Common Mistakes in Selecting Terms in Regression

When trying to find a regression model, it is usually desirable to include as few terms as possible while still giving a good model. Various procedures have been developed to help try to decide which explanatory variables can be dropped without important loss of information. However, these procedures are often used inappropriately. Here are some common mistakes that may occur in variable selection.

Assuming linearity is preserved when terms are dropped

Problems with stepwise model selection procedures