In this paper it is argued that when the population regression coeffic
ient is of interest, the use of sampling weights can be desirable in r
egression models with complex survey data. A two-step ML estimator is
proposed as an alternative to OLS and weighted least squares. Specific
ation tests are given. The ML estimator does well in simulations, incl
uding several cases where it is based on a misspecified model. The spe
cification tests are effective in selecting the best estimator. As an
example, the methods are used to estimate the returns to education usi
ng data from the Canadian Survey of Consumer Finances. (C) 1998 Elsevi
er Science S.A. All rights reserved.