You are on page 1of 29

# Blocher−Stout−Cokins−Chen

:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision
Making

Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

Cost Estimation 189

The learning rate is Y/a, and the learning index (b) can be determined from the learning
rate using an algebraic approach.6 This form of the learning model is very general and will
allow consideration of other learning assumptions, in addition to the doubling-of-output base
commonly used.

LIMITATIONS OF LEARNING CURVE ANALYSIS
Although learning curve analysis can signiﬁcantly enhance the ability to predict costs when learning occurs, three inherent limitations and problems are associated with the use of this method.
The ﬁrst and key limitation of using learning curves is that the approach is most appropriate
for labor-intensive contexts that involve repetitive tasks performed for long production runs for
which repeated trials improve performance, or learning. When the production process is designed to maximize ﬂexibility and very fast set-up times for manufacturing machinery using robotics and computer controls as many manufacturers now do, the manufacturing setting requires
relatively little repetitive labor and consequently relatively little opportunity for learning.
A second limitation is that the learning rate is assumed to be constant (average labor time
decreases at a ﬁxed rate as output doubles). In actual applications, the decline in labor time
might not be constant. For example, the learning rate could be 80 percent for the ﬁrst 20,000
units, 90 percent for the next 35,000 units, and 95 percent thereafter. Such differences indicate
the need to update projections based on the observed progression of learning.
Third, a carefully estimated learning curve might be unreliable because the observed change
in productivity in the data used to ﬁt the model was actually associated with factors other than
learning. For example, the increase in productivity might have been due to a change in labor
mix, a change in product mix, or some combination of other related factors. In such cases, the
learning model is unreliable and produces inaccurate estimates of labor time and cost.

Appendix B
LEARNING OBJECTIVE 6
Use statistical measures to
evaluate a regression analysis.

Regression Analysis
This appendix uses an example to explain the development of a regression estimate and the
related statistical measures. Then we interpret the statistical measures to assess the precision
and reliability of the regression.

THE REGRESSION ESTIMATE
To illustrate the manner in which a regression estimate is obtained, we use the data in Exhibit
6.4. Recall that regression analysis ﬁnds the unique line through the data that minimizes the
sum of the squares of the errors, where the error is measured as the difference between the
values predicted by the regression and the actual values for the dependent variable. In this
example, the dependent variable, supplies expense (Y), is estimated with a single independent
variable, production level (X). The regression for the three data points is
Y = a + b ⫻ X = \$220 + \$0.75 ⫻ X
6

To determine the learning index (b) for a given learning rate, ﬁrst develop a linear expression for the general model by taking
the natural log of both sides of the equation.
ln(Y ) = ln (a ) + b ⫻ ln ( X )
so that:
b =

ln (Y) − ln (a) ln (Y / a)
=
ln (X)
ln (X)

Thus, if we consider the changes in Y/a as X increases, the index b simpliﬁes to the ratio of the learning rate to the rate of
increase in output, or
b =

ln (learningrate)
ln (percent increase inoutput /100)

For example, to calculate the learning index for the doubling-output assumption (200 percent), we use:
b = ln (learningrate)/ ln (2)
And, for a learning rate of 80 percent, the learning index is therefore ln (.8)/ ln (2)= −.3219. The index is negative because
average unit labor time decreases with increasing output.

Blocher−Stout−Cokins−Chen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

190 Part Two

II. Planning and Decision
Making

Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

EXHIBIT 6B.1 Variance Components for Regression Analysis: Total Variance, Regression Variance, and Error Variance
1

2

Dependent
Variable
Y
250
310
325

Variance Components
6

3

4

5

Independent
Variable
X

Mean of Y
(YM)

Regression
Prediction for Y
(YE)

Total
Variance of Y
(T) = (Y − YM)

Regression
Variance
(R) = (YE − YM)

Error
Variance
(E) = (Y − YE)

7

50
100
150

295
295
295

257.5
295.0
332.5

(45)
15
30

(37.5)
0.0
37.5

(7.5)
15.0
(7.5)

The intercept term, labeled a, and the coefﬁcient of the independent variable, labeled b,
are obtained from a set of calculations performed by spreadsheet and other programs and are
described in basic textbooks on probability and statistics. The calculations themselves are
beyond the scope of this text. Our focus is on the derivation and interpretation of the statistical
measures that tell management accountants something about the reliability and precision of
the regression.

STATISTICAL MEASURES
The statistical measures of the reliability and precision of the regression are derived from an
analysis of the variance of the dependent variable. Variance is a measure of the degree to which
the values of the dependent variable vary about its mean. The term analysis of variance is used
because the regression analysis is based on a separation of the total variance of the dependent
variable into error and explained components. The underlying concept is that in predicting individual values for the dependent variable, the regression is explaining changes (i.e., variance)
in the dependent variable associated with changes in the independent variable. The variance in
the dependent variable that is not explained is called the residual, or error variance. Thus, the
regression’s ability to correctly predict changes in the dependent variable is a key measure of its
reliability and is measured by the proportion of explained to error variances. Based on the data
in Exhibit 6.4, Exhibit 6B.1 shows how the variance measures are obtained.
The ﬁrst two columns of Exhibit 6B.1 show the data for the independent (X) and dependent (Y) variables. Column (3) shows the mean of the dependent variable (YM), and column
(4) the regression prediction (YE) for each of the points. The last three columns indicate the
three variance measures. Column (5) shows the total variance, or variance of the dependent
variable, measured as the difference between each data point and the mean of the dependent
variable (Y − YM). Column (6) shows the variance explained by the regression (YE − YM), and

EXHIBIT 6B.2
\$400
Dependent Variable (supplies expense)

Variance Components for
Regression Analysis

The regression line
E ⴝ 15
T ⴝ 15
Rⴝ0

350
R ⴝ 37.5

E ⴝ 7.5
R ⴝ 37.5

300
T ⴝ 30
250

E ⴝ 7.5
T ⴝ 45

200

50

Mean of the
dependent
variable

Key:
Actual data point
R ⴝ regression distance (from mean to line)
E ⴝ error distance (from line to data point)
T ⴝ total distance (from mean to data point)

100
150
Independent Variable (units of output)

The degrees of freedom for each component of variance represent the number of independent choices that can be made for that component. how likely the regression is to continue to provide accurate predictions over time and for different levels of the independent variables. explained variance.3.8A and 6. The analysis of variance table serves as a useful basis to discuss the key statistical measures of the regression. The standard error is interpreted as a range of values around the regression estimate such that we can be approximately 67 percent conﬁdent the actual value lies in this range (see Exhibits 6. respectively.5 (45)2 + (15)2 + (30)2 = 3.3. (Y − YE).4 Six Key Statistical Measures Precision 1. Precision of the Regression The standard error of the estimate (SE) is a useful measure of the accuracy of the regression’s estimates. Precision of the regression (measured by the standard error of the estimate) Reliability 2. This is best illustrated in the analysis of variance table in Exhibit 6B. exists between the conﬁdence level and the width of the interval. Mean squared variance is the ratio of the amount of variance of a component to the number of degrees of freedom for that component. and error variance. Precision refers to the ability of the regression to provide accurate estimates—how close the regression’s estimates are to the unknown true value.37 EXHIBIT 6B. An inverse relationship.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.0 Cost Estimation 191 Degrees of Freedom Mean Squared Variance 1 1 2 2. 2008 Appendix B: Regression Analysis Chapter 6 EXHIBIT 6B.575. Fourth Edition II. mean squared variance. Statistical reliability (F-statistic) 4. Goodness of ﬁt (R-squared) 3.150.52 + 152 + 7. The third column shows the degrees of freedom for each component. The ﬁrst two columns of the table show the type and amount of variance for each of the three variance terms. The value of the SE for a given regression can be obtained directly from the analysis of variance table as follows: SE = Mean Square error = 337.812. which represents the number of independent choices that can be made for that component.5 1.5 = 18. Reliability refers to the conﬁdence the user can have that the regression is valid.52 + 02 + 37. and therefore a trade-off.52 = 337. The error degrees of freedom equal the total less the explained degrees of freedom. one measure refers to the precision of the regression and ﬁve measures refer to the reliability of the regression. and the total degrees of freedom is always equal to the number of data points less 1.0 column (7) shows the error variance. The measures in these last three columns are squared and summed to arrive at the desired values for total variance. Nonindependence of errors (Durbin-Watson statistic) .3 Analysis of Variance Table for Regression Analysis The analysis of variance table separates the total variance of the dependent variable into both error and explained variance components.8B). The analysis of variance table separates the total variance of the dependent variable into both error and explained components. is the ratio of the amount of the variance of a component (in the second column) to the number of degrees of freedom (in the third column).5 7. Reliability of precision (rank-order correlation) 6. Planning and Decision Making © The McGraw−Hill Companies. Of the six principal measures in Exhibit 6B. These terms are illustrated in Exhibit 6B.5 337. The sum of the error and explained variance terms equals total variance. The three variance terms are the basic elements of the statistical analysis of the regression. Source of Variance Variance of Each Component of the Regression (also called sum of squares) Explained (regression) Error Total 37. the number of degrees of freedom for the explained variance component is always equal to the number of independent variables.2 and the values calculated in Exhibit 6B.4.52 = 2.812. Statistical reliability for each independent variable (t-value) 5. Thus. that is. The fourth column.

