Will anybody buy? Logistic regression - Improving the User Experience through Practical Data Analytics - page 271

CHAPTER

11

Will anybody buy?
Logistic regression

11.1 INTRODUCTION
Logistic regression is an extension of “regular” linear regression. It is used when the
dependent variable, Y, is categorical. We now introduce binary logistic regression, in
which the Y variable is a “Yes/No” type variable. We will typically refer to the two
categories of Y as “1” and “0,” so that they are represented numerically. However,
the two categories can be virtually anything, such as “adopted the search engine vs.
did not adopt the search engine” or “completed a task vs. did not complete a task”
or, in the world of general database marketing, “responded to an offer (i.e., made a
purchase) vs. did not respond to the offer.” In these situations, regular linear regres-
sion (whether simple or multiple) is not appropriate.
It should be noted that the goal of binary logistic regression is the same as in
the two previous chapters—to ind the best itting, simplest model, to understand
the relationship between the Y and the X's, and to be able to reach appropriate
statistical conclusions. Not only does binary logistic regression allow you to assess
how well your set of variables predicts your categorical dependent variable and
determine the “goodness-of-it” of your model as does regular linear regression,
but also it provides a summary of the accuracy of the classiication of cases, which
helps you determine the percent of predictions made from this model/equation that
will be correct.
But, once we account for the difference between regular linear regression,
which assumes that Y is continuous (and, for our purposes, bell-shaped), and logis-
tic regression, which assumes that the Y is a set of categories (only two categories,
if we are discussing binary logistic regression), the logistic regression process fol-
lows the same general principles as used in regular linear regression.
Binary logistic regression was used originally only in epidemiologic research,
but is now routinely used in many ields, including general business and marketing.
Its use has mushroomed in the past two decades. A classic early application was in
inance, where a “1” stood for “person paid back the loan,” and a “0” stood for “per-
son defaulted on the loan.”