . .849 Adj SS 128. βk..317 187. j ..k ..24. Wiley. p 4. p2 αˆ i = yi....676 16.. k = 1...Solutions from Montgomery.25 1..000 0... 2..764 34. − y. p k =1 p αˆ i + pτˆ j + p i =1 ∑ βˆ k = y. Consider a p x p Latin square with rows (αi).2.. i = 1.. (2008) Design and Analysis of Experiments. ..567 4. p μˆ = βˆk = yi.. j = 1. columns (βk).764 34. .

(a) Set up the normal equations for this model...] The p x p Latin square contains only p observations for each treatment.... C. j ..k − 2 y' .h + y.. ( p − 1)( p − 2) 4. ρˆ h = y.. + y' ... + yijk p p where R is all terms without yijk...... τˆ j = y. Note that α i ( h) and β k ( h ) are row and column effects in the hth square. The appropriate model is y ijkh = μ + ρ h + α i( h ) + τ j + β k ( h ) + ( τρ ) jh + ε ijkh ⎧ i = 1... ∑ j τˆ j = 0 . j. and ∑ (τˆρ ) ∑ h h i jh μˆ = y.. + y' . αˆ i ( h ) = yi..2. p ⎪ j = 1. ∑ j (τˆρ ) jh ∑ = 0 for each h. and ( τρ) jh is the interaction between treatments and squares.. j . Assume ρˆ h = 0 .. D. n where yijkh is the observation on treatment j in row i and column k of the hth square. j ....h ⎛ ^ ⎞ ⎜ τρ ⎟ = y.. To obtain more replications the experimenter may use several squares..Solutions from Montgomery. ⎝ ⎠ jh 4-30 = 0 for each j....h − y.2. or y ijk = ( ) p y' i . [See Cochran and Cox (1957). Designs involving several Latin squares..2... Wiley..′ + yijk ) 2 +R p2 ∂SS E = 0 . + y' . ∑ k .. j .h − y.k 2 .2. (2008) Design and Analysis of Experiments.. From y ijk ( p − 1)( p − 2) = p ( ) − ( y′ 2 ) p + yijk ) 2 + p ( 2 y...25.... − y.. and ρ h is the effect of the hth square.. 2 .k − 2 y' . and solve for estimates of the model parameters. ijk . we obtain: ∂y ijk p y' i ....h − y. j .. − y.. αˆ i (h ) = 0 .h βˆk ( h ) = y.. John (1971).. say n. NY Let yijk be missing. Then ( y′ + y ) − ( y′ − 2 SS E = y 2 ijk i .. p ⎪⎩ h = 1.kh − y. p ⎪ ⎨ ⎪k = 1.. + y' ... and βˆ k (h ) = 0 that appropriate side conditions on the parameters are for each h. It is immaterial whether the squares used are the same are different.

..2 h ∑ p − np n(p-1) Columns y. (2008) Design and Analysis of Experiments... Minitab Output General Linear Model Factor Type Levels Values Catalyst fixed 5 A B C D E Batch random 5 1 2 3 4 5 Day random 4 1 2 3 4 4-31 . The data can be analyzed as a balanced incomplete block design with a = b = 5... Develop an appropriate analysis for the remaining data.26..... υ1 = p − 1 υ 2 = ( p − 2)( p − 1) 4. Treatment x Squares ∑ Rows yi2. Suppose that in Problem 4.. Two methods of analysis exist: (1) Use the general regression significance test.2j .27. C. 2 np2-1 np 4. Source SS DF y.. 2 y. For the fixed effects model use: Φ2 = ∑ pτ = ∑ τ pσ σ 2 j 2 2 j 2 . r = k = 4 and λ = 3.h − p-1 2 y. Using either approach will yield the same analysis of variance..20 the data taken on day 5 were incorrectly analyzed and had to be discarded..2j . Wiley. υ1 = p − 1 υ 2 = ( p − 2)( p − 1) For the random effects model use: λ = 1+ pσ τ2 σ2 . D.Solutions from Montgomery. 2 Treatments ∑ np − np Squares ∑p y.2 h 2 y.2 h ∑ p − np n(p-1) Error subtraction p − − SSTreatments − SS Squares np 2 n(p-1)(p-2) ∑∑∑∑ y 2 ijkh Total (p-1)(n-1) − 2 y. Discuss how the operating characteristics curves in the Appendix may be used with the Latin square design..h y. NY (b) Write down the analysis of variance table for this design.. or (2) recognize that the design is a Youden square.2kh y. n-1 np 2 2 y......

