A CRITIQUE OF THE STATISTICAL ANALYSIS OF THE OEO EXPERIMENT IN PERFORMANCE CONTRACTING

The experimental students in the OEO experiment were generally lower scoring and less affluent than control students. OEO's statistical analysis used a variety of statistical procedures to adjust for the initial differences in the experimental groups. A reanalysis of the data using a chi‐square technique demonstrated that the adjusted posttest differences between the experimental groups were biased against the performance contractors because the differences were not independent of the pretest means. These biases are discussed in terms of differential growth rates, specification error, and errors of measurement. The random assignment of schools to the experimental and control conditions would have avoided many of these problems.