Prelude

Mail-order sales
company uses many different mailed catalogs to sell different types
A of merchandise. One catalog that features home goods, such as
bedspreads and pillows, was mailed to 200,000 people who were not
current customers.1 The response variable is whether or not the person
places an order. Logistic regression is used to model the probability p
of a purchase as a function of five explanatory variables. These are the
number of purchases within the last 24 months from a home gift catalog,
the proportion of single people in the zip code area based on census data,
the number of credit cards, a variable that distinguishes apartment dwellers
from those who live in single-family homes, and an indicator of whether or
not the customer has ever made a purchase from a similar type of catalog.
The fitted logistic model is then used to estimate the probability that a
large collection of potential customers will make a purchase. Catalogs are
sent to those whose estimated probability is above some cutoff value.

17-2
 CHAPTER
17
Logistic Regression*

17.1 The Logistic Regression

Model
17.2 Inference for Logistic
Regression
17.3 Multiple Logistic
Regression

* This chapter requires the material on binomial distributions

in the optional Chapter 5.
17-4 CHAPTER 17 䡲 Logistic Regression

Introduction
The simple and multiple linear regression methods we studied in Chapters 10
and 11 are used to model the relationship between a quantitative response
variable and one or more explanatory variables. In this chapter we describe
similar methods for use when the response variable has only two possible
values: customer buys or does not buy, patient lives or dies, candidate
accepts job or not.
In general, we call the two values of the response variable “success”
and “failure” and represent them by 1 (for a success) and 0. The mean is
then the proportion of ones, p ⳱ P(success). If our data are n independent
observations with the same p, this is the Binomial setting (page 319). What
is new in this chapter is that the data now include an explanatory variable
x. The probability p of a success depends on the value of x. For example,
suppose we are studying whether a customer makes a purchase (y ⳱ 1) or
not (y ⳱ 0) after being offered a discount. Then p is the probability that
the customer makes a purchase, and possible explanatory variables include
(a) whether the customer has made similar purchases in the past, (b) the
type of discount, and (c) the age of the customer. Note that the explanatory
variables can be either categorical or quantitative. Logistic regression2 is a
statistical method for describing these kinds of relationships.

17.1 The Logistic Regression Model

In general, the data for logistic regression are n independent observations,
each consisting of a value of the explanatory variable x and either a success
or a failure for that trial. For example, x may be the age of a customer,
and “success” means that this customer made a purchase. Every observation
may have a different value of x. To introduce logistic regression, however, it
is convenient to start with the special case in which the explanatory variable
x is also a yes-or-no variable. The data then contain a number of outcomes
(success or failure) for each of the two values of x. There are just two values
of p, one for each value of x. The count of successes for each value of x
has a Binomial distribution, so that we are on familiar ground. Here is an
example.
CASE 17.1

BINGE DRINKERS

Exercise 8.14 (page 517) describes a survey of 17,096 students in U.S.

four-year colleges. A student who reports drinking five or more drinks in a
row three or more times in the past two weeks is called a “frequent binge
drinker.” The researchers were interested in estimating the proportion of
students who are frequent binge drinkers.
 17.1 The Logistic Regression Model 17-5

One promising explanatory variable is the gender of the student. We

indicator variable express gender numerically using an indicator variable,

x⳱ 1
0 冦 if the student is a man
if the student is a woman
The sample contained 7180 men and 9916 women. The probability
that a randomly chosen student is a frequent binge drinker has two values,
p1 for men and p0 for women. The number of men in the sample who
are frequent binge drinkers has the Binomial distribution B(7180, p1 ). The
count of frequent binge drinkers among the women has the B(9916, p0 )
distribution.

Binomial distributions and odds

We begin with a review of some ideas associated with Binomial distributions.

EXAMPLE 17.1 Proportion of binge drinkers

In Chapter 8 we used sample proportions to estimate population proportions.

CASE 17.1
The binge-drinking study found that 1630 of the 7180 men in the sample were
frequent binge drinkers, as were 1684 of the 9916 women. Our estimates of the
two population proportions are
1630
Men: pˆ 1 ⳱ ⳱ 0.2270
7180
and
1684
Women: pˆ 0 ⳱ ⳱ 0.1698
9916
That is, we estimate that 22.7% of college men and 17.0% of college women are
frequent binge drinkers.

odds Logistic regression works with odds rather than proportions. The odds
are the ratio of the proportions for the two possible outcomes. If p is the
probability of a success, then 1 ⫺ p is the probability of a failure, and
p probability of success
ODDS ⳱ ⳱
1⫺p probability of failure
A similar formula for the sample odds is obtained by substituting pˆ for p in
this expression.

EXAMPLE 17.2 Odds of being a binge drinker

The proportion of frequent binge drinkers among the men in the study is
CASE 17.1

pˆ 1 ⳱ 0.227, so the proportion of men who are not frequent binge drinkers is

1 ⫺ pˆ 1 ⳱ 1 ⫺ 0.2270 ⳱ 0.7730
17-6 CHAPTER 17 䡲 Logistic Regression

The estimated odds of a male student being a frequent binge drinker are therefore
pˆ 1
ODDS ⳱
1 ⫺ pˆ 1
0.2270
⳱ ⳱ 0.2937
0.7730
For women, the odds are

pˆ 0
ODDS ⳱
1 ⫺ pˆ 0
0.1698
⳱ ⳱ 0.2045
1 ⫺ 0.1698

When people speak about odds, they often round to integers or fractions.
Since 0.205 is approximately 1/5, we could say that the odds that a female
college student is a frequent binge drinker are 1 to 5. In a similar way, we
could describe the odds that a college woman is not a frequent binge drinker
as 5 to 1.
17.1 Successful franchises and exclusive territories. In Case 9.1 (page 549) we
APPLY YOUR
studied data on the success of 170 franchise firms and whether or not the
KNOWLEDGE
owner of a franchise had an exclusive territory. Here are the data:

Observed numbers of firms

Exclusive territory
Success Yes No Total
Yes 108 15 123
No 34 13 47
Total 142 28 170

What proportion of the exclusive-territory firms are successful? Find the

proportion for the firms that do not have exclusive territories. Convert each
of these proportions to odds.
17.2 No Sweat labels and gender. In Example 9.2 (page 548) we examined the
relationship between gender and the use of clothing labels that indicated
that the garments were manufactured under fair working conditions. Here
are the data:

Gender
Label user Women Men Total
Yes 63 27 90
No 233 224 457
Total 296 251 547
 17.1 The Logistic Regression Model 17-7

Find the proportion of men who are label users; do the same for women.
Restate each of these proportions as odds.

