BlocherStoutCokinsChen:

Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
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 signicantly enhance the ability to predict costs when learning occurs, three inherent limitations and problems are associated with the use of this method.
The rst 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 exibility 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 xed 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 rst 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 t 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 nds 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, rst 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 simplies 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.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

190 Part Two

II. Planning and Decision

Making

The McGrawHill
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

Dependent
Variable
Y
250
310
325

Variance Components
6

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)

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 coefcient 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
regressions 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 rst 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 15
T 15
R0

350
R 37.5

E 7.5
R 37.5

300
T 30
250

E 7.5
T 45

200

50

Mean of the
dependent
variable

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

100
150
Independent Variable (units of output)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

EXHIBIT 6B.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.
The degrees of freedom
for each component of variance
represent the number of
independent choices that can be
made for that component.
Mean squared variance
is the ratio of the amount of
variance of a component to the
number of degrees of freedom
for that component.

Source of Variance

Variance of Each Component

of the Regression (also
called sum of squares)

Explained (regression)
Error
Total

37.52 + 02 + 37.52 = 2,812.5

7.52 + 152 + 7.52 = 337.5
(45)2 + (15)2 + (30)2 = 3,150.0

Cost Estimation 191

Degrees
of Freedom

Mean
Squared
Variance

1
1
2

2,812.5
337.5
1,575.0

column (7) shows the error variance, (Y YE). The measures in these last three columns are
squared and summed to arrive at the desired values for total variance, explained variance, and
error variance, respectively. The sum of the error and explained variance terms equals total variance. These terms are illustrated in Exhibit 6B.2 and the values calculated in Exhibit 6B.3.
The three variance terms are the basic elements of the statistical analysis of the regression.
This is best illustrated in the analysis of variance table in Exhibit 6B.3. The analysis of variance table separates the total variance of the dependent variable into both error and explained
components. The rst two columns of the table show the type and amount of variance for
each of the three variance terms. The third column shows the degrees of freedom for each
component, which represents the number of independent choices that can be made for that
component. Thus, the number of degrees of freedom for the explained variance component
is always equal to the number of independent variables, and the total degrees of freedom is
always equal to the number of data points less 1. The error degrees of freedom equal the total
less the explained degrees of freedom.
The fourth column, mean squared variance, 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).
The analysis of variance table serves as a useful basis to discuss the key statistical measures of the regression. Of the six principal measures in Exhibit 6B.4, one measure refers
to the precision of the regression and ve measures refer to the reliability of the regression.
Precision refers to the ability of the regression to provide accurate estimateshow close
the regressions estimates are to the unknown true value. Reliability refers to the condence
the user can have that the regression is valid; that is, how likely the regression is to continue to provide accurate predictions over time and for different levels of the independent
variables.

Precision of the Regression

The standard error of the estimate (SE) is a useful measure of the accuracy of the regressions
estimates. The standard error is interpreted as a range of values around the regression estimate
such that we can be approximately 67 percent condent the actual value lies in this range (see
Exhibits 6.8A and 6.8B). An inverse relationship, and therefore a trade-off, exists between the
condence level and the width of the interval. 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
= 337.5 = 18.37

EXHIBIT 6B.4
Six Key Statistical Measures

Precision
1. Precision of the regression (measured by the standard error of the estimate)
Reliability
2. Goodness of t (R-squared)
3. Statistical reliability (F-statistic)
4. Statistical reliability for each independent variable (t-value)
5. Reliability of precision (rank-order correlation)
6. Nonindependence of errors (Durbin-Watson statistic)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

192 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

A condence interval
is a range around the regression
line within which the
management accountant can be
condent the actual value of the
predicted cost will fall.

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, as
illustrated in the preceding formula for SE.
The standard error of the estimate can also be used to develop condence intervals for the
accuracy of the prediction, as illustrated in Exhibits 6.8A and B. A condence interval is a
range around the regression line within which the management accountant can be condent
the actual value of the predicted cost will fall. A 67 percent condence 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; a 95 percent condence interval would be determined from
2 standard error distances. Condence intervals are useful and precise tools for management
accountants to describe the degree of precision obtained from the regression.

Goodness of Fit (R-squared)

R-squared (also called the coefcient of determination) is a direct measure of the explanatory
power of the regression. It measures the percent of variance in the dependent variable that can
be explained by the independent variable. R-squared is calculated in Exhibit 6B.3:
R2 =
=

of squares (explained )
of squares (total)
2, 812.5
= .892
3, 150

The explanatory power of the regression improves as the explained sum of squares increases
relative to the total sum of squares. A value close to 1 reects a good-tting regression with
strong explanatory power.

Statistical Reliability (F-Statistic)

The F-statistic
is a useful measure of the
statistical reliability of the
regression.

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. If
only a small number of data points are used, it is possible to have a relatively high R-squared
(if the regression is a good t to the data points), but this offers relatively little condence that
a statistical relationship exists because of the small number of data points.
The larger the F, the lower the risk that the regression is statistically unreliable. The determination of an acceptable F-value depends on the number of data points, but the required
F-value decreases as the number of data points increase. Most regression software programs
show the F-value and the related p-value, which should be less than approximately 5 percent.
The F-statistic can be obtained from the analysis of variance table as follows:
Mean square (explained )
Mean square error
2,, 812.5
= 8.333
=
337.5

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, it has
an interpretation very much like that of the F-statistic. The t-value equals the ratio of the coefcient of the independent variable to the standard error of the coefcient for that independent variable. The standard error of the coefcient is not the same as the standard error of the
estimate, but it is interpreted in the same way. However, the SE cannot be obtained directly
from the analysis of variance table. For the data in Exhibit 6B.1, the value of the standard error
for the coefcient is .2598.7 The -value is thus
t = .75/.2598 = 2.8868
7

The standard error of the coefcient is calculated as follows:

Standarderror = SE/(Std.deviationof theinddependent variable)
=

18.37
(50 100)2 + (100 100)2 + (150 100)2

= .2598

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

Appendix B: Regression
Analysis

The McGrawHill
Companies, 2008

Chapter 6

Cost Estimation 193

EXHIBIT 6B.5

Dependent Variable

Nonconstant Variance

Independent Variable

A t-value larger than 2.0 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. Regression software such as Excel shows the 95 percent condence range
for the coefcient of each of the independent variables. The range of the standard error of the
estimate should be relatively small. A small range provides condence in the accuracy of the
coefcients value.
Nonconstant variance
is the condition when the
variance of the errors is not
constant over the range of the
independent variable.

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.

Reliability of Precision (Nonconstant Variance)

For certain sets of data, the standard error of the estimate varies over the range of the independent variable. The variance of the errors is not constant over the range of the independent variable. This is the case, for example, when the relationship between the independent
and dependent variables becomes less stable over time. This type of behavior is illustrated in
Exhibit 6B.5.
If there is nonconstant variance, the SE value provided by the regression is not uniformly
accurate over the range of the independent variable. To detect nonconstant variance, we calculate the rank-order correlation between the position of the data and the size of the error. 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 high rank-order correlation is evidence of
nonconstant variance. For the data in Exhibit 6B.1, the Spearman rank-correlation coefcient
is .125, a relatively small correlation that indicates little evidence of nonconstant variance. The
calculation of rank-order correlation is beyond the scope of this introductory chapter but can
be found in many statistics texts.
To x the problem of nonconstant variance, management accountants should transform the
dependent variable with the log or square root to see whether this improves the behavior of
the errors. If it does not x the condition, management accountants should be very cautious in
interpreting the SE value.
Nonindependent Errors (Durbin-Watson Statistic)

The Durbin-Watson statistic

is a measure of the extent of
nonlinearity in the regression.

