Given the thousands of Head Start centers across the country and the 1 million children served by Head Start each year, researchers and policymakers alike are interested to know who benefits the most from Head Start. Some have argued that “skill begets skill” (Heckman, 2008) with early skills allowing children to take better advantage of later opportunities. However, it may also be that children who enter Head Start with the lowest levels of skills have the most to gain from the program and thus should see the greatest program impact. This paper tests these competing questions: Does “skill beget skill” in early intervention preschool education? Or do those who have “most to gain” benefit most from Head Start? 

The Head Start Impact Study is a random assignment study of the impacts of a nationally representative set of Head Start programs on children and families (treatment = 2,783; control=1,884). There were no statistically significant differences found in the demographic makeup of the groups. We focused on three outcomes: children’s academic achievement (a composite measure created from Woodcock Johnson Applied Problems, Woodcock Johnson Letter-Word, and Peabody Picture Vocabulary Test; Mather & Woodcock, 2001; Dunn & Dunn, 1997), children’s parent-rated problem behaviors, and children’s parent-rated positive behaviors. All models were run using jackknife replicate weights to account for the clustered sampling design and participant non-response.

To examine how program effects varied by children’s initial skills, we randomly selected 30% of the control group to develop a prediction model for children’s outcomes at the end of the program year (spring) in the absence of treatment, an approach similar to the use of propensity scores (Yoshikawa, Magnuson, Bos, & Hsueh, 2003).  Children’s outcomes were regressed on their initial score at beginning of the program year (fall) and a number of baseline covariates.  The predicted scores from these models were then applied to experimental group members and the remaining members of the control to identify children’s predicted performance on the outcomes in the absence of Head Start treatment.  We next divided children (excluding the 30% of the control group used to develop the prediction model) into quartiles of predicted scores.  Validity checks were conducted to examine the balance of predicted scores across experimental and control youth.  The checks indicated few significant differences by experimental status within quartile on both fall scores and the other baseline covariates.  Lastly, we conducted regression models within each quartile predicting children’s outcome scores from treatment condition, controlling for their predicted scores. 

Our results indicate that Head Start participation predicted higher academic achievement, with effects concentrated within both the highest and lowest quartiles. Head Start also predicted lower levels of problem behavior, but the effect was concentrated among children who had relatively low levels of problem behavior at the start of the program. No impact was found for positive behavior. We thus found some support for the “skill begets skill” effect but stronger support for the “more to gain” effect of Head Start.