ABSTRACT
The method of least squares and its generalizations have served statisticians well for many years. Most of the statistical computer packages are based on these theories, and thus will be used for many years to come and rightfully so. Statisticians should question, however, whether the assumptions underlying these theories are met in each application. Statisticians are doing this more often now with various procedures that involve examination of the residuals from the fitted model-whether the model is a one-sample estimation problem or a multiple linear regression problem. For example, if the residuals are skewed to the right, transformations like the square root and the logarithm can make the underlying distribution more symmetric and, in particular, more normal.
