Will anybody buy? Logistic regression - Improving the User Experience through Practical Data Analytics - page 297

• a person who is of the opinion that having Strong Credibility is more impactful in

making a decision about his/her likelihood of paying for an online news experience
is more likely, indeed, to be willing to pay for an online news experience; and

• a person who is of the opinion that having In-depth Analysis of Business Stories

is more impactful in making a decision about his/her likelihood of paying for an
online news experience is more likely, indeed, to be willing to pay for an online
news experience.

The other 8 of the 13 X variables are not signiicant, although two are close: age
( p -value = 0.062) and national ( p -value = 0.058).
Of course, you're very excited that the model fared well, and that so many of your
variables have an impact on its success.
But, before blurting out these indings to anyone who will listen, you need to do one
more thing—we need to perform a stepwise (binary logistic) regression . Why? As we
noted in Chapter 10, it is possible that two (or rarely, but possibly, more than two) of
the nonsigniicant variables in the multiple regression are highly overlapping, masking
the real signiicance of both of them. The stepwise regression will reveal if this is the
case and provide a more accurate accounting of which variables are really signiicant.

11.5.1 STEPWISE REGRESSION ANALYSIS OF THE
CHARLESTONGLOBE.COM DATA
To perform stepwise regression, we have to return to a previous step, when we chose
the “Method” dropdown in SPSS. This is shown in Figure 11.17 (see oval in Figure
11.17 ). As you can see, there are various options available.
The option that is most similar to the “Stepwise Regression” command of Chapter
10 is “Forward: LR.” This is the option you should use. The “LR” stands for “Like-
lihood Ratio,” a term involved in the process of using the “maximum likelihood”
criterion as discussed earlier in the sidebar on page 275.
After we click on “Forward: LR” and then “OK,” you get the stepwise output
shown in Figure 11.18 .
Let's begin with the “Variables in the Equation” section at the bottom of the out-
put. You can see in the third (bottom) section that there were ive steps. (To brush up
on stepwise regression, refer back to Chapter 10.) As we noted in discussing stepwise
regression in Chapter 10, where it was irst introduced, it is only the last step that
really matters. In this case, it is step 5. In step 5, there are still ive variables, same
as in the earlier multiple regression. But there are major differences in the output.
First, the Unique Local Content Not Available on a Local Newspaper Site has
dropped out of the regression model. What do we make of that? Well, regression
analysis (linear or logistic) is a complex technique, with different patterns of over-
lap when different sets of variables are in the equation. As a consequence, different
p -values and different benchmarks have to be overcome to be signiicant at 0.05.
(If we forced it to enter as a sixth variable, its p -value would be 0.113, which is, of
course, not less than 0.05, which is why it did not enter; still, 0.113 is not considered
that much above 0.05.)