Experiments Versus Quasi- Experiments: Do They Yield the Same Answer?

Life would be ever so much easier if quasi-experiments yielded just as good causal inferences as randomized experiments. Of course, the term "quasi-experiment " covers a multitude of designs. Here it refers to the workhorse of the quasi-experimental design literature: the nonequivalent control group design that includes a treatment group, a control group not receiving treatment, and a posttest for both, but where the assignment of subjects to conditions is not controlled by the researcher, and is certainly not random. The latter comparison to randomized experiments is generally of most interest. For the assessment of treatment outcome, randomized experiments are widely acknowledged to have many important advantages. Most salient, statistical theory suggests that randomized experiments yield unbiased estimates of treatment effects. For this reason, randomized experiments are usually viewed as the gold standard against which to compare the results of other methods for assessing treatment outcome. If quasi-experiments did as well as randomized experiments, they could often be substituted for randomized experiments, which in many situations would make the logistics of experimentation considerably easier for both researcher and subject. Unfortunately, relatively few researchers have tried to compare results from randomized experiments to those from quasiexperiments; those who have explored the issue have found inconsistent results. In the medical and surgical literatures, for example, research suggests that randomized trials of medical innovations yield smaller estimates of the effectiveness of the innovation (Colditz et al. 1988; Gilbert et al. 1978). In psychotherapy research, the findings suggest that random assignment may make little difference to outcome (Smith et al. 1980). Becker's (1990) study of Scholastic Aptitude Test (SAT) coaching found that randomized trials yielded larger effect sizes than quasi-experiments. In reality, of course, each of these studies operationalized the question slightly differently. Some included only sequential assignment of subjects to conditions in the quasiexperimental