Did welfare reform lead some American families to work less?

The landmark US welfare reform of 1996 provided strong incentives for poor women to work while receiving assistance – but it also provided incentives for some women to reduce their earnings to qualify for benefits. This research develops a new approach to detecting this ‘welfare opt-in’ effect and uses it to analyze data from a large randomized evaluation of welfare reform in Connecticut: the “Jobs First” program. The results reveal that the Jobs First program induced a substantial fraction of the women who were capable of lifting their families out of poverty without assistance to opt for welfare instead.

Twenty years ago, President Clinton signed into law the Personal Responsibility and Work Opportunity Act of 1996 (PRWORA), which promised to ‘end welfare as we know it’ by instituting time limits on welfare participation and providing strong financial incentives to work.

Key among the financial incentives were changes to rules detailing how welfare benefits would be phased out with higher earnings. Prior to the Act, a woman’s welfare benefits could be reduced by as much as one dollar for every extra dollar she earned, equivalent to a 100% tax rate on earned income. After the Act, provisions for phasing out benefits were set by the states, which often lowered the effective tax rates on earnings dramatically.

Connecticut’s Jobs First welfare reform lowered the effective tax rate on earnings to zero

The consensus view among economists is that the 1996 reform was successful at getting welfare recipients to work (Blank, 2002; Grogger and Karoly, 2002; and Meyer and Rosenbaum, 2001). But it is less clear whether welfare reform affected how much potential welfare recipients earn while working (see for example, Meyer, 2002; and Chetty, 2012).

Tax rates and work effort

In theory, reductions in tax rates can have offsetting effects on work effort. On the one hand, a tax reduction implies that the returns to working go up, which could lead to an increase in work effort – what’s known as the ‘substitution effect’. On the other hand, tax reductions make working households wealthier, which could lead them to reduce their work effort – the ‘income effect’.

The relative strength of these effects is ambiguous and could vary from person to person. If some people have relatively stronger substitution effects, while others have stronger income effects, PRWORA could induce a ‘reshuffling’ of earnings levels among households. Such reshuffling can make it difficult to infer how a policy reform affects behavior.

Suppose, for example, that we conduct an experiment that randomly assigns women on welfare to either a high tax policy H or a low tax policy L. Inspecting the results of the experiment, we find that under both policies, half the women earn $100 a month while the other half earn $200 a month. What should we conclude from this evidence about the behavioral responses to a reduction in tax rates?

The simplest interpretation would be that such a reduction yields no effect on anyone’s earnings. But suppose the women who earn $100 a month under policy H earn $200 a month under policy L (the substitution effect dominates), while the women who earn $200 a month under policy H earn $100 a month under policy L (the income effect dominates). Switching from policy H to policy L changes everyone’s earnings level by $100 but leaves the distribution of earnings unaffected.

Jobs First successfully encouraged many women to work but it also encouraged some women to reduce their earnings

Because our experiment cannot tell us what any individual would do if exposed to the other policy regime, there is no way to distinguish between these two interpretations of the results. This is a problem because evidence of any behavioral response to reform, even reshuffling, would typically be taken to indicate that the tax reduction increased economic efficiency. For this reason, the magnitude of such ‘intensive margin’ responses is a key input to the optimal design of welfare and other transfer programs (Saez, 2002).

Our research analyzes data from a large-scale randomized evaluation of Connecticut’s ‘Jobs First’ implementation of PRWORA. We develop a new approach to measuring earnings responses to welfare reform that avoids invoking ad hoc assumptions about the skills, opportunities or preferences of the people being analyzed.

Our approach exploits elementary economic reasoning to restrict the set of earnings responses consistent with rationality. Exploiting these restrictions, we find that while Jobs First successfully encouraged many women to work, it also encouraged some women to reduce their earnings.

The Jobs First experiment

Between December 1996 and February 1997, a sample of welfare applicants and recipients were assigned at random either to Connecticut’s pre-reform welfare rules or to the new Jobs First rules. The assignment lasted for five years during which the welfare participation and earnings histories of the individuals involved in the experiment were recorded.

Connecticut’s Jobs First policy lowered the effective tax rate on earnings to zero. But to ensure that the program remained means-tested, households became ineligible for assistance when their earnings exceeded the poverty line.

Figure 1 depicts the benefit schedule under pre-reform rules and Jobs First rules. Jobs First’s unreduced grant amount provides an incentive to work while on assistance. But the abrupt discontinuation of benefits at the federal poverty line provides strong incentives for women with higher earnings opportunities to take jobs paying below the poverty line.

Figure 1: Monthly earnings plus welfare transfer under pre-PRWORA and post-PRWORA (Jobs First) policy rules


Notes: The figure (not drawn to scale) depicts the sum of monthly earnings and welfare transfers under, respectively, pre-PRWORA policy rules (dashed blue line) and Jobs First policy rules (dashed red line) for a woman who in 1997 had two children and had been working on welfare for more than four months.

