Page 120 - Statistics II for Dummies

104
Part II: Using Different Types of Regression to Make Predictions
Getting a Kick out of Estimating

Punt Distance

Before you jump into a model selection procedure to predict y by using a set
of x variables, you have to do some legwork. The variable of interest is y, and
that’s a given. But where do the x variables come from? How do you choose
which ones to investigate as being possible candidates for predicting y? And
how do those possible x variables interact with each other toward making
that prediction?

You must answer all these questions before using any model selection pro-
cedure. However, this part is the most challenging and the most fun; a com-
puter can’t think up x variables for you!

Suppose you’re at a football game and the opposing team has to punt the
ball. You see the punter line up and get ready to kick the ball, and some
questions come to you: “Gee, I wonder how far this punt will go? I wonder
what factors influence the distance of a punt? Can I use those factors in a
multiple regression model to try to estimate punt distance? Hmm, I think
I’ll consult my Statistics II For Dummies book on this and analyze some data
during halftime. . . .”

Well, maybe that’s pushing it, but it’s still an interesting line of question-
ing for football players, golfers, soccer players, and even baseball players.
Everyone’s looking for more distance and a way to get it.

In the following sections, you can see how to identify and assess different x
variables in terms of their potential contribution to predicting y.

Brainstorming variables
and collecting data

Starting with a blank slate and trying to think of a set of x variables that may
be related to y may sound like a daunting task, but in reality, it’s probably
not as bad as you think. Most researchers who are interested in predicting
some variable y in the first place have some ideas about which variables may
be related to it. After you come up with a set of logical possibilities for x, you
collect data on those variables, as well as on y, to see what their actual rela-
tionship with y may be.

7/23/09 9:27:03 PM
11_466469-ch06.indd 104 7/23/09 9:27:03 PM
11_466469-ch06.indd 104