Model for logistic regression

In Chapter 8 we learned how to compare the proportions of frequent
binge drinkers among college men and women using z tests and confidence
intervals. Logistic regression is another way to make this comparison, one
that extends to more general settings with a success-or-failure response
variable.
In simple linear regression we modeled the mean ␮ of the response
variable y as a linear function of the explanatory variable: ␮ ⳱ ␤0 Ⳮ ␤1 x.
When y is just 1 or 0 (success or failure), the mean is the probability p of
a success. Logistic regression models the mean p in terms of an explanatory
variable x. We might try to relate p and x as in simple linear regression:
p ⳱ ␤0 Ⳮ ␤1 x. Unfortunately, this is not a good model. Whenever ␤1 苷 0,
extreme values of x will give values of ␤0 Ⳮ ␤1 x that fall outside the range
of possible values of p, 0 ⱕ p ⱕ 1.
The logistic regression model removes this difficulty by working with the
log odds natural logarithm of the odds, p/(1 ⫺ p). We use the term log odds for this
transformation. As p moves from 0 to 1, the log odds moves through all
negative and positive numerical values. We model the log odds as a linear
function of the explanatory variable:

冢1 ⫺ p 冣 ⳱ ␤ Ⳮ ␤ x
p
log 0 1

Figure 17.1 graphs the relationship between p and x for some different
values of ␤0 and ␤1 . The logistic regression model uses natural logarithms.
Most calculators and statistical software systems have a built-in function for
the natural logarithm, often labeled “ln.”

p
1.0
0.9 β 0 = – 4.0 β 0 = – 8.0
0.8 β 1 = 2.0 β 1 = 1.6
0.7 β 0 = – 4.0
0.6 β1 = 1.8
0.5
0.4
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7 8 9 10
x

FIGURE 17.1 Plot of p versus x for selected values of ␤ 0 and ␤ 1 .

17-8 CHAPTER 17 䡲 Logistic Regression

Returning to the binge-drinking study, we have for men

log(ODDS) ⳱ log(0.2937) ⳱ ⫺1.23

and for women

log(ODDS) ⳱ log(0.2045) ⳱ ⫺1.59

Verify these results with your calculator, remembering that “log” is the
natural logarithm. Here is a summary of the logistic regression model.

LOGISTIC REGRESSION MODEL

The statistical model for logistic regression is

冢1 ⫺ p 冣 ⳱ ␤ Ⳮ ␤ x
p
log 0 1

where p is a Binomial proportion and x is the explanatory variable.

The parameters of the logistic model are ␤0 and ␤1 .

17.3 Log odds for exclusive territories. Refer to Exercise 17.1. Find the log odds
APPLY YOUR
for the franchises that have exclusive territories. Do the same for the firms
KNOWLEDGE
that do not.
17.4 Log odds for No Sweat labels. Refer to Exercise 17.2. Find the log odds for
the women. Do the same for the men.

Fitting and interpreting the logistic regression model

We must now fit the logistic regression model to data. In general, the
data consist of n observations on the explanatory variable x, each with a
success-or-failure response. Our binge-drinking example has an indicator (0
or 1) explanatory variable. Logistic regression with an indicator explanatory
variable is a special case but is important in practice. We use this special case
to understand a little more about the model.

EXAMPLE 17.3 Logistic model for binge drinking

In the binge-drinking example, there are n ⳱ 17, 096 observations. The
CASE 17.1

explanatory variable is gender, which we coded using an indicator variable with

values x ⳱ 1 for men and x ⳱ 0 for women. There are 7180 observations with
x ⳱ 1 and 9916 observations with x ⳱ 0. The response variable is also an
indicator variable: y ⳱ 1 if the student is a frequent binge drinker and y ⳱ 0 if
not. The model says that the probability p that a student is a frequent binge
drinker depends on the student’s gender (x ⳱ 1 or x ⳱ 0). There are two
possible values for p, say p1 for men and p0 for women. The model says that
 17.1 The Logistic Regression Model 17-9

for men

冢1 ⫺ p 冣 ⳱ ␤ Ⳮ ␤
p1
log 0 1
1

and for women

冢1 ⫺ p 冣 ⳱ ␤
p0
log 0
0

Note that there is a ␤1 term in the equation for men because x ⳱ 1, but it is
missing in the equation for women because x ⳱ 0.

In general, the calculations needed to find estimates b0 and b1 for the

parameters ␤0 and ␤1 are complex and require the use of software. When the
explanatory variable has only two possible values, however, we can easily
find the estimates. This simple framework also provides a setting where we
can learn what the logistic regression parameters mean.

EXAMPLE 17.4 Parameter estimates for binge drinking

For the binge-drinking example, we found the log odds for men,

CASE 17.1
log
冢 pˆ 1
1 ⫺ pˆ 1 冣
⳱ ⫺1.23

and for women,

log
冢 pˆ 0
1 ⫺ pˆ 0 冣
⳱ ⫺1.59

To find estimates b0 and b1 of the model parameters ␤0 and ␤1 , we match the

male and female model equations in Example 17.3 with the corresponding data
equations. Because

log 冢1 ⫺ p 冣 ⳱ ␤
p0
0
0 and log
冢 pˆ 0
1 ⫺ pˆ 0 冣
⳱ ⫺1.59

the estimate b0 of the intercept is simply the log(ODDS) for the women,
b0 ⳱ ⫺1.59

Similarly, the estimated slope is the difference between the log(ODDS) for the men
and the log(ODDS) for the women,
b1 ⳱ ⫺1.23 ⫺ (⫺1.59) ⳱ 0.36

The fitted logistic regression model is

log(ODDS) ⳱ ⫺1.59 Ⳮ 0.36x
17-10 CHAPTER 17 䡲 Logistic Regression

The slope in this logistic regression model is the difference between the
log(ODDS) for men and the log(ODDS) for women. Most people are not
comfortable thinking in the log(ODDS) scale, so interpretation of the results
in terms of the regression slope is difficult. Usually, we apply a transformation
to help us. The exponential function (e x key on your calculator) reverses the
natural logarithm transformation. That is, continuing Example 17.4,

ODDS ⳱ e⫺1.58Ⳮ0.36x ⳱ e⫺1.59 ⫻ e0.36x

From this, the ratio of the odds for men (x ⳱ 1) and women (x ⳱ 0) is

ODDSmen
⳱ e0.36 ⳱ 1.43
ODDSwomen

The transformation e0.36 undoes the logarithm and transforms the logistic
odds ratio regression slope into an odds ratio, in this case, the ratio of the odds that a
man is a frequent binge drinker to the odds that a woman is a frequent binge
drinker. We can multiply the odds for women by the odds ratio to obtain
the odds for men:

ODDSmen ⳱ 1.43 ⫻ ODDSwomen

The odds for men are 1.43 times the odds for women.
Notice that we have chosen the coding for the indicator variable so that
the regression slope is positive. This will give an odds ratio that is greater
than 1. Had we coded women as 1 and men as 0, the signs of the slope would
be reversed, the fitted equation would be log(ODDS) ⳱ ⫺1.23 ⫺ 0.36x, and
the odds ratio would be e⫺0.36 ⳱ 0.70. The odds for women are 70% of the
odds for men.
It is of course often the case that the explanatory variable is quantitative
rather than an indicator variable. We must then use software to fit the
logistic regression model. Here is an example.