000 10.167 11.017 F 7.65 0. workplace (α. five standing times.04 P 0.Solutions from Montgomery. (A. using Adjusted SS for Tests Source Catalyst Batch Day Error Total DF 4 4 3 8 19 Seq SS 119.133 Adj MS 30. NY Analysis of Variance for Time.785 0. A fourth factor.05) and draw conclusions.400 46.05) and draw conclusions.29.700 3. δ) may be introduced and another experiment conducted. ε).950 32.21 the engineer suspects that the workplaces used by the four operators may represent an additional source of variation. γ.042 2.001 0.58 P 0. β. β. D. Minitab Output General Linear Model Factor Type Levels Values Time fixed 5 A B C D Catalyst random 5 a b c d Batch random 5 1 2 3 4 Acid random 5 1 2 3 4 E e 5 5 Analysis of Variance for Yield. E) and five catalyst concentrations (α.100 5.667 6.73 0.443 4.850 F 14. δ.667 6. yielding the Graeco-Latin square that follows. five acid concentrations.729 0. Order of Assembly 1 2 3 4 1 Cβ=11 Bα=8 Aδ=9 Dγ=9 2 Bγ=10 Cδ=12 Dα=11 Aβ=8 4-32 Operator 3 Dδ=14 Aγ=10 Bβ=7 Cα=18 4 Aα=8 Dβ=12 Cγ=15 Bδ=6 .598 0. (2008) Design and Analysis of Experiments. The Graeco-Latin square that follows was used. γ.000 10.000 24.950 32.400 46. The yield of a chemical process was measured using five batches of raw material.800 12. Analyze the data from this experiment (use α = 0. Suppose that in Problem 4.48 0.43 1.000 24.500 6.800 436.317 4.28.800 Adj MS 85. C.550 Adj SS 120. using Adjusted SS for Tests Source Time Catalyst Batch Acid Error Total DF 4 4 4 4 8 24 Seq SS 342. D.000 Adj SS 342. Acid Concentration Batch 1 2 3 4 5 1 Aα=26 Bβ=16 Cγ=19 Dδ=16 Eε=13 2 Bγ=18 Cδ=21 Dε=18 Eα=11 Aβ=21 3 Cε=20 Dα=12 Eβ=16 Aγ=25 Bδ=13 4 Dβ=15 Eγ=15 Aδ=22 Bε=14 Cα=17 5 Eδ=10 Aε=24 Bα=17 Cβ=17 Dγ=14 The Minitab output below identifies standing time as having a significant effect on yield.133 170.008 0.51 0. Wiley. C.800 11.646 4. Analyze the data from this experiment (use α = 0.800 12.917 2.000 2. B.

