Graphical Displays for Alternate Regression Fits

Abstract

Regression is probably the single most widely used technique of data analysis. Multiple linear regression, where y is fitted by Σc _j x _j , is its most widely used special case. Techniques of robust regression, where fitting is to be effective in a more or less wide variety of uncomfortable circumstances, including the too-frequent occurence of large deviations, have been explored over the last few decades. Attention may also be needed to: (a) variability that depends systematically upon y and the x’s; (b) re-expression (transformation) of either y or one or more x’s; (c) data that can be divided into two or more portions, each of which can be much better fitted separately than together.

Prepared in connection with research at Princeton University sponsored by the Army Research Office (Durham), DAAL03-88-K-0045.