Incentives and the Work Decisions of Welfare Recipients

Kevin Duncan

KEVIN DUNCAN [*]

Evidence from the Panel Survey of Income Dynamics, 1981-1988

ABSTRACT. Can work replace welfare? This important policy issue is examined through evidence derived from the Panel Survey of Income Dynamics (1981-1988). Results indicate that welfare recipients supplement the earnings from low paying occupations with public assistance. Further, this income strategy responds to changes in economic incentives. For example, higher wages are associated with a greater probability of working without a welfare subsidy while lower wages are associated with a higher probability of mixing welfare and work. These results underscore the importance of liveable wage levels in providing for a durable substitution of work for welfare. The results also support Solow's recommendation of "packaging" welfare and work for those who find their occupational choices limited to the lower end of the labor market.

I

Introduction

THE ONGOING WELFARE POLICY DEBATE is split between the view that welfare use breeds dependence and the belief that welfare use, and poverty in general, can be attributed to limited labor market opportunities [see Harrison (1978), Mead (1986), Murray (1984) and Solow (1998)]. Current policy, with emphasis on work requirements and time limits, is consistent with the view that welfare use results in dependency and the best way to end dependency is for recipients to work their way off welfare. However, several studies have found that those who work their way off welfare experience high rates of recidivism. For example, Bane and Ellwood (1986) find that 40 percent of those who leave welfare through work return for another welfare spell. [1] These authors also find that another 40 percent of recipients who worked their way off welfare have incomes below the poverty line in the years after leaving welfare. These findings imply that public policy should not focus exclusively on employment as welfare reform, since mer ely having a job does not guarantee an end for the need, or use of welfare.

Harrison (1978, 1972) provides an explanation of the relationship between the jobs that welfare recipients are likely to hold and repeated welfare use. Put simply, Harrison's explanation is based on the proposition that there are "good" jobs, which offer high pay and employment stability, and there are "bad" jobs, characterized by low wages and high turnover. Since there are not enough "good" jobs to go around, someone will always have to work the "bad" jobs. Further, since discrimination limits access to well paying jobs, some (particularly blacks and women) have constrained labor market choices. Since low paying and unstable jobs do not by themselves adequately provide for a family's needs, the focus is on the income strategies individuals employ. According to Harrison, individuals with limited labor market options will develop strategies of mixing welfare and work to provide for subsistence. This analysis is useful to the ongoing policy debate because welfare is viewed as more than a function of inadequat e work related skills and family choice. It is also a symptom of the occupational structure. Also, this analysis suggests that replacing welfare with work may prove inadequate if the jobs welfare recipients fill do not provide for subsistence.

Recent studies on the working poor help to illustrate Harrison's views on welfare use. For example, Gardner and Herz (1992) note that for about half of the poor women maintaining families, it is some combination of low pay, insufficient hours of work, or periodic unemployment that pushes their income below the poverty line. Faced with these employment and earnings prospects, Harrison's analysis implies that there are two motivations for individuals to mix welfare and work. Welfare can be used to fill an income gap caused by a spell of unemployment. As empirical support for this kind of income strategy, Harrison finds that 92 percent of the respondents from the Panel Survey of Income Dynamics (PSID) who received AFDC also worked at some time over the 1968 to 1972 period. The present study uses the PSID to examine the work and welfare decisions of AFDC recipients over the 1980-1987 period and finds that 80 percent of those who received welfare also worked at some time. While the updated percentage measuring the extent of mixing strategies remains high, it suggests a decrease in the use of this strategy over time. This change can be attributed to the enforcement of the Omnibus Budget Reconciliation Act of 1981 (OBRA) which significantly affected the incentives of this income strategy. The impact of this policy is discussed in greater detail below.

Individuals employed in low paying jobs may also use welfare to supplement their earned income. Data from the subsample of the PSID respondents who participated in welfare at least once over the 1980-1987 period indicate that when individuals worked without a welfare supplement, they earned an average real wage of $5.36 per hour (in 1982 dollars). This wage rate is 67 percent of the $7.99 average real wage earned by U.S. private, non agricultural workers over this period. However, those who mixed welfare and work held jobs offering an average real wage of $4.05 which is 51 percent of the U.S. average real wage for this period. [2] These wage differences not only illustrate the incentive to mix welfare and work, but also point out the need for supplemental income for those who hold low paying occupations.