833 0.. 25 4 24 ... ∑ 5 i =1 25 y2 1 5 2 y.333 2.02 0.500 Adj MS 31. there are only three degrees of freedom for error... Construct a 5 x 5 hypersquare for studying the effects of five factors..616 0.. (2008) Design and Analysis of Experiments.500 19.. Latin letters = factor 3.500 0. ∑ 5 m =1 25 2 y..843 Method and workplace do not have a significant effect on assembly time...500 9. ∑ 5 l =1 25 Greek Letters Numbers 4 4 SSE by subtraction Error 5 Total DF y2 1 5 2 yi.30.000 7... Exhibit the analysis of variance table for this design.. so the test is not very sensitive.000 7...69 0. columns = factor 2. − .. − .Solutions from Montgomery. NY Minitab Output General Linear Model Factor Type Levels Values Method fixed 4 A B C D Order random 4 1 2 3 4 Operator random 4 1 2 3 4 Workplac random 4 a b c d Analysis of Variance for Time.. j ....27 P 0. D...500 19..996 0. 1 5 2 y.167 F 3.. C. The analysis of variance table is: SS Source Rows Columns Latin Letters 4 4 4 y2 1 5 2 y.. using Adjusted SS for Tests Source Method Order Operator Workplac Error Total DF 3 3 3 3 3 15 Seq SS 95.500 0..500 27... − .m − . Wiley. − ∑ 5 j =1 25 5 5 5 5 ∑∑∑∑∑ y i =1 j =1 k =1 l =1 m =1 4-33 2 ijklm y2 − ...47 0.k .2 l . Three 5 x 5 orthogonal Latin Squares are: αβγδε γδεαβ εαβγδ βγδεα δεαβγ ABCDE BCDEA CDEAB DEABC EABCD 12345 45123 23451 51234 34512 Let rows = factor 1. Greek letters = factor 4 and numbers = factor 5.167 6.. 4.000 Adj SS 95.167 0..500 150.. ∑ 5 k =1 25 5 y2 1 y. However...500 27.

Compare your results to the approximate analysis of these data given in Table 4.17.00* 0. Batch 1 2 3 4 1 C=10 B=7 A=5 D=10 (32) Square 1 . Consider the data in Problems 4.08 0. the data was transformed by multiplying by 10 and substracting 95.=109 y. analyze the data using the method developed in Problem 4.Operator 2 3 4 D=14 A=7 B=8 C=18 D=11 A=8 B=10 C=11 D=9 A=10 B=12 C=14 (52) (41) (36) Batch 1 2 3 4 1 C=11 B=8 A=9 D=9 (37) Square 2 .3.Operator 2 3 4 B=10 D=14 A=8 C=12 A=10 D=12 D=11 B=7 C=15 A=8 C=18 B=6 (41) (49) (41) Assembly Methods A B C D Source Assembly Methods Squares AxS Assembly Order (Rows) Operators (columns) Error Total SS 159. D.25.=68 y. C.7.50 45.50 8.79 F0 14.50 2..50 303. the data in Problems 4. 4-34 .75 3.00 70. Analyze this data by using the exact analysis of the missing value problem discussed in Section 4.77 Significant at 1%.1.=90 DF 3 1 3 6 6 12 31 MS 53.2.1. Suppressing the Greek letters in 4. Consider the randomized block design with one missing value in Problem 4.4.29. Wiley.=65 y. (2008) Design and Analysis of Experiments.17.75 19...17 and 4..92 3.4. To simplify the calculations.29.50 Row Total (39) (44) (35) (46) 164=y…1 Row Total (43) (42) (42) (41) 168=y…2 Totals y. NY 4.32.17 11.21 and 4.25 0. 4.31.Solutions from Montgomery.

