APA Dictionary of Psychology

a procedure used to estimate regression coefficients and other parameters when error variances are correlated or show heterogeneity of variance such that a conventional ordinary least squares regression procedure will produce biased results.