The remainder of the paper is organized along the following lines. The data used in the empirical analysis is discussed in section II. As mentioned above, The Omnibus Budget Reconciliation Act of 1981 included significant welfare policy changes that, in particular, affected work incentives and the income strategies of welfare recipients. The implications of this policy are discussed in section III. Harrison's analysis implies that the work and welfare decisions are affected by wage and AFDC benefit levels. In sections IV and V these implied hypotheses are tested by examining the effect of wage and AFDC levels on the probabilities that a welfare participant will work without AFDC, receive only AFDC or combine the two. Finally, the paper concludes with a discussion of policy recommendations and implications for future research.

II

Data

AS IS THE CASE WITH THE STUDIES BY BANE AND EllWOOD AND HARRISON, the data for this study are derived from the Panel Survey of Income Dynamics (PSID). The PSID contains human capital, income and demographic information on over 7000 families. Over sampling of low income families has made this longitudinal sample well suited for tracking the experiences of welfare recipients. This Study uses a subsample of female headed households who received one dollar or more of AFDC income between 1980 and 1987 (corresponding to PSID survey years 1981-1988) and who participated in the PSID every year over the period. [3]

III

Policy Changes, 1980-1987

SINCE THE PERIOD OF THE STUDY OVERLAPS the enforcement of the Omnibus Budget Reconciliation Act of 1981 (OBRA), the effect of this policy on the work behavior of welfare recipients can be examined. OBRA reduced the cost of AFDC by reducing the number of individuals who could qualify for the program. This was achieved, in part, by increasing the benefit-loss rate which indicates the reduction in the basic welfare benefit for those recipients who also receive income from work. [4] In addition to affecting program costs, changes in this rate have a profound effect on the work incentives of welfare recipients. The example presented in Table 1 illustrates how AFDC subsidies were determined before and after OBRA and how changes in benefit-loss rate affected work incentives and the income strategies of working welfare recipients. Assume that the state in question offers a basic AFDC benefit (for nonworking recipients before OBRA) of $300 per month. If a recipient works, the basic benefit is reduced or "taxed" based on a percentage (the benefit-loss rate) of the earned income. The first "$30 and one third" of earned income was exempt from the "tax." So, with an earned income of $200 per month, "taxable" income was $103 (or $200 - $30 - $67 = $103). With this "taxable" income and a benefit loss rate of .67 the "tax" or reduction in the basic benefit is $69 (or $103 x .67 = $69). So, the actual AFDC subsidy for this working recipient is $231 (or $300 - $69 = $231). This is added to the wage income of $200 for a total income from work and welfare of $431. This example demonstrates the logic and incentive behind mixing welfare and work before OBRA. Those working at a low paying job could increase their family income by also receiving welfare. Or, from the other point of view, those on welfare could increase total income by working.

Next, consider total income determination after OBRA. This act limited the "$30 and one third" exemption to four months and increased the benefit-loss rate to 100%. [5] Note that working through the example in the manner described above, yields a lower income ($397) from work and welfare during the first four months when the exemption applies. This reduction is due to the higher benefit-loss rate (100%) which results in a higher benefit "tax" ($103 instead of $69). Also, when the four-month exemption expires total income from welfare and work decreases to $300 which is exactly the income level if the recipient had not worked at all and only received the basic AFDC benefit. Therefore, OBRA significantly reduced the incentive to work while receiving AFDC as total income could be increased through mixing only for a short period of time.

This example suggests a decrease in the mixing of welfare and work corresponding to the enforcement of OBRA. Table 2 reports the distribution of welfare recipients by source of income for each year between 1980 and 1987. [6] OBRA took effect in 1982 and the data indicate that those who received income from (or mixed) welfare and work were the ones to leave welfare. In 1982 the percent of the PSID sample who mixed decreased, the number working increased and the percent receiving only AFDC remained unchanged. Mixers who had incomes near the threshold level, where an individual no longer qualifies for assistance, may have been better off to leave welfare through work rather than remain on aid and only receive the basic benefit. While OBRA eliminated the incentive to simultaneously mix welfare and work, relying on welfare during spells of unemployment remained a viable income strategy. This may explain why some (18 percent) of the sample continued to receive AFDC and wage income in 1982.

The economic expansion and other OBRA provisions can explain the decrease in the percent receiving only AFDC and the increase in the percent working beginning in 1984. Economic expansion increased job opportunities which were associated with more work and mixing relative to receiving only AFDC. OBRA also allowed states to require work in return for welfare (workfare). As more states took advantage of this option after 1982 the expected result is reduced AFDC use. [7] To separate the effect of policy changes and business cycle trends and to measure the effect of other factors which contribute to the labor market or welfare choices of an individual, the results of logit estimations are discussed in the following section.

