Panel Paper: Elusive Longer-Run Impacts of Head Start: Replications and Extensions

Saturday, November 10, 2018
Tyler - Mezz Level (Marriott Wardman Park)

*Names in bold indicate Presenter

Remy J.C. Pages, Greg J. Duncan and Drew H. Bailey, University of California, Irvine


Background

Building on the approach of Currie and Thomas (1993), Deming (2009) estimated long-term impacts of the Head Start using sibling data from the NLSY79 through 2004, and found differences in early-adult outcomes for siblings who did and did not attend the program. Head Start participants had higher rates of high school graduation, college attendance (at least one year), and scored higher on a summary index of young adult outcomes. Deming’s work provides an important population-based perspective on the efficacy of early-childhood education programs.

Objective and Method

We first replicated Deming’s design and findings (siblings 4 years old by 1990; N = 1,251); then extended his work in three ways. 1) By replicating the analysis over siblings 4 years old between 1992 to 2000 (N = 2,144); 2) by matching and appending 10 additional years of biennial survey data (i.e., up to 2014) on the siblings included in Deming’s analysis; and 3) by combining both samples (N = 3,738).

Our design allowed for the estimation of impacts on longer run outcomes: completed years of schooling, college graduation and earnings (computed as the log of 1994-2014 averaged earnings, in 2014 USD, adjusted for age and survey-year). Head Start impacts were estimated using a no preschool attendance counterfactual, and standard errors clustered at the family level.

Preliminary Results

We found that for the new complement sample, estimated Head Start impacts were mostly non-significant; some were negative: e.g., on the Adulthood index, -.143 sd (p < .05). For Deming’s extended sample, estimated impacts on the Adulthood index were significant but smaller: .166 sd (p < .05) vs. Deming’s (2009) estimate of .228 (p < .01). Considering longer run outcomes: for the complement sample, all Head Start estimates were null or negative; for Deming’s extended sample, Head Start participation yielded an estimated .3 year more completed schooling but no significant effect on college graduation or earnings.

Our overall combined sample reflected these opposite trends, with Head Start participation showing no significant effects on adult outcomes (for both, age 19+ and 25+) except for low maternal AFQT children on the Adulthood index at age 25 or older: .180 sd (p < .05). Results likely reflect a combination of the heterogeneous effects of Head Start by family characteristics (the complement sample was less disadvantaged than Deming’s original sample), and changing counterfactual conditions (the complement sample attended school more recently).