The Head Start Impact Study (HSIS) was designed to provide an estimate of the average impact of Head Start, with relatively less attention to heterogeneity of impact, particularly across sites. Indeed, the study of variation in impact in the HSIS and other preschool studies has typically addressed moderation by child and family risk (rather than neighborhoods).  Yet, information on variation by neighborhood context is especially critical for large-scale programs such as Head Start, which serve families from a large and diverse range of communities. 

In this paper, we capitalize on the addition of geocodes for Head Start centers in which children were randomly assigned to address questions about the moderating role of neighborhood characteristics.  Specifically, we explore the extent to which impacts of assignment to Head Start on outcomes for children vary by levels of neighborhood disadvantage, community crime, and the availability of alternative child care options.  To date, we have matched center geocodes to census data to assess neighborhood disadvantage (i.e., poverty).  Survey data from parents aggregated to the center level assess neighborhood crime. Future analyses will also include the availability of alternative child care arrangements at the neighborhood level. 

Our analytic approach utilizes OLS regressions with random slopes in which children’s spring PPVT scores are modeled as a function of their fall scores, a set of center group fixed effects (in which random assignment took place), a set of child-level covariates, the random assignment dummy representing the impact of assignment to Head Start, the neighborhood dimensions of interest (poverty, crime, child care availability), and the interactions between assignment to Head Start and neighborhood dimensions. 

Preliminary results demonstrate a significant main effect of crime and poverty on children’s residualized PPVT scores, (b=3.93, SE=1.14, p<.01 for crime and b=-2.59, SE=0.70, p<.01 for poverty), where children who were randomized in center groups in high crime neighborhoods showed significantly higher spring PPVT scores and children who were randomized in center groups in low poverty neighborhoods showed significantly higher spring PPVT scores, net of their fall scores.  More importantly given our question regarding treatment heterogeneity, there was a significant interaction between children’s assignment to Head Start and their center group neighborhood’s level of crime, b=0.14, SE=0.06, p<.05, where the positive impacts of assignment to Head Start on children’s residualized PPVT scores were concentrated among children in high-crime environments, as well as a significant interaction between children’s assignment to Head Start and their center group’s neighborhood level of poverty, b=-0.13, SE=0.07, p<.10, where the positive impacts of assignment to Head Start on children’s residualized PPVT scores were concentrated among children in low poverty environments.  Notably, results were robust to the inclusion of the interaction of urbanicity and random assignment to Head Start in the model.  Preliminary analyses on other child cognitive outcomes do not show this same pattern of moderation. 

Given wide variation across children and programs nationally, understanding heterogeneity of impact in the HSIS is critical.  This study provides key information about how neighborhood characteristics may moderate the impacts of this highly-watched program.