25 = 39. β 3 = . β ) − R(μ . Now R(μ . τˆ 2 = . β2 = . βˆ 4 = . β3 = . β ) = 145.78 j =1 With 7 degrees of freedom.22 which is identical to SSE obtained in the approximate analysis. NY μ : 15μˆ +4τˆ1 +3τˆ2 +4τˆ4 τ1 : 4 μˆ +4τˆ1 τ2 : 3μˆ τ3 : 4 μˆ τ4 : 4 μˆ β1 : 4 μˆ +τˆ1 +τˆ2 +τˆ3 +τˆ4 β2 : 4 μˆ +τˆ1 +τˆ2 +τˆ3 +τˆ4 β3 : 3μˆ +τˆ1 +τˆ3 +τˆ4 β4 : 4 μˆ +τˆ1 +τˆ3 +τˆ4 +4 βˆ1 + βˆ +4 βˆ2 + βˆ + βˆ1 + βˆ + βˆ2 + βˆ + βˆ1 +4 βˆ + βˆ2 1 +3τˆ2 +4τˆ3 1 +4τˆ4 +τˆ2 ∑ τˆ = ∑ βˆ Applying the constraints μˆ = +4τˆ3 i j 2 2 +3βˆ3 + βˆ 3 +4 βˆ4 + βˆ = 17 + βˆ4 + βˆ =1 4 = −2 + βˆ4 = 15 =3 4 + βˆ3 + βˆ 3 = −4 1 +4 βˆ2 = −3 +3βˆ3 =6 +4 βˆ4 = 18 = 0 .τ . β4 = 36 36 36 36 36 1 36 36 36 36 R(μ .Solutions from Montgomery. β ) = μˆ y. In general.. β ) = R(μ ..25 .00 − 138. The normal equations used are: μ : 15μˆ +4 βˆ1 +4 βˆ2 Applying the constraint μˆ = β1 : 4 μˆ β2 : 4 μˆ β3 : 3μˆ β4 : 4 μˆ ∑ βˆ j +3βˆ3 +4 βˆ4 +4 βˆ1 = 17 = −4 +4 βˆ2 = −3 +3βˆ3 =6 +4 βˆ4 = 18 = 0 . β ) = μˆ y . j = 99. τˆ 1 = . β1 = . + 16 16 16 16 16 4 ∑ βˆ j j =1 with 4 degrees of freedom. To test Ho: τ i = 0 the reduced model is yij = μ + β j + ε ij . we obtain: 19 ˆ 13 53 −35 ˆ −31 ˆ .78 = 6.τ . (2008) Design and Analysis of Experiments.78 − 99. β ) = 138. Wiley. we obtain: 41 94 ˆ −14 −24 −59 24 ˆ 121 −77 ˆ −68 ˆ . j .53 = SS Treatments 4-35 y .j i =1 = 138. R(τ μ . β = . + 4 4 ∑ τˆ y + ∑ βˆ y i i.τ . β2 = . SS E = ∑∑ y 2 ij − R(μ . τˆ 4 = . τˆ 3 = . C. ∑∑ y 2 ij = 145.00 . D. the SSE in the exact and approximate analyses will be the same.

D. The normal equations used are: Model Restricted to β j = 0 : μ : 15μˆ +4τˆ1 +3τˆ2 +4τˆ3 +4τˆ4 = 17 τ 1 : 4μˆ +4τˆ1 =3 τ 2 : 3μˆ +3τˆ2 =1 τ 3 : 4μˆ +4τˆ3 = −2 τ 4 : 4μˆ +4τˆ4 = 15 Applying the constraint ∑τˆ = 0 .53 78. β ) is used to test Ho: τ i = 0 . Source Tips Blocks Error Total DF 3 3 8 14 SS(exact) 39. An engineer is studying the mileage performance characteristics of five types of gasoline additives.Solutions from Montgomery. 4. τˆ4 = 12 12 12 12 12 R(μ .05) and draw conclusions.τ ) = R(μ . The sum of squares for blocks is found from the reduced model y ij = μ + τ i + ε ij .78 − 59. because of a time constraint. Additive 1 2 3 4 5 1 14 12 13 11 2 17 14 11 12 4-36 Car 3 14 13 11 10 4 13 13 12 12 5 12 10 9 8 . C. R (β μ .53 6.. In the road test he wishes to use cars as blocks. R (τ μ . SST ≠ SSTips + SS Blocks + SS E .83 i =1 with 4 degrees of freedom.33.τ ) = 138.73 Note that for the exact analysis.22 125.τ ) = μˆ y. τˆ3 = . (2008) Design and Analysis of Experiments. he must use an incomplete block design.95 6. NY with 7-4=3 degrees of freedom. however. τˆ1 = . Analyze the data from this experiment (use α = 0. Wiley.95 = SS Blocks with 7-4=3 degrees of freedom. we obtain: i μˆ = 13 −4 −9 −19 32 . β ) − R(μ .83 = 78.74 SS(approximate) 39.τ . = 59.98 79. He runs the balanced design with the five blocks that follow.22 125. τˆ2 = . + 4 ∑ τˆ y i i.

