Quasi-Maximum Likelihood Estimation with Complex Survey Data,” Unpublished mimeo

Large-scale surveys such as the Current Population Survey, the Panel Study of Income Dynamics, and the Health and Retirement Study use complex sample designs. The resulting data violate the familiar i.i.d. assumption, but existing econometric practice ignores this fact. As a result, researchers often use inconsistent parameter estimators and, almost without exception, they inappropriately estimate their estimator’s covariance matrices and therefore carry out potentially misleading hypothesis tests. This paper explains how the population means, the population linear regression coefficients, and other population parameter values can be consistently estimated in the context of the complex survey design commonly used in large-scale surveys, and it derives the asymptotic distributions of the estimators. It also explains how to achieve consistent estimation of the estimators ’ covariance matrices and how to perform appropriate statistical inference