Dorsett, Dovalee et T. Webster, J., Guidelines for Variable Selection Problems When Dummy Variables are Used, American statistician , 37(4A), 1983, pp. 337-339
When dummy variables are used to incorporate categorical information into a regression equation, it is standard procedure to use one less dummy variable than the number of categories considered.This work illustrates the pitfall of this approach when the forward selection procedure is used to identify categories that differ from a norm.In addition, the relation between zero residuals and potential outliers is demonstrated for regression studies with dummy variables.