Estimating the Mediating Effect of Intervening Variables in Pooled Cross-Sectional and Time Series Designs

Abstract

An important contribution of evaluation research is information regarding the extent to which the effect of an intervention is mediated by characteristics of the unit in which it is introduced and/or variations in the way the intervention is implemented in different units. It is shown in this article that by extending the type of model specified by Berk and his associates (1979), this type of information can be obtained from pooled cross-sectional and time series designs. However, the conventional estimation procedure for pooled cross-sectional and time series data, such as that embedded in PROC TSCSREG in SAS, is not appropriate for this purpose because it produces inefficient estimates of the coefficients which are of central interest in this type of problem, that is, those coefficients that characterize the effect of the intervention and the extent to which it is mediated by characteristics of the unit in which it is introduced and/or variations in the way the intervention is implemented. Two alternative estimation procedures are developed that result in efficient estimates of the coefficients that are of central interest in this type of problem.