Association for Public Policy Analysis & Management

This paper investigates the local deterrence effect of immigration raids for Hispanic families on Head Start enrollment. Head Start is the largest federal early childcare education program in the U.S. and provides education, health, and other comprehensive services to low income families. Children in mixed-status families—those with at least one undocumented family member—are particularly vulnerable if this deterrence effect exists since they face multiple disadvantages, and the benefits of preschool are especially strong for English learners. At the same time, Head Start administrators have expressed concerns that immigration enforcement around their centers has hindered efforts to engage Hispanic families.

For this study, we created a comprehensive panel of nationwide immigration raids, program-level Head Start enrollment, public school-level enrollment, other locally enforced immigration laws/policies, and county-level demographics over the late 2000s. The most novel of these data sources is county locations and dates of federal immigration enforcement raids on workplaces, homes, and communities that were conducted between 2006-2008. These were obtained by cross-checking raid listings from three immigrants-rights organizations—that track immigration raids through a variety of sources, including Freedom of Information Act requests, news tracking, and immigrant networks. These data include date, location, and different types of raids (e.g., workplace, community, and residential).

There are a two primary challenges to identifying the deterrence effect of raids on Head Start enrollment: the non-random location of raids, and separating the deterrence effect from an out-migration mobility effect. We address these challenges through a combination of local-area matching with a triple-difference design. First, for the matching, 17% of Head Start counties (249 counties) experienced a raid over our period of interest. To ensure we compare enrollment from similar counties, for each “ever-raid” county, we use the donor pool of “never-raid” counties to identify a match that is similar on both Head Start enrollment patterns and other countywide demographic characteristics up until the first raid. This is done using covariate matching techniques, and since it is done separately for each ever-raid county, this ensures that pre-raid levels and trends are balanced across ever-raid and never-raid counties.

Second, for the triple-difference design, we include Hispanic students enrolled in first grade as an additional counterfactual. Specifically, we conceptualize this as a stacked difference-in-differences design that allows us to separate the deterrence effect from the mobility effect. Because first grade school attendance is mandatory, a difference-in-differences estimate for first-grade Hispanic students will pick up any mobility effects caused from the raids, whereas, the difference-in-differences estimate for Hispanic Head Start students picks up both the deterrent and the mobility effect. By subtracting these two estimates in a triple-difference model, assuming the same mobility effect for both groups, the deterrent effect is isolated.

Preliminary findings suggest that county-level Hispanic enrollment does decrease following immigration raids in a county. Based on a general difference-in-differences estimate of Head Start students only, Hispanic enrollment drops by 9 percentage points following a raid in a county. Our next steps are to implement the triple difference research design and additional robustness checks.