TANF Data Collaborative Pilot: Longitudinal Data for Understanding Participant Actions in New York

The project aims to understand factors influencing participants in New York's Temporary Assistance (TA) program, including their departure or return, subgroup tendencies, and long-term benefit receipt. The goal is to develop a predictive tool to assist staff in identifying potential long-term participants and improving decision-making on service provision and interventions. 

The project analyzed administrative data from New York's Temporary Assistance (TA) program and Unemployment Insurance system. A longitudinal spell file covered TA participants from August 2005 to January 2020, tracking their public assistance experiences. Focusing on a cohort of adults with children who started receiving benefits between July 1 and September 30, 2016, in select counties, the team conducted survival analyses using Cox proportional hazards regression. They identified factors influencing the likelihood of leaving or returning to TA and developed a logistic regression model to predict long-term participants. Inputs included personal characteristics, employment data, county information, and unemployment rates. Cross-validation ensured model reliability and alignment with outcomes. The project aimed to understand participant characteristics, aid early identification of long-term participants, inform service provision, interventions, and staffing decisions, and generate evidence on intervention impacts. 

The project identified earning history and prior cash assistance receipt as the primary predictors of when participants would leave or return to TA. The model effectively predicted low-risk long-term participants but underestimated high-risk cases. Refinements are planned to address biases to ensure fair utilization of the model's results. 

This project serves as a replicable model for staff from other agencies looking for examples of efforts that support the use of administrative data for learning and improvement. Policymakers, researchers, and organizations seeking to expand the use of data in state TANF agencies may find interest in this project.