IV

Logit Models

HARRISON'S ANALYSIS IMPLIES that the strategy of mixing welfare and work is affected by wage and AFDC benefit levels. That is, lower AFDC levels encourage more mixing and higher wages encourage more work without welfare. To estimate the effect of AFDC benefits and wages on the relative probabilities of receiving only AFDC, mixing or working without welfare the PSID data were pooled over the 1980-1987 period. The general forms of the logit equations used to predict the activity of female AFDC recipients are:

[ln.sub.e] ([P.sub.1i]/[P.sub.0i]) = [[beta].sub.0] + [[beta].sub.1] Real [Wage.sub.i] + [[beta].sub.2] [Z.sub.i] + [[beta].sub.3] [X.sub.i] (1)

[ln.sub.e] ([P.sub.2i]/[P.sub.0i]) = [[beta].sub.0] + [[beta].sub.1] Real [AFDC.sub.i] + [[beta].sub.3] [Z.sub.i] + [[beta].sub.4] [X.sub.i] (2)

where [P.sub.0i] refers to the probability that the ith respondent mixed welfare and work (received wage and AFDC income in the same year). [P.sub.1i] refers to the probability of working (without AFDC). [P.sub.2i] refers to the probability of receiving only AFDC (without work). Real Wage, from equation 1, is the inflation adjusted hourly wage earned by the individual(1982 as the base year). Similarly, Real AFDC, from Equation 2, is the inflation adjusted AFDC income received when the individual was on welfare. Z is a vector of respondent characteristics that affect labor market participation and X is a vector of factors, such as time, that measures policy changes. Employment conditions that influence welfare participation are also included in X. Hence, these specifications provide an opportunity to test the effect of wage and AFDC levels on work and AFDC decisions, holding all other factors constant.

The equations were estimated using a sub-sample of PSID respondents who received at least one dollar of AFDC income over the 1980-1987 period. The results from this sample are reported below. Equation 1 was also estimated using a sample of those PSID respondents who did and did not participate in AFDC. While these results are not reported, they are discussed. The equations were also estimated with dummy state of residence variables to control for differences in welfare policies between states. The results of these estimations, while not presented here, are also discussed below.

V

Results

THE MEANS OF THE INDEPENDENT VARIABLES are reported in Table 3. [8] These means are for individuals in each of the categories (working only, mixing welfare and work and receiving AFDC benefits only). Results reported in Table 3 suggest that welfare recipients who work are older and are more likely to have been married but less likely to have young children, reside in an SMSA and to be black. These individuals are also more likely to reside in states with lower unemployment rates, to have high school degrees and to have experienced fewer years of assistance. The means for years of AFDC across the categories indicate movement between welfare and labor market states over the period.

Coefficient estimates for the logit model are reported in Table 4. A coefficient under the column [P.sub.1]/[P.sub.0] (Equation 1) can be interpreted as a change in the natural log of the odds ratio of working relative to mixing welfare and work, given a one unit change in the independent variable. The coefficient for the real wage is positive and significant at the .01 level indicating that an increase in the real wage increases the likelihood that a welfare recipient will work without a welfare subsidy. This finding did not change when the estimate of equation 1 included additional independent variables for the respondent's state of residence. The odds ratio was also estimated using a sample of PSID respondents who did and did not participate in AFDC over the period. Results of this estimation are similar to those reported here.

This result is consistent with studies by Boskin and Nold (1975), Hutchens (1981) and Plotnik (1983) who report that higher wage rates are associated with lower probabilities of welfare use and a greater probability of working. However, none of these studies examine the relation between wages, work and mixing welfare and work. The results reported above identify mixing as an important step in the transition on and off welfare.

The remaining results reported under [P.sub.2]/[P.sub.0] indicate the effect of other labor force participation variables on the relative odds of working and mixing. Age, having never married, SMSA residence, a high school degree, race and the unemployment rate are not significantly related to this odds ratio. However, the presence of young children and another year of welfare are associated with a lower probability of working relative to mixing. The coefficients and t-values for the linear and quadratic terms for Years AFDC indicate that another year of AFDC decreases the probability of working relative to mixing at a decreasing rate. [9]

