Head Start is the largest publicly funded preschool program in the U.S. and one of its primary goals is to improve the school readiness of low-income children.  As has been widely reported, the first randomized trial of Head Start in the program’s history found some evidence that it is achieving this goal.  Receiving one year of Head Start had small impacts on children’s cognitive outcomes, with impacts cognitive impacts concentrated in the language and literacy domain.  However, effects largely faded out by the end of first grade (U.S. Department of Health and Human Services, 2010).  In explaining these results, some have pointed to the variation in quality across Head Start centers.  For example, fewer than 1 in 20 4-year-olds in the treatment group were in centers with an “excellent” quality rating and only about half were in centers with recommended pupil/staff  ratios (National Forum on Early Childhood Policy and Programs, 2010).  Given the variation in center quality, by extension, there may be considerable variation in Head Start’s impacts on children.  That is, under certain conditions and for certain children, Head Start may be much more effective than it is on average.

In the current paper, we use data from the Head Start treatment year to address this possibility.  Specifically, we will examine whether there is statistically significant variation in Head Start’s impacts on children’s cognitive outcomes across individual children, subgroups of children, and Head Start centers.  To do so, we use a new and innovative methodology for estimating sources of variation in program impacts (Bloom, 2012).  Outcomes used in our study include the Peabody Picture Vocabulary Test-III (Dunn & Dunn, 1997), the Woodcock-Johnson Letter-Word Identification subscale, the Woodcock-Johnson Oral Comprehension subscale, and the Woodcock-Johnson Applied Problems subscale (Woodcock, McGrew, & Mather, 2001).  These tests measure children’s receptive vocabulary, early reading, text comprehension, and early mathematics, respectively.  Our sample includes all three-year-old and four-year-old children randomized to either receive Head Start in the treatment year or to a business-as-usual control condition and who were tested at the end of the first year of the study (N=approximately 3,505 children in 180 center groups). 

Analysis is ongoing, but preliminary work suggests that there is statistically significant variation in Head Start impacts across center groups on children’ early reading and text comprehension skills (p<.05).  There are no statistically significant differences across center groups on children’s receptive vocabulary or early mathematics skills.  At APPAM, we will also present impact variation results across individuals and subgroups.  Findings will be discussed in the context of currently evolving policy proposals to combine early childhood funding streams to expand preschool access to all U.S. 4-year-old children.