it is possible to have a relatively high R-squared (if the regression is a good ﬁt to the data points). Most regression software programs show the F-value and the related p-value. A 67 percent conﬁdence interval is determined by taking the regression line and identifying a range that is 1 standard error distance on either side of the regression line. The determination of an acceptable F-value depends on the number of data points. The larger the F.1. The F-statistic can be obtained from the analysis of variance table as follows: Mean square (explained ) Mean square error 2. Goodness of Fit (R-squared) R-squared (also called the coefﬁcient of determination) is a direct measure of the explanatory power of the regression. 812. Statistical Reliability (F-Statistic) The F-statistic is a useful measure of the statistical reliability of the regression. Statistical reliability asks whether the relationship between the variables in the regression actually exists or whether the correlation between the variables is a chance relationship of the data at hand.75/.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. which should be less than approximately 5 percent. The standard error of the coefﬁcient is not the same as the standard error of the estimate. The t-value equals the ratio of the coefﬁcient of the independent variable to the standard error of the coefﬁcient for that independent variable. as illustrated in the preceding formula for SE..5 F= Statistical Reliability for Each of the Independent Variables (t-value) The t-value is a measure of the reliability of each independent variable and as such. The standard error of the estimate can also be used to develop conﬁdence intervals for the accuracy of the prediction. the SE cannot be obtained directly from the analysis of variance table.2598.7 The -value is thus t = . but it is interpreted in the same way. Fourth Edition 192 Part Two II. However.deviationof theinddependent variable) = 18. 150 The explanatory power of the regression improves as the explained sum of squares increases relative to the total sum of squares. The F-statistic is a useful measure of the statistical reliability of the regression.5 = . but the required F-value decreases as the number of data points increase. but this offers relatively little conﬁdence that a statistical relationship exists because of the small number of data points.8868 7 The standard error of the coefﬁcient is calculated as follows: Standarderror = SE/(Std. R-squared is calculated in Exhibit 6B.37 (50− 100)2 + (100− 100)2 + (150 − 100)2 = . 812.2598 .892 3. as illustrated in Exhibits 6. If only a small number of data points are used. it has an interpretation very much like that of the F-statistic.8A and B. 2008 Appendix B: Regression Analysis Planning and Decision Making A conﬁdence interval is a range around the regression line within which the management accountant can be conﬁdent the actual value of the predicted cost will fall.2598 = 2. a 95 percent conﬁdence interval would be determined from 2 standard error distances. For the data in Exhibit 6B. Planning and Decision Making © The McGraw−Hill Companies. the value of the standard error for the coefﬁcient is . The precision and accuracy of the regression improve as the variance for error is reduced and as the number of data points increases because the number of degrees of freedom increases.333 = 337. Conﬁdence intervals are useful and precise tools for management accountants to describe the degree of precision obtained from the regression. A conﬁdence interval is a range around the regression line within which the management accountant can be conﬁdent the actual value of the predicted cost will fall. the lower the risk that the regression is statistically unreliable.5 = 8.3: R2 = = Σ of squares (explained ) Σ of squares (total) 2. A value close to 1 reﬂects a good-ﬁtting regression with strong explanatory power. It measures the percent of variance in the dependent variable that can be explained by the independent variable.

If it does not ﬁx the condition. the SE value provided by the regression is not uniformly accurate over the range of the independent variable. A common method that detects nonlinearity is the Durbin-Watson (DW) statistic. The calculation of rank-order correlation is beyond the scope of this introductory chapter but can be found in many statistics texts.0 indicates that the independent variable is reliable at a risk level less than approximately 5 percent and is therefore a reliable independent variable to include in the regression.5 Dependent Variable Nonconstant Variance Independent Variable A t-value larger than 2. The rank-order correlation is a statistic that measures the degree to which two sets of numbers tend to have the same order or rank. a relatively small correlation that indicates little evidence of nonconstant variance. management accountants should be very cautious in interpreting the SE value. the errors are systematically related to each other. Nonindependent Errors (Durbin-Watson Statistic) The Durbin-Watson statistic is a measure of the extent of nonlinearity in the regression. The relationship between the variables might also be inherently nonlinear. Then the regression is unreliable and subject to greater than expected estimation errors. we calculate the rank-order correlation between the position of the data and the size of the error. or rank. and cyclical inﬂuences. The range of the standard error of the estimate should be relatively small. One type of nonlinearity (nonindependence of errors) is illustrated in Exhibit 6B. A small range provides conﬁdence in the accuracy of the coefﬁcient’s value. when the relationship between the independent and dependent variables becomes less stable over time. that is. as when learning occurs or a multiplicative rather than an additive relationship exists (such as predicting payroll costs from hours worked and wage rates). Regression software such as Excel shows the 95 percent conﬁdence range for the coefﬁcient of each of the independent variables.1. Nonconstant variance is the condition when the variance of the errors is not constant over the range of the independent variable.125. This is the case. If the data are nonlinear because of seasonality or a cyclical pattern. To detect nonconstant variance. A key assumption of regression is that the relationship between the independent and dependent variables is linear. Planning and Decision Making Appendix B: Regression Analysis © The McGraw−Hill Companies. 2008 Chapter 6 Cost Estimation 193 EXHIBIT 6B. for example. The rank-order correlation is a statistic that measures the degree to which two sets of numbers tend to have the same order. If there is nonconstant variance. It is calculated from the amount and change of the errors over the range of the independent variable. The variance of the errors is not constant over the range of the independent variable. the standard error of the estimate varies over the range of the independent variable. for example. . are not independent. the Spearman rank-correlation coefﬁcient is . management accountants should transform the dependent variable with the log or square root to see whether this improves the behavior of the errors. seasonality. To ﬁx the problem of nonconstant variance. This type of behavior is illustrated in Exhibit 6B.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.5.6. This assumption is violated frequently because ﬁnancial data are often affected by trend. Fourth Edition II. For the data in Exhibit 6B. A relatively high rank-order correlation is evidence of nonconstant variance. Reliability of Precision (Nonconstant Variance) For certain sets of data.