9106 F 9.0167 76.55 0.06 4.9333 8. The Minitab General Linear Model procedure is a widely available package with this capability. The output from this routine for Problem 4. (2008) Design and Analysis of Experiments. Construct a set of orthogonal contrasts for the data in Problem 4.33 follows.2333 10.8083 0. respectively.41 0.0167 Adj MS 8. C. D. using Adjusted SS for Tests Source Additive Car Error Total DF 4 4 11 19 Seq SS 31.19 33. Minitab Output General Linear Model Factor Type Levels Values Additive fixed 5 1 2 3 4 5 Car random 5 1 2 3 4 5 Analysis of Variance for Mileage.03 1.33.34.2333 10.001 0.10 4.81 9.426 1.7333 35.67 P 0. One possible set of orthogonal contrasts is: H0 H0 H0 H0 : μ 4 + μ5 = μ1 + μ 2 : μ1 = μ 2 : μ 4 = μ5 : 4 μ3 = μ 4 + μ5 + μ1 + μ 2 (1) (2) (3) (4) The sums of squares and F-tests are: Brand -> Qi 1 33/4 2 11/4 3 -3/4 4 -14/4 5 -27/4 (1) (2) (3) (4) -1 1 0 -1 -1 -1 0 -1 0 0 0 4 1 0 -1 -1 1 0 1 -1 ∑ ci Qi -85/4 22/4 -13/4 -15/4 Contrasts (1) and (2) are significant at the 1% and 5% levels. The adjusted sums of squares are the appropriate sums of squares to use for testing the difference between the means of the gasoline additives. Compute the sum of squares for each contrast.Solutions from Montgomery.21 .001 4. The gasoline additives have a significant effect on the mileage.9500 Adj SS 35. Wiley.7000 35. NY There are several computer software packages that can analyze the incomplete block designs discussed in this chapter. 4-37 SS F0 30.

10 168.12 P 0. Seven different hardwood concentrations are being studied to determine their effect on the strength of the paper produced.43 394. τ\$1 = −9 / 8 . the analyst uses the balanced incomplete block design that follows.62 394. However the pilot plant can only produce three runs each day. Minitab Output General Linear Model Factor Type Levels Values Concentr fixed 7 2 4 6 8 10 12 14 Days random 7 1 2 3 4 5 6 7 Analysis of Variance for Strength.070 4. τ\$ 2 = −7 / 8 .35.29 Adj SS 1317.57 2600. The output from this routine for Problem 4.68 21.42 3. Analyze this experiment (use α = 0. Analyze the data in Example 4. D. The adjusted sums of squares are the appropriate sums of squares to use for testing the difference between the means of the hardwood concentrations. The Minitab General Linear Model procedure is a widely available package with this capability. τ\$ 4 = 20 / 8 .10 168.Solutions from