EXAMPLE 17.5 Is the quality of the product acceptable?

The CHEESE data set described in the Data Appendix includes a response variable
called “Taste” that is a measure of the quality of the cheese in the opinions of
several tasters. For this example, we will classify the cheese as acceptable (y ⳱ 1) if
Taste ⱖ 37 and unacceptable (y ⳱ 0) if Taste ⬍ 37. This is our response variable.
The data set contains three explanatory variables: “Acetic,” “H2S,” and “Lactic.”
Let’s use Acetic as the explanatory variable. The model is

冢1 ⫺ p 冣 ⳱ ␤ Ⳮ ␤ x
p
log 0 1

where p is the probability that the cheese is acceptable and x is the value of Acetic.
The model for estimated log odds fitted by software is

log(ODDS) ⳱ b0 Ⳮ b1 x ⳱ ⫺13.71 Ⳮ 2.25x

 17.1 The Logistic Regression Model 17-11

The odds ratio is eb1 ⳱ 9.49. This means that if we increase the acetic acid content
x by one unit, we increase the odds that the cheese will be acceptable by about 9.5
times. (See Exercise 17.7.)

17.5 Fitted model for exclusive territories. Refer to Exercises 17.1 and 17.3.
APPLY YOUR
Find the estimates b0 and b1 and give the fitted logistic model. What is
KNOWLEDGE
the odds ratio for exclusive territory (x ⳱ 1) versus no exclusive territory
(x ⳱ 0)?
17.6 Fitted model for No Sweat labels. Refer to Exercises 17.2 and 17.4. Find
the estimates b0 and b1 and give the fitted logistic model. What is the odds
ratio for women (x ⳱ 1) versus men (x ⳱ 0)?
17.7 Interpreting an odds ratio. If we apply the exponential function to the fitted
model in Example 17.5, we get
ODDS ⳱ e⫺13.71Ⳮ2.25x ⳱ e⫺13.71 ⫻ e2.25x
Show that for any value of the quantitative explanatory variable x, the odds
ratio for increasing x by 1,
ODDSxⳭ1
ODDSx
is e2.25 ⳱ 9.49. This justifies the interpretation given at the end of
Example 17.5.

SECTION 17.1 SUMMARY

䡵 Logistic regression explains a success-or-failure response variable in
terms of an explanatory variable.
䡵 If p is a proportion of successes, then the odds of a success are p/(1 ⫺ p),
the ratio of the proportion of successes to the proportion of failures.
䡵 The logistic regression model relates the proportion of successes in the
population to an explanatory variable x through the logarithm of the odds
of a success:

冢 冣
p
log ⳱ ␤0 Ⳮ ␤1 x
1⫺p
That is, each value of x gives a different proportion p of successes. The
data are n values of x, with observed success or failure for each. The
model assumes that these n success-or-failure trials are independent, with
probabilities of success given by the logistic regression equation. The
parameters of the model are ␤0 and ␤1 .
䡵 The odds ratio is the ratio of the odds of a success at x Ⳮ 1 to the odds
of a success at x. It is found as e␤1 , where ␤1 is the slope in the logistic
regression equation.
䡵 Software fits the data to the model, producing estimates b0 and b1 of the
parameters ␤0 and ␤1 .
17-12 CHAPTER 17 䡲 Logistic Regression

17.2 Inference for Logistic Regression

Statistical inference for logistic regression with one explanatory variable
is similar to statistical inference for simple linear regression. We calculate
estimates of the model parameters and standard errors for these estimates.
Confidence intervals are formed in the usual way, but we use standard
Normal zⴱ -values rather than critical values from the t distributions. The
ratio of the estimate to the standard error is the basis for hypothesis tests.
Software often reports the test statistics as the squares of these ratios, in
which case the P-values are obtained from the chi-square distribution with
1 degree of freedom.

CONFIDENCE INTERVALS AND SIGNIFICANCE TESTS

FOR LOGISTIC REGRESSION
An approximate level C confidence interval for the slope ␤1 in the
logistic regression model is
b1 ⫾ zⴱ SEb1
The ratio of the odds for a value of the explanatory variable equal to
x Ⳮ 1 to the odds for a value of the explanatory variable equal to x is
the odds ratio e␤1 . A level C confidence interval for the odds ratio is
obtained by transforming the confidence interval for the slope,

(eb1 ⫺z SEb1 , eb1Ⳮz SEb1 )

ⴱ ⴱ

In these expressions zⴱ is the standard Normal critical value with

area C between ⫺zⴱ and zⴱ .
To test the hypothesis H0: ␤1 ⳱ 0, compute the test statistic
2
2
X ⳱
b1
冢 冣
SEb1

In terms of a random variable ␹ 2 having the ␹ 2 distribution with 1

degree of freedom, the P-value for a test of H0 against Ha: ␤1 苷 0 is
approximately P(␹ 2 ⱖ X2 ).

We have expressed the null hypothesis in terms of the slope ␤1 because

this form closely resembles the hypothesis of “no linear relationship” in
simple linear regression. In many applications, however, the results are
expressed in terms of the odds ratio. A slope of 0 is the same as an odds ratio
of 1, so we often express the null hypothesis of interest as “the odds ratio is
1.” This means that the two odds are equal and the explanatory variable is
not useful for predicting the odds.
 17.2 Inference for Logistic Regression 17-13

EXAMPLE 17.6 Computer output for binge drinking

Figure 17.2 gives the output from SPSS and SAS for the binge-drinking study. The

CASE 17.1
parameter estimates given by SPSS are b0 ⳱ ⫺1.587 and b1 ⳱ 0.362, more exact
than we calculated directly in Example 17.4. The standard errors are 0.027 and
0.039. A 95% confidence interval for the slope is

b1 ⫾ zⴱ SEb1 ⳱ 0.362 ⫾ (1.96)(0.039)

⳱ 0.362 ⫾ 0.076
We are 95% confident that the slope is between 0.286 and 0.438. The SPSS output
provides the odds ratio 1.435 (under the heading “Exp(B)”) but does not give the
confidence interval for e␤1 . This is easy to compute from the interval for the
slope ␤1 :
b1 ⫺zⴱ SEb b1Ⳮzⴱ SEb
(e 1 , e 1 ) ⳱ (e0.286 , e0.438 )
⳱ (1.33, 1.55)
We conclude, “College men are more likely to be frequent binge drinkers than
college women (odds ratio ⳱ 1.44, 95% CI ⳱ 1.33 to 1.55).”