A key assumption of regression is that the relationship between the independent and dependent
variables is linear. If the data are nonlinear because of seasonality or a cyclical pattern, for
example, the errors are systematically related to each other, that is, are not independent. This
assumption is violated frequently because nancial data are often affected by trend, seasonality, and cyclical inuences. The relationship between the variables might also be inherently
nonlinear, as when learning occurs or a multiplicative rather than an additive relationship exists (such as predicting payroll costs from hours worked and wage rates). Then the regression
is unreliable and subject to greater than expected estimation errors. One type of nonlinearity
(nonindependence of errors) is illustrated in Exhibit 6B.6.
A common method that detects nonlinearity is the Durbin-Watson (DW) statistic. It is calculated from the amount and change of the errors over the range of the independent variable.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

194 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

EXHIBIT 6B.6

Dependent Variable

Nonindependence of Errors

Independent Variable

EXHIBIT 6B.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 t

R-squared

Should be
approximately
.75 or better

Add or delete independent variables

If DW is poor, could need transforms
(lag, log, rst differences, . . .)
Correct measurement errors in the
data, for example, cutoff errors, or
reporting lags

Inaccurate estimates

Statistical reliability
for the regression

F-statistic

Depends on
sample size

Increase sample size

Other changes as suggested for
reliabilitygoodness of t (see above)

Inaccurate estimates

Statistical reliability
for the independent
variables

t-value

Should be
greater than 2.0

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 t 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 2.0
and 3.0

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. The exact values for a specic regression depend on a number of factors including the sample size and the number of
independent variables. A recent study of regression analysis applied to 20 different overhead cost accounts showed that most of the R-squared values fall between .83 and .93. 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. See G. R. Cluskey Jr., Mitchell H. Raiborn, and
Doan T. Modianos, Multiple-Cost Flexible Budgets and PC-Based Regression Analysis, Journal of Cost Management, JulyAugust 2000, pp. 3547.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

Appendix B: Regression
Analysis

The McGrawHill
Companies, 2008

Chapter 6

Cost Estimation 195

The DW value falls between zero and 4.0; with 20 or more data points, a value of DW between
approximately 1.0 and 3.0 indicates little chance of a nonlinearity as described earlier; values
less than 1.0 or greater than 3.0 should indicate the need to study the data and to choose appropriate xes if necessary.
The problem of nonindependent errors usually can be xed by deseasonalizing the data,
using a dummy variable for seasonality, or using an index to remove the trend. Alternatively,
what may be required is to convert a multiplicative relationship to an equivalent additive (that
is, linear) relationship by taking the logarithm of the independent and dependent variables.
The statistical measures, their indicators, and ways to x the underlying conditions are summarized in Exhibit 6B.7.

TIME SERIES AND CROSS-SECTIONAL REGRESSION

Time-series regression
is the application of regression
analysis to predict future
amounts, using prior periods
data.
Cross-sectional regression
estimates costs for a particular
cost object based on information
on other cost objects and other
variables, where the information
for all variables is taken from the
same period of time.

Key Terms

Comments on Cost
Management
in Action

The two examples used in this chapter, the Ben Garcia case and the WinDoor Inc. case, are
both examples of what is called time-series regression. Time-series regression is the application of regression analysis to predict future amounts, using prior periods data. In contrast,
cross-sectional regression estimates costs for a particular cost object based on information
on other cost objects and variables, where the information for all variables is taken from the
same period of time. For example, suppose a residential home builder uses regression to estimate the cost of constructing a new home, and the builder knows that the main cost driver
for building cost is the size of the home, in square feet of oor space. 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. The regression equation that the builder develops is then used to predict the cost of homes to be built, based on
the expected size of the new home in square feet. All of the statistical measures of reliability
and precision explained above apply equally to both types of regression, except for the issue,
nonindependence of errors, which applies only in time-series regressions.

analysis of variance table, 191

condence interval, 192
correlation, 180
cost estimation, 171
cross-sectional regression, 195
degrees of freedom, 191
dependent variable, 176
dummy variable, 179
Durbin-Watson statistic, 193
rst difference, 185
F-statistic, 192
high-low method, 173

independent variable, 176

learning curve analysis, 186
learning rate, 186
least squares regression, 176
mean absolute percentage
error (MAPE), 173
mean squared variance, 191
multicollinearity, 180
multiple linear regression, 183
nonconstant variance, 193
outlier, 177
p-Value, 180

rank-order correlation, 193

regression analysis, 176
R-squared, 179
simple linear regression, 183
standard error of the
estimate, 180
time-series regression, 195
trend variable, 185
t-value, 179
work measurement, 176
work sampling, 176

Using Regression to Estimate the Value of Commercial Real Estate

Estimating Real Estate Values for Apartment Buildings and Ofce Buildings As expected, real estate appraisers performing regression analysis to appraise the value of an apartment building or an ofce building
use as the dominant independent variable the propertys past, current, and expected future net operating
income (NOI). That is, the chief determinant of the value of the property is its ability to produce cash ows
and prots. Other variables regarding the property include its size (as measured by the number of units,
number of square feet, number of two-bedroom and one-bedroom apartments, etc.), its age, and the relevant
vacancy rate in the property and in the submarket area where it is located. Since the regression analysis is
usually built from actual sales numbers over a period of time, these appraisers also use a trend variable to tie
the sales price of the property to the year it was sold.
Estimating Real Estate Values for Warehouses and Manufacturing Plants Similarly, real estate appraisers
have developed regression analyses for warehouses and manufacturing plants using their size, age, and location. The NOI variable is usually not relevant. They also use a trend variable to distinguish sales of properties in different years. For example, an analysis of sales value (per square foot) of industrial properties in the
Los Angeles area in the early 1990s showed a signicant trend variable ($2.83 per square foot per year); the

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

196 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

coefcient on the trend variable was negative because prices were falling during that period. A signicant
size variable ($2.43 per square foot, per 100,000 square feet of space) indicated that larger buildings had
on average lower sales prices per square foot. Age was also a factor, the coefcient being $0.41 per square
foot per year of age. The location variable was also signicant, showing that properties in certain counties
in the Los Angeles area (Orange County, San Bernadino, etc.) were predicted to have as much as a $2.32
difference in value per square foot.
Sources: Stephen T. Crosson, Charles G. Dannis, and Thomas G. Thibodeau, Regression Analysis: A Cost-Effective Approach
for the Valuation of Commercial Property, Real Estate Finance, Winter 1996; Maxwell O. Ramsland Jr. and Daniel E. Markham,
Market-Supported Adjustments Using Multiple Regression Analysis, The Appraisal Journal, April 1998, pp. 18191; and Stephen
C. Kincheloe, Linear Regression Analysis of Economic Variables in the Sales Comparison and Income Approaches, The Appraisal
Journal, October 1993.

Self-Study
Problems
(For solutions, please turn
to the end of the chapter.)

1. Using the High-Low Method

Hectors Delivery Service uses four small vans and six pickup trucks to deliver small packages in the
Charlotte, North Carolina, metropolitan area. Hector spends a considerable amount of money on the gas, oil,
and regular maintenance of his vehicles, which is done at a variety of service stations and repair shops. To
budget his vehicle expenses for the coming year, he gathers data on his expenses and number of deliveries
for each month of the current year.

