The Impact of Immigration Policy on Student Enrollment

• In Event: Public School Student Disenrollment: Causes, Effects, and Potential Solutions

Abstract

The 2025-26 school year unfolded against a backdrop of substantially intensified federal immigration enforcement, raising questions about whether and how immigrant-connected families have engaged with public institutions, including schools. English Learner (EL) students offer a natural lens for examining whether enrollment patterns have departed from historical trends, given that this group is disproportionately composed of children in immigrant households. EL status is not synonymous with immigrant or foreign-born status because most EL students are U.S.-born citizens, but EL enrollment remains a useful observable proxy in a period of heightened enforcement. Because enforcement intensified nationally, credibly identifying the causal effect of enforcement on enrollment would require strong assumption. I therefore take a different approach and ask a narrower but tractable question, namely whether 2025-26 EL enrollment departs from what past trends would lead us to expect. To answer it, I conduct a series of prediction exercises using district-by-year enrollment data spanning the mid-1990s through 2025-26 across 24 states.

I estimate district-specific trend models on data through 2024-25 and generate out-of-sample predictions for 2025-26. To benchmark the resulting prediction errors, I implement a leave-one-out procedure across all prior years, which constructs a historical distribution of prediction errors against which the 2025-26 deviations can be compared. This approach sidesteps the need for a formal control group while allowing me to ask whether 2025-26 is an outlier relative to each district's own prediction history. I discuss results here for the ten states where 2025-26 EL enrollment data is already available but anticipate that this number will grow as additional states release current school year data.

I find aggregate EL enrollment shortfalls of approximately 9% in Texas (roughly 126,000 students below prediction), 9% in Massachusetts, 8% in Connecticut, 6% in New York, and 4 to 5% in Rhode Island, Virginia, and Washington. In Texas, Massachusetts, and Connecticut, 2025-26 is the single worst prediction year in the historical record. In Texas and New York, more than one-third of districts record their largest-ever prediction error in 2025-26. Placebo analyses using White and Black student enrollment show deviations well within the historical prediction-error distribution, suggesting the EL-specific pattern is not an artifact of broader enrollment dynamics or model misspecification.

I pair the LOO validation exercise with a panel event-study specification that incorporates district fixed effects and district-specific linear time trends, clustering standard errors at the district level. Results corroborate the prediction-based findings.While I document shortfalls rather than estimate causal effects of specific policies, the magnitudes are large, the pattern is subgroup-specific, and the timing is striking. Because school funding formulas in most states are driven by enrollment counts, shortfalls of the magnitudes I document carry potentially substantial implications for district budgets in subsequent years, particularly for the EL programs and services supported by supplementary categorical funding. For the final presentation, I will expand the analysis to additional states as data become available and introduce proxies for local vulnerability to immigration enforcement to examine geographic heterogeneity in the documented shortfalls.

Authors

• Andrew Mitchell Camp, Brown University
  Presenting Author

• Christopher Cleveland, Brown University
  Non-Presenting Co-Author