ABSTRACT
Screening data prior to formal regression analysis can eliminate costly errors that the formal analysis might not even detect. One of the easiest methods of spotting errors in collecting or recording data is to visually scan the data base. This chapter discusses the transformations of response and predictor variables. More critical and unrealistic is the frequent use of the fitted model to describe physical phenomena as though it unerringly defined a functional expression relating the response and predictor variables. Predictor variables can erroneously be eliminated from fitted models because they are not properly defined. A variable that is intrinsically nonnumerical is referred to as a categorical or qualitative variable. Data smoothing techniques aid in specifying the form of response or predictor variables by reducing the variability of plotted observations and enabling trends to be more clearly recognized. Interaction terms in a regression model are products of two or more predictor variables.