January
February
March
April
May
June
July
August
September
October
November
December

Total Vehicle
Expenses

Total
Deliveries

$145,329
133,245
123,245
164,295
163,937
176,229
180,553
177,293
155,389
150,832
152,993
201,783

5,882
5,567
5,166
6,621
6,433
6,681
7,182
6,577
5,942
5,622
5,599
7,433

Required Use the high-low estimation method to determine the relationship between the number of deliveries and the cost of maintaining the vehicles.

2. Using Regression Analysis

George Harder is the Imperial Foods Companys plant manager of one of the processing plants. George is
concerned about the increase in plant overhead costs in recent months. He has collected data on overhead
costs for the past 24 months and has decided to use regression to study the factors inuencing these costs.
He has also collected data on materials cost, direct labor-hours, and machine-hours as potential independent
variables to use in predicting overhead.
George runs two regression analyses on these data, 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

.65
$12,554

12%
2.0
4.5

Required Which of the two regressions is better and why?

Regression 2
(labor-hours and
machine-hours)
.58
$13,793

14%
1.6
3.8
1.4

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

Cost Estimation 197

3. Using Both High-Low and Regression

John Meeks Company is a medium-size manufacturing company with plants in three small mid-Atlantic
towns. The company makes plastic parts for automobiles and trucks, primarily door panels, exterior trim,
and related items. The parts have an average cost of $5 to $20. 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.
Currently, management is reviewing the incidence of scrap and waste in the manufacturing process at
one of its plants. Meeks denes scrap and waste as any defective unit that is rejected for lack of functionality
or another aspect of quality. The plants have a number of different inspection points, and failure or rejection
can occur at any inspection point. The number of defective units is listed in the following table; management
estimates the cost of this waste in labor and materials is approximately $10 per unit.
An unfavorable trend appears to exist with regard to defects, and management has asked you to investigate and estimate the defective units in the coming months. A rst step in your investigation is to identify the
cost drivers of defective parts, to understand what causes them, and to provide a basis on which to estimate
future defects. For this purpose, you have obtained these recent data on the units produced, the units shipped,
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
1,335
1,610
1,405
1,511
1,600

Jul
Aug
Sep
Oct
Nov
Dec

75
81
70
79
82
70

962
1,052
1,104
1,224
1,261
1,020

70
76
80
89
92
74

1,570
1,910
2,011
2,230
2,300
1,849

Jan 2008
Feb
Mar
Apr
May
Jun

67
72
85
75
81
85

850
916
1,107
968
1,037
1,103

62
67
80
70
76
80

1,549
1,669
2,012
1,756
1,889
1,650

92
100
91
101
105
88

1,208
1,310
1,380
1,536
1,580
1,270

87
95
101
111
115
92

2,187
2,387
2,514
2,787
2,310
2,311

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 t for this purpose.

Questions

6-1
6-2
6-3
6-4
6-5
6-6
6-7
6-8
6-9
6-10

Dene cost estimation and explain its purpose.

Explain the assumptions used in cost estimation.
List the three methods of cost estimation. Explain the advantages and disadvantages of each.
Explain the implementation problems in cost estimation.
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.
How is cost estimation used in activity-based costing?
Explain how to choose the dependent and independent variables in regression analysis.
What are nonlinear cost relationships? Give two examples.
List four advantages of regression analysis.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

198 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 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.
How do we know when high correlation exists? Is high correlation the same as cause and effect?
What does the coefcient 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 cost driver is number of labor-hours. Wallace determines that the
high and low costs are $25,830 and $18,414, respectively, and the values for the cost driver are 3,495
and 1,958 hours, respectively. What is the variable cost per hour?
Carter Dry Cleaning has developed two regression analyses for cost estimation. The accounting
manager has presented statistical measures for both of these regressions. Regression A has an
R-squared value of .53 and a t-value of 1.08 and Regression B has an R-squared of .89 and a t-value
of 2.17. What do these statistical measures indicate about the regressions? Which regression should
Carter Dry Cleaning use for cost estimation?
Williams Inc. produces uorescent lightbulbs for commercial use. The accounting manager is attempting to estimate the total cost for the next quarter using the high-low method. He has compiled
data and found the high and low costs are $10,000 and $6,000 and the associated cost drivers are
7,000 and 3,000 packs, respectively. He has also determined the variable cost to be $1.00 per pack.
What is the value for a (the xed quantity)?
Grant Healthcare provides plastic gloves for hospitals. They are attempting to forecast costs for future production. Their dependent variable is labor expense. List some possible independent variables
for a regression analysis of nancial data.
Smith Glass Co. produces industrial glass for factories. 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

3,197
46,835

4,105
53,227

5,056
49,734

3,586
43,649

Which two years should Smith select for the high-low method analysis and why?
Johnson Plastics Inc. produces jewel cases for CDs. The accounting manager has calculated a
regression to determine future production costs. The regression estimate for 2008 is $5,000 with an
R-squared of .9, a t-value of 2.5, and a standard error of $400. Within what interval would she be
reasonably (67 percent) condent that the actual values will fall?
Peppers Lockdown produces keys for houses and cars. As they were planning for next years production, they decided to implement a high-low system to forecast future costs. The accounting manager
determined that the years to be used are 2003, with total production of 2,500,000 keys at a total cost
of $10,000, and 2006, with total production of 3,000,000 keys at a total cost of $20,000. What is the
variable cost per key?
Power Drink Inc. produces sports drinks. The accounting manager has decided to implement a highlow costing system to predict future materials handling costs. She has provided you with the following table of costs for the last ve years. Which two years should be used for this method?

Production hours
Handling cost ($)

6-22

2003
5,683
50,457

2003

2004

2005

2006

2007

100,000
456,233

138,679
498,672

98,843
507,284

203,517
601,678

188,352
544,314

Jamison Construction has implemented a costing system by use of a regression analysis of past
costs. The variable cost per hour of labor is $35 and the xed cost was determined to be $125,000.
If Jamison projects it will be working 200,000 hours in 2008, what is the projected total cost?
Curry Rubber manufactures rubber bands for commercial retail companies. The accounting manager has created a cost projection formula through a regression analysis of past data. He presents this
to you to evaluate the reliability of the regression. You notice that the formula has an R-squared of .6,
a t-value of 2.3, and a standard error of $200,000. The estimate for next quarter costs is $2,584,072.
What do these statistics tell you about the reliability of his regression analysis?

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 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. Miller has compiled data and
decided to use the high-low costing method. Their low value is $250,000 for 5,000 hours and their
high value is $400,000 for 8,000 hours. What is their variable cost per hour?
Sanders Bears produces stuffed animals. They are in the process of implementing a cost forecasting
system using the high-low method. They have found the variable cost per animal to be $2 and their
high and low costs used were $80,000 for 120,000 animals and $40,000 for 100,000 animals. What
is the value of the xed cost for their formula?
Cost Classication: Match each cost to the appropriate cost behavior pattern shown in the graphs (a)
through (l). Any graph can t two or more patterns.

1.
2.
3.
4.

5.
6.
7.
8.
9.

10.
11.
12.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)

The cost of lumber used to manufacture wooden kitchen tables.

The cost of order llers in a warehouse. When demand increases, the number is increased, and
when demand falls off, the number is decreased.
The salary of the plants quality control inspector, who inspects each batch of products.
The cost of water and sewer service to the manufacturing plant. The local municipality charges
a xed rate per gallon for usage up to 10,000 gallons, and a higher charge per gallon for usage
above that point.
The cost of an Internet connection of $23 per month.
The cost of an Internet connection of $10 per month plus $2 per hour of usage above 10
hours.
The cost to make copies of a given document at a printing shop, which reduces the per-copy
charge for customers who make more than 100 copies of the document.
The total cost of manufacturing a new camera over its entire life cycle.
To discourage excess usage and to level the demand, especially in peak load times, the local
electric utility increases the per-kilowatt-hour charge for each additional 5,000 kilowatt-hours
usage.
A clothing store in the SunnyVale Mall pays a xed rental charge of $1,000 per month plus
2 percent of gross sales receipts.
The cost of repair for a machine used in manufacturing.
A shoe store in the SunnyVale Mall pays 6 percent of gross sales receipts, up to a maximum
of $3,000 per month as a rental charge.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

200 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 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. The variable costs for each
computer are $150 and the xed costs total $62,250 per month.

