Can Variation in Head Start Parenting Effects be Explained by Variation in Family-Level Characteristics?

• In Event: 3-024 - Associations Between Early Education and Care and Parenting Attitudes and Behaviors With Three Diverse Samples

Integrative Statement

In addition to providing low-income children with early care and education services, a primary goal Head Start (HS) is to help parents foster their children’s development. Prior studies suggest that HS participation is associated with increased parent reading and math-related activities and reduced harsh punishment (Gelber & Isen, 2013; Puma et al., 2010). However, this literature almost exclusively examines the average effect of HS, ignoring potential heterogeneity of effects across programs that could elucidate the conditions under which parenting is most enhanced by HS. The present study exploits the random assignment and multi-site design of the Head Start Impact Study to 1) quantify variation in program impacts on cognitively stimulating and socioemotional interactions with children across 297 HS sites, and 2) investigate whether variation in program impacts is due to variation in the types of families who attend different centers.

Following prior investigations (Bloom et al., 2017; Bloom & Weiland, 2015), I examine variation in the effect of random assignment to HS (an average effect of “intent to treat” [ITT]) using a 2-level random coefficients model, and variation in the effect of enrolling in HS (the “Local Average Treatment Effect” [LATE]) using two-stage least squares. In subsequent models, I enter each family characteristic at level 2 of the ITT model to estimate whether these characteristics explain the average HS treatment effect.

Findings reveal significant variation in program impacts across a range of parenting behaviors after one year of HS (Table 1). For instance, parents randomly assigned to HS engage in more cognitive stimulation than non-HS parents after one year of the program (average effect=.19), but this impact varied significantly across programs (SD=.29), with effects ranging from -0.38 to 0.76. All effects were slightly larger in LATE models estimating the impact of enrollment in (rather than random assignment to) HS (Table 1).

I also examine whether variation in program effects is explained by variation in average household risk and the percentage of dual-language learners (DLLs) within a center. Household risk represents a sum of 5 binary items—food stamps receipt, neither parent graduated high school, neither parent worked, mother was a teenager, mother was single. In Table 2, coefficients on the Treatment*HouseholdRisk (Treatment*DLL) interactions can be interpreted as the rate of change in the average HS ITT effect size per point increase in average family risk (percentage point increase in DLLs). Results show that the representation of these family-level characteristics within centers have small but significant effects on the ITT treatment effect for parents’ math activities and their odds of giving children a time-out. For example, for every point increase in a center’s average household risk, the positive HS effect for parents’ math activities increased by .43. Higher average household risk was associated with greater HS impacts compared to lower risk, whereas higher center representation of DLLs was associated with weaker HS impacts compared to lower representations. Results from the present study add to our knowledge about whether and under what conditions HS can impact parents’ interactions with their children.

Author

• Christina Padilla, Georgetown University
  Presenting Author