Missing Variables and Two-Stage Least-Squares Estimation from More than One Data Set

In a situation when no single sample inc1udes all the endogenous variables of a simultaneous equation model but there are two (or more) non-overlapping samples and each variable is included in at least one, then it is possible to pool the data and estimate the model consistently by a two-stage least-squares procedure. The asymptotic variances of the estimates are not always larger than those which would have been obtained with TSLS from one complete sample. It is also shown that under certain assumptions the same approach can be applied to an ordinary regression model.