Modianos. for example. “Multiple-Cost Flexible Budgets and PC-Based Regression Analysis. ﬁrst differences.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. and Doan T. A recent study of regression analysis applied to 20 different overhead cost accounts showed that most of the R-squared values fall between . The values for the standard error of the estimates averaged 12 percent of the mean of the dependent variable. with most falling between 5 percent and 20 percent. . The exact values for a speciﬁc regression depend on a number of factors including the sample size and the number of independent variables.83 and . Cluskey Jr. See G. log. cutoff errors. . pp. could need transforms (lag.0 • Delete or transform the independent variable • Inaccurate estimates Precision of the regression Standard error of the estimates (SE) Should be small relative to the dependent variable • Same considerations as for reliability— goodness of ﬁt above (see above) • Inaccurate estimates Reliability of precision (nonconstant variance) Rank-order correlation Should be small • Square root or log transform the dependent variable • Add a dummy variable • SE is unreliable Reliability— Potential nonlinearity (nonindependence of errors) Durbin-Watson statistic (DW) Between 2. July–August 2000. Mitchell H.0 and 3. . R.75 or better • Add or delete independent variables • If DW is poor.93.. 35–47.0 For certain series: • Deseasonalize • Detrend • Use dummy variable for shift For nonlinear relationship • Log transform • Some other nonlinear transform • • Inaccurate estimates SE is unreliable * The values shown here are useful for a wide range of regressions. Fourth Edition 194 Part Two II.7 Summary of Statistical Measures Measure Concerns Statistical Measure What Is an OK Value?* What Is the Right Fix If Not OK? Consequence If Not Fixed Reliability— Goodness of ﬁt R-squared Should be approximately . Planning and Decision Making © The McGraw−Hill Companies. 2008 Appendix B: Regression Analysis Planning and Decision Making EXHIBIT 6B. Raiborn. .6 Dependent Variable Nonindependence of Errors Independent Variable EXHIBIT 6B.) • Correct measurement errors in the data.” Journal of Cost Management. or reporting lags • Inaccurate estimates Statistical reliability for the regression F-statistic Depends on sample size • Increase sample size • Other changes as suggested for reliability—goodness of ﬁt (see above) • Inaccurate estimates Statistical reliability for the independent variables t-value Should be greater than 2.

number of two-bedroom and one-bedroom apartments. etc. That is. 195 degrees of freedom. 2008 Chapter 6 Cost Estimation 195 The DW value falls between zero and 4. are both examples of what is called time-series regression.0 indicates little chance of a nonlinearity as described earlier. an analysis of sales value (per square foot) of industrial properties in the Los Angeles area in the early 1990s showed a signiﬁcant trend variable (−\$2. 193 ﬁrst difference. except for the issue. Alternatively. The regression equation that the builder develops is then used to predict the cost of homes to be built. using prior period’s data.). and location. Cross-sectional regression estimates costs for a particular cost object based on information on other cost objects and other variables. a value of DW between approximately 1. 186 learning rate. 180 multiple linear regression. number of square feet. 179 Durbin-Watson statistic. its age. 193 outlier. based on the expected size of the new home in square feet. Estimating Real Estate Values for Warehouses and Manufacturing Plants Similarly. values less than 1.83 per square foot per year). what may be required is to convert a multiplicative relationship to an equivalent additive (that is. the chief determinant of the value of the property is its ability to produce cash ﬂows and proﬁts. Time-series regression is the application of regression analysis to predict future amounts. age. their indicators. 191 multicollinearity. 180 time-series regression.0 and 3. 180 rank-order correlation. the Ben Garcia case and the WinDoor Inc. suppose a residential home builder uses regression to estimate the cost of constructing a new home. 191 conﬁdence interval. in square feet of ﬂoor space. the . using a dummy variable for seasonality. cross-sectional regression estimates costs for a particular cost object based on information on other cost objects and variables. 173 independent variable. these appraisers also use a trend variable to tie the sales price of the property to the year it was sold. The problem of nonindependent errors usually can be ﬁxed by deseasonalizing the data. 179 work measurement. In contrast. 176 R-squared.7. and expected future net operating income (NOI). TIME SERIES AND CROSS-SECTIONAL REGRESSION Time-series regression is the application of regression analysis to predict future amounts. and ways to ﬁx the underlying conditions are summarized in Exhibit 6B. 191 dependent variable. nonindependence of errors. where the information for all variables is taken from the same period of time. 185 t-value.0 or greater than 3.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.0. 176 Using Regression to Estimate the Value of Commercial Real Estate Estimating Real Estate Values for Apartment Buildings and Ofﬁce Buildings As expected. 192 correlation. 177 p-Value. 185 F-statistic. with 20 or more data points. real estate appraisers have developed regression analyses for warehouses and manufacturing plants using their size. Planning and Decision Making Appendix B: Regression Analysis © The McGraw−Hill Companies. 183 nonconstant variance. 176 dummy variable. 176 work sampling. 180 cost estimation. 179 simple linear regression. 183 standard error of the estimate. The NOI variable is usually not relevant. real estate appraisers performing regression analysis to appraise the value of an apartment building or an ofﬁce building use as the dominant independent variable the property’s past. current. using prior period’s data. case. or using an index to remove the trend. For example. They also use a trend variable to distinguish sales of properties in different years. For example. linear) relationship by taking the logarithm of the independent and dependent variables. 192 high-low method. 193 regression analysis. 171 cross-sectional regression. and the relevant vacancy rate in the property and in the submarket area where it is located. 176 mean absolute percentage error (MAPE). The builder develops a regression model using the cost of homes built previously that year as the dependent variable and the size in square feet of these homes as the independent variable. All of the statistical measures of reliability and precision explained above apply equally to both types of regression. 195 trend variable. 173 mean squared variance. Key Terms Comments on Cost Management in Action The two examples used in this chapter.0 should indicate the need to study the data and to choose appropriate ﬁxes if necessary. where the information for all variables is taken from the same period of time. Other variables regarding the property include its size (as measured by the number of units. and the builder knows that the main cost driver for building cost is the size of the home. 176 learning curve analysis. 186 least squares regression. The statistical measures. Since the regression analysis is usually built from actual sales numbers over a period of time. which applies only in time-series regressions. Fourth Edition II. analysis of variance table.

182 6. please turn to the end of the chapter. he gathers data on his expenses and number of deliveries for each month of the current year. Using Regression Analysis George Harder is the Imperial Foods Company’s plant manager of one of the processing plants. “Market-Supported Adjustments Using Multiple Regression Analysis. “Regression Analysis: A Cost-Effective Approach for the Valuation of Commercial Property.832 152.4 .5 Required Which of the two regressions is better and why? Regression 2 (labor-hours and machine-hours) .41 per square foot per year of age. Charles G.” Real Estate Finance. Dannis.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.245 123.6 3. A signiﬁcant size variable (−\$2. San Bernadino. George runs two regression analyses on these data. and machine-hours as potential independent variables to use in predicting overhead.621 6.882 5. Planning and Decision Making © The McGraw−Hill Companies. with the following results: Regression 1 (labor-hours only) R-squared Standard error Standard error as a percent of the dependent variable t-values Materials cost Labor-hours Machine-hours . per 100.329 133. oil. October 1993. Crosson. pp.567 5. To budget his vehicle expenses for the coming year. Markham.993 201. direct labor-hours. and Thomas G.554 12% 2. which is done at a variety of service stations and repair shops.43 per square foot.622 5. Thibodeau.433 Required Use the high-low estimation method to determine the relationship between the number of deliveries and the cost of maintaining the vehicles. and regular maintenance of his vehicles. and Stephen C.166 6. Winter 1996. 2.” The Appraisal Journal.295 163.553 177. The location variable was also signiﬁcant. Self-Study Problems (For solutions. Maxwell O. North Carolina. 2008 Appendix B: Regression Analysis Planning and Decision Making coefﬁcient on the trend variable was negative because prices were falling during that period. He has collected data on overhead costs for the past 24 months and has decided to use regression to study the factors inﬂuencing these costs. Fourth Edition 196 Part Two II. Kincheloe.000 square feet of space) indicated that larger buildings had on average lower sales prices per square foot.8 1.793 14% −1. and Daniel E.) 1. etc.293 155. metropolitan area. “Linear Regression Analysis of Economic Variables in the Sales Comparison and Income Approaches.58 \$13.577 5.433 6.937 176. He has also collected data on materials cost. Age was also a factor. George is concerned about the increase in plant overhead costs in recent months.681 7.389 150.229 180. showing that properties in certain counties in the Los Angeles area (Orange County.) were predicted to have as much as a \$2.0 4. the coefﬁcient being −\$0. Hector spends a considerable amount of money on the gas. 181–91.” The Appraisal Journal.32 difference in value per square foot. April 1998.65 \$12. January February March April May June July August September October November December Total Vehicle Expenses Total Deliveries \$145.599 7. Sources: Stephen T.245 164.783 5. Using the High-Low Method Hector’s Delivery Service uses four small vans and six pickup trucks to deliver small packages in the Charlotte. Ramsland Jr.942 5.

