Linking School Meals Policy to Medicaid Spending
An independent published methods example linking Arkansas school records and Medicaid claims to study policy adoption with stacked difference-in-differences.
Home / Case Studies / Linking School Meals Policy to Medicaid Spending
Source published August 5, 2026 · Resource reviewed September 7, 2026
The research question
Can education policy be evaluated using children’s Medicaid spending?
Data files and cohort construction
An honest broker linked Arkansas education records to All-Payer Claims Database enrollment and claims for 2013–2020. School meal adoption came from NCES. The sample required at least ten Medicaid enrollment months per academic year and excluded inconsistent identifiers, grade trajectories, and non-PASSE commercial coverage. Four adoption cohorts contributed two pre-adoption and two post-adoption years. The matched analysis included 131,851 children.
Measures and analytic design
Stacked difference-in-differences combined cohort-specific comparisons, using coarsened exact matching on prior spending trends. Total spending was log-transformed after flooring values below $1. Service categories separated inpatient, emergency, outpatient, and pharmacy spending; CCSR diagnoses and GPI drug groups supported clinical breakdowns. Standard errors clustered by school and student.
Robustness checks and limitations
Checks relaxed payer restrictions, excluded PASSE participants, and removed the pandemic-overlapping cohort. Event studies and alternative specifications assessed sensitivity. The pooled two-year spending association was not significant; the second-year estimate was lower. Nonrandom adoption, linkage selection, and single-state coverage limit interpretation.
Practical application: plan the linkage before modeling
The following are FastHSR implementation considerations, not additional procedures claimed for the published study. Begin with a data map that separates student identity, school attendance, coverage, services, and policy dates. Assign each field a source and observation period. Report unmatched records and excluded children before presenting an effect estimate. A successful file join does not establish that the matched population represents every child served.
Keep the policy clock aligned with the claims clock
An academic year crosses calendar years. Choose one consistent spending window and make coverage eligibility refer to that same window. Track school moves and policy changes explicitly. For each adoption cohort, draw a simple timeline showing the comparison population and the pre- and post-policy periods. Count unique children separately from rows in the stacked analysis because one child may contribute to more than one comparison.
How to use the result responsibly
This approach can inform evaluations of programs outside the clinic when exposure can be linked to healthcare records. Keep spending and health outcomes distinct: lower payments alone do not establish better health or adequate access. Present the overall prespecified comparison alongside year-specific and service-specific estimates. Before projecting a budget impact, account for program costs, population differences, coverage changes, and the uncertainty around the estimate.
Source article and supplement
Sundell J, Kaur H, Lawson GM, et al. Universal free school meals and Medicaid spending: evidence from linked education and claims data in Arkansas. Health Affairs Scholar. 2026;4(8):qxag196. doi:10.1093/haschl/qxag196.
The full article was reviewed. The separate supplement download could not be retrieved during this scan; descriptions of its contents rely on the article. Use the journal’s supplement access point to verify exact specifications before replication.
Frequently asked question
Why link school records to claims?
School records identify attendance and policy exposure; claims and enrollment records supply healthcare spending and the periods in which it can be observed.