A Post-Hoc Procedure That Can Be Better than Random Assignment to Treatment.

An analytical procedure that uses fall kindergarten assessment data to retroactively create equivalent comparison groups for longitudinal research and evaluation studies was designed and tested, and then analytically and empirically compared with random assignment to treatment. Students entering kindergarten in Hawaii are administered the Peabody Picture Vocabulary Test (Revised) and the Missouri Kindergarten Inventory of Developmental Skills. In 1990, 8,909 matched pairs of students (i.e., students with 1986 fall kindergarten data and 1990 third-grade data) with data on these tests and on the Stanford Achievement Test in grade 3 were identified, and retroactive equivalent groups were created to compare a group who had received some sort of specific treatment with equivalents. Given the distribution of pretreatment scores for the treatment group, along with socioeconomic data and ethnic distribution, an equivalent distribution was created by filling the slots with appropriate individuals, also choosing the schools so as to ensure equivalence in distribution of school attrition rates. The net result was two groups with equivalent distributions. Had the design relied on random assignment to treatment, there would have been non-trivial probabilities that the groups would differ substantively on one or more of the variables. The method appeared feasible for conducting reality-based research when data allow for selection of a comparison group retroactively. Three tables present study data. (SLD)