This woman’s monthly federal poverty line amounted to $1,111 and her base welfare grant amounted to $543. Under the pre-PRWORA rules, her welfare transfer phased out smoothly with earnings and was exhausted at $834. In contrast, the woman’s Jobs First welfare transfer was constant at $543 but fell to zero at earnings levels above the federal poverty line. For example, if the woman earned $200 per month, she would have received a welfare transfer that was $80 (17%) larger under Jobs First than under pre-PRWORA rules; if she earned $800 per month, she would have received a transfer that was $518 (2,098%) larger under Jobs First than under pre-PRWORA rules.
[pull_out]Welfare opt-in behavior – reducing earnings to gain eligibility for benefits – is surprisingly common[/pull_out]

Visually, the Jobs First reform induced an upward pivoting of the woman’s budget line past the $90 earnings amount that was also disregarded under pre-PRWORA rules. Notice that under Jobs First, if the woman earns between the federal poverty line and $1,654, she can increase the sum of earnings and welfare transfers by taking up welfare assistance and earning less.

The purple arrows represents instances of ‘welfare opt-in’, whereby a woman who would earn above the federal poverty line under the pre-PRWORA rules responds to the reform by reducing her earnings in order to establish eligibility.

Hence, in addition to encouraging some women to work, Jobs First may have also had the unintended consequence of leading some women to reduce their earnings in order to gain eligibility for welfare benefits – what’s known as the ‘welfare opt-in’ effect (Ashenfelter, 1983).

By virtue of the randomized assignment of households to policy regimes, we can compute the effect of welfare reform on the entire distribution of earned income (Bitler et al, 2006). But measuring what fraction of women lowered their earnings to establish eligibility for welfare benefits is a more difficult endeavor. Because each woman either faced Jobs First or pre-reform rules, we cannot directly infer what she would have done had she been exposed to the policy to which she was not assigned. In other words, data on her counterfactual choice are missing.

Revealed preference restrictions

In a nutshell, our study simplifies this ‘missing data’ problem by ruling out labor supply responses that would be irrational given the incentives provided by the Jobs First reform. To deduce which responses are irrational, we rely on the idea of revealed preference.

The logic of revealed preference is quite simple. Suppose you are choosing between two cars – a minivan and an SUV – and that (at current prices) you prefer the SUV. Revealed preference says that to be persuaded to buy the minivan, you would either need the price of the minivan to fall or the price of the SUV to rise. Specifically, there would be a violation of revealed preference if you could be persuaded to buy the minivan by raising its price or by lowering the price of the SUV.

Violating revealed preference is clearly irrational. And while there is good evidence that violations sometimes occur for complicated choices involving uncertain gambles (Choi et al, 2014), they are thought to be rare for more straightforward decisions. In fact, many apparent violations of revealed preference derive from non-equivalent comparisons over time (Blundell et al, 2008). For example, you may prefer the SUV one year but the next year you have an extra child, which leads you to choose the minivan when facing the same prices.

In our setting, the options in question are how much to work and whether to participate in welfare given the policy rules. By studying the results of a randomized experiment, we place restrictions on what options a given woman would pair across the two experimental policy regimes at a given point in time.

Ruling out irrational responses, we find that the impact of reform on the number of women with earnings above the federal poverty line can serve as a robust indicator of opt-in behavior. Specifically, a negative impact signals welfare opt-in.

Sharp provisions for phasing out benefits can significantly depress the earnings of disadvantaged people

To develop an intuitive understanding of this result, consider a ‘nearly poor’ woman who would forfeit assistance under pre-reform rules by earning 10% above the poverty line. We can group her possible responses to welfare reform into three categories:

  • First, she could continue to earn 10% above the poverty line under Jobs First.
  • Second, she could reduce her earnings below the poverty line in order to receive welfare benefits under Jobs First (the purple arrows in Figure 1).
  • Third, she could reduce (or increase) her earnings while remaining off assistance.

Note that the third option was available under pre-reform rules. Presumably, welfare reform didn’t affect the relative value of working different amounts while off welfare. Therefore, the fact that the woman earns 10% above the poverty line under pre-reform rules reveals that she prefers that level of earnings to the other available earnings options.

Thus, engaging in the third kind of response would violate revealed preference and would also presumably be costly. Ruling this third response out implies that opt-in responses are the only way of rationalizing reform-induced reductions in the fraction of women with earnings above the poverty line.

When applying this reasoning to the Job First evaluation, we have to contend with some complications arising from how welfare policy is implemented in practice. One is that some women may respond to the pressure to find a job by leaving welfare entirely and earning amounts above the poverty line – note that this would lead to an increase in the number of women earning above the poverty line.

