Results from the Head Start Impact Study (HSIS) reveal positive impacts of the offer to enroll in Head Start.  We extend these results by examining whether the offer of Head Start was differentially beneficial for students grouped by the care setting they would have received absent the Head Start offer.  Answers to this question are policy relevant in that they may inform future efforts to target high quality, comprehensive services such as Head Start. 

Parents’ decisions regarding the type of childcare they choose for their children flow from a complex set of family, community and policy factors (Webber, 2011). Chaudry, Henly and Myers (2010) conceive of parents’ decisions about childcare not as choices but rather, as accommodations to a dynamic confluence of employment, family, and market factors that constrain families’ abilities to select and make use of optimal care for their young children. In HSIS, both treatment and control group students took up different types of care settings such as Head Start, other center care, parent care, and home daycares. Given this, in the first step of our analysis, we build a comprehensive model to predict child-care choice based on a robust set of baseline covariates.  This has the important additional benefit of providing information on relevant predictors of child-care choice. Analyses conducted to date indicate that characteristics such as child age, disability status, gender, pre-academic skills, and mother’s marital status, among others, are important predictors for the type of care a child received 

Using our predictive model, we employ a “principal stratification” framework coupled with Bayesian analytic methods to stratify children into latent subgroups defined by the pair of which child care setting each child would experience under the randomized offer of Head Start and under no such offer. By then examining treatment-control differences in outcomes for only those children who, for example, would enroll in Head Start if offered the opportunity, but who otherwise would be cared for by a parent, we can estimate the impact of Head Start across the alternate care-type settings.  In particular, we compare performance on outcomes such as the Peabody Picture Vocabulary Test and Woodcock-Johnson III assessments. This approach has two main advantages: (1), we obtain valid estimates of the causal effects of interest within each “principal stratum;” and (2), we can incorporate substantive assumptions directly into the model.  We can also address the differential attrition of students from the treatment and control groups in the HSIS by including a classification category to handle missingness of care setting information.