Montgomery. C.05) and draw conclusions. μ : 12 μˆ +3τˆ1 +3τˆ2 +3τˆ3 +3τˆ4 +3βˆ1 +3βˆ2 +3βˆ3 +3βˆ4 = 870 τ1 : 3μˆ = 218 τ2 : 3μˆ τ3 : 3μˆ τ4 : 3μˆ β1 : 3μˆ +τˆ1 β2 : 3μˆ +τˆ1 β3 : 3μˆ β4 : 3μˆ Applying the constraints +3τˆ1 + βˆ1 +3τˆ2 2 +3τˆ3 +τˆ1 +τˆ3 +τˆ2 +τˆ3 +τˆ2 +τˆ3 +τˆ2 ∑ τ\$i = ∑ β\$ j + βˆ2 + βˆ +3τˆ4 + βˆ1 + βˆ +τˆ4 +3βˆ1 + βˆ2 + βˆ4 + βˆ + βˆ3 + βˆ 4 = 216 3 + βˆ3 1 + βˆ4 +τˆ4 = 224 +3βˆ3 = 207 +3βˆ4 = 0 . (2008) Design and Analysis of Experiments.5 using the general regression significance test. we obtain: μ\$ = 870 / 12 .36.002 0. 4-38 = 222 = 221 +3βˆ2 +τˆ4 = 214 = 218 .57 65.07 F 10.35 follows.57 Adj MS 219. Hardwood Concentration (%) 2 4 6 8 10 12 14 1 114 126 2 Days 4 3 120 137 5 120 6 7 117 119 117 129 141 134 149 150 145 143 118 120 123 130 136 127 There are several computer software packages that can analyze the incomplete block designs discussed in this chapter. As days may differ. τ\$ 3 = −4 / 8 . NY 4. Wiley. using Adjusted SS for Tests Source Concentr Days Error Total DF 6 6 8 20 Seq SS 2037.

4-39 . β\$4 = 0 / 8 4 4 i =1 j =1 R ( μ . β ) is used to test Ho: τ i = 0 . R(τ μ . βˆ3 = −31/ 8 .75 = SS Treatments with 7 – 4 = 3 degrees of freedom. β1 = .75 = 3.τ . j = 63. β4 = 12 6 6 6 6 R(μ . β ) = μˆ y . we obtain: 870 ˆ 13 ˆ −21 ˆ 7 1 . β ) = 63152. y ij = μ + τ i + ε ij . βˆ2 = . ∑ ∑ y ij2 = 63. R (τ μ . The normal equations used are: Model Restricted to β j = 0 : μ : 12μˆ +3τˆ1 +3τˆ2 τ 1 : 3μˆ +3τˆ1 τ 2 : 3μˆ +3τˆ2 τ 3 : 3μˆ τ 4 : 3μˆ +3τˆ3 +3τˆ4 +3τˆ3 +3τˆ4 = 870 = 218 = 214 = 216 = 222 The sum of squares for blocks is found as in Example 4.130. + ∑τˆi yi ..152. βˆ2 = 24 / 8 . The sum of squares for blocks is found from the reduced model y ij = μ + τ i + ε ij . β ) − R(μ .00 SS E = ∑ ∑ y ij2 − R ( μ . + ∑ βˆ j y. To test Ho: τ i = 0 the reduced model is yij = μ + β j + εij . (2008) Design and Analysis of Experiments.Solutions from Montgomery.5.00 j =1 with 4 degrees of freedom. C. β ) = R(μ . NY β\$1 = 7 / 8 .75 with 7 degrees of freedom. + 4 ∑ βˆ j y . Wiley. We may use the method shown above to find an adjusted sum of squares for blocks from the reduced model.. β ) = 63156.75 − 63130. β ) = μˆ y.00 = 22.00 − 63152. D. The normal equations used are: Applying the constraint μˆ = μ : 12μˆ +3βˆ1 +3βˆ2 +3βˆ3 +3βˆ4 = 870 β1 : 3μˆ = 221 β2 : 3μˆ β3 : 3μˆ β4 : 3μˆ ∑ βˆ j +3βˆ1 +3βˆ2 = 224 +3βˆ3 = 207 +3βˆ4 = 218 = 0 . τ .156. j = 63.τ . β3 = .25 .

