Assessing the Evaluations of Early Childhood Education Programs

Chapter 19 - Ludwig and Miller
Introduction to the Original Evaluation (Excerpt)

Head Start began in 1965 with the goal of helping low-income children achieve school readiness "by enhancing the social and cognitive development of children through the provision of educational, health, nutritional, social and other services to enrolled children and families."1 Currently, the program serves approximately 905,000 children at a cost of around $7 billion per year.

Jens Ludwig of the University of Chicago and Douglas Miller of the University of California (Davis) used a regression-discontinuity design to estimate the impact of Head Start funding and participation on the health and educational attainment of children enrolled in the program between 1965 and 1983. The authors report a 50 to 100 percent increase in levels of Head Start participation and funding in the 228 poorest counties in the country as compared to the 349 next poorest counties during this time period. They also find a 33 to 50 percent decrease in mortality rates of children between the ages of five to nine living in the poorest counties due to diseases targeted by Head Start's health program. Finally, they find suggestive evidence for a positive impact of Head Start on high school completion and college attendance.

These findings are often cited as evidence that Head Start can have lasting impacts, however, their lack of generalizability to today's current Head Start population cannot be ignored. The counties included in the study were not only the poorest in the country, but were also primarily located in the South. During the course of the time period studied many other social programs, such as Medicaid and WIC, were implemented that presumably would have influenced health and education outcomes as well. Moreover, the study measures the impacts of the Head Start program as it existed over 40 years ago. There have been many programmatic changes during this time, such as the increasing use of commercially available early childhood curriculums and focus on school readiness.

Additionally, the nonparametric analytic approaches utilized by the authors created large standard errors, required large sample sizes and caused them to choose a large bandwidth in order to minimize bias and increase the precision of their estimates. The size of the bandwidth seems to indicate that there was a lack of data right at the cut-point. This is problematic given that regression discontinuity designs are contingent on this point. This is the point used to approximate randomization and that serves to highlight the differences between the program and comparison groups.

s Finally, the authors were also interested in examining whether there were differential impacts on blacks and whites, but the large standard errors of these estimates made the findings non-significant. These methods also require large sample sizes to achieve precision, which leads to questions regarding the use of NELS data as an estimator for educational achievement. The NELS data contained relatively small sample sizes and also produced estimates with large standard errors. There is some indication that this was not an ideal data set to use for the questions being studied by the authors.

Continue Reading


This report is open to public comments, subject to review by the forum moderator and our discussion guidelines. To leave a comment, fill in the form below, or email us at

There are currently no comments on this document

Comment Form
Send to:
* Name: (required)
* Email: (required)
Organization or Affiliation:

Thank you for your submission.