Beta weights, unit weights, and other weights.

Abstract

A simple example of the use of different weights with just two predictors was presented. The replacement of the beta weights by unit weights resulted in a very slight loss of predictive validity, and even the arbitrary reversal of the beta weights resulted in a smaller loss than might have been expected. A relationship demonstrated by Green was used to help explain these results. In many situations, it appears worthwhile to follow the approach taken here to determine how great a loss of predictive validity would result from the use of unit weights in place of beta weights.