Since the rate of unemployment (measuring the business cycle) is held constant, the year dummy variables capture the effect of factors such as policy changes or welfare preferences (or stigma) that change over time. [10] The arrangement of the year dummy variables (1982 is the reference year) allows us to observe changes in the odds ratio before and after OBRA legislation took effect. Results from the year dummy variables indicate that the odds of working relative to mixing are significantly lower in 1980 and 1981 relative to the base year of 1982. After 1982 the odds ratios are higher and the difference is statistically significant beginning in 1984. This finding is consistent with the results discussed above (see Table 2) which indicate that OBRA had the effect of encouraging the substitution of work for mixing welfare and work. Also, changes in the magnitude and significance level of the dummy year variables after 1982 suggest that changes in welfare stigma and other OBRA provisions had the effect of incr easing the odds of working. Indeed, the odds of working relative to mixing welfare and work is very sensitive to time. Solving the equation with the independent variables set at their means and year equal to 1982 yields a relative odds ratio of .56 in that year. Holding all else constant, the relative odds of working increases to .94 in 1987.

The column of coefficients under [P.sub.2] /[P.sub.0] in Table 4 refers to the natural log of the odds ratio of receiving AFDC only and mixing welfare and work. An increase in AFDC income is associated with an increase in the odds of receiving only AFDC relative to mixing. This finding is consistent with other studies that have examined the effect of benefit guarantees on welfare participation. For example, Ermisch and Wright (1991) find that a higher benefit guarantee reduces the probability that a single parent will work. The results reported above indicate that as benefits rise, individuals reduce mixing actively and are increasingly likely to receive only AFDC. Once again, this result identifies the role of mixing in the transition on and off welfare. This finding did not change when we estimated the odds ratio using different specifications of equation 2 which included the dummy variables for state of residence to control for differences in welfare policies between states.

The other variables, measuring labor force participation, have a mixed effect on the odds of receiving only AFDC relative to combining welfare and work. For example, older recipients are significantly more likely to receive only AFDC while those with a high school degree are significantly less likely to do so. Also, an increase in the rate of unemployment of the state of residence is associated with a significantly higher odds ratio. Only at a lower level of confidence can we say that more years of AFDC are associated with a greater likelihood of receiving AFDC relative to mixing. Finally, the results for the year dummy variables indicate that the odds ratio for years other than 1982 is not significantly different than the base year. The higher benefit-loss rate introduced in 1982 discouraged mixing and had the effect of creating an incentive to move into some other welfare or labor market state. The results of the year dummy variables from equations 1 and 2 indicate that this "tax" change introduced by OBRA had the effect of moving recipients into work instead of receiving only AFDC.

The logit equations were also estimated with multiplicative terms (between the Real Wage and Years of AFDC and Real AFDC and Years AFDC)to test hypotheses from the welfare dependence perspective. For example, AFDC use may change the labor-leisure preferences of recipients in such a way that individuals come to prefer AFDC income to wage income. If this is the case, the effect of the real wage on the relative odds of working and mixing will be a decreasing function of years on welfare. According to the labor-leisure explanation offered above, the coefficient for the interaction term between Real Wage and Years AFDC is expected to be negative. The coefficient for the interaction term is negative, but fails to achieve conventional levels of statistical significance. [11] This result indicates that, given a wage rate, the willingness to work does not vary with time spent on welfare.

Similarly, the welfare dependency view suggests that time on welfare may also shift labor-leisure preferences away from mixing welfare and work and toward receiving only AFDC. If this is the case, the effect of AFDC on the odds of receiving AFDC relative to mixing should be an increasing function of time spent on welfare. Or, the interaction between Real AFDC and Years AFDC should be positive. However, the interaction between Years AFDC and real AFDC income is negative. [12] This finding is not consistent with the notion (stated above) that welfare participation alters the labor-leisure preferences away from work. On the contrary, it suggests that those who have spent more time on welfare are less likely to receive only AFDC relative to mixing. This result may be due to the desire of long term participants to maintain some work activity as part of their income strategy.

VI

Conclusions and Policy Implications

IN THE 1970s HARRISON EMPHASIZED the importance of a wage policy as a complement to welfare reform. This recommendation remains valid. Since working welfare recipients are likely to hold minimum wage jobs, establishing and maintaining a meaningful level of this wage rate would reduce dependency. Others have recommended wage subsidies to address the low wages of working welfare recipients. For example, Solow (1998) has criticized recent welfare reform on the grounds that work requirements will further depress the already low wages of unskilled workers. He recommends a policy of job creation along with the "packaging" of work with a welfare subsidy to provide an acceptable living standard for those receiving low wages. For Solow, such a policy is consistent with the pervasive social value of self-sufficiency and the reality of limited altruism. Snower (1994) and Phelps (1994) also recommend wage subsidies as a substitute for the welfare state. These studies and the present one suggest that if incentives, either in the form of higher wages or wage subsidies, are directed to work activity welfare recipients will, quite rationally, choose employment over public assistance.