It is standard to use 95% confidence intervals, and software often reports

these intervals. A 95% confidence interval for the odds ratio also provides a

SPSS

Variables in the Equation

B S.E. Wald df Sig. Exp(B)
Step 1 GENDERM 0.362 0.039 86.611 1 0.000 1.435
Constant –1.587 0.027 3520.069 1 0.000 0.205
a Variable(s) entered on step 1: GENDERM

SAS
The LOGISTIC Procedure
Analysis of Maximum Likelihood Estimates
Standard Wald
Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept 1 –1.5868 0.0267 3520.3120 < 0.001

genderm 1 0.3617 0.0388 86.6811 < 0.001

Odds Ratio Estimates

Point 95% Wald
Effect Estimate Confidence Limits

genderm 1.436 1.330 1.549

FIGURE 17.2 Logistic regression output from SPSS and SAS for the
binge-drinking data, for Example 17.6.
17-14 CHAPTER 17 䡲 Logistic Regression

test of the null hypothesis that the odds ratio is 1 at the 5% significance level.
If the confidence interval does not include 1, we reject H0 and conclude that
the odds for the two groups are different; if the interval does include 1, the
data do not provide enough evidence to distinguish the groups in this way.

CASE 17.1
17.8 Read the output. Examine the SAS output in Figure 17.2. Report
APPLY YOUR
the estimates of ␤0 and ␤1 with the standard errors as given in
KNOWLEDGE
this display. Also report the odds ratio with its 95% confidence
interval as given in this output.
17.9 Inference for exclusive territories. Use software to run a logistic regression
analysis for the exclusive territory data of Exercise 17.1. Summarize the
results of the inference.
17.10 Inference for No Sweat labels. Use software to run the logistic regression
analysis for the No Sweat label data of Exercise 17.2. Summarize the results
of the inference.

Examples of logistic regression analyses

The following example is typical of many applications of logistic regres-
sion. It concerns a designed experiment with five different values for the
explanatory variable.

EXAMPLE 17.7 Effectiveness of an insecticide

An experiment was designed to examine how well the insecticide rotenone kills an
aphid, called Macrosiphoniella sanborni, that feeds on the chrysanthemum plant.3
The explanatory variable is the concentration (in log of milligrams per liter) of the
insecticide. About 50 aphids were exposed to each of five concentrations. Each
insect was either killed or not killed. Although logistic regression uses the log odds
of the proportion killed, software requires that we enter just the concentration, the
number of insects, and the number killed. Here are the data, along with the results
of some calculations:

Concentration Number Number Proportion

x (log scale) of insects killed killed pˆ log odds

0.96 50 6 0.1200 ⫺1.9924

1.33 48 16 0.3333 ⫺0.6931
1.63 46 24 0.5217 0.0870
2.04 49 42 0.8571 1.7918
2.32 50 44 0.8800 1.9924

Least-squares regression of log odds on log concentration gives the fit illustrated in

Figure 17.3. There is a clear linear relationship. The logistic regression fit for the
proportion killed appears in Figure 17.4. It is a transformed version of Figure 17.3
with the fit calculated using the logistic model rather than by least-squares.
 17.2 Inference for Logistic Regression 17-15

Log odds of percent killed

1

–1

–2
0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4
Log concentration

FIGURE 17.3 Plot of log odds of percent killed versus log concentration
for the insecticide data, for Example 17.7.

100
90
80
70

Percent killed
60
50
40
30
20
10
0
0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4
Log concentration

FIGURE 17.4 Plot of the percent killed versus log concentration with the
logistic fit for the insecticide data, for Example 17.7.
17-16 CHAPTER 17 䡲 Logistic Regression

When the explanatory variable has several values, we can use graphs
like those in Figures 17.3 and 17.4 to visually assess whether the logistic
regression model seems appropriate. Just as a scatterplot of y versus x in
simple linear regression should show a linear pattern, a plot of log odds
versus x in logistic regression should be close to linear. Just as in simple
linear regression, outliers in the x direction should be avoided because they
may overly influence the fitted model.
The graphs strongly suggest that insecticide concentration affects the kill
rate in a way that fits the logistic regression model. Is the effect statistically
significant? Suppose that rotenone has no ability to kill Macrosiphoniella
sanborni. What is the chance that we would observe experimental results
at least as convincing as what we observed if this supposition were
true? The answer is the P-value for the test of the null hypothesis that
the logistic regression slope is zero. If this P-value is not small, our
graph may be misleading. As usual, we must add inference to our data
analysis.

EXAMPLE 17.8 Does concentration affect the kill rate?

Figure 17.5 gives the output from SPSS, SAS, and Minitab for logistic regression
analysis of the insecticide data. The model is

冢1 ⫺ p 冣 ⳱ ␤ Ⳮ ␤ x
p
log 0 1

where the values of the explanatory variable x are 0.96, 1.33, 1.63, 2.04, 2.32.
From the SPSS output we see that the fitted model is
log(ODDS) ⳱ b0 Ⳮ b1 x ⳱ ⫺4.89 Ⳮ 3.11x
or
pˆ
⳱ e⫺4.89Ⳮ3.11x
1 ⫺ pˆ

Figure 17.4 is a graph of the fitted pˆ given by this equation against x, along
with the data used to fit the model. SPSS gives the statistic X2 under the heading
“Wald.” The null hypothesis that ␤1 ⳱ 0 is clearly rejected (X2 ⳱ 64.23,
P ⬍ 0.001).
The estimated odds ratio is 22.39. An increase of one unit in the log
concentration of insecticide (x) is associated with a 22-fold increase in the odds
that an insect will be killed. The confidence interval for the odds is given in the SAS
output: (10.470, 47.896).
Remember that the test of the null hypothesis that the slope is 0 is the same
as the test of the null hypothesis that the odds are 1. If we were reporting the
results in terms of the odds, we could say, “The odds of killing an insect increase
by a factor of 22.3 for each unit increase in the log concentration of insecticide
(X2 ⳱ 64.23, P ⬍ 0.001; 95% CI = 10.5 to 47.9).”