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., a contact lens manufacturer:

Output in Units

Fixed Costs

250
300
350
400

$4,750
4,750
4,750
4,750

Variable Costs

Total Costs

$ 7,500
9,000
10,500
12,000

$12,250
13,750
15,250
16,750

Required
1. Graph total cost, total variable costs, and total xed costs.
2. Graph the per-unit total cost, per-unit variable cost, and per-unit xed cost.
3. Discuss the behavior of the xed, variable, and total cost.
6-29

Cost Estimation, Average Cost Maribeths Cafe bakes croissants that it sells to local restaurants
and grocery stores in the Raleigh, North Carolina, area. The average costs to bake the croissants are
$0.55 for 500 and $0.50 for 600.

Required If the total cost function for croissants is linear, 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. The agencys argument is that a positive linear relationship exists between advertising and sales in the sailboard industry. 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

$2,500
3,000
3,500
4,000
4,500
5,000
5,500

$ 95,000
110,000
124,000
138,000
143,000
147,000
150,000

Required
1. Graph annual sales and advertising expense.
2. Do the data prove Sues 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.

SE
R-squared
t-values:
Labor-hours
Machine-hours
Required

Regression 1

Regression 2

Regression 3

33,844
0.55

45,383
0.35

31,044
0.58

1.1

1.9
0.8

2.3

Which regression would you choose and why?

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

6-32

Cost Estimation 201

Cost Estimation: High-Low Method Horton Manufacturing Inc. produces blinds and other window
treatments for residential homes and ofces. The owner is concerned about the maintenance costs
for the production machinery, as maintenance costs for the previous scal year were higher than
he expected. He has asked you to assist him in estimating his future maintenance costs so that he
can better predict his rms protability. Together, you have determined that the best cost driver for
maintenance costs is machine-hours. These data are from the previous scal year for maintenance
expense and machine-hours:

Month
1
2
3
4
5
6

Expense

Hours

Month

Expense

Hours

$2,625
2,670
2,720
2,822
2,855
3,005

1,499
1,590
1,605
1,655
1,775
1,880

7
8
9
10
11
12

$2,865
2,905
2,780
2,570
2,590
2,890

1,785
1,805
1,695
1,410
1,550
1,405

Required What is the cost equation for maintenance cost using the high-low method? Graph the data
points to check for outliers.
6-33

Cost Estimation, High-Low Method Ethan Manufacturing Inc. produces oor mats for automobiles.
The owner, Joseph Ethan, has asked you to assist him in estimating his maintenance costs. Together,
you and Joseph determine that the single best cost driver for maintenance costs is machine-hours.
These data are from the previous scal 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

$2,600
2,760
2,910
3,020
3,100
3,070
3,010
2,850
2,620
2,220
2,230
2,450

1,690
1,770
1,850
1,870
1,900
1,880
1,860
1,840
1,700
1,100
1,300
1,590

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 patients stay
in the hospital. The researchers used regression analysis and found the following results:
Total Cost for Patient = Constant, plus
a length of stay (measured in days), plus
b presence of one or more complications (= 1 if true, 0 if false), plus
c use of a laparoscope (= 1 if true, 0 if false)
Where:
a, b, c are coefcients of the regression model, and
The laparoscope is an instrument somewhat like a miniature telescope with a ber optic system which brings light into the abdomen. It is about as big around as a fountain pen and twice
as long.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

202 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

The research, based on 57 patients, showed the following regression results:

R-squared: 53%
constant term: $3,719
Coefcients and t-values for independent variables:
Length of Stay
coefcient
t-value

$ 861
10.76

Complication

Laparoscope

$1,986
4.89

$ 908
2.54

Required
1. What is the estimated cost for a patient who has complications and stays in the hospital two days, and
whose surgery requires a laparoscope?
2. Which, if any, dummy variables are used in this regression?
3. Comment on the statistical measures for the model.
6-35

Analysis of Regression Results; Appendix (Continuation of 6-34) The following table shows the
regression results presented by the researchers in the study described in Exercise 6-34. The righthand column shows the results for the laparoscopic surgery. The left-hand column shows the results
for the sample of patients who were treated without the laparoscopic surgery, and the related costs.

Coefcients for independent variables

Regression intercept
Length of stay
Coefcient*
Standard error for the coefcient
Number of complications
Coefcient
Standard error for the coefcient
Laparascopic
Coefcient
Standard error for the coefcient
R-squared

Not Laparoscopic

Laparoscopic

$ 8,043

$ 3,719

Not signicant
Not applicable

861
80

3,393
1,239

1,986
406

Not applicable
Not applicable
0.11

908
358
0.53

*
Note: All independent variables are signicant at the level of p = .05 (and t-value >2) except for the length of stay variable in the nonlaparoscopic
condition.

Required
1. Which of the two regressions has the better reliability and precision in estimating cost? Why?
2. Interpret the values of each coefcient and the standard error for each coefcient.
3. What are the t-values for each of the independent variables for each treatment condition?

Problems

6-36

Cost Estimation, High-Low Method Jay Bauer Company specializes in the purchase, renovation,
and resale of older homes. Jay Bauer employs several carpenters and painters to do the work for
him. It is essential for him to have accurate cost estimates so he can determine total renovation costs
before he purchases a piece of property. If estimated renovation costs plus the purchase price of a
house are higher than its estimated resale value, the house is not a worthwhile investment.
Jay has been using the homes interior square feet for his exterior paint cost estimations. Recently he decided to include the number of openingsthe total number of doors and windows in a
houseas a cost driver. Their cost is signicant because they require time-consuming preparatory
work and careful brushwork. The rest of the house usually is painted either by rollers or spray guns,
which are relatively efcient ways to apply paint to a large area. Jay has kept careful records of these
expenses on his last 12 jobs:

House

Square Feet

Openings

Cost

1
2
3

2,600
3,010
2,800

13
15
12

$3,300
3,750
3,100
(Continued)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

4
5
6
7
8
9
10
11
12

2,850
4,050
2,700
2,375
2,450
2,600
3,700
2,650
3,550

12
19
13
11
11
10
16
13
16

Cost Estimation 203

3,150
4,700
3,250
2,800
2,800
2,875
4,100
3,200
3,950

Required
1. Using the high-low cost estimation technique, determine the cost of painting a 3,300-square-foot house
with 14 openings. Also determine the cost for a 2,400-square-foot house with 8 openings.
2. Plot the cost data against square feet and against openings. Which variable is a better cost driver? Why?
6-37

Cost Estimation, Machine Replacement, Ethics SpectroGlass Company manufactures glass for
ofce buildings in Arizona and Southern California. As a result of age and wear, a critical machine in
the production process has begun to produce quality defects. SpectroGlass is considering replacing the
old machine with a new machine, either brand A or brand B. 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)
4,000
7,000
9,000
14,000
16,000
24,000

