How Important was the Economy to the Effectiveness of Welfare Reform?

This blog is based on an article in the Journal of Social Policy by Marilyn Edelhoch, Cynthia Flynn and Qiduan Li. Click here to access the article.

The short answer is “Very Important”. The purpose of this study was to answer two research questions. First, what was the impact of participation in South Carolina’s Temporary Assistance to Needy Families (TANF) programme, Family Independence (FI), on participants’ earnings? Second, how did the impact vary as the economy moved from relative health and stability to deep recession?

In this paper, we report results of a quasi-experimental study assessing the impact of the economy in good times and bad on employment trajectories, drawing on administrative data over the years of the Great Recession (2007-2009). During the recession, unemployment rates in South Carolina increased from less than 6% to over 12%. By tracking employment outcomes, we determined that under normal economic conditions, South Carolina’s TANF programme was helpful to participants compared to similar programme applicants who did not participate. When the economy started to deteriorate, programme effects lessened and at the bottom of the recession, there was no difference between programme participants and non-participants in employment outcomes.

Over recent decades, the United States, Canada, and most European countries implemented major welfare reforms with the primary goal of moving employable recipients into the workforce. The reformed welfare system in the United States, TANF, was implemented in 1996. Based on the first welfare leaver surveys in the country, Edelhoch and Liu reported in 1999 that welfare reform was the “roughly right” policy in South Carolina. Most leavers described being better off than they were on welfare. Subsequent studies in SC and elsewhere showed that TANF programmes had generally positive average effects on employment and earnings, although the gains were not evenly distributed across subgroups.

The use of large administrative databases in evaluation and policy research has grown in popularity given the expense and difficulties of implementing randomized controlled trials. Not only does our paper help to answer important questions about the efficacy of welfare reform, but also, we have illustrated the use of latent growth curve (LGC) modeling in evaluating an employment programme using longitudinal administrative data. This statistical modelling technique has proven valuable across a variety of disciplines for conducting economical and methodologically sound evaluations. LGC modeling allowed us to determine that the effects we observed were statistically significant (i.e., not the result of chance).

For this study, we selected three cohorts from applicants who applied for FI before, at the beginning of, and at the height of the recession. Our treatment group came from applicants who had been approved to participate in FI. Our comparison group came from applicants who did not complete the application, whose resources or income exceeded the limit, or who were deemed ineligible. From the agency and state administrative databases, we extracted Unemployment Insurance (UI) earnings data over seven quarters for each applicant, including three quarters prior to intake, the quarter of intake, and three quarters after intake. We used propensity scores to match the treatment and comparison group members on four quarters of UI earnings, local unemployment rates, and a number of individual and family demographic variables.

Figures 1 A-C shows the earnings trajectories over seven-quarter spans for FI entrants (the matched treatment group) and non-entrants (the matched comparison group) against state unemployment rates by cohort. For both the entrants and non-entrants, the earnings trajectories display a decreasing pre-intake trend, a turning point, and an increasing post-intake trend. The decreases in earnings before intake are remarkably similar between the entrants and non-entrants. After intake, the increase in earnings of the non-entrants represents a ‘normal recovery’ that would have occurred for the entrants had they not joined FI.

If the FI programme did indeed make a positive impact, we should see UI earnings increase more rapidly for the entrants compared to the non-entrants. As expected, this was the case for the pre-recession cohort. Although the average earnings of entrants were initially the same as those of the non-entrants, their quarterly earnings were $305 more than the non-entrants by the end of the follow-up period (Figure 1 A).

Figures 1 B-C illustrates the effect of the deterioration of the economy on the earnings of the early recession cohort and the deep recession cohort, with either little or no difference between the entrants and non-entrants in post-intake earning trajectories. By the end of the follow-up period, the advantage of the entrants over the non-entrants was only $108 for Cohort 2008 and was literally zero for Cohort 2009.

We then employed a multiple-group, two-piece latent growth curve (LGC) model to formally test whether what we see in Figures 1A-C below arose by chance. The LGC model showed that South Carolina’s FI programme successfully increased earnings amongst participants before the recession. Program impact lessened as the economy weakened. During the period when South Carolina was in deep recession, the FI programme did not result in higher earnings for the participants relative to those of the comparison group who were without FI assistance.

In summary, this study found that the impact of South Carolina’s ‘work-first’ approach depended on the economic environment and the availability of employment opportunities. During good economic times, the program worked much as intended; (however, we have noted that the benefits of the program were not evenly distributed; subsequent studies showed that more disadvantaged clients in particular suffered greater deprivations). The findings of this study also underscore the need for governments to support and sustain assistance such that programme expectations, funding, and practices conform to the context in which the programmes operate. For example, state unemployment rates can vary widely; in September 2020, rates ranged from 3.5% in Nebraska and 4.1% in South Dakota to 11% in California, 12.6% in Nevada and 13.2% in Hawaii. During periods of economic recession and/or instability, and especially exacerbated by crises such as covid-19, assistance programmes should provide a safety net to prevent destitution and hardships, and then, provide job skills training, education, and supports to improve future employability and prospects for self-sufficiency.