17.11 Find the 95% confidence interval for the slope. Using the information in the
APPLY YOUR
output of Figure 17.5, find a 95% confidence interval for ␤1 .
KNOWLEDGE
 17.2 Inference for Logistic Regression 17-17

SPSS

Variables in the Equation

B S.E. Wald df Sig.
Exp(B) 95.0% C.I. for EXP(B)
Lower Upper
LCONC 3.109 0.388 64.233 1 0.000 22.394 10.470 47.896
Constant –4.892 0.643 57.961 1 0.000 0.008

SAS
The LOGISTIC Procedure S.E.
Analysis of Maximum Likelihood Estimates
0.388
0.643 Standard Wald
Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept 1 –4.8923 0.6426 57.9606 < 0.001

lconc 1 3.1088 0.3879 64.2332 < 0.001

Odds Ratio Estimates

Point 95% Wald
Effect Estimate Confidence Limits

lconc 22.394 10.470 47.896

Minitab
Logistic Regression Table
Odds 95% CI
Predictor Coef StDev Z P Ratio Lower Upper
Constant –4.8923 0.6426 –7.61 0.000
lconc 3.1088 0.3879 8.01 0.000 22.39 10.47 47.90

FIGURE 17.5 Logistic regression output from SPSS, SAS, and Minitab for the insecticide
data, for Example 17.8.

17.12 Find the 95% confidence interval for the odds ratio. Using the estimate b1
and its standard error, find the 95% confidence interval for the odds
ratio and verify that this agrees with the interval given by SAS.
17.13 X 2 or z. The Minitab output in Figure 17.5 does not give the value of X2 .
The column labeled “Z” provides similar information.
(a) Find the value under the heading “Z” for the predictor lconc. Verify that
Z is simply the estimated coefficient divided by its standard error. This
is a z statistic that has approximately the standard Normal distribution
if the null hypothesis (slope 0) is true.
(b) Show that the square of z is X2 . The two-sided P-value for z is the same
as P for X2 .

In Example 17.5 we studied the problem of predicting whether or not

the taste of cheese was acceptable using Acetic as the explanatory variable.
We now revisit this example to include the results of inference.
17-18 CHAPTER 17 䡲 Logistic Regression

Logistic Regression Table

Odds 95% CI
Predictor Coef StDev Z P Ratio Lower Upper
Constant –13.705 5.932 –2.31 0.021
acetic 2.249 1.027 2.19 0.029 9.48 1.27 70.96

FIGURE 17.6 Logistic regression output from Minitab for the cheese data with Acetic as
the explanatory variable, for Example 17.9.

EXAMPLE 17.9 Predicting acceptable taste in cheese

Figure 17.6 gives the output from Minitab for a logistic regression analysis using
Acetic as the explanatory variable. The fitted model is
log(ODDS) ⳱ b0 Ⳮ b1 x ⳱ ⫺13.705 Ⳮ 2.249x
This agrees up to rounding with the result reported in Example 17.5.
From the output we see that because P ⳱ 0.029, we can reject the null
hypothesis that the slope ␤1 ⳱ 0. The value of the test statistic is z ⳱ 2.19,
calculated from the estimate b1 ⳱ 2.249 and its standard error SEb1 ⳱ 1.027.
Minitab reports the odds ratio as 9.48, with a 95% confidence interval of
(1.27, 70.96).

We estimate that increasing the acetic acid content of the cheese by one
unit will increase the odds that the cheese will be acceptable by about 9
times. The data, however, do not give us a very accurate estimate. The
odds ratio could be as small as a little more than 1 or as large as 71
with 95% confidence. We have evidence to conclude that cheeses with
higher concentrations of acetic acid are more likely to be acceptable, but
establishing the true relationship accurately would require more data.

SECTION 17.2 SUMMARY

䡵 Software fits the data to the model, producing estimates b0 and b1 of the
parameters ␤0 and ␤1 . Software also produces standard errors for these
estimates.
䡵 A level C confidence interval for the intercept ␤0 is
b0 ⫾ zⴱ SEb0

A level C confidence interval for the slope ␤1 is

b1 ⫾ zⴱ SEb1

A level C confidence interval for the odds ratio e␤1 is obtained by

transforming the confidence interval for the slope,

(eb1 ⫺z SEb1 , eb1Ⳮz SEb1 )

ⴱ ⴱ
 17.3 Multiple Logistic Regression 17-19

In these expressions zⴱ is the standard Normal critical value with area

C between ⫺zⴱ and zⴱ .
䡵 The null hypothesis that x does not help predict p in the logistic
regression model is H0: ␤1 ⳱ 0 or H0: e␤1 ⳱ 1 in terms of the odds ratio.
To test this hypothesis, compute the test statistic
2
2
X ⳱
b1
SEb1冢 冣
䡵 In terms of a random variable ␹ 2 having a ␹ 2 distribution with 1
degree of freedom, the P-value for a test of H0 against Ha: ␤1 苷 0 is
approximately P(␹ 2 ⱖ X2 ).

17.3 Multiple Logistic Regressionⴱ

The CHEESE data set includes three explanatory variables: Acetic, H2S,
and Lactic. Example 17.9 examines the model where Acetic alone is used
to predict the odds that the cheese is acceptable. Do the other explanatory
variables contain additional information that will give us a better prediction?
multiple logistic We use multiple logistic regression to answer this question. Generating the
regression computer output is easy, just as it was when we generalized simple linear
regression with one explanatory variable to multiple linear regression with
more than one explanatory variable in Chapter 11. The statistical concepts
are similar, although the computations are more complex. Here is the
analysis.

EXAMPLE 17.10 Multiple logistic regression

As in Example 17.9, we predict the odds that the cheese is acceptable. The
explanatory variables are Acetic, H2S, and Lactic. Figure 17.7 gives the output.
From the Minitab output, we see that the fitted model is

log(ODDS) ⳱ b0 Ⳮ b1 Acetic Ⳮ b2 H2S Ⳮ b3 Lactic

⳱ ⫺14.26 Ⳮ 0.584 Acetic Ⳮ 0.685 H2S Ⳮ 3.47 Lactic
When analyzing data using multiple regression, we first examine the hypothesis
that all of the regression coefficients for the explanatory variables are zero. We do
the same for logistic regression. The hypothesis
H0: ␤1 ⳱ ␤2 ⳱ ␤3 ⳱ 0
is tested by a chi-square statistic with 3 degrees of freedom. This is given in the last
line of the output. Minitab calls the statistic “G.” The value is G ⳱ 16.33, with

*This optional section uses ideas from Chapter 11.

