![[ Visit Client Website ]](images/banner.gif)
Childcare subsidy receipt is shaped by a complex set of family eligibility, provider participation, policy changes, and local childcare market conditions. Yet state agencies often lack small-area forecasting tools to identify where subsidy receipt is likely to rise or fall over time. Most prior studies have focused on descriptive, individual-level patterns of subsidy use, offering limited insight into the geographic and structural factors shaping utilization. Because subsidy receipts likely reflect nonlinear interactions among potential predictors, conventional analytical approaches may not fully capture these dynamics. Guided by the early care and education access framework, this study examines whether linked administrative data and machine learning can improve predictions of geographic variation in subsidy receipt across Massachusetts and identify the factors most strongly associated with predictive performance.
Methods
We used a longitudinal census tract-by-month dataset linking child care subsidy records, provider licensing data, tract-level demographic indicators, and policy timing measures to predict the monthly number of children receiving subsidies in each census tract. The analytic sample included 148,632 tract-month observations. Predictors included provider-market characteristics, including licensed childcare capacity, subsidy density, local demographics, seasonality, and policy shifts. We estimated fixed-effects models using the full sample as the baseline and an XGBoost model using a 10% subsample split into training (80%) and test (20%) sets. To minimize data leakage, we excluded closely related predictors from the same outcome family from the XGBoost model. Model performance was evaluated using root mean squared error (RMSE), mean absolute error (MAE), and R-squared. Feature importance and SHAP analyses were used to interpret the XGBoost results.
Results
The fixed-effects model achieved a within-R2 of 0.828 (RMSE = 6.18, MAE = 3.06), with prior-month subsidy count and local provider-market conditions as the strongest predictors of within-tract variation over time. The XGBoost model showed strong predictive performance in forecasting local monthly subsidy receipt with high precision (RMSE = 4.31; MAE = 1.66; R2 = 0.995). The strongest predictors were provider-market conditions, including licensed childcare capacity (importance = 0.098), concentration of subsidized children (importance = 0.078), lagged licensed capacity (importance = 0.061), preschool capacity (importance = 0.057), and school-age capacity (importance = 0.050). Even after removing several dominant proxy-like predictors, model performance remained strong (RMSE = 10.65; MAE = 4.63; R2 = 0.968), suggesting that prediction was driven not only by near-outcome indicators but also by broader structural features of local childcare markets.
Conclusions
Because the two models were estimated on different samples in this analysis, comparisons are descriptive. Still, findings highlight the promise of linked administrative data and machine learning for improving local prediction of subsidy receipt and developing actionable forecasting tools for subsidy systems. Results also suggest that subsidy receipt is shaped not only by family demand, but also by the structure and capacity of local childcare markets. For social policy research, this shifts attention from retrospective description to proactive planning. Forecasting tools may help state agencies identify emerging access gaps, anticipate changes in local demand, and target reimbursement, outreach, and provider support to strengthen equitable access to subsidized care.