• Unintended consequences: Did welfare reform kill some participants?

    The Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA, also known as ‘Welfare Reform’) changed US welfare by limiting how long clients could receive it and introducing a work requirement. How well has welfare reform worked for recipients?

    Supporters of the law pointed to many positive changes that followed welfare reform, including reductions of welfare case loads and rates of child poverty. However, it is difficult to know whether these positive effects are due to welfare reform or the economic growth since the mid-1990s.

    Peter Muennig, Zohn Rosen and Elizabeth Ty Wilde report on the long term effects of a randomized controlled trial of welfare reform in Florida. From 1994 to 1999, Florida randomly selected a group of welfare recipients into either the Family Transition Program (FTP), a program that included many features of the later PRWORA, or the then-standard Aid for Families with Dependent Children welfare program (AFDC). Muennig and his colleagues looked at Social Security Records for 3,224 participants in the experiment to establish whether participants had died in the 18 years following the randomization.

    What they found was that

    participants in the experimental group had a 16 percent higher mortality rate than members of the control group (hazard ratio: 1.16; 95% confidence interval: 1.14, 1.19; p < 0.01). This amounts to nine months of life expectancy lost between the ages of thirty and seventy for people in FTP.

    This was a surprise: Muennig et al. had expected that the increased employment associated with FTP would lead to improved health. Instead, FTP harmed the participants. Muennig and his colleagues speculate that

    the increased mortality rate could be related to the psychological stress associated with losing welfare benefits, although we cannot measure this directly. Psychological stress is associated with poor health and a shorter lifespan in both humans and animals. It is also possible that employment in low-wage jobs carries hazards (for example, occupational risks) that are not present in most other jobs.

    We have to be careful about generalizing from this result. It’s just one experiment and the FTP may have been different from welfare reforms implemented in other states. Moreover, FTP might have benefited the participants or their children in other ways that compensated for the loss of life.

    Nevertheless, the study has some important lessons.

    First, experiments really clarify the consequences of social policy. Because participants were randomly selected into the reform versus standard welfare conditions, and because it’s likely that few if any participants were lost to follow-up, it’s possible to say with considerable confidence that welfare reform killed some Florida recipients.

    Second, we should do more long term follow-ups of the participants in policy experiments. For this to happen,

    New studies should also require that participants’ identifiers be preserved. Such identifiers are central to understanding the long-term economic and health impacts of social policies because they make it possible to collect data on participants’ earnings and disability and mortality status inexpensively and with great accuracy over decades of follow-up research.

    Finally, policies have unintended consequences. We won’t find these unintended consequences unless we look for them, so we should make a point

    to include health measures in nonmedical social experiments, which would better define both the intended and unintended consequences of such policies on population health.

    I’ve noticed that many of my conservative friends have a core belief that ‘well-intentioned’ liberal policies of the New Deal or Great Society had unintended harmful effects on the recipients when those policies were put in practice. This study shows that ‘unintended consequences’ do not have a partisan affiliation. Liberal policies have unintended consequences and conservative policies have them too. Every social policy should be evaluated with an open mind using the strongest possible empirical design.


    • I’m surprised that health insurance coverage was not reported as a potential mechanism in this study. Welfare de-linked Medicaid from TANF, and although eligibility was nominally preserved, Medicaid participation declined markedly after PRWORA (I know this Florida reform pre-dates that policy, but still interesting). For the adults on TANF and Medicaid, that loss of health insurance could have made a big difference (as they transitioned into service jobs without good benefits).

      • Interesting possibility, Brendan. However, wouldn’t be the case that both groups would have been delinked as of 1996? So the FTP would have had only two extra years of de-linkage (assuming FTP actually did delink). Thinking about RAND and Oregon, would that be enough time to have a big mortality effect?

    • A study this important should include more of its results in the paper. There isn’t a single figure in the paper! Even the appendix doesn’t have their survival analysis regression, just a pair of survival curves. It’s hard to know whether to believe this study’s conclusions without the evidence presented more thoroughly, even though its an RCT.

      • Agree. I understand that putting more regression details in the main article goes against the Health Affairs style, but they could have put them in the Appendix.

    • Wow I would have expected it to go the other way. I would have guessed that forcing people to work would have increased life expectancy.

      My assumption would have been based on these:



      I guess that I need to reconsider a bit. Only one data point but that is a big effect and it was random.

    • Maybe I missed it– dd the low-wage workers get Medicaid?

    • It would be really interesting to know the cause. For example, if the deaths were due to occupational hazards, you could argue that someone else would probably have those jobs if they didn’t, so perhaps those deaths would have happened anyway, just to different people. If the deaths were due to traffic accidents driving to work, you could argue that it’s the same risk anyone else who works faces every day. On the other hand, it the deaths were related to increased stress or poor health due to greater poverty or insecurity, then welfare reform looks a little less humane.

      • Well, as far as deaths from low-wage jobs, when I was a nursing-home ombudsman for several years, a number of the certified nursing assistants were involved in traffic accidents (many worked two jobs and were very weary) — and two died from them.

        You don’t have to die directly on the job, or even die from a cause like say heart disease people think of as related to job stress — you may die because of a traffic accident.

        That no you wouldn’t have died in if not working (and probably wouldn’t have died in if working only ONE job, too).

    • Would we all live longer if we were on welfare?

    • “it’s possible to say with considerable confidence that welfare reform killed some Florida recipients”

      Technically, given the interpretation, welfare reform removed something that had a protective effect on recipients’ mortality. “Killed” is an active word will elicit tribal reactions from readers with ideological biases toward programs like these.

      Ishmael’s question is a great one. Add virtually unlimited technological capabilities and a beneficent AI, and you have the basis for Iain Banks’ Culture series of science fiction novels.