17-20 CHAPTER 17 䡲 Logistic Regression

SPSS
Omnibus Tests of Model Coefficients
Chi-square df Sig.
Model 16.334 3 0.001

Variables in the Equation

B S.E. Wald df Sig. Exp(B) 95.0% C.I. for EXP(B)
Lower Upper
ACETIC 0.584 1.544 0.143 1 0.705 1.794 0.087 37.001
H2S 0.685 0.404 2.873 1 0.090 1.983 0.898 4.379
LACTIC 3.468 2.650 1.713 1 0.191 32.084 0.178 5776.637
Constant –14.260 8.287 2.961 1 0.085 0.000

SAS
Testing Global Null Hypothesis: BETA = 0
Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 16.3344 3 0.0010

S.E.
Analysis of Maximum Likelihood Estimates
0.388
0.643 Standard Wald
Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept 1 –14.2604 8.2869 2.9613 0.0853

acetic 1 0.5845 1.5442 0.1433 0.7051
h2s 1 0.6848 0.4040 2.8730 0.0901
lactic 1 3.4684 2.6497 1.7135 0.1905

Odds Ratio Estimates

Point 95% Wald
Effect Estimate Confidence Limits

acetic 1.794 0.087 37.004

h2s 1.983 0.898 4.379
lactic 32.086 0.178 >999.999

Minitab
Logistic Regression Table
Odds 95% CI
Predictor Coef StDev Z P Ratio Lower Upper
Constant –14.260 8.287 –1.72 0.085
acetic 0.584 1.544 0.38 0.705 1.79 0.09 37.01
h2s 0.6849 0.4040 1.69 0.090 1.98 0.90 4.38
lactic 3.468 2.650 1.31 0.191 32.09 0.18 5777.85

Log-Likelihood = –9.230
Test that all slopes are zero: G = 16.334, DF = 3, P-Value = 0.001

FIGURE 17.7 Logistic regression output from SPSS, SAS, and Minitab for the cheese data with
Acetic, H2S, and Lactic as the explanatory variables, for Example 17.10.
 17.3 Multiple Logistic Regression 17-21

3 degrees of freedom. The P-value is 0.001. We reject H0 and conclude that one or
more of the explanatory variables can be used to predict the odds that the cheese is
acceptable.
Next, examine the coefficients for each variable and the tests that each of these
is 0 in a model that contains the other two. The P-values are 0.71, 0.09, and 0.19.
None of the null hypotheses, H0: ␤1 ⳱ 0, H0: ␤2 ⳱ 0, and H0: ␤3 ⳱ 0, can be
rejected. That is, none of the three explanatory variables adds significant predictive
ability once the other two are already in the model.

Our initial multiple logistic regression analysis told us that the explana-
tory variables contain information that is useful for predicting whether or
not the cheese is acceptable. Because the explanatory variables are correlated,
however, we cannot clearly distinguish which variables or combinations of
variables are important. Further analysis of these data using subsets of the
three explanatory variables is needed to clarify the situation. We leave this
work for the exercises.

SECTION 17.3 SUMMARY

䡵 In multiple logistic regression the response variable has two possible
values, as in logistic regression, but there can be several explanatory
variables.
䡵 As in multiple regression there is an overall test for all of the explanatory

variables. The null hyphothesis that the coefficients for all of the
explanatory variables are zero is tested by a statistic that has a distribution
that is approximately ␹ 2 with degrees of freedom equal to the number of
explanatory variables. The P-value is approximately P(␹ 2 ⱖ X2 ).
䡵 Hypotheses about individual coefficients, H0 : ␤j ⳱ 0 or H1 : e␤j ⳱ 1, in
terms of the odds ratio, are tested by a statistic that is approximately ␹ 2
with 1 degree of freedom. The P-value is approximately P(␹ 2 ⱖ X2 ).

STATISTICS IN SUMMARY
Logistic regression is much like linear regression. We use one or more
explanatory variables to predict a response in both instances. For logistic
regression, however, the response variable has only two possible values.
The statistical inference issues are quite similar. We can test the significance
of an explanatory variable and give confidence intervals for its coefficient
in the model. When there are several explanatory variables, we can test
the null hypothesis that all of their coefficients are zero and the null
hypothesis that a single coefficient is zero when the other variables are
in the model. Here are the skills you should develop from studying this
chapter.
17-22 CHAPTER 17 䡲 Logistic Regression

A. PRELIMINARIES
1. Recognize the logistic regression setting: a success-or-failure response
variable and a straight-line relationship between the log odds of a
success and an explanatory variable x.
2. If the data contain several observations with the same or similar values
of the explanatory variable, compute the proportion of successes, the
odds of a success, and the log odds for each different value. Plot
the log odds versus the explanatory variable. The relationship should
be approximately linear.

B. INFERENCE USING SOFTWARE

1. Recognize which type of inference you need in a particular logistic
regression setting.
2. Explain in any specific logistic regression setting the meaning of the
slope ␤1 in the logistic regression model and of the odds ratio e␤1 .
3. Understand software output for logistic regression. Find in the output
estimates of all parameters in the model and their standard errors.
4. Carry out tests and calculate confidence intervals for the logistic
regression slope.
5. Estimate the odds ratio and give a 95% confidence interval for this
quantity.

CHAPTER 17 REVIEW EXERCISES

17.14 Holiday purchases. A poll of 811 adults aged 18 or older asked about
purchases that they intended to make for the upcoming holiday season.4
One of the questions asked about what kind of gift they intended to buy for
the person on whom they intended to spend the most. Clothing was the first
choice of 487 people.
(a) What proportion of adults said that clothing was their first choice?
(b) What are the odds that an adult will say that clothing is his or her first
choice?
(c) What proportion of adults said that something other than clothing was
their first choice?
(d) What are the odds that an adult will say that something other than
clothing is his or her first choice?
(e) How are your answers to parts (a) and (d) related?
17.15 Stock options. Different kinds of companies compensate their key employ-
ees in different ways. Established companies may pay higher salaries, while
new companies may offer stock options that will be valuable if the company
succeeds. Do high-tech companies tend to offer stock options more often
than other companies? One study looked at a random sample of 200 com-
panies. Of these, 91 were listed in the Directory of Public High Technology
Corporations, and 109 were not listed. Treat these two groups as SRSs
of high-tech and non-high-tech companies. Seventy-three of the high-tech
 Chapter 17 Review Exercises 17-23