Machine A
Estimated Total Costs
$ 54,600
78,800
90,300
114,900
132,400
210,000

Machine B
Estimated Total Costs
$ 70,000
100,000
115,000
137,000
146,000
192,000

Required
1. If SpectroGlasss output is expected to be 22,000 square yards, which machine should it purchase? At
15,000 square yards?
2. As a cost analyst at SpectroGlass, you have been assigned to complete requirement 1. A production
supervisor comes to you to say that the nature of the defect is really very difcult to detect and that most
customers will not notice it, so he questions replacing it. 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. What do you say?
3. Assume that brand A is manufactured in Germany and brand B is manufactured in Canada. As a U.S.
based rm, what considerations are important to SpectroGlass, in addition to those already mentioned in
your answer to requirement 1?
6-38

Cost Estimation, High-Low Method Antelope Park Amoco (APA) in Antelope Park, Alaska, has
noticed that utility bills are substantially higher the colder the average monthly temperature is. The
only thing in the shop that uses natural gas is the furnace. Because of prevailing low temperatures,
the furnace is used every month of the year (though less in the summer months and very little
in August). Everything else in the shop runs on electricity, and electricity use is fairly constant
throughout the year.
For a year, APA has been recording the average daily temperature and the cost of its monthly
utility bills for natural gas and electricity.
Average Temperature
January
February
March
April
May
June

31 F
41
43
44
46
50

Utility Cost
$760
629
543
410
275
233
(Continued)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

204 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 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. The forecast for January is a near record average temperature of 10F; temperatures in February are
expected to average 40F.
6-39 to 6-43 Regression Analysis Problems 6-39 through 6-43 are based on Armer Company, which is
accumulating data to use in preparing its annual prot plan for the coming year. The cost behavior
pattern of the maintenance costs must be determined. The accounting staff has suggested the use of
linear regression to derive an equation for maintenance hours and costs. 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
4,800
400

Average cost per hour ($43,200/4,800) = $9.00

a (intercept)
b coefcient
Standard error of the estimate
R-squared
t-value for b

Maintenance Costs
$ 4,200
3,000
3,600
2,820
4,350
2,960
3,030
4,470
4,260
4,050
3,300
3,160
$43,200
3,600

684.65
7.2884
34.469
.99724
60.105

Required (6-39) If Armer Company uses the high-low method of analysis, the equation for the relationship
between hours of activity and maintenance cost follows:
a. y = 400 + 9.0x
b. y = 570 + 7.5x
c. y = 3,600 + 400x
d. y = 570 + 9.0x
e. None of the above

(CMA Adapted)
Required (6-40) Based on the data derived from the regression analysis, 420 maintenance hours in a month
mean that maintenance costs should be budgeted at
a. $3,780
b. $3,461

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

Cost Estimation 205

c. $3,797
d. $3,746
e. None of the above

(CMA Adapted)
Required (6-41) The coefcient of determination for Armers regression equation for the maintenance
activities is
a. 34.469/49.515
b. .99724
c. square root of .99724
d. (.99724)2
e. None of the above

(CMA Adapted)
Required (6-42) The percent of the total variance that can be explained by the regression equation is
a. 99.724%
b. 69.613%
c. 80.982%
d. 99.862%
e. None of the above

(CMA Adapted)
Required (6-43) At 400 hours of activity, Armer management can be approximately two-thirds condent
that the maintenance costs will be in the range of
a. $3,550.50 to $3,649.53
b. $3,551.37 to $3,648.51
c. $3,586.18 to $3,613.93
d. $3,565.54 to $3,634.47
e. None of the above

(CMA Adapted)
6-44

Regression Analysis Whittenberg Distributors, a major retailing and mail-order operation, has
been in business for the past 10 years. During that time, its mail-order operations have grown from
a sideline to represent more than 80 percent of the companys annual sales. Of course, the company
has suffered growing pains. At times, overloaded or faulty computer programs resulted in lost sales,
and scheduling temporary workers to augment the permanent staff during peak periods has always
been a problem.
Peter Bloom, manager of mail-order operations, has developed procedures for handling most
problems. However, he is still trying to improve the scheduling of temporary workers to take customer telephone orders. Under the current system, Peter keeps a permanent staff of 60 employees who
handle the base telephone workload and supplements this staff with temporary workers as needed.
The temporary workers are hired on a daily basis; he determines the number needed for the next day
the afternoon before based on his estimate of the upcoming telephone volume.
Peter has decided to try regression analysis to improve the hiring of temporary workers. By
summarizing the daily labor-hours into weekly totals for the past year, he determined the number
of workers used each week. In addition, he listed the number of orders processed each week. After
entering the data into a spreadsheet, Peter ran two regressions. Regression 1 related the total number
of workers (permanent staff plus temporary workers) to the number of orders received. Regression
2 related only temporary workers to the number of orders received. The output of these analyses
follows:
Regression model: W = a + b T
where:
W = workers; T = telephone orders

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

206 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

Regression 1
a
b
Standard error of the estimate
t-value
Coefcient of determination
Durbin Watson statistic

21.938
.0043
3.721
1.95
.624
1.33

Regression 2
46.569
.0051
1.495
2.04
.755
1.67

Required
1. Peter Bloom estimates that Whittenberg Distributors will receive 12,740 orders during the second week
of December.
a. Predict the number of temporary workers needed for this week using regression 1. Round your
answer to the nearest whole number.
b. Using regression 2, predict the number of temporary workers needed during this week. Round your
answer to the nearest whole number.
2. Which of the two regression analyses appears to be better? Explain your answer.
3. Describe at least three ways that Peter Bloom could improve his analysis to make better predictions than
either of these regression results provides.
(CMA Adapted)
6-45

Regression Analysis Pilot Shop is a catalog business providing a wide variety of aviation products
to pilots throughout the world. Maynard Shephard, the recently hired assistant controller, has been
asked to develop a cost function to forecast shipping costs. The previous assistant controller had
forecast shipping department costs each year by plotting cost data against direct labor-hours for the
most recent 12 months and visually tting a straight line through the points. The results were not
satisfactory.
After discussions with the shipping department personnel, Maynard decided that shipping costs
could be more closely related to the number of orders lled. He based his conclusion on the fact that
10 months ago the shipping department added some automated equipment. Furthermore, he believes
that using linear regression analysis will improve the forecasts of shipping costs. Cost data for the
shipping department have been accumulated for the last 25 weeks. He ran two regression analyses of
the data, one using direct labor-hours, and one using the number of cartons shipped. The information
from the two linear regressions follows:

Equation
R-squared
Standard error of the estimate
t-value

Regression 1

Regression 2

SC = 804.3 + 15.68DL
.365
2.652
1.89

SC = 642.9 + 3.92NR
.729
1.884
3.46

where:
SC = total shipping department costs
DL = total direct labor-hours
NR = number of cartons shipped
Required
1. Identify which cost function (regression 1 or regression 2) Pilot Shop should adopt for forecasting total
shipping department costs and explain why.
2. If Pilot Shop projects that 600 orders will be lled the coming week, calculate the total shipping department costs using the regression you selected in requirement 1.
3. Explain two or three important limitations of the regression you selected in requirement 1, and identify
one or two ways to address the limitations. Specically include in your discussion the effect, if any, of
the global nature of Pilot Shops business.
(CMA Adapted)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

Appendix B: Regression
Analysis

The McGrawHill
Companies, 2008

Chapter 6

6-46

Cost Estimation 207

Analysis of Regression Results Rock n Roll Heaven is an outdoor pavilion that presents musical
performers throughout a six-month season, from late spring to early fall. Rock n Roll presents a
diverse venue of artists in a set of approximately 40 events each season. In order to better project its
costs and expected attendance, Rock n Roll uses regression analysis to project expected ticket sales
for upcoming events for each performer. The regression results shown below are derived from the
three most recent seasons. The dependent variable for Rock n Roll is the number of paying tickets
holders for each event, and the independent variables are
1. Whether or not this particular performer appeared at Rock n Roll previously (a dummy variable,
0 if no and 1 if yes).
2. The spending on advertising targeted to the performers appearance.
3. The performers local sales of CDs in the most recent year prior to their appearance.
4. The number of television appearances for the performer in the most recent year.
5. The number of public performances in the United States by the performer in the recent year.

