Estimation of Linear Models with Incomplete Data

In estimating linear models, it is often necessary to combine data from two or more samples or subsamples, each with a somewhat different set of variables. Conventional methods for doing this are statistically inefficient or have unknown statistical properties. This paper describes a maximum likelihood method that is both consistent and efficient and that can be implemented with LISREL. Unlike other maximum likelihood algorithms, this method can estimate overidentified models, and it produces consistent estimates of standard errors. Using this approach, one can