Can you relate in multiple ways? Multiple linear regression and stepwise regression - Improving the User Experience through Practical Data Analytics - page 239

CHAPTER

10

Can you relate in
multiple ways? Multiple
linear regression and
stepwise regression

10.1 INTRODUCTION
When there is more than one independent variable (X), we call the regression analy-
sis by the term “multiple regression.” With one X, we referred to the regression
analysis in the previous chapter as “simple regression.” Using Excel or SPSS, we
generate the best-itting (least-squares) line, 1 as we did in the previous chapter. This
line would be, say with seven X's:

Yc=a+b1*X1+b2*X2+b3*X3+⋯+b7*X7 .
Finding this line by hand would be prohibitive, so statistical software must be
used unless you are a mathematics prodigy, and even then we would use the soft-
ware!! (For simple regression, statistical software should be used, but doing it by
hand with a calculator would be feasible, although not preferred, if the number of
data points were not exceedingly large.)

10.2 CASE STUDY: DETERMINING THE IDEAL SEARCH
ENGINE AT BEHEMOTH.COM
Let's return to our favorite employment Web site, Behemoth.com, where you were
hired as usability researcher. Behemoth is one of the largest job search engines in
the world, with over a million job openings at any time, over 1 million resumes in
the database, and over 63 million job seekers per month. Behemoth has over 5000
employees spread around the world.
You'll recall from Chapter 4 that you had to deliver the hard, cold facts about
“Novix” to your creative director, Hans Blitz: despite spending $80 million on a new
search engine, your current one still performed better than the new one in a head-to-
head usability test.
You'll also recall from Chapter 9 the UX Director Hans Blitz challenged you to
prove to him that Behemoth spent 80 million getting rid a feature that clients actually
wanted: the ability to perform a Boolean search.

1 Actually, in multidimensions, it would not be, literally, a “line.” It would be a plane or hyperplane.
However, for simplicity, we shall continue to call it the least-squares line .