Independent Variables
Regression intercept
Attendance at prior concert
Coefcient
t-value
Spending on advertising
Coefcient
t-value
Performers CD sales
Coefcient
t-value
Television appearances
Coefcient
t-value
Other public performances
Coefcient
t-value
R-squared
Standard error of the estimate

Results
1,224
3,445
4.11
0.113
1.88
0.00044
1.22
898
2.4
1,233
3.7
0.88
2,447

Required
1. Using the above regression, what attendance would be predicted for a performer who had appeared at
Rock n Roll previously, had six other public performances but no TV appearances, and had CD sales of
$10 million, and Rock n Roll planned to spend $35,000 on advertising?
2. Evaluate the precision and reliability of the regression results shown above. What changes, if any, do you
propose for the regression? Which variables should be deleted, and which do you think should be added,
and why?

6-47

Correlation Analysis; Appendix B PolyChem is a large manufacturer of packaging materials for

supermarkets and other retail applications; the packages are used by customers to carry away their
purchases. PolyChem has succeeded for many years by providing a high-quality product and superior customer service. Recently, additional competitors have entered the market, both local and
foreign, and PolyChem is nding that it must increasingly compete on price. PolyChems strategy
for dealing with the increased competition is to market its products to smaller retailers that would
appreciate the rms quality and service, as well as the rms ability to customize the product
adding different designs and colors to the packaging material. Until recently, the rm determined
product costs based on simple averages of materials purchases, plant labor, and overhead. The
rms management is now interested in improving, if possible, the accuracy of its cost information. As a start, Cheryl Greenberg, the management accountant, obtains the following sample of
data (Table 1) from the plant manager, showing the machine number, the order size (quantity, in
thousands), the machine setup time (in hours per unit; setup time also includes cleanup time after
the order is run), run time (the operating time for the machine to produce the order), and a measure

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

208 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 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).
Cheryl wants to run some regression analyses to better understand this data and, as a rst step,
obtains a correlation analysis which shows the simple correlation between each of the variables in
Table 1. The results are shown in Table 2. 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, as follows: (correlation between two variables)2 = the R-squared for simple regression analysis between
these two variables. To illustrate, note that the correlation between machine number and order
quantity = .33919. The R-squared for the regression between these two variables (with either as
the dependent variable) is (.33919)2=.1151. Cheryl also recalls that a negative correlation means
that the two variables are inversely relatedwhen one increases, the other decreases.
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

0.002
0.001
0.005
0.004
0.002
0.002
0.004
0.003
0.014
0.011
0.004
0.005
0.046
0.002
0.002
0.007
0.008
0.008
0.005

0.042
0.043
0.042
0.040
0.035
0.040
0.040
0.041
0.041
0.038
0.043
0.035
0.041
0.043
0.043
0.042
0.042
0.043
0.047

Correlation Results for PolyChems Plant Data

Number
Order size
Complexity
Setup time
Run time

Number

Order Size

Complexity

Setup Time

1
0.33919
0.071001
0.03805
0.346651

1
0.07095
0.20952
0.80882

1
0.4521388
0.140537

1
0.06534

Run Time

Note: Correlations with absolute value > .4 are statistically signicant at p < .10; correlations with absolute value > .5 are statistically signicant at
p < .05.

Required
1. Analyze the ndings in Table 2 and assess how, if at all, order size and complexity affect setup time and
run time. What other ndings in Table 2 are of particular interest?
2. How can your analysis in 1 above help PolyChem become more competitive?
(CMA Adapted)
6-48

Regression Analysis United States Motors Inc. (USMI) manufactures automobiles and light trucks
and distributes them for sale to consumers through franchised retail outlets. As part of the franchise
agreement, dealerships must provide monthly nancial statements following the USMI accounting

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

Cost Estimation 209

procedures manual. USMI has developed the following nancial prole of an average dealership that
sells 1,500 new vehicles annually.
AVERAGE DEALERSHIP FINANCIAL PROFILE
Composite Income Statement
Sales
Cost of goods sold

$30,000,000
24,750,000

Gross prot
Operating costs
Variable expenses

$ 5,250,000
862,500

Mixed expenses
Fixed expenses

2,300,000
1,854,000

Operating income

233,500

USMI is considering a major expansion of its dealership network. The vice president of marketing has asked Jack Snyder, corporate controller, to develop some measure of the risk associated with
the addition of these franchises. Jack estimates that 90 percent of the mixed expenses shown are variable for purposes of this analysis. He also suggested performing regression analyses on the various
components of the mixed expenses to more denitively determine their variability.
Required
1. Calculate the composite dealership prot if 2,000 units are sold.
2. Assume that regression analyses were performed on the separate components of the mixed expenses
and that a coefcient of determination value of .60 was determined as applicable to aggregate mixed
expenses over the relevant range.
a. Dene the term relevant range.
b. Explain the signicance of an R-squared value of .60 to USMIs analysis.
c. Describe the limitations that may exist in applying the composite-based relationships to specic new
dealerships that have been proposed.
e. Dene the standard error of the estimate.
3. 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 $4,500,000. If the projected annual sales
for a dealership total $28,500,000, determine the approximate 95 percent condence range for Jacks
prediction of sales.
4. What is the strategic role of regression analysis for USMI?
(CMA Adapted)
6-49

Cost Estimation, High-Low Method, Regression Analysis DVD Express is a large manufacturer of
affordable DVD players. Management recently became aware of rising costs resulting from returns
of malfunctioning products. As a starting point for further analysis, Bridget Forrester, the controller,
wants to test different forecasting methods and then use the best one to forecast quarterly expenses
for 2007. 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

$15,000
17,500
18,500
18,600

1
2
3
4

$16,200
17,800
18,800
17,700

1
2
3
4

$16,600
18,100
19,000
19,200

The result of a simple regression analysis using all 12 data points yielded an intercept of $16,559.09 and a
coefcient for the independent variable of $183.22. (R-squared = .27, t = 1.94, SE = 1128).
Required
1. Calculate the quarterly forecast for 2007 using the high-low method and regression analyses. Recommend which method Bridget should use and explain why.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

210 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

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, High-Low Method, Regression Analysis Clothes for U is a large merchandiser of
apparel for budget-minded families. Management recently became concerned about the amount of
inventory carrying costs and transportation costs between warehouses and retail outlets. As a starting point in further analyses, Gregory Gonzales, the controller, wants to test different forecasting
methods and then use the best one to forecast quarterly expenses for 2007. 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

$12,500
11,300
11,600
13,700
12,900
12,100
11,700
14,000
13,300
12,300
12,100
14,600

The results of a simple regression analysis using all 12 data points yielded an intercept of $11,854.55 and a
coefcient for the independent variable of $126.22 (R-squared = .19, t = 1.5, SE = 974).
Required
1. Calculate the quarterly forecasts for 2007 using the high-low method and regression analysis. Recommend which method Gregory should use and explain why.
2. 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, preWorld War II aircraft. They will be built completely from scratch to the exact specications used for the originals. As further authentication, the aircraft will be made using the technology
and manufacturing processes available when the originals were built. Each of the 16 will be own to
Air Force and aviation museums throughout the country for exhibition. Aviation enthusiasts can also
visit the production facility to see exactly how such aircraft were built in 1938.
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. The information includes some of the old cost
data from the builders of the original aircraft. The available information is for the total accumulated
time as the rst, eighth, and thirty-second aircraft, respectively, were completed.
Output

