Identification of individual aberrations in least squares regression

Abstract

In single equation LS regression the common practice is. to test goodness-of- fit by the standard error o f estimate s and probable absence, of residual autoregression by the Durbin-Watson d, or the more recent count Of sign changes r. With a wide choice of causative (or independent) variables (indvars) and With access to a computer, a multitude of regressions can be produced, one for each set o f indvars selected. We usually pick the regression with the lowest s and a satisfactory d or r as the 'best', unless there are very compelling a priori reasons for picking some other set. Truth to say, there is still much empiricism in regression practice; in it art has a place as well as science.