Working with state agencies to create public resources for modeling SNAP errors

• In Event: Leveraging Data and Partnerships to Help States Navigate SNAP Policy Changes

Abstract

The One Big Beautiful Bill (H.R. 1) passed in July 2025 dramatically altered how the supplemental nutrition assistance program (SNAP) is implemented by states. Under H.R. 1, states with payment error rates above 6% need to pay back some share of the benefit costs to the federal government. The cost to states, beginning FY 2028, will reach hundreds of millions for many and is projected to exceed $1B in at least two states. 

As part of a wider effort coordinated by the Safety Net Response Network, Better Government Lab has been providing code, data, and technical guides to states that are trying to reduce SNAP error rates. Rather than increasing scrutiny of all cases, which results in more burden on clients and staff, many states are using historical data and predictive modeling to identify cases which are more likely to have errors. These modeling practices not only increase the precision and yield of cases identified for secondary review, but can inform how application and case processing practices can be altered to minimize errors. A major challenge is that each state only has data on a few hundred errors in recent years.  

To help address this, Better Government Lab (BGL) is using national quality control microdata to provide guidance to agencies on feature importance, modeling approaches, and potential prioritization rules to identify specific kinds of errors (such as underpayment errors or earned income errors). Along with the Response Network, BGL has convened an ongoing data modeling workshop for 15 states, which has led to rapid advances in modeling performance for several states.  

This talk will review what worked in terms of convening state agencies, surprises in which operational, systems and household model features were most helpful, and which approaches proved useful in facilitating exchange and translation to practice. Learn more about the work by exploring resources including workshop summaries and data visualization tools, and code for state and national-level regression trees.

https://www.bettergovernmentlab.org/resources/snap-quality-control-error-viewer 
https://github.com/giannella/snap_qc

Author

• Eric Giannella, Georgetown University
  Presenting Author