Catching families before they slip through the cracks: Using early warning systems to increase preschool enrollment

• In Event: Policies on Early Childhood Access, Skills, and Opportunity

Abstract

Equitable access to early childhood education and care (ECEC) remains elusive despite the expansion of universal pre-K programs. While these programs guarantee free seats for young children, participation often falls short of universal. This may be, at least in part, because families must navigate complex application, assignment, and waitlist processes that can ultimately lead to non-enrollment. These choice-based procedures may unintentionally reproduce the very inequities that universal programs aim to reduce. 

In this study, and in partnership with New York City Public Schools, I investigate whether predictive modeling can help early childhood choice-based systems identify and support applicant families at risk of not completing enrollment. I use data from admissions cycles between 2018 and 2024 in the 3-K and pre-K systems. I leverage available information about applicant families’ neighborhood conditions, application behaviors, and assignment outcomes, to train and validate a model for 3-K and pre-K applicants separately. I use a standard supervised machine learning workflow for model training and evaluation. 

I examine whether three supervised machine learning approaches, elastic net logistic regression, Random Forests, and XGBoost, alongside logistic regression as a benchmark, can provide sufficiently precise predictions of non-enrollment and serve as useful tools in future admissions cycles. I train and validate models using time-aware cross-validation and assess them using various metrics (e.g., precision, recall, PR-AUC, ROC-AUC, and calibration) to identify the best-performing model for each system. After identifying these models, I report on the factors that are most predictive of non-enrollment using variable importance plots and SHAP values for 3-K and pre-K applicants. 

My findings contribute to the ECEC access literature by illustrating how, throughout the length of the choice process, neighborhood conditions interact with system-level features and assignment outcomes to ultimately shape enrollment outcomes. My study also contributes to the early warning systems literature by adapting predictive methods typically applied in K-12 dropout contexts to ECEC. If the best-performing models are sufficiently predictive, this study’s findings could provide NYCPS with two early warning system tools that could support targeted outreach during ongoing enrollment processes and even inform decision-making for the upcoming 2-K program rollout. Ultimately, this work offers both theoretical and practical pathways for strengthening universal pre-K systems’ ability to deliver on their promise of equitable access.

Author

• Berta Bartoli, New York University
  Presenting Author