ABSTRACT
This chapter focuses on variable selection techniques when a pool of predictor variables are available, some of which will collectively enable adequate prediction of the response variable. Included in the pool of potential predictor variables are any transformations of the raw predictor variables that are necessary. Selection of a subset of predictor variables for use in a prediction equation when there are many potential variables available for use is a common problem encountered in regression analysis. Often a data analyst collects observations on a large number of predictor variables when it is unknown which specific ones are most influential on a response. Variable selection and variable specification are two closely-related considerations in fitting regression models. One of the most comprehensive yet cumbersome ways of selecting subset regressions is to compute all possible subset regressions. Stepwise regression procedures are selection techniques that sequentially add or delete single predictor variables to the prediction equation.
