Files
Abstract
A complex survey design often necessitates the use of some approxima— tion technique when calculating regression standard errors. Two appro— ximation procedures, the Taylor expansion method (TEM) and balanced repeated replication (BRR) technique, were compared to ordinary least squares (OLS) and weighted least squares (WLS). While the regression coefficients were similar across all procedures, TEM and BRR produced larger standard error estimates than did OLS and WLS.