Insights / Research BriefJul 28, 2023

Certification And Recertification In Welfare Programs: What Happens When Automation Goes Wrong?

Derek Wu, Bruce D. Meyer
Enrollment in SNAP, TANF, and Medicaid fell by 15%, 24%, and 4% one year after automation, likely attributable to insufficient personalized assistance from caseworkers, lower tolerance for application and recertification errors, and delays and technical glitches at overwhelmed call centers.

While safety net programs are key to many American livelihoods—Medicaid and the Supplemental Nutrition Assistance Program (SNAP) alone reach 23% and 10% of the US population—some eligible individuals miss out because of administrative burdens. Application can be time-consuming, and many well-intentioned applicants may otherwise struggle with completion. While these and other burdens may also screen out less needy individuals, understanding why needy applicants do not finish the job is key to ensuring the effectiveness of these programs.

The effects of administrative burdens can hinge critically on the context in which they appear. A given set of burdens, for example, may occur at initial application and/or during the recertification process, and their effects may vary based on populations served. Despite the importance of these programs, and despite the awareness of the differential effects of these burdens, little is known about how they differ across programs and application stages.

This paper addresses that gap by investigating how barriers to enrollment affect the take-up of three important safety net programs—SNAP, Medicaid, and Temporary Assistance for Needy Families (TANF)1—and the types of individuals that are screened out at a given application stage. The authors analyze administrative burdens associated with the automation of welfare caseworker assistance, which states have increasingly adopted to allow individuals to apply and recertify remotely. Automation is usually meant to make things more efficient (more productive and less expensive), but is that what happened in this case? 

To answer that question, the authors examine the experience of Indiana that, in 2006, awarded a 10-year, $1.3 billion contract to the IBM Corp. to automate caseworker assistance for the state’s welfare services. Expected to lower administrative costs and increase convenience for clients and operators alike, IBM’s automated system began rolling out to counties in late 2007. However, just two years later, the system was terminated after reaching only 59 out of Indiana’s 92 counties. The authors develop a framework to compare the effects of barriers to enrollment on take-up and targeting among initial applicants and recertifiers. Using administrative welfare records covering nearly 3 million recipients, the authors find the following:

  • Significant declines in SNAP (15%), TANF (24%), and Medicaid (4%) enrollments one year after the rollout of IBM’s automated system. 
  • These effects can likely be attributed to insufficient personalized assistance from caseworkers, lower tolerance for application and recertifications errors, and delays and technical glitches at overwhelmed call centers.
  • Decreases in entry rates for all three programs are similar in magnitude, while increases in exit rates are largest for TANF and smallest (and statistically insignificant) for Medicaid. The larger effects on TANF exit can be explained by higher transaction costs stemming from shorter TANF recertification intervals and fewer avenues for assistance (vice-versa for Medicaid).
  • Overall, IBM’s automated system screens out individuals who appear less needy, as those remaining on the program rolls typically have lower pre-treatment incomes, fewer years of education, higher per-person benefits, and higher disability levels. 
  • However, these overall effects conceal striking and novel differences across application stages, with more needy individuals screened out at exit, and less needy individuals screened out at entry. 
  • At the county level, proportional enrollment reductions are also largest in earlier-treated and higher-poverty counties, as well as in lower-unemployment counties whose residents may have more earnings to verify.

Bottom line: Efficiency, in this case, has a cost. The move toward automation and the shift toward remote determinations of eligibility for government programs introduces the possibility of unintended consequences. Further, there are different effects at different stages, depending on the types of clients appearing at a given stage. Likewise, policymakers are well-advised to adapt procedures differentially across stages, depending on whom they target.

Finally, the natural experiment setting and groundbreaking data sources in this paper can be used to examine whether welfare cuts induce individuals to increase their short-run labor supply and potentially induce greater self-sufficiency in the long run. One can also examine the extent to which individuals turn to other forms of insurance as a result of sharp negative income shocks, among other issues. These further analyses could provide more direct evidence on the mechanisms underlying the long-term effects on SNAP and Medicaid enrollment, identified in this paper, that persist beyond the life of IBM’s automated system.

1 TANF is a cash welfare program aimed at families with children, reaching less than 1% of the population, with federal expenditures of $14 billion in 2019.