052 1.610 1. 2008 Appendix B: Regression Analysis Chapter 6 Cost Estimation 197 3. Explain the advantages and disadvantages of each.405 1. Explain the implementation problems in cost estimation. primarily door panels.387 2.310 2.012 1. management is reviewing the incidence of scrap and waste in the manufacturing process at one of its plants. management estimates the cost of this waste in labor and materials is approximately \$10 per unit.270 87 95 101 111 115 92 2.300 1. and related items.380 1.310 1.600 Jul Aug Sep Oct Nov Dec 75 81 70 79 82 70 962 1.335 1.570 1. you have obtained these recent data on the units produced.580 1.889 1.514 2. Meeks deﬁnes scrap and waste as any defective unit that is rejected for lack of functionality or another aspect of quality. Planning and Decision Making © The McGraw−Hill Companies.104 1. List four advantages of regression analysis.311 Jul Aug Sep Oct Nov Dec Required Use the high-low method and regression analysis to estimate the defective units in the coming months and to determine which method provides the best ﬁt for this purpose. Explain the assumptions used in cost estimation. The number of defective units is listed in the following table. and the cost of sales since these numbers are easily available and relatively reliable on a monthly basis: Units Produced (000s) Cost of Sales (000s) Units Shipped (000s) Defective Units Jan 2007 Feb Mar Apr May Jun 55 58 69 61 65 69 \$ 689 737 886 768 828 878 50 53 64 56 60 64 856 1.011 2. Currently. What are the six steps in cost estimation? Which one is the most important? Why? Contrast the use of regression analysis and the high-low method to estimate costs. and to provide a basis on which to estimate future defects.787 2. The plants have a number of different inspection points.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.650 92 100 91 101 105 88 1. The company makes plastic parts for automobiles and trucks. and management has asked you to investigate and estimate the defective units in the coming months. The company has a steady demand for its products from both domestic and foreign automakers and has experienced growth in sales averaging between 10 and 20 percent over the last 8 to 10 years.187 2. Fourth Edition II. Using Both High-Low and Regression John Meeks Company is a medium-size manufacturing company with plants in three small mid-Atlantic towns. What are nonlinear cost relationships? Give two examples. the units shipped. A ﬁrst step in your investigation is to identify the cost drivers of defective parts. exterior trim.208 1.511 1. The parts have an average cost of \$5 to \$20.037 1. List the three methods of cost estimation.549 1. and failure or rejection can occur at any inspection point.020 70 76 80 89 92 74 1.756 1. An unfavorable trend appears to exist with regard to defects.261 1.536 1. Questions 6-1 6-2 6-3 6-4 6-5 6-6 6-7 6-8 6-9 6-10 Deﬁne cost estimation and explain its purpose.910 2. to understand what causes them.669 2. For this purpose.230 2. How is cost estimation used in activity-based costing? Explain how to choose the dependent and independent variables in regression analysis. .224 1.107 968 1.849 Jan 2008 Feb Mar Apr May Jun 67 72 85 75 81 85 850 916 1.103 62 67 80 70 76 80 1.

The accounting manager has created a cost projection formula through a regression analysis of past data.00 per pack.227 5.000 456.000 hours in 2008. As they were planning for next year’s production.000 keys at a total cost of \$20.649 Which two years should Smith select for the high-low method analysis and why? Johnson Plastics Inc.000. a t-value of 2. She has provided you with the following table of costs for the last ﬁve years. produces ﬂuorescent lightbulbs for commercial use.000 keys at a total cost of \$10.000.105 53. The regression estimate for 2008 is \$5.958 hours. Smith would like to forecast data using the high-low method and has compiled the following data from prior results: 6-15 6-16 6-17 6-18 Total operating hours Inspection cost (\$) 6-19 6-20 6-21 6-23 2004 2005 2006 2007 3.056 49.584.678 188.000.679 498. and the values for the cost driver are 3. If Jamison projects it will be working 200. they decided to implement a high-low system to forecast future costs. The accounting manager determined that the years to be used are 2003. The cost driver is number of labor-hours.000 and 3.457 2003 2004 2005 2006 2007 100. Planning and Decision Making © The McGraw−Hill Companies. produces jewel cases for CDs. respectively.500. with total production of 2. 2008 Appendix B: Regression Analysis Planning and Decision Making Brief Exercises 6-11 6-12 6-13 Explain what dummy variables are and how they are used in regression analysis.835 4.672 98. Regression A has an R-squared value of . The accounting manager has decided to implement a highlow costing system to predict future materials handling costs.9.89 and a t-value of 2. and 2006. What is the value for a (the ﬁxed quantity)? Grant Healthcare provides plastic gloves for hospitals.08 and Regression B has an R-squared of . Which two years should be used for this method? Production hours Handling cost (\$) 6-22 2003 5.000 and the associated cost drivers are 7. What is the variable cost per key? Power Drink Inc. You notice that the formula has an R-squared of . Their dependent variable is labor expense.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. with total production of 3.414. List some possible independent variables for a regression analysis of ﬁnancial data.843 507.17. They are attempting to forecast costs for future production.495 and 1. What do these statistical measures indicate about the regressions? Which regression should Carter Dry Cleaning use for cost estimation? Williams Inc. How do we know when high correlation exists? Is high correlation the same as cause and effect? What does the coefﬁcient of determination (R-squared) measure? 6-14 Wallace Heating is attempting to estimate the production cost for heating ducts for the coming year using the high-low method. The accounting manager has presented statistical measures for both of these regressions. Wallace determines that the high and low costs are \$25. He has also determined the variable cost to be \$1.000 packs. The accounting manager is attempting to estimate the total cost for the next quarter using the high-low method. Fourth Edition 198 Part Two II. and a standard error of \$400. The accounting manager has calculated a regression to determine future production costs. He presents this to you to evaluate the reliability of the regression.197 46. He has compiled data and found the high and low costs are \$10.586 43.314 Jamison Construction has implemented a costing system by use of a regression analysis of past costs. respectively.233 138.830 and \$18.000 and \$6. Smith Glass Co. What do these statistics tell you about the reliability of his regression analysis? .000.072. what is the projected total cost? Curry Rubber manufactures rubber bands for commercial retail companies. The variable cost per hour of labor is \$35 and the ﬁxed cost was determined to be \$125. Within what interval would she be reasonably (67 percent) conﬁdent that the actual values will fall? Peppers Lockdown produces keys for houses and cars.734 3.5. and a standard error of \$200.683 50.352 544.000.6. produces sports drinks.000 with an R-squared of .517 601. produces industrial glass for factories. The estimate for next quarter costs is \$2.3.284 203. respectively. a t-value of 2. What is the variable cost per hour? Carter Dry Cleaning has developed two regression analyses for cost estimation.53 and a t-value of 1.

Their low value is \$250. 8. The cost of repair for a machine used in manufacturing. and when demand falls off.000 for 8. Any graph can ﬁt two or more patterns. 2. 1. . 6.000 per month as a rental charge.000 per month plus 2 percent of gross sales receipts. the local electric utility increases the per-kilowatt-hour charge for each additional 5. The salary of the plant’s quality control inspector. The total cost of manufacturing a new camera over its entire life cycle. the number is increased. 9. What is the value of the ﬁxed cost for their formula? Cost Classiﬁcation: Match each cost to the appropriate cost behavior pattern shown in the graphs (a) through (l). The cost to make copies of a given document at a printing shop.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. 11. Planning and Decision Making © The McGraw−Hill Companies. 5.000 hours and their high value is \$400. who inspects each batch of products. To discourage excess usage and to level the demand. up to a maximum of \$3. The cost of an Internet connection of \$10 per month plus \$2 per hour of usage above 10 hours.000 for 100. The local municipality charges a ﬁxed rate per gallon for usage up to 10. They have found the variable cost per animal to be \$2 and their high and low costs used were \$80. They are in the process of implementing a cost forecasting system using the high-low method. 3.000 for 5. Fourth Edition II.000 gallons. 7. 2008 Appendix B: Regression Analysis Chapter 6 6-24 6-25 Exercises 6-26 Cost Estimation 199 Miller Landscaping is attempting to project costs for future quarters. The cost of water and sewer service to the manufacturing plant. the number is decreased. A clothing store in the SunnyVale Mall pays a ﬁxed rental charge of \$1. Miller has compiled data and decided to use the high-low costing method. A shoe store in the SunnyVale Mall pays 6 percent of gross sales receipts. 12.000 hours.000 kilowatt-hours’ usage. 10. especially in peak load times. 4.000 animals and \$40. The cost of order ﬁllers in a warehouse. and a higher charge per gallon for usage above that point. which reduces the per-copy charge for customers who make more than 100 copies of the document. When demand increases. (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) The cost of lumber used to manufacture wooden kitchen tables. What is their variable cost per hour? Sanders Bears produces stuffed animals.000 animals.000 for 120. The cost of an Internet connection of \$23 per month.