companies and 75 of the non-high-tech companies offered incentive stock

options to key employees.5
(a) What proportion of the high-tech companies offer stock options to their
key employees? What are the odds?
(b) What proportion of the non-high-tech companies offer stock options to
their key employees? What are the odds?
(c) Find the odds ratio using the odds for the high-tech companies in the
numerator. Interpret the result in a few sentences.
17.16 Log odds for high-tech and non-high-tech firms. Refer to the previous
exercise.
(a) Find the log odds for the high-tech firms. Do the same for the non-high-
tech firms.
(b) Define an explanatory variable x to have the value 1 for high-tech firms
and 0 for non-high-tech firms. For the logistic model, we set the log
odds equal to ␤0 Ⳮ ␤1 x. Find the estimates b0 and b1 for the parameters
␤0 and ␤1 .
(c) Show that the odds ratio is equal to eb1 .
17.17 Do the inference. Refer to the previous exercise. Software gives 0.3347 for
the standard error of b1 .
(a) Find the 95% confidence interval for ␤1 .
(b) Transform your interval in (a) to a 95% confidence interval for the odds
ratio.
(c) What do you conclude?
17.18 Suppose you had twice as many data. Refer to Exercises 17.15 to 17.17.
Repeat the calculations assuming that you have twice as many observations
with the same proportions. In other words, assume that there are 182
high-tech firms and 218 non-high-tech firms. The numbers of firms offering
stock options are 146 for the high-tech group and 150 for the non-high-tech
group. The standard error of b1 for this scenario is 0.2366. Summarize your
results, paying particular attention to what remains the same and what is
different from what you found in Exercises 17.15 to 17.17.
17.19 Bicycle accidents and alcohol. In the United States approximately 900 people
die in bicycle accidents each year. One study examined the records of 1711
bicyclists aged 15 or older who were fatally injured in bicycle accidents
between 1987 and 1991 and were tested for alcohol. Of these, 542 tested
positive for alcohol (blood alcohol concentration of 0.01% or higher).6
(a) What proportion of the bicyclists tested positive for alcohol?
(b) What are the odds that a fatally injured bicyclist will test positive for
alcohol?
(c) What proportion of the bicyclists did not test positive for alcohol?
(d) What are the odds that a fatally injured bicyclist will not test positive
for alcohol?
(e) How are your answers to parts (a) and (d) related?

17.20 Healthy companies versus failed companies. Case 7.2 (page 476) compared
the mean ratio of current assets to current liabilities of 68 healthy firms
17-24 CHAPTER 17 䡲 Logistic Regression

with the mean ratio for 33 firms that failed. Here we analyze the same
data with a logistic regression. The outcome is whether or not the firm
is successful, and the explanatory variable is the ratio of current assets to
current liabilities. Here is the output from Minitab:

Logistic Regression Table

Odds 95% CI
Predictor Coef StDev Z P Ratio Lower Upper
Constant –2.6293 0.6950 –3.78 0.000
ratio 2.6865 0.5632 4.77 0.000 14.68 4.87 44.27

(a) Give the fitted equation for the log odds that a firm will be successful.
(b) Describe the results of the significance test for the coefficient of the ratio
of current assets to current liabilities.
(c) The odds ratio is the estimated amount that the odds of being successful
would increase when the current assets to current liabilities ratio is
increased by one unit. Report this odds ratio with the 95% confidence
interval.
(d) Write a short summary of this analysis and compare it with the analysis
of these data that we performed in Chapter 7. Which approach do you
prefer?
17.21 Blood pressure and cardiovascular disease. There is much evidence that high
blood pressure is associated with increased risk of death from cardiovascular
disease. A major study of this association examined 3338 men with high
blood pressure and 2676 men with low blood pressure. During the period
of the study, 21 men in the low-blood-pressure and 55 in the high-blood-
pressure group died from cardiovascular disease.
(a) Find the proportion of men who died from cardiovascular disease in the
high-blood-pressure group. Then calculate the odds.
(b) Do the same for the low-blood-pressure group.
(c) Now calculate the odds ratio with the odds for the high-blood-pressure
group in the numerator. Describe the result in words.
17.22 Do the inference and summarize the results. Refer to the previous exercise.
Computer output for a logistic regression analysis of these data gives the
estimated slope b1 ⳱ 0.7505 with standard error SEb1 ⳱ 0.2578.
(a) Give a 95% confidence interval for the slope.
(b) Calculate the X2 statistic for testing the null hypothesis that the slope is
zero and use Table F to find an approximate P-value.
(c) Write a short summary of the results and conclusions.
17.23 Transform to the odds. The results describing the relationship between
blood pressure and cardiovascular disease are given in terms of the change
in log odds in the previous exercise.
(a) Transform the slope to the odds and the 95% confidence interval for
the slope to a 95% confidence interval for the odds.
(b) Write a conclusion using the odds to describe the results.

17.24 Do syntax textbooks have gender bias? To what extent do syntax textbooks,
which analyze the structure of sentences, illustrate gender bias? A study of
 Chapter 17 Review Exercises 17-25

this question sampled sentences from 10 texts.7 One part of the study
examined the use of the words “girl,” “boy,” “man,” and “woman.” We
will call the first two words juvenile and the last two adult. Here are data
from one of the texts:

Gender n X (juvenile)
Female 60 48
Male 132 52

(a) Find the proportion of the female references that are juvenile. Then
transform this proportion to odds.
(b) Do the same for the male references.
(c) What is the odds ratio for comparing the female references to the male
references? (Put the female odds in the numerator.)
17.25 Do the inference and summarize the results. The data from the study of
gender bias in syntax textbooks given in the previous exercise are analyzed
using logistic regression. The estimated slope is b1 ⳱ 1.8171 and its standard
error is SEb1 ⳱ 0.3686.
(a) Give a 95% confidence interval for the slope.
(b) Calculate the X2 statistic for testing the null hypothesis that the slope is
zero and use Table F to find an approximate P-value.
(c) Write a short summary of the results and conclusions.
17.26 Transform to the odds. The gender bias in syntax textbooks is described in
the log odds scale in the previous exercise.
(a) Transform the slope to the odds and the 95% confidence interval for
the slope to a 95% confidence interval for the odds.
(b) Write a conclusion using the odds to describe the results.
17.27 Analysis of a reduction in force. To meet competition or cope with economic
slowdowns, corporations sometimes undertake a “reduction in force” (RIF),
where substantial numbers of employees are terminated. Federal and various
state laws require that employees be treated equally regardless of their age.
In particular, employees over the age of 40 years are in a “protected” class,
and many allegations of discrimination focus on comparing employees over
40 with their younger coworkers. Here are the data for a recent RIF:

Over 40
Terminated No Yes
Yes 7 41
No 504 765

(a) Write the logistic regression model for this problem using the log odds
of a termination as the response variable and an indicator for over and
under 40 years of age as the explanatory variable.
17-26 CHAPTER 17 䡲 Logistic Regression

