Abstract: This paper is concerned with the estimation of correlated response variance from unbalanced data in complex surveys. The Minimum Variance Quadratic Unbiased Estimator (MINQUE) proposed by Rao (1971, 1972) is often used for variance components estimations. However, for unbalanced data the equations for obtaining the MINQU estimates are very difficult to solve. In this paper we propose a constrained MINQU estimator by substituting robust unbiased estimates for the random residual errors in the MINQUE equations to obtain estimates of the correlated response variance. We demonstrate through numerical and empirical studies that the constrained MINQUE is efficient compared to the full MINQU estimator. We also point out that the constrained MINQU estimator can be used in other cases where reducing the complexity of the MINQUE equations is desired.
Key words and phrases: Correlatedresponsevariance, MinimumVari anceQuadratic Unbiased Estimator (MINQUE).