Another complication is that some women drawing assistance turn out to have administrative earnings above the federal poverty line, presumably because they under-reported their earnings to the welfare agency.

These complications imply that we cannot estimate the fraction of nearly poor women who opted-in to welfare. But we can obtain a lower bound on the fraction of women who respond to welfare reform in this manner.

Our lower bound comes from focusing on the impact of reform on the fraction of women with earnings above the poverty line that are not receiving assistance. If the impact is negative, it means that more of these women were induced to opt-in to welfare than were pushed off welfare into earning amounts above the poverty line.

Empirical findings

Applying our revealed preference approach, we find that welfare opt-in behavior is surprisingly common: at least one fifth of the women in the experiment who would have earned above the poverty line reduced their earnings in response to the Jobs First reform.

Moreover, the magnitude of the earnings reductions involved was often substantial: we find that at least 17% of women who would have earned amounts greater than 1.2 times the poverty line were induced by the Jobs First reform to lower their earnings below the poverty line. Since these figures are only lower bounds, they suggest that earnings reductions are potentially a common response among welfare recipients capable of lifting themselves out of poverty.

Understanding individuals’ behavioral responses to tax changes is key to the design of welfare reform

By eliminating benefit taxes on low earnings amounts, Jobs First also encouraged many women who would have worked off welfare at low earnings levels under pre-reform rules to receive assistance under Jobs First. We find that at least 32% of these ‘working poor’ women were induced to take up assistance by reform.

On the other hand, the work requirements associated with welfare reform may have led some women to be kicked off welfare because they failed to search actively for work. Such responses are potentially a concern if the women involved have work disabilities making them unable to work. But we find that at most 14% of the women who would have been on the welfare rolls under pre-reform rules were induced by reform to leave welfare.

Policy relevance

While our findings are specific to the sample of women in Connecticut’s Jobs First experiment, the welfare opt-in results indicate that sharp provisions for phasing out benefits can significantly depress the earnings of disadvantaged people even when the net effect of the program is to get more people working. These earnings reductions are inefficient insofar as they cost taxpayers money and trap low-skilled workers in jobs that they otherwise wouldn’t want.

An important question for policy-makers is whether these inefficiencies can be diminished by adopting smooth benefit phase in and phase out provisions such as those specified by the Earned Income Tax Credit (EITC). By phasing benefits out more gradually, such schemes replace a large distortion concentrated over a small number of relatively high earners with a small distortion spread over a larger number of people.

There are reasons to suspect that such a tradeoff could be worthwhile: economists often find disproportionately large behavioral responses to stronger incentives (Chetty, 2012). This could be because the program rules that generate strong incentives (for example, the Jobs First eligibility thresholds) are more salient or because households deem adjusting to weaker incentives to be ‘not worth the trouble’.

More research on these questions is necessary to inform the optimal design of welfare and other transfer programs.

This column summarizes ‘Bounding the Labor Supply Responses to a Randomized Welfare Experiment: A Revealed Preference Approach’ by Patrick Kline (University of California, Berkeley) and Melissa Tartari (University of Chicago),  American Economic Review, 106 (4): 972-1014

Further reading

Ashenfelter, Orley (1983) ‘Determining Participation in Income-Tested Social Programs’, Journal of the American Statistical Association 78(383): 517-25.

Blank, Rebecca M (2002) ‘Evaluating Welfare Reform in the United States’, Journal of Economic Literature 40(4): 1105-66.

Blundell, Richard, Martin Browning, and Ian Crawford (2008) ‘Best Nonparametric Bounds on Demand Responses’, Econometrica 76(6): 1227-62.

Bitler, Marianne, Jonah Gelbach, and Hilary Hoynes (2006) ‘What Mean Impacts Miss: Distributional Effects of Welfare Reform Experiments’, American Economic Review 96(4): 988-1012.

Chetty, Raj (2012) ‘Bounds on Elasticities with Optimization Frictions: A Synthesis of Micro and Macro Evidence on Labor Supply’, Econometrica 80(3): 969-1018.

Choi, Syngjoo, Shachar Kariv, Wieland Muller, and Dan Silverman (2014) ‘Who is (More) Rational?’, American Economic Review 104(6): 1518-50.

Grogger, Jeffrey, and Lynn Karoly (2005) Welfare Reform: Effects of a Decade of Change, Harvard University Press.

Meyer, Bruce, and Dan Rosenbaum (2001) ‘Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers’, Quarterly Journal of Economics 116(3): 1063-1114.

Meyer, Bruce (2002) ‘Labor Supply at the Intensive and Extensive Margins: the EITC, Welfare, and Hours Worked’, American Economic Review 92(2): 373-79.

Saez, Emmanuel (2002) ‘Optimal Income Transfer Programs: Intensive Versus Extensive Labor Supply Responses’, Quarterly Journal of Economics 117(3): 1039-73.