(b) Explain the assumption concerning Binomial distributions in terms of

the variables in this exercise. To what extent do you think that these
assumptions are reasonable?
(c) Software gives the estimated slope b1 ⳱ 1.3504 and its standard error
SEb1 ⳱ 0.4130. Transform the results to the odds scale. Summarize the
results and write a short conclusion.
(d) If additional explanatory variables were available, for example, a per-
formance evaluation, how would you use this information to study the
RIF?
17.28 Analysis of the time to repair golf clubs. The Ping Company makes custom-
built golf clubs and competes in the $4 billion golf equipment industry. To
improve its business processes, Ping decided to seek ISO 9001 certification.8
As part of this process, a study of the time it took to repair golf clubs sent to
the company by mail determined that 16% of orders were sent back to the
customers in 5 days or less. Ping examined the processing of repair orders
and made changes. Following the changes, 90% of orders were completed
within 5 days. Assume that each of the estimated percents is based on a
random sample of 200 orders. Use logistic regression to examine how the
odds that an order will be filled in 5 days or less has improved. Write a short
report summarizing your results.
17.29 Know your customers. To devise effective marketing strategies it is helpful to
know the characteristics of your customers. A study compared demographic
characteristics of people who use the Internet for travel arrangements and
of people who do not.9 Of 1132 Internet users, 643 had completed college.
Among the 852 nonusers, 349 had completed college. Model the log odds
of using the Internet to make travel arrangements with an indicator variable
for having completed college as the explanatory variable. Summarize your
findings.
17.30 Does income relate to use of the Internet? The study mentioned in the
previous exercise also asked about income. Among Internet users, 493
reported income of less than $50,000 and 378 reported income of $50,000
or more. (Not everyone answered the income question.) The corresponding
numbers for nonusers were 477 and 200. Repeat the analysis using an
indicator variable for income of $50,000 or more as the explanatory variable.
What do you conclude?
17.31 Gender, alcohol, and fatal bicycle accidents. A study of alcohol use and
deaths due to bicycle accidents collected data on a large number of fatal
accidents.10 For each of these, the individual who died was classified accord-
ing to whether or not there was a positive test for alcohol and by gender.
Here are the data:

Gender n X (tested positive)

Female 191 27
Male 1520 515

Use logistic regression to study the question of whether or not gender is

related to alcohol use in people who are fatally injured in bicycle accidents.
 Chapter 17 Case Study Exercises 17-27

CHAPTER 17 CASE STUDY EXERCISES

CASE STUDY 17.1: Predict whether or not the product will be acceptable. In
Examples 17.5, 17.9, and 17.10 we analyzed data from the CHEESE data set
described in the Data Appendix. In Examples 17.5 and 17.9, we used Acetic as
the explanatory variable. Use each of the other two explanatory variables alone
in a logistic regression to predict acceptable taste and summarize the results.
With three explanatory variables, there are three different logistic regressions that
include two of these. Run these models as well, and summarize the results. Write a
report.

CASE STUDY 17.2: Predict whether or not the GPA will be 3.0 or better. In
Case 11.2 we used multiple regression methods to predict grade point average using
the CSDATA data set described in the Data Appendix. The explanatory variables
were the two SAT scores and three high school grade variables. Let’s define success
as earning a GPA of 3.0 or better. So, we define an indicator variable, say HIGPA,
to be 1 if the GPA is 3.0 or better and 0 otherwise. Examine logistic regression
models for predicting HIGPA using the two SAT scores and three high school grade
variables. Summarize all of your results and compare them with what we found
using multiple regression to predict the GPA.

CASE STUDY 17.3: Analyze a Simpson’s paradox data set. In Exercise 2.85 (page
153) we studied an example of Simpson’s paradox, the reversal of the direction
of a comparison or an association when data from several groups are combined
to form a single group. The data concerned two hospitals, A and B, and whether
or not patients undergoing surgery died or survived. Here are the data for all
patients:

Hospital A Hospital B
Died 63 16
Survived 2037 784
Total 2100 800

And here are the more detailed data where the patients are categorized as being in
good condition or poor condition before surgery:

Good Condition Poor Condition

Hospital A Hospital B Hospital A Hospital B
Died 6 8 Died 57 8
Survived 594 592 Survived 1443 192
Total 600 600 Total 1500 200
17-28 CHAPTER 17 䡲 Logistic Regression

Use a logistic regression to model the odds of death with hospital as the explanatory
variable. Summarize the results of your analysis and give a 95% confidence interval
for the odds ratio of Hospital A relative to Hospital B. Then rerun your analysis using
the hospital and the condition of the patient as explanatory variables. Summarize
the results of your analysis and give a 95% confidence interval for the odds ratio of
Hospital A relative to Hospital B. Write a report explaining Simpson’s paradox in
terms of the results of your analyses.

CASE STUDY 17.4: Compare the homes in two zip codes. The HOMES data set
described in the Data Appendix gives data on the selling prices and characteristics
of homes in several zip codes. In Case 11.3 we used multiple regression models to
predict the prices of homes in zip code 47904. For this case study we will look at
only the homes in zip codes 47904 and 47906. Define a response variable that has
the value 0 for the homes in 47904 and 1 for the homes in 47906. Prepare numerical
and graphical summaries that describe the prices and characteristics of homes in
these two zip codes. Then explore logistic regression models to predict whether or
not a home is in 47906. Summarize your results in a report.

Notes for Chapter 17

1. This example describes an analysis done by Christine Smiley of Kestenbaum Con-
sulting Company, Chicago, Illinois.
2. Logistic regression models for the general case where there are more than two possible
values for the response variable have been developed. These are considerably more
complicated and are beyond the scope of our present study. For more information
on logistic regression, see A. Agresti, An Introduction to Categorical Data Analysis,
Wiley, 1996; and D. W. Hosmer and S. Lemeshow, Applied Logistic Regression,
Wiley, 1989.
3. This example is taken from a classical text written by a contemporary of R. A. Fisher,
the person who developed many of the fundamental ideas of statistical inference that
we use today. The reference is D. J. Finney, Probit Analysis, Cambridge University
Press, 1947. Although not included in the analysis, it is important to note that the
experiment included a control group that received no insecticide. No aphids died in
this group. We have chosen to call the response “dead.” In the text the category is
described as “apparently dead, moribund, or so badly affected as to be unable to
walk more than a few steps.” This is an early example of the need to make careful
judgments when defining variables to be used in a statistical analysis. An insect
that is “unable to walk more than a few steps” is unlikely to eat very much of a
chrysanthemum plant!

4. The poll is part of the American Express Retail Index Project and is reported in
Stores, December 2000, pp. 38–40.
5. Based on Greg Clinch, “Employee compensation and firms’ research and develop-
ment activity,” Journal of Accounting Research, 29 (1991), pp. 59–78.
6. Data from Guohua Li and Susan P. Baker, “Alcohol in fatally injured bicyclists,”
Accident Analysis and Prevention, 26 (1994), pp. 543–548.
7. From Monica Macaulay and Colleen Brice, “Don’t touch my projectile: gen-
der bias and stereotyping in syntactic examples,” Language, 73, no. 4 (1997),
pp. 798–825.
 Notes for Chapter 17 17-29

8. Based on Robert T. Driescher, “A quality swing with Ping,” Quality Progress,

August 2001, pp. 37–41.

9. From Karin Weber and Weley S. Roehl, “Profiling people searching for and purchas-
ing travel products on the World Wide Web,” Journal of Travel Research, 37 (1999),
pp. 291–298.

10. See Note 6.