The agency’s argument is that a positive linear relationship exists between advertising and sales in the sailboard industry.750 Required 1.000 147. Graph total cost.55 for 500 and \$0. Graph the per-unit total cost.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.500 12. total variable costs.1 1.500 \$ 95.8 2.000 3. area.383 0. Sue Lawson presents these data taken from industry data for stores similar in size and market share to Kansas City Sailboards: Advertising Expense Annual Sales \$2. Fourth Edition 200 Part Two II.55 45. 3. 2. Required If the total cost function for croissants is linear.000 5.000 110.750 4. Planning and Decision Making © The McGraw−Hill Companies.750 Variable Costs Total Costs \$ 7. and total cost.000 Required 1.000 4.50 for 600.500 5. 6-29 Cost Estimation. Discuss the behavior of the ﬁxed.500 4.3 Which regression would you choose and why? .500 9. a contact lens manufacturer: Output in Units Fixed Costs 250 300 350 400 \$4.000 124. and total ﬁxed costs. what will be the average cost to bake 560? 6-30 Cost Estimation Using Graphs Lawson Advertising Agency is trying to persuade Kansas City Sailboards Company to spend more money on advertising.750 15. Required What is the average cost per unit if the facility normally expects to operate at 80 percent of capacity? 6-28 Cost Relationships The following costs are for Optical View Inc.000 150.500 3. The average costs to bake the croissants are \$0. and per-unit ﬁxed cost.044 0. Graph annual sales and advertising expense.250 13.000 10. The variable costs for each computer are \$150 and the ﬁxed costs total \$62. Do the data prove Sue’s point? 6-31 Analysis of Regression Results Wang Manufacturing uses regression analysis to predict manufacturing overhead costs based on labor-hours and/or machine-hours and has developed the three following regression equations.250 16. 2.9 0.750 4..250 per month. per-unit variable cost.000 138.750 4.58 1. North Carolina.000 \$12. SE R-squared t-values: Labor-hours Machine-hours Required Regression 1 Regression 2 Regression 3 33. variable. Average Cost Maribeth’s Cafe bakes croissants that it sells to local restaurants and grocery stores in the Raleigh.844 0.35 31. 2008 Appendix B: Regression Analysis Planning and Decision Making 6-27 Cost Relationships Comptech hired Erwin & Associates to design a new computer-aided manufacturing facility that has the capacity to produce 250 computers per month.000 143.

100 1. Joseph Ethan.870 1. The researchers used regression analysis and found the following results: Total Cost for Patient = Constant.700 1.655 1.900 1. b.010 2.550 1. Fourth Edition II.780 2.590 What is the cost equation for maintenance cost using the high-low method? Interpreting Regression Results Recent research into the cost of various medical procedures has shown the impact of certain complications encountered in surgery on the total cost of patient’s stay in the hospital.890 1. .590 1. has asked you to assist him in estimating his maintenance costs.070 3. you have determined that the best cost driver for maintenance costs is machine-hours. It is about as big around as a fountain pen and twice as long. High-Low Method Ethan Manufacturing Inc. These data are from the previous ﬁscal year for maintenance expense and machine-hours: Month 1 2 3 4 5 6 Expense Hours Month Expense Hours \$2. produces blinds and other window treatments for residential homes and ofﬁces. Together. These data are from the previous ﬁscal year for maintenance expense and machine-hours: Required 6-34 Month Maintenance Expense Machine-Hours 1 2 3 4 5 6 7 8 9 10 11 12 \$2.880 7 8 9 10 11 12 \$2. as maintenance costs for the previous ﬁscal year were higher than he expected.499 1. Together.805 1. The owner is concerned about the maintenance costs for the production machinery.860 1.410 1.880 1.020 3.690 1.230 2.450 1. 0 if false) Where: • a.670 2.785 1.605 1. plus b ⫻ presence of one or more complications (= 1 if true.905 2.850 1.620 2. He has asked you to assist him in estimating his future maintenance costs so that he can better predict his ﬁrm’s proﬁtability.695 1. c are coefﬁcients of the regression model.850 2.220 2.720 2.775 1.005 1.570 2. Planning and Decision Making © The McGraw−Hill Companies. 6-33 Cost Estimation.910 3.855 3.300 1. The owner.865 2. produces ﬂoor mats for automobiles.405 Required What is the cost equation for maintenance cost using the high-low method? Graph the data points to check for outliers.100 3. 0 if false).590 2.840 1.760 2.600 2. plus c ⫻ use of a laparoscope (= 1 if true.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. and • The laparoscope is an instrument somewhat like a miniature telescope with a ﬁber optic system which brings light into the abdomen.770 1.625 2. you and Joseph determine that the single best cost driver for maintenance costs is machine-hours.822 2. plus a ⫻ length of stay (measured in days). 2008 Appendix B: Regression Analysis Chapter 6 6-32 Cost Estimation 201 Cost Estimation: High-Low Method Horton Manufacturing Inc.

76 Complication Laparoscope \$1. If estimated renovation costs plus the purchase price of a house are higher than its estimated resale value. Recently he decided to include the number of openings—the total number of doors and windows in a house—as a cost driver.11 908 358 0.986 406 Not applicable Not applicable 0. Jay has kept careful records of these expenses on his last 12 jobs: House Square Feet Openings Cost 1 2 3 2. showed the following regression results: R-squared: 53% constant term: \$3. and resale of older homes.05 (and t-value >2) except for the length of stay variable in the nonlaparoscopic condition. Interpret the values of each coefﬁcient and the standard error for each coefﬁcient.750 3. Required 1. Their cost is signiﬁcant because they require time-consuming preparatory work and careful brushwork. It is essential for him to have accurate cost estimates so he can determine total renovation costs before he purchases a piece of property. Comment on the statistical measures for the model.800 13 15 12 \$3. What are the t-values for each of the independent variables for each treatment condition? Problems 6-36 Cost Estimation.53 * Note: All independent variables are signiﬁcant at the level of p = .010 2. the house is not a worthwhile investment. and the related costs. if any. Coefﬁcients for independent variables Regression intercept Length of stay Coefﬁcient* Standard error for the coefﬁcient Number of complications Coefﬁcient Standard error for the coefﬁcient Laparascopic Coefﬁcient Standard error for the coefﬁcient R-squared Not Laparoscopic Laparoscopic \$ 8. High-Low Method Jay Bauer Company specializes in the purchase. The rest of the house usually is painted either by rollers or spray guns. Jay has been using the home’s interior square feet for his exterior paint cost estimations.239 1.986 4. Jay Bauer employs several carpenters and painters to do the work for him. Appendix (Continuation of 6-34) The following table shows the regression results presented by the researchers in the study described in Exercise 6-34. renovation.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. Which.89 \$ 908 2. What is the estimated cost for a patient who has complications and stays in the hospital two days.600 3. which are relatively efﬁcient ways to apply paint to a large area. The left-hand column shows the results for the sample of patients who were treated without the laparoscopic surgery. and whose surgery requires a laparoscope? 2. Planning and Decision Making © The McGraw−Hill Companies. Fourth Edition 202 Part Two II. based on 57 patients. 3.719 Not signiﬁcant Not applicable 861 80 3. 2008 Appendix B: Regression Analysis Planning and Decision Making The research. 6-35 Analysis of Regression Results.719 Coefﬁcients and t-values for independent variables: Length of Stay coefﬁcient t-value \$ 861 10.100 (Continued) . dummy variables are used in this regression? 3.300 3. Which of the two regressions has the better reliability and precision in estimating cost? Why? 2. The righthand column shows the results for the laparoscopic surgery.393 1.54 Required 1.043 \$ 3.

