Adjusting for Confounding Factors in Quasi-Experiments: Another Reanalysis of the Westinghouse Head Start Evaluation.

Evaluations of social programs based upon quasi-experimental designs are typically plagued by problems of nonequivalence between the experimental and comparison group prior to the experiment. In such settings it is extremely difficult, if not impossible, to isolate the effects of the program from the confounding effects associated with the relevant preexisting differences between the groups. A classic occurrence of the problem was in the 1969 large-scale quasiexperimental evaluation of the Westinghouse Head Start program, their negative findings having been used to justify phasing out the summer programs. A portion of the data using Sorbom's statistical adjustment is reanalyzed. This approach improves upon previous analyses in the following ways: (1) it recognizes that the Head Start and comparison groups are separate and distinct populations; (2) it offers a statistical test of the null hypothesis that the two groups are equal on a latent factor called the socio-economic advantage; and (3) a goodness of fit statistic providing an overall test of the assumptions of the model indicates that the model fits the data better than any previous model. Results do not support the strong inferences drawn by the original evaluators. (Author/RL)