Total Hours

1
8
32

250
1,458
4,724

Required
1. If Soren Industries expects that the time spent per unit will be the same as it was in 1938, how many
hours will it take to build the 16 aircraft for the Air Force Museum Foundation?
2. What is the role of learning curves in Soren Industries business for contracts such as this?
652

Learning Curves Ben Matthews and David Everhart work for a landscaping company in Twin
Cities, Oklahoma. Their principal job is to lay railroad ties to line the sidewalks around apartment
complexes and to install ower boxes. The rst time Ben and David undertook one of these projects,

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

Appendix B: Regression
Analysis

The McGrawHill
Companies, 2008

Chapter 6

Cost Estimation 211

they spent 17 hours. Their goal by the end of the summer was to be able to nish an apartment complex in 8 hours, one working day. They performed eight of these jobs and had an 80 percent learning
curve. Assume that all apartment complexes are approximately the same size.
Required Did they reach their goal? If not, what would the learning rate have to have been for them to
have accomplished their goal?
6-53

Learning Curves Emotional Headdress (EH) is a Des Moines, Iowa, manufacturer of avant garde
hats and headwear. On March 11, 2007, the company purchased a new machine to aid in producing various established product lines. Production efciency on the new machine increases with the
workforce experience. It has been shown that as cumulative output on the new machine increases,
average labor time per unit decreases up to the production of at least 3,200 units. As EHs cumulative output doubles from a base of 100 units produced, the average labor time per unit declines by
15 percent. EHs production varies little from month to month and averages 800 hats per month.
Emotional Headdress has developed a new style of mens hat, the Morrisey, to be produced on
the new machine. One hundred Morrisey hats can be produced in a total of 25 labor-hours. All other
direct costs to produce each Morrisey hat are $16.25, excluding direct labor cost. EHs direct labor
cost per hour is $15. Fixed costs are $8,000 per month, and EH has the capacity to produce 3,200
hats per month.

Required
1. Emotional Headdress wishes to set the selling price for a Morrisey hat at 125 percent of the hat production cost. At the production level of 100 units, what is the selling price?
2. The company has received an order for 1,600 Morrisey hats from Smiths, Inc. Smiths is offering $20 for
each hat. Should the company accept Smiths order and produce the 1,600 hats? Explain.
6-54

Learning Curves Hauser Company, a family-owned business, engineers and manufactures a line of
mopeds and dirt bikes under the trade-name Trailite. The company has been in business for almost
20 years and has maintained a protable share of the recreational vehicle market due to its reputation for high-quality products. In addition, Hausers engineering department has kept the company
in the forefront by incorporating the latest technology in the Trailite bikes. Most subassembly work
for the bikes is subcontracted to reliable vendors. However, the nal assembly and inspection of all
products is performed at Hausers plant. Hauser recently developed a new braking system for the
Trailite Model-500 dirt bike. Because of the companys current availability of production capacity,
Jim Walsh, production manager, recommended that the rst lot of the new braking system be manufactured in-house rather than by subcontractors. This 80-unit production run has now been completed. The cumulative average labor-hours per unit for the braking system was 60 hours. Hausers
experience with similar products indicates that a learning curve of 80 percent is applicable and that
the learning factor can be expected to extend only through the fourth production run (80 per batch).
Hausers direct labor cost is $14.50 per direct labor-hour. Its management must decide whether to
continue producing the braking system in its own plant or to subcontract this work. Joyce Lane,
Hausers purchasing agent, has received a proposal from MACQ, a company specializing in component assembly. MACQ has done work in the past for Hauser and has proved to be efcient and
reliable. The terms of MACQs proposal are negotiable, and before beginning discussions with them,
Joyce has decided to conduct some relevant nancial analysis.

Required
1. Hauser Company has an immediate requirement for a total of 1,000 units of the braking system. Determine Hausers future direct labor costs to produce the required braking system units if it manufactures
the units in-house.
2. A consultant has advised Joyce that the learning rate for this application might be closer to 75 percent.
What is the effect on projected costs of using a 75 percent learning curve as opposed to an 80 percent
learning curve?
3. What conditions in a manufacturing plant, if present, would offset the potential benets of the learning
curve? What is the strategic role of learning curve analysis for Hauser Company?
(CMA Adapted)
6-55

Cost Estimation, Regression Analysis Plantcity is a large nursery and retail store specializing in
house and garden plants and supplies. Jean Raouth, the assistant manager, is in the process of budgeting monthly supplies expense for 2007. She assumes that in some way supplies expense is related

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

212 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

to sales, either in units or in dollars. She has collected these data for sales and supplies expenses for
June 2004 through December 2006, and has estimated sales for 2007:

Date

Supplies Expense

Sales Units

Jun 2004
Jul
Aug
Sep
Oct
Nov
Dec

$2,745
3,200
3,232
2,199
2,321
3,432
4,278

354
436
525
145
199
543
1,189

$2,009
2,190
2,878
1,856
2,168
2,152
2,463

Jan 2005
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

2,310
2,573
2,487
2,484
3,384
2,945
2,758
3,394
2,254
2,763
3,245
4,576

212
284
246
278
498
424
312
485
188
276
489
1,045

1,999
2,190
1,894
2,134
3,210
2,850
2,265
2,435
1,893
2,232
3,004
3,309

Jan 2006
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

2,103
2,056
3,874
2,784
2,345
2,912
2,093
2,873
2,563
2,384
2,476
3,364

104
167
298
398
187
334
264
333
143
245
232
1,122

2,195
2,045
2,301
2,345
1,815
2,094
1,934
2,054
1,977
1,857
2,189
3,433

180
230
190
450
350
350
450
550
300
300
450
950

$1,600
2,000
1,900
2,400
2,300
2,300
2,500
3,000
2,500
2,500
3,200
3,900

Jan 2007 (estimated)

Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

Sales Dollars

Required
1. Develop the regression that Jean should use based on these data and using the regression procedure in
Excel or an equivalent regression software program. Evaluate the reliability and precision of the regression you have chosen.
2. What are the predicted monthly gures 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. A major cost driver in
either type of facility is the capacity of the plant. For example, the capacity of a recycling plant is

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 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. These plants can
vary in size from a few hundred TPD to as many as several thousand TPD. 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. The dependent variable of the regression is the actual construction
cost of each project, while the independent variable is the TPD for the plant. Below is a sample of
some recent projects and the related construction costs (in thousands).

