Abstract

Compared the use of unit regression weights with least squared error multiple regression weights. 100 sample correlation matrices computed from randomly drawn samples from a multivariate normal distribution were employed, with the sample size varying from 25-1,000 and number of variables from 2-10. In addition, correlation matrices were also sampled from published data. The results of comparing the efficiency of the 2 methods of weighting are presented for correlation matrices with and without the effect of suppressor variables. Use of these findings is discussed