A Comparison of Covariance Estimators for Complete and Incomplete Panel Data Models.

A fixed effects panel model with randomly missing data is considered. The incomplete data are characterized as a set of blocks of complete data. A covariance (CV) estimator for the incomplete data is developed. It is a linear combination of the CB estimators of each of the complete blocks, with weights inversely proportional to their respective variances. It is efficient relative to several possible complete data CV estimators which can be used on the truncated data. Efficiency comparisons are also made between the selection of complete data CV estimators. The asymptotic behavior of the incomplete data CV estimator is investigated using two possible characterizations of a large sample; first, the number of individuals increasing in each block of fixed time dimension, and second, the number of blocks increasing with the time and cross-section dimensions fixed in each block. The incomplete data CV estimator is extended to allow for fixed time effects. Copyright 1991 by Blackwell Publishing Ltd