In the text I have recommended against the use of stepwise procedures for multiple regression, whereby we systematically hunt among the variables to predict some sort of optimal equation.

a) Explain why I would make such a recommendation.

b) How, then, could I justify asking you to do just that in Exercise 11.12?

Exercise 11.12

The data set Harass.dat, included on this book’s website, contains data on 343 cases created to replicate the results of a study of sexual harassment by Brooks and Perot (1991). The variables are, in order, Age, Marital Status (1 = married, 2 = not married), Feminist Ideology, Frequency of the Behavior, Offensiveness of the behavior, and the dependent variable, whether or not the subjects Reported incidents of sexual harassment (0 = no, 1 = yes). For each variable, higher numbers represent more of the property. Technically, this is a problem that might be better approached with what is called logistic regression, because the dependent variable is a dichotomy. However we can get a very close approximation to the optimal solution by using plain old linear multiple regression instead. Use multiple regression to predict whether or not sexual harassment will be reported, based on the variables you have here. Find a model that does not have many nonsignificant predictors.