Estimation and hypothesis testing for seemingly unrelated regressions: A sociological application

Sociologists frequently use ordinary least squares (OLS) to estimate a series of regression equations from data on the same observational entities. Such "seemingly unrelated regressions" are linked by correlations among the disturbances. In this paper we review three techniques for estimating "seemingly related regressions"-OLS, Zellner's generalized least-squares method, and maximum likelihood estimation-and present anl illustrative sociological example employing each technique. We discuss the conditions under which the non-OLS estimation procedures offer advantages for efficiency of estimation and hypothesis testing.