Panel data analysis uncovers regional variations in food subsidy utilization across Massachusetts, highlighting targeted vendor allocation strategies.
Key Points
To estimate the heterogeneous impacts of adding new food vendors on program utilization in the Massachusetts Healthy Incentives Program using panel causal inference.
Formulated the panel clustering estimator (PaCE), which clusters units with similar treatment effects using a regression tree and exploits low-rank panel structure for estimation.
Established mathematical convergence guarantees and evaluated estimator performance using semisynthetic numerical experiments against existing panel estimators.
Applied PaCE to empirical panel records from the Massachusetts Healthy Incentives Program across geographic zip codes.
Theoretical results proved the asymptotic convergence of PaCE estimates to true treatment effects.
Semisynthetic simulations demonstrated superior accuracy over existing panel benchmarks for average and heterogeneous effects using compact regression trees of 40 or fewer leaves.
Empirical evaluation uncovered marked regional heterogeneity in how vendor additions alter program utilization across Massachusetts zip codes.