what considerations are important to SpectroGlass.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.700 2. Planning and Decision Making © The McGraw−Hill Companies.150 4.875 4. The only thing in the shop that uses natural gas is the furnace. so he questions replacing it.950 Required 1.600 3.700 3. determine the cost of painting a 3.000 square yards? 2. For a year.400 210.– based ﬁrm. Fourth Edition II. APA has been recording the average daily temperature and the cost of its monthly utility bills for natural gas and electricity. Assume that brand A is manufactured in Germany and brand B is manufactured in Canada. Machine Replacement. 2008 Appendix B: Regression Analysis Chapter 6 4 5 6 7 8 9 10 11 12 2.375 2.000 146. Average Temperature January February March April May June ° 31 F 41 43 44 46 50 Utility Cost \$760 629 543 410 275 233 (Continued) .550 12 19 13 11 11 10 16 13 16 Cost Estimation 203 3. Plot the cost data against square feet and against openings.S. Using the high-low cost estimation technique.000 16.300 114.850 4. has noticed that utility bills are substantially higher the colder the average monthly temperature is. 2.200 3.000 Required 1. The manufacturer of each machine has provided SpectroGlass these data on the cost of operation of its machine at various levels of output: Output (square yards) 4. the furnace is used every month of the year (though less in the summer months and very little in August).800 2. SpectroGlass is considering replacing the old machine with a new machine.100 3.000 115.000 Machine B Estimated Total Costs \$ 70.800 2.400-square-foot house with 8 openings.650 3. Alaska. A production supervisor comes to you to say that the nature of the defect is really very difﬁcult to detect and that most customers will not notice it. which machine should it purchase? At 15. High-Low Method Antelope Park Amoco (APA) in Antelope Park.050 2. As a cost analyst at SpectroGlass. in addition to those already mentioned in your answer to requirement 1? 6-38 Cost Estimation.000 14. He suggests that you modify your calculations to justify keeping the present machine to keep things the way they are and save the company some money. Which variable is a better cost driver? Why? 6-37 Cost Estimation. What do you say? 3.000 24.700 2. Ethics SpectroGlass Company manufactures glass for ofﬁce buildings in Arizona and Southern California. As a U.450 2. a critical machine in the production process has begun to produce quality defects. and electricity use is fairly constant throughout the year. Because of prevailing low temperatures.300-square-foot house with 14 openings. If SpectroGlass’s output is expected to be 22.000 192.000 100. As a result of age and wear.000 7.800 90.250 2.900 132. you have been assigned to complete requirement 1. Everything else in the shop runs on electricity.000 Machine A Estimated Total Costs \$ 54.600 78.000 square yards. Also determine the cost for a 2.000 137. either brand A or brand B.000 9.

temperatures in February are expected to average 40°F. 6-39 to 6-43 Regression Analysis Problems 6-39 through 6-43 are based on Armer Company.260 4. y = 570 + 7.2884 34.469 .300 3. The forecast for January is a near record average temperature of 10°F. which is accumulating data to use in preparing its annual proﬁt plan for the coming year.65 7. Planning and Decision Making © The McGraw−Hill Companies. The accounting staff has suggested the use of linear regression to derive an equation for maintenance hours and costs.200 3.0x b.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. None of the above (CMA Adapted) Required (6-40) Based on the data derived from the regression analysis.350 2. Data regarding the maintenance hours and costs for the last year and the results of the regression analysis follow: Hours of Activity January February March April May June July August September October November December Sum Average 480 320 400 300 500 310 320 520 490 470 350 340 4. y = 3. \$3.600 684.461 .00 a (intercept) b coefﬁcient Standard error of the estimate R-squared t-value for b Maintenance Costs \$ 4.050 3.600 + 400x d.160 \$43.780 b.030 4.200/4.5x c. 2008 Appendix B: Regression Analysis Planning and Decision Making Average Temperature Utility Cost 53 60 50 40 30 20 220 210 305 530 750 870 July August September October November December Required Use the high-low method to estimate utility cost for the upcoming months of January and February.820 4. y = 570 + 9.960 3.200 3.105 Required (6-39) If Armer Company uses the high-low method of analysis. y = 400 + 9. The cost behavior pattern of the maintenance costs must be determined. the equation for the relationship between hours of activity and maintenance cost follows: a.000 3.0x e.800 400 Average cost per hour (\$43.99724 60. 420 maintenance hours in a month mean that maintenance costs should be budgeted at a.800) = \$9. \$3.600 2. Fourth Edition 204 Part Two II.470 4.

10.041 0.5 are statistically signiﬁcant at p < .002 0.1151. Fourth Edition 208 Part Two II.038 0.80882 1 0.33919)2=. dealerships must provide monthly ﬁnancial statements following the USMI accounting . Required 1.041 0.047 Correlation Results for PolyChem’s Plant Data Number Order size Complexity Setup time Run time Number Order Size Complexity Setup Time 1 −0.06534 Run Time 1 Note: Correlations with absolute value > .035 0.040 0.041 0.043 0.20952 −0. order size and complexity affect setup time and run time. Table 1 Table 2 Plant Data for PolyChem Machine Number Order Size Order Complexity Setup time Per Unit Run time 2 2 2 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 480 489 480 180 2160 1377 120 540 360 1080 300 2400 81 360 120 120 60 240 60 1 1 2 1 1 1 2 1 2 2 1 2 2 1 1 2 2 1 2 0.001 0. To illustrate.035 0.07095 −0. note that the correlation between machine number and order quantity = −.004 0.140537 1 −0.043 0.004 0. 2008 Appendix B: Regression Analysis Planning and Decision Making of the complexity of the order based on a subjective rating where 1 = less complex and 2 = more complex (complexity relates to the number and type of images and colors printed on the packaging material).Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.071001 −0.007 0.4521388 −0. Cheryl understands that each of the correlation numbers in Table 2 is equivalent to the R-squared for a simple linear regression between the variable.03805 0.005 0. The R-squared for the regression between these two variables (with either as the dependent variable) is (−.002 0. Cheryl wants to run some regression analyses to better understand this data and.05. Cheryl also recalls that a negative correlation means that the two variables are inversely related—when one increases.043 0. correlations with absolute value > .005 0.002 0. as follows: (correlation between two variables)2 = the R-squared for simple regression analysis between these two variables.008 0. as a ﬁrst step.33919.004 0. if at all. the other decreases.4 are statistically signiﬁcant at p < .043 0.040 0. The results are shown in Table 2. How can your analysis in 1 above help PolyChem become more competitive? (CMA Adapted) 6-48 Regression Analysis United States Motors Inc.008 0.014 0.346651 1 −0. Analyze the ﬁndings in Table 2 and assess how. Planning and Decision Making © The McGraw−Hill Companies.046 0.042 0.011 0.002 0. (USMI) manufactures automobiles and light trucks and distributes them for sale to consumers through franchised retail outlets. obtains a correlation analysis which shows the simple correlation between each of the variables in Table 1.042 0.042 0.33919 0. As part of the franchise agreement.043 0.003 0.042 0.040 0.002 0.005 0. What other ﬁndings in Table 2 are of particular interest? 2.

22.000.800 18.500.300.559.000.000 24. Deﬁne the standard error of the estimate.750.200 The result of a simple regression analysis using all 12 data points yielded an intercept of \$16.94.000 Gross proﬁt Operating costs Variable expenses \$ 5. The vice president of marketing has asked Jack Snyder.500 new vehicles annually. As a starting point for further analysis. AVERAGE DEALERSHIP FINANCIAL PROFILE Composite Income Statement Sales Cost of goods sold \$30.854. Bridget Forrester.000. Planning and Decision Making © The McGraw−Hill Companies.000 862. 2. Management recently became aware of rising costs resulting from returns of malfunctioning products. Recommend which method Bridget should use and explain why. . corporate controller. Assume that regression analyses were performed on the separate components of the mixed expenses and that a coefﬁcient of determination value of .60 to USMI’s analysis. Regression Analysis DVD Express is a large manufacturer of affordable DVD players.09 and a coefﬁcient for the independent variable of \$183.500 Mixed expenses Fixed expenses 2. Describe the limitations that may exist in applying the composite-based relationships to speciﬁc new dealerships that have been proposed.27. b. t = 1.500 18.000 Operating income \$ 233. The regression equation that Jack Snyder developed to project annual sales of a dealership has an Rsquared of 60 percent and a standard error of the estimate of \$4. determine the approximate 95 percent conﬁdence range for Jack’s prediction of sales. wants to test different forecasting methods and then use the best one to forecast quarterly expenses for 2007.000 1. (R-squared = . Explain the signiﬁcance of an R-squared value of . What is the strategic role of regression analysis for USMI? (CMA Adapted) 6-49 Cost Estimation.500. Required 1. c.500 USMI is considering a major expansion of its dealership network. 3.600 1 2 3 4 \$16.100 19. Deﬁne the term relevant range.000 17. Jack estimates that 90 percent of the mixed expenses shown are variable for purposes of this analysis.250.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. the controller.000 units are sold.200 17.000 19. SE = 1128). 2008 Appendix B: Regression Analysis Chapter 6 Cost Estimation 209 procedures manual. Fourth Edition II. 4. to develop some measure of the risk associated with the addition of these franchises.700 1 2 3 4 \$16. e. Calculate the quarterly forecast for 2007 using the high-low method and regression analyses.60 was determined as applicable to aggregate mixed expenses over the relevant range.500 18. Calculate the composite dealership proﬁt if 2. He also suggested performing regression analyses on the various components of the mixed expenses to more deﬁnitively determine their variability. Required 1. High-Low Method.600 18. The relevant data for the previous three years follows: 2004 Quarter Return Expenses 2005 Quarter Return Expenses 2006 Quarter Return Expenses 1 2 3 4 \$15. USMI has developed the following ﬁnancial proﬁle of an average dealership that sells 1. a. If the projected annual sales for a dealership total \$28.800 17.