β ) = μˆ y . + ∑ βˆ j y . j j =1 and the sum of squares we need is: a b b i =1 j =1 j =1 R(τ μ . j ⎥ ⎢ y . + − ⎢ k − k − k k ⎥⎥ i =1 j =1 ⎢ ⎢ ⎥ ⎣ ⎦ ∑ ∑ a 1 a ⎛ ⎞ a ⎛ kQ R (τ μ . a ⎡ ⎤ y n ij τˆ i ⎢ .Solutions from Montgomery. − ∑ nij y.. β ) = ∑τˆi ⎜ yi . We may use the general regression significance test to derive the computational formula for the adjusted treatment sum of squares.β ) = μˆ y . + b ∑n y ij . + R (τ μ . j kμˆ y . NY k 4. j i =1 and from this we have: ky.37. We will need the following: τˆ i = kQi . i =1 j y. Prove that ∑ a i =1 Qi2 (λa ) is the adjusted sum of squares for treatments in a BIBD.2j − ky. j μˆ − y. β ) = μˆ y . Wiley.35)..j 2 2 ⎥ a b y.2j k The normal equation for β is.τ .. j − ∑ y . j i =1 a b ∑ τˆ y + ∑ βˆ i i. j i =1 τˆ i y i . + ∑ τˆ i y i . D. (2008) Design and Analysis of Experiments. j a ∑ n τˆ ij i i =1 therefore. j ⎟ = ∑ Qi ⎜ i k i =1 i =1 ⎝ ⎠ i =1 ⎝ λ a 4-40 ∑ a ⎛ Qi2 ⎞ ⎞ k = ⎜ ⎟ ≡ SSTreatments ( adjusted ) ∑ ⎟ ⎠ i =1 ⎝ λ a ⎠ . β : kμˆ + a ∑ n τˆ ij i + kβˆ j = y. C. j βˆ j = y. kQ = kyi . − (λa ) i R(μ . from equation (4.

intrablock and combined estimates is: Parameter τ1 τ2 τ3 τ4 τ5 Intrablock 2. Find a BIBD for this experiment with six blocks. The design is discussed by John (1971. D. with each main effect and interaction successively confounded (7 replications) forming the 14 blocks. An experimenter wishes to compare eight treatments in blocks of four runs.80 4-41 Interblock -1.63 . k = 3. b = 14. An experimenter wishes to compare four treatments in blocks of two runs. A summary of the interblock. (2008) Design and Analysis of Experiments.38. It may be generated from a 23 factorial design confounded in two blocks of four observations each. pg. a = 4. 473).33. NY 4.40.91 and σˆ β2 = 2.33 uses σˆ 2 = 0.80 9. λ = 3. Find a BIBD with 14 blocks and λ = 3.93 -1.80 . pg. C. The interblock analysis for Problem 4. and r = 2 4. r = 2 and k = 4.39. Perform the interblock analysis for the design in Problem 4. The design follows: Blocks 1 2 3 4 5 6 7 8 9 10 11 12 13 14 1=(I) X 2=a 3=b X X X 5=c X X X X X X X X X X X X X X X X X X X X X X 7=bc X 8=abc X X X X X X X X X X X 6=ac X X X X X 4=ab X X X X X X X X X X X X X 4.20 -1. 222) and Cochran and Cox (1957. Treatment 1 2 3 4 Block 1 X X Block 2 X Block 3 X Block 4 Block 5 X X X X X Block 6 X X X Note that the design is formed by taking all combinations of the 4 treatments 2 at a time.20 -5.20 0.Solutions from Montgomery. Wiley. b = 6.73 -0. The design has parameters a = 8. The parameters of the design are λ = 1.20 -0.80 0.