Commerce, CA
Hudson Falls, NY
Layton, UT
Oxford Township, NJ
Savannah, GA
Poughkeepsie, NY
Panama City, FL
Ronkonkoma, NY
Okahuma,FL
Spokane, WA
Arlington, VA
Camden, NJ
York, PA
Bridgeport, CT
Chester, PA

TPD

Cost

360
400
420
450
500
506
510
518
528
800
975
1,050
1,344
2,250
2,688

$ 59,369
77,013
50,405
75,779
87,439
57,463
60,730
84,457
88,119
152,902
127,021
163,395
139,302
344,852
448,073

Required
1. Develop a regression model using Excel or an equivalent system to predict the cost of a proposed new
plant in Babylon, New York, which will have a required capacity of 750 TPD. What is the predicted cost
for the Babylon plant, using your regression?
2. Evaluate the precision and reliability of the regression you have developed. How could it be improved?
6-57

Regression Analysis: Cross-Sectional Analysis; Calculation of a Regression Equation Jim

Manzano is the general partner of an investment group that owns a number of commercial and industrial properties, including a chain of 15 convenience stores located in the greater metropolitan
area of Cleveland, Ohio. Jim is concerned about the recent increase in inventory theft and waste
(he calls it spoilage) in his stores. Spoilage has increased by more than 20 percent in each of the
past two years. In some stores, the main reason is theft; in others, it is damage and vandalism; and
in still others, merchandise actually does spoil and must be thrown out. 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 oor space, number of employees, and total sales) and to the
location of the store (location 1 is an area where few arrests for theft, disorderly conduct, or vandalism are made, and location 3 is for areas with high arrests). Jim is not sure, but he suspects, based
on his experience managing convenience stores, that a relationship exists among these factors. A
colleague told him that a type of regression called cross-sectional regression would suit his needs.
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, different stores). 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. This approach is often
used in cases similar to Jims in which the accuracy or reasonableness of the reported dependent
variable is a concern. In effect, the cross-sectional regression develops a model that represents the
overall patterns in all the data, and the unusual stores will be identied by the largest error terms in
the regression. The following data are for the most recent months operations:

Store
Number
1
2
3
4
5

Inventory
Spoilage
$ 1,512
3,005
1,686
1,908
2,384

Square
Footage
2,400
3,900
3,200
3,400
3,750

Number of
Employees
8
10
12
12
9

Location

Sales

1
2
1
1
2

$ 312,389
346,235
376,465
345,723
453,983
(Continued)

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

214 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 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

4,806
2,253
1,443
3,755
1,023
1,552
2,119
5,506
3,034
772
$36,758

4,800
3,500
3,000
5,550
2,250
2,500
3,500
7,500
5,700
2,200
57,150

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
502,984
325,436
253,647
562,534
287,364
198,374
333,984
673,345
588,947
225,364
$5,786,774

Required
1. Using Excel or an equivalent software program, prepare a regression analysis that predicts inventory
spoilage at each of the 15 stores. Use any of the four potential independent variables (or a combination)
you think appropriate and explain your answer. Also evaluate the precision and reliability of the regression you select.
2. Using the regression equation you developed in requirement 1, determine which of the 15 stores might
have inventory spoilage that is out of line relative to the entire chain of stores. Explain your choice.
658

Regression Analysis in Tax Court Cases Since at least the late 1960s, the court systems in the
United States and elsewhere have accepted regression analysis as evidence in court cases. In many
instances, however, because of limitations or errors in developing the regression analysis, tax courts
question or deny the regression evidence. A study was performed recently to determine the factors
in the regression analysis that the court considered in determining whether regression evidence was
admissible.

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. Using the High-Low Method

Begin by graphing the data to determine whether there are any unusual (i.e., seasonal) patterns or outliers
in Exhibit 1.
The graph shows no unusual patterns or outliers, 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)
($201,783 $123,245)/(7,433 5,166) = $34.644 per delivery
To determine the intercept
$201,783 7,433 $34.644 = $55,726
$123,245 5,166 $34.644 = $55,726
The estimation equation is
Vehicle costs = $55,726 + $34.644 Number of deliveries/Month
Note that the intercept is a negative number, which simply means that the relevant range of 5,166 to 7,433
deliveries is so far from the zero point (where the intercept is) that the intercept cannot be properly interpreted as a xed cost. The estimation equation therefore is useful only within the relevant range of approximately 5,000 to 7,500 deliveries and should not be used to estimate costs outside that range.

2. Using Regression Analysis

All relevant criteria favor the rst regression based on higher R-squared and t-values and lower standard
error. Moreover, the sign on the materials cost variable in regression 2 is negative, which is difcult to
explain. This variable should have a direct relationship with overhead; thus, the sign of the variable should
be positive. The reason for the improvement of regression 1 over regression 2 might be that machine-hours

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

EXHIBIT 1

Cost Estimation 215

$225,000

Total Vehicle Expenses

Plot of Data for Hectors

Delivery Service

150,000

75,000
4,500

5,000

5,500

6,000

6,500

7,000

7,500

8,000

Total Deliveries

are highly correlated with either materials costs, labor-hours, or both, thus causing multicollinearity. By
excluding machine-hours as an independent variable, George reduced or removed the multicollinearity, and
the regression improved as a result. He should therefore use regression 1.

3. Using Both High-Low and Regression

Begin by graphing the data for the number of defective units, as shown in Exhibit 2. The objective is to
identify any unusual patterns that must be considered in developing an estimate.
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. Knowing that the production level also has been
increasing (as measured either by cost of sales, units produced, or units shipped), we now want to determine
whether the relationship between defects and production level (Exhibit 3) has changed.
We begin with units produced as the independent variable, since it should have the most direct relationship with defects; the other independent variables can be tried later. The second graph (Exhibit 3) makes
clear that a relationship exists between units produced and the number of defects.
The next step is to quantify this relationship with the high-low method and regression analysis. We begin
with the high-low analysis. For Exhibit 3, we identify February 2007 and December 2008 as representative
low and high periods, respectively.

EXHIBIT 2

3000

Defective Units from

January 2007 to December
2008

2500

Defects

2000

1500

1000

500

0
1

11

13
Months

15

17

19

21

23

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

216 Part Two

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Planning and Decision Making

EXHIBIT 3

3000

Defective Units vs.

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, but they produce a line that is representative of the data):
slope = (2,311 1,335)/(88 58) = 32.533
And
Intercept = 2,311 32.533 88 = 1,335 32.533 58 = 552
Thus, the estimation equation is
Number of defects = 552 + 32.533 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. Regression is thus performed to
provide a more precise estimate. Thus, the next step is to obtain a regression analysis from the previous data
and to assess the precision and reliability of the regression estimate. The regression can be completed with
a spreadsheet program or any of a number of available software systems. The results for three regression
analyses are presented in Exhibit 4. The dependent variable in each case is the number of defective units.
Regression 1 has the following independent variables: cost of sales, units shipped, and units produced.
R-squared and SE are OK, but we observe that all three t-values are less than 2.0, indicating unreliable
independent variables. Because we expect correlation among these variables and because of the low tvalues, we suspect multicollinearity among these variables. To reduce the effect of multicollinearity, we try
regression 2, which removes the variable units shipped, since that variable is likely to be least associated
with defective units and has among the lowest of the t-values. R-squared for regression 2 is essentially the
same as for regression 1, although SE improves very slightly, and the t-value for cost of sales is now OK.
The results of regression 3, with the cost of sales variable only, show that SE and the t-value improve again
while R-squared is unchanged. Because it has the best SE and t-values, and a very good R-squared, the third
regression is the best choice.

BlocherStoutCokinsChen:
Cost Management: A
Strategic Emphasis, Fourth
Edition

II. Planning and Decision

Making

The McGrawHill
Companies, 2008

Appendix B: Regression
Analysis

Chapter 6

EXHIBIT 4
Regressions for the Number
of Defects

Intercept

Coefcient of
Independent Variable

t-value for
Independent Variable

38.974 (units shipped)

2.849 (units produced)
4.702 (cost of sales)

.44
.38
.72

2.230 (units produced)

1.837 (cost of sales)

.309
4.54

Cost Estimation 217

R-squared

Standard Error of
the Estimate

.883

161

.881

158

.881

155

Regression 1
103.20

Regression 2
92.24

Regression 3
43.95
1.720 (cost of sales)

12.77