Output Total Hours 1 8 32 250 1. How does your analysis in requirement 1 change if Clothes for U is involved in global sourcing of products for its stores? 6-51 Learning Curves The Air Force Museum Foundation has commissioned the purchase of 16 Four F Sixes. The relevant data for the previous three years follows: Quarter Warehouse and Transportation Expense (\$000) 1/2004 2 3 4 1/2005 2 3 4 1/2006 2 3 4 \$12. the aircraft will be made using the technology and manufacturing processes available when the originals were built. respectively. Planning and Decision Making © The McGraw−Hill Companies. 2. the controller. .19. eighth. how many hours will it take to build the 16 aircraft for the Air Force Museum Foundation? 2. How does your analysis in requirement 1 change if DVD Express manufactures its products in multiple global production facilities to serve the global market? 6-50 Cost Estimation.724 Required 1.500 11. Aviation enthusiasts can also visit the production facility to see exactly how such aircraft were built in 1938.300 12. Their principal job is to lay railroad ties to line the sidewalks around apartment complexes and to install ﬂower boxes. t = 1. The available information is for the total accumulated time as the ﬁrst. Oklahoma.700 12. They will be built completely from scratch to the exact speciﬁcations used for the originals. Management recently became concerned about the amount of inventory carrying costs and transportation costs between warehouses and retail outlets. What is the role of learning curves in Soren Industries’ business for contracts such as this? 6–52 Learning Curves Ben Matthews and David Everhart work for a landscaping company in Twin Cities.600 The results of a simple regression analysis using all 12 data points yielded an intercept of \$11. 2008 Appendix B: Regression Analysis Planning and Decision Making 2.5. The information includes some of the old cost data from the builders of the original aircraft.458 4.854. Recommend which method Gregory should use and explain why.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.300 12. Gregory Gonzales.22 (R-squared = . Each of the 16 will be ﬂown to Air Force and aviation museums throughout the country for exhibition. Calculate the quarterly forecasts for 2007 using the high-low method and regression analysis. SE = 974).300 11.900 12. and thirty-second aircraft.55 and a coefﬁcient for the independent variable of \$126. High-Low Method. If Soren Industries expects that the time spent per unit will be the same as it was in 1938. The ﬁrst time Ben and David undertook one of these projects. As a starting point in further analyses. Required 1. wants to test different forecasting methods and then use the best one to forecast quarterly expenses for 2007.600 13.000 13. Regression Analysis Clothes for U is a large merchandiser of apparel for budget-minded families.100 11. As further authentication. Soren Industries wants to bid on the aircraft contract and asked for and received certain cost information about the Four F Sixes from the Air Force. pre–World War II aircraft. Fourth Edition 210 Part Two II.100 14.700 14. were completed.

189 \$2.094 1.300 2.254 2. 2.576 212 284 246 278 498 424 312 485 188 276 489 1.310 2.484 3.873 2. either in units or in dollars.476 3.190 2.600 2.210 2.384 2. and has estimated sales for 2007: Date Supplies Expense Sales Units Jun 2004 Jul Aug Sep Oct Nov Dec \$2.432 4.874 2.900 Jan 2007 (estimated) Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Sales Dollars Required 1.300 2.500 2.364 104 167 298 398 187 334 264 333 143 245 232 1.345 1.999 2.000 1. A major cost driver in either type of facility is the capacity of the plant.878 1.758 3.856 2.265 2.045 1.122 2.054 1.912 2. the capacity of a recycling plant is . For example.394 2.487 2.500 3.093 2. She has collected these data for sales and supplies expenses for June 2004 through December 2006.763 3.573 2.893 2.345 2.056 3.004 3.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.195 2.945 2.435 1.321 3.309 Jan 2006 Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 2. Evaluate the reliability and precision of the regression you have chosen.433 180 230 190 450 350 350 450 550 300 300 450 950 \$1.894 2.850 2.200 3.815 2.103 2.384 2.168 2.200 3.232 3.245 4.745 3.977 1.199 2. Planning and Decision Making © The McGraw−Hill Companies.400 2.563 2.152 2.134 3.045 2.190 1.500 3.900 2.301 2.232 2. Fourth Edition 212 Part Two II.278 354 436 525 145 199 543 1. 2008 Appendix B: Regression Analysis Planning and Decision Making to sales.784 2.463 Jan 2005 Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 2. Develop the regression that Jean should use based on these data and using the regression procedure in Excel or an equivalent regression software program. What are the predicted monthly ﬁgures for supplies expense for 2007? 6-56 Cross-Sectional Regression Analysis WasteTec is a large construction company that specializes in the construction of large wastewater treatment plants and recycling plants.857 2.934 2.009 2.189 3.000 2.

CA Hudson Falls. New York.439 57.463 60.200 3. using your regression? 2.750 Number of Employees 8 10 12 12 9 Location Sales 1 2 1 1 2 \$ 312. PA Bridgeport. Below is a sample of some recent projects and the related construction costs (in thousands). WA Arlington.021 163. and in still others. Regression analysis is a useful method to estimate the cost of a new plant by using a regression equation developed from prior plant construction projects. Jim has collected data on spoilage at each of his stores in the recent month and is looking for patterns of spoilage relative to store size (measured by square feet of ﬂoor space. the cross-sectional regression develops a model that represents the overall patterns in all the data. Ohio. and total sales) and to the location of the store (location 1 is an area where few arrests for theft. This approach is often used in cases similar to Jim’s in which the accuracy or reasonableness of the reported dependent variable is a concern. FL Ronkonkoma. or vandalism are made. Planning and Decision Making © The McGraw−Hill Companies.779 87.512 3.900 3. the main reason is theft. Jim is not sure. VA Camden. it is damage and vandalism. but he suspects.FL Spokane. How could it be improved? 6-57 Regression Analysis: Cross-Sectional Analysis. and the unusual stores will be identiﬁed by the largest error terms in the regression. based on his experience managing convenience stores. NY Panama City. that a relationship exists among these factors.384 Square Footage 2. Develop a regression model using Excel or an equivalent system to predict the cost of a proposed new plant in Babylon.050 1.369 77.902 127. NY Layton. The dependent variable of the regression is the actual construction cost of each project. GA Poughkeepsie.073 Required 1. disorderly conduct.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis.344 2. A colleague told him that a type of regression called “cross-sectional” regression would suit his needs. which will have a required capacity of 750 TPD.250 2. merchandise actually does spoil and must be thrown out.983 (Continued) .395 139.400 3.465 345. What is the predicted cost for the Babylon plant. NJ York. In effect. NY Okahuma. Calculation of a Regression Equation Jim Manzano is the general partner of an investment group that owns a number of commercial and industrial properties. Commerce.686 1.005 1. PA TPD Cost 360 400 420 450 500 506 510 518 528 800 975 1.730 84. in others.688 \$ 59. different stores).302 344. while the independent variable is the TPD for the plant.723 453. including a chain of 15 convenience stores located in the greater metropolitan area of Cleveland.405 75. Fourth Edition II.013 50. CT Chester. The objective of the cross-sectional regression is to compare the actual known value for the dependent variable to the predicted value as a basis for assessing the reasonableness of the actual value.908 2. NJ Savannah. The cross-sectional regression takes data from a single time period and determines predictions for the dependent variable at different cost objects (in this case. Evaluate the precision and reliability of the regression you have developed. Jim is concerned about the recent increase in inventory theft and waste (he calls it “spoilage”) in his stores. and location 3 is for areas with high arrests). number of employees.235 376. UT Oxford Township. These plants can vary in size from a few hundred TPD to as many as several thousand TPD.400 3. Spoilage has increased by more than 20 percent in each of the past two years.457 88.852 448.389 346. The following data are for the most recent month’s operations: Store Number 1 2 3 4 5 Inventory Spoilage \$ 1. 2008 Appendix B: Regression Analysis Chapter 6 Cost Estimation 213 measured by the number of tons of water per day (TPD) that the plant can process. In some stores.119 152.