41. so a balanced design with a −1 7 7 these parameters cannot exist.14 -0.36 -0. These conditions imply that λ = r ( k − 1) 8(3) 24 = = . j .92 1.21 21. D. Wiley. and combined estimates is give below Parameter Intrablock -12.Solutions from Montgomery. r = 8. j = (k − 1) yi .29 10. ⎟ ⎟ j =1 ⎠ b ∑ y i . and b = 16 does not exist. − (λa ) k (λa ) 2 .35 uses σˆ 2 = 21. and kQi = kyi . Therefore.29 -8.67 -6.03 4. and Qi = y i .57 10. Show that the variance of the intra block estimators { τ\$ i } is Note that τˆ i = kQi 1 .07 ] ( 6 ) = = 19. k = 4. since λ (a − 1) = r (k − 1) .86 τ1 τ2 τ3 τ4 τ5 τ6 τ7 Interblock -11.12 .21 -22.43. NY 4. ⎛ nij y.61 14. 4. contains r observations. intrablock. j − yi . (2008) Design and Analysis of Experiments. σ β2 = ⎣ a ( r − 1) 7 ( 2) A summary of the interblock. and the quantity in the parenthesis is the sum of r(k-1) observations. − j =1 ∑ ⎞ nij y.57 2.76 10.71 Combined -12. we have: 4-42 . Perform the interblock analysis for the design in Problem 4. note that: 1 k2 [r (k − 1)σ 2 {(k − 1) + 1}] = r (k − 1)σ 2 k ⎛ k ⎞ ⎛ k ⎞ r (k − 1) 2 kr (k − 1) 2 V (τˆ i ) = ⎜ ⎟ V (Q )i = ⎜ ⎟ σ = σ a λ k ⎝ ⎠ ⎝ λa ⎠ (λa )2 2 2 However.42.38 -7. Verify that a BIBD with the parameters a = 8.68 − 21. not including treatment i.43 -8. − ⎜ ⎜ j =1 ⎝ b b ∑ k ( (a − 1) )σ 2 nij y . V (kQi ) = k 2V (Qi ) = r (k − 1)2σ 2 + r (k − 1)σ 2 or V (Qi ) = To find V (τˆ i ) .71 -5.79 9.71 13.79 -4. which is not an integer.35.07 and ⎡ MS Blocks ( adj ) − MS E ⎤⎦ ( b − 1) [ 65. The interblock analysis for problem 4. C.

In the balanced case.5 1. (Hint: In the extended incomplete block design. plus a block from the balanced incomplete block design with k* = k-a=9-5=4 and λ*=3.4.2.5 Note that r=9. suppose we have a=5 treatments. The difference between this and the residual sum of squares is due to interaction.4. and r= r* + b = 4+5=9.3.2j y .2.) As an example of an extended incomplete block design.3.3.5 1.4.5 1.4. we have λ = 2r-b+λ*. The analysis of variance table is shown below: Source Treatments (adjusted) Blocks Interaction Error Total SS k Qi2 ∑ aλ y .3.. Extended incomplete block designs. and λ = 2rb+λ*=18-5+3=16. (2008) Design and Analysis of Experiments. b=5 blocks and k=9.3.2. Write out the statistical analysis. The design is: Block 1 2 3 4 5 Complete Treatment 1.5 1.2.4.44.2.4.2. Wiley. N ∑ − ∑∑ 4-43 DF a-1 b-1 (a-1)(b-1) b(k-a) N-1 .5 1.4.3. A design could be found by running all five treatments in each block.3.5 1.2. the {τˆ i } are not independent.4.Solutions from Montgomery. Occasionally the block size obeys the relationship a < k < 2a. since the augmenting incomplete block design has r*=4.2 k N Subtraction [SS between repeat observations] y2 y ij2 − . An extended incomplete block design consists of a single replicate or each treatment in each block along with an incomplete block design with k* = k-a.. Since some treatments are repeated in each block it is possible to compute an error sum of squares between repeat observations. NY V (τˆ i ) = k (a − 1) λa 2 σ2 ( ) Furthermore.5 Incomplete Treatment 2. and λ*.3.2.4 1. C.3. D.5 1. this is required to show that V τˆ i − τˆ j = 2k 2 σ λa 4. the incomplete block design will have parameters k* = k-a. r* = r-b.