A recently published paper on Head Start, by Pedro Carneiro and Rita Ginja, presents evidence that Head Start has sizable long-run behavioral benefits, compared to no preschool, even though cognitive impacts fade.

The paper’s methodology is not a random assignment experiment, but rather relies on a natural experiment. Head Start uses an income cutoff for eligibility. If we compare children who just make the income cut-off for eligibility for Head Start, versus children who just miss the cut-off, we are comparing children who are likely very similar in unobserved as well as observed characteristics. Therefore, it is more plausible that any future changes in outcomes for these children are due to the child’s participation in Head Start.

Carneiro and Ginja in this paper focus on outcomes for males participating in Head Start, because it turns out that their estimation strategy works well for males but not females. Specifically, they find that close to the income cutoff for Head Start, there is a big jump in Head Start participation among males as income changes from just above to just below the cut-off, whereas there is no abrupt jump for females. This may be due to programs using some allowed exceptions to income rules to admit some female children to Head Start who are slightly above the income cutoff.

Comparing males just below or just above the Head Start eligibility cutoff at age 4, Carneiro and Ginja find that males who are just under the income cutoff, versus males who are just above, show reduced obesity, improved health and improved behavior at ages 12-13, reduced obesity and depression at ages 16-17, and reduced arrests and convictions and reduced “idleness” (being neither employed nor in school) at ages 20-21. In contrast, there are no signs of statistically or substantively large improvements in test scores at ages 12-13.

These effects are large enough that it seems likely that Head Start would pass a benefit-cost test simply based on the health improvements and crime reductions. Carneiro and Ginja estimate a benefit-cost ratio of 9 to 1.

The findings are consistent with the notion that Head Start, and perhaps other pre-K programs, have much of their long-term benefits by improving behavior, not necessarily by improving scores on standardized IQ tests. “Soft skills” matter a lot to long-run success, and pre-K programs such as Head Start can improve these skills.

Another interesting study finding is that Head Start’s effects appear to be greater for students participating in Head Start in the 1980s than for students participating in the 1990s. As Carneiro and Ginja note in an appendix, this difference may be occurring because of changes in the alternatives to Head Start. Carneiro and Ginja find evidence that in the past, increased Head Start participation came at the expense of reductions in informal child care arrangements, whereas more recently increased Head Start participation reduces participation in other preschools. Therefore, the reduction in Head Start effects in recent years does not necessarily mean preschool does not work; it means that Head Start is not clearly better in outcomes than other preschools.

Carneiro and Ginja’s study may help reconcile the results of the Head Start random assignment experiment with other studies finding long-run effects of Head Start. The studies finding long-run effects of Head Start focus on more behavioral outcomes such as work and educational attainment, whereas the random assignment Head Start experiment focused more on test scores. Furthermore, the studies finding long-run effects of Head Start are for children participating in Head Start in the 1980s or earlier, whereas the Head Start random assignment study began in 2002. The impacts we find for Head Start may be sensitive to what preschool alternatives we’re comparing the program with, and with whether we focus on long-run behavioral change or shorter-run effects on standardized tests.