2. which simply means that the relevant range of 5. In many instances.644 = − \$55. Fourth Edition 214 Part Two II.984 673. seasonal) patterns or outliers in Exhibit 1.034 772 \$36. Explain your choice.783 − \$123. Using the High-Low Method Begin by graphing the data to determine whether there are any unusual (i.783 − 7.250 2..774 Required 1. the sign of the variable should be positive. Use any of the four potential independent variables (or a combination) you think appropriate and explain your answer.506 3.000 5.166 ⫻ \$34.644 = − \$55. Using the regression equation you developed in requirement 1.647 562.984 325.364 \$5. because of limitations or errors in developing the regression analysis.150 Number of Employees 10 8 10 15 15 9 16 15 16 8 173 Location 3 1 1 2 1 1 2 3 2 1 Sales 502.166) = \$34.552 2.726 + \$34. The estimation equation therefore is useful only within the relevant range of approximately 5. 6–58 Regression Analysis in Tax Court Cases Since at least the late 1960s.e. Using Regression Analysis All relevant criteria favor the ﬁrst regression based on higher R-squared and t-values and lower standard error. determine which of the 15 stores might have inventory spoilage that is out of line relative to the entire chain of stores. thus.023 1.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. 2.644 ⫻ Number of deliveries/Month Note that the intercept is a negative number. Moreover.500 deliveries and should not be used to estimate costs outside that range.433 ⫻ \$34. the court systems in the United States and elsewhere have accepted regression analysis as evidence in court cases. Using Excel or an equivalent software program.534 287.758 4. This variable should have a direct relationship with overhead.245 − 5.000 to 7.500 5. so the high-low estimate can be determined directly from the low point (March) and the high point (December) as follows: To determine the slope of the line (unit variable cost) (\$201.800 3.550 2.806 2.345 588.253 1. which is difﬁcult to explain.500 7. the sign on the materials cost variable in regression 2 is negative.755 1.436 253.786.726 \$123.644 per delivery To determine the intercept \$201.200 57. however.166 to 7. prepare a regression analysis that predicts inventory spoilage at each of the 15 stores.443 3.433 − 5. Planning and Decision Making © The McGraw−Hill Companies.364 198. Required What factors regarding the development of a regression analysis do you suspect the tax courts considered in determining the acceptability of a regression analysis as evidence? Solutions to SelfStudy Problems 1. tax courts question or deny the regression evidence.500 3.947 225.700 2.726 The estimation equation is Vehicle costs = − \$55.500 3.433 deliveries is so far from the zero point (where the intercept is) that the intercept cannot be properly interpreted as a ﬁxed cost.245)/(7. 2008 Appendix B: Regression Analysis Planning and Decision Making Store Number 6 7 8 9 10 11 12 13 14 15 Totals Inventory Spoilage Square Footage 4. A study was performed recently to determine the factors in the regression analysis that the court considered in determining whether regression evidence was admissible.119 5. Also evaluate the precision and reliability of the regression you select.374 333. The graph shows no unusual patterns or outliers. The reason for the improvement of regression 1 over regression 2 might be that machine-hours .

labor-hours. By excluding machine-hours as an independent variable.000 5. The next step is to quantify this relationship with the high-low method and regression analysis. We begin with units produced as the independent variable. units produced.500 7. 2008 Appendix B: Regression Analysis Chapter 6 EXHIBIT 1 Cost Estimation 215 \$225. thus causing multicollinearity. or units shipped).000 4.000 6. Planning and Decision Making © The McGraw−Hill Companies.500 5. respectively. we now want to determine whether the relationship between defects and production level (Exhibit 3) has changed.500 6. Fourth Edition II. Exhibit 2 shows that the number of defective units varies considerably from month to month and that a steady increase has occurred over the past two years.000 Total Deliveries are highly correlated with either materials costs. we identify February 2007 and December 2008 as representative low and high periods.500 8. the other independent variables can be tried later. or both. and the regression improved as a result. Using Both High-Low and Regression Begin by graphing the data for the number of defective units. Knowing that the production level also has been increasing (as measured either by cost of sales. EXHIBIT 2 3000 Defective Units from January 2007 to December 2008 2500 Defects 2000 1500 1000 500 0 1 3 5 7 9 11 13 Months 15 17 19 21 23 . He should therefore use regression 1.000 Total Vehicle Expenses Plot of Data for Hector’s Delivery Service 150. For Exhibit 3. 3. George reduced or removed the multicollinearity. We begin with the high-low analysis. The objective is to identify any unusual patterns that must be considered in developing an estimate.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. The second graph (Exhibit 3) makes clear that a relationship exists between units produced and the number of defects.000 7.000 75. since it should have the most direct relationship with defects. as shown in Exhibit 2.

533 And Intercept = 2. and units produced. and the t-value for cost of sales is now OK. To reduce the effect of multicollinearity.533 ⫻ Production level The high-low estimate is subject to the limitations of subjectivity in the choice of high and low points and because it uses only those two data points to develop the estimate. the estimation equation is Number of defects = − 552 + 32. units shipped.335 − 32.533 ⫻ 88 = 1. the next step is to obtain a regression analysis from the previous data and to assess the precision and reliability of the regression estimate. but we observe that all three t-values are less than 2.311 − 1.0. and a very good R-squared. The results of regression 3. Thus.Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. R-squared for regression 2 is essentially the same as for regression 1. with the cost of sales variable only. Regression 1 has the following independent variables: cost of sales.533 ⫻ 58 = − 552 Thus.335)/(88 − 58) = 32. show that SE and the t-value improve again while R-squared is unchanged. indicating unreliable independent variables. since that variable is likely to be least associated with defective units and has among the lowest of the t-values. which removes the variable units shipped. we try regression 2. Production Level from January 2007 to December 2008 2500 Defects 2000 1500 1000 500 40 50 60 70 80 90 100 110 Units Produced We calculate the high-low estimate as follows (these two points are not the absolute lowest and highest points. we suspect multicollinearity among these variables. Because we expect correlation among these variables and because of the low tvalues. R-squared and SE are OK. 2008 Appendix B: Regression Analysis Planning and Decision Making EXHIBIT 3 3000 Defective Units vs. Fourth Edition 216 Part Two II. Because it has the best SE and t-values. but they produce a line that is representative of the data): slope = (2. The results for three regression analyses are presented in Exhibit 4. The regression can be completed with a spreadsheet program or any of a number of available software systems. although SE improves very slightly. the third regression is the best choice. Planning and Decision Making © The McGraw−Hill Companies. The dependent variable in each case is the number of defective units. .311 − 32. Regression is thus performed to provide a more precise estimate.

Blocher−Stout−Cokins−Chen: Cost Management: A Strategic Emphasis. Fourth Edition II.720 (cost of sales) 12.44 −.38 .20 Regression 2 92.230 (units produced) 1.837 (cost of sales) −.881 158 .95 1.702 (cost of sales) −.849 (units produced) 4.54 Cost Estimation 217 R-squared Standard Error of the Estimate .881 155 Regression 1 103.309 4.974 (units shipped) −2. 2008 Appendix B: Regression Analysis Chapter 6 EXHIBIT 4 Regressions for the Number of Defects Intercept Coefﬁcient of Independent Variable t-value for Independent Variable −38. Planning and Decision Making © The McGraw−Hill Companies.77 .883 161 .24 Regression 3 43.72 −2.