Current policy, as reflected by the Personal Responsibility and Work Opportunity Act of 1996, requires participation in work activities in exchange for welfare. Similar to policy changes in the 1980s, work requirements will likely discourage welfare use as part of the income strategy of welfare recipients. If so, Solow's warning of former welfare recipients entering already saturated conditions at the lower end of the labor market is of particular concern for those interested in reducing poverty and welfare use. While the economic expansion of the 1990s provided some of the income and employment needs of those moving off the welfare roles, an important question for future research, and for policy makers, concerns the effectiveness of work requirements in reducing welfare use and poverty in an economic environment characterized by slower job growth.

Finally, high rates of welfare recidivism have been noted in the recent literature. Outside of measuring the time periods between welfare spells, little has been done to examine the cause of this behavior. Job characteristics likely influence recidivism and future research should examine the role of wages, type of occupation and employment instability in explaining recurrent welfare use.

(*.) Kevin Duncan is an associate professor of economics at the University of Southern Colorado, Pueblo, CO 81001. E-mail: duncan@uscolo.edu. Professor Duncan's specialities are occupational sex segregation, earnings disparity, and welfare use. He published "Gender Differences in the Effect of Education on the Slope of Experience-Earnings Profiles," The American Journal of Economics and Sociology 55 (1996):457-471. He is currently working on a paper entitled "Child Quality and Racial Earnings Differences."

Notes

(1.) Weeks (1991) also finds high rates of recidivism (35 percent) among welfare participants in Washington state. Finally, Blank and Ruggles (1994) find that 20 percent of all welfare spells end with a relatively quick return (6-9 months) to public assistance.

(2.) Mixing welfare and work is defined as receiving AFDC and wage income in the same year. It is not possible with the PSID to determine if wage and AFDC income is received simultaneously. However, the difference in wages between those who work and those who mix strongly suggests that those receiving $4.05 per hour are also receiving AFDC to supplement their income.

(3.) In the present study a welfare recipient is defined as an individual who received AFDC income and does not include those who received other forms of public assistance.

(4.) The benefit-loss rate measures the reduction in AFDC benefits as income from work increases. Prior to 1982 this rate was .67 indicating that for every dollar earned from work, AFDC benefits decreased by $.67. In 1982 this rate increased to 1.00 indicating that benefits decreased by one dollar for every dollar earned. The income level where a recipient no longer qualifies for assistance is given by the ratio of the basic benefit to the benefit-loss rate. Increasing the benefit-loss rate has the effect of lowering the level of the qualifying income. Hence, fewer qualify for assistance and program costs are reduced.

(5.) For ease of presentation the numerical example assumes that the "30 and one third" exemption expires in four months. In fact, OBRA limited the "30" exemption to eight months and the "one third" disregard to four months. After these periods the implicit "tax" applies to all income. See Nightengale and Burbridge (1987) for a discussion of changes under OBRA.

(6.) Data in Table 2 reports the percentage distribution (by category and year) of the PSID subsample of welfare participants. An individual who participated in AFDC at some time over the 1980-1987 period could have received income from working (without welfare) mixing welfare and work, or received only AFDC.

(7.) Shiller and Brasher (1993) find that the adoption of workfare policies by states after 1982 had the effect of reducing welfare use.

(8.) The logit equations also include the square of Years AFDC and year dummy variables. The averages of these variables are not reported in Table 3.

(9.) Years of AFDC is derived by summing the number of years that a respondent received at least $1.00 of AFDC. Since we do not use the PSID prior to 1980, we do not include information on assistance prior to 1980, consequently if those who received assistance tend to have a long history (more than eight years) of welfare, the analysis will overstate the effect of another year of AFDC on the odds ratios. This problem is unavoidable for even if the PSID was used from its inception (1968) we would still not be able to measure the welfare experience of individuals whose activity precedes 1968.

(10.) See Moffit (1983) for a discussion of welfare stigma.

(11.) The coefficient and t-value (in parentheses) for the interaction term between Real Wage and Years AFDC is reported below:

[delta](Prob. Work)/[delta](Prob. Mix) = .140 - .011 Years AFDC (2.28) (- .69)

(12.) The coefficient and t-value (in parentheses) for the interaction term between Real Wage and Years AFDC is reported below:

[delta](Prob.AFDC)/[delta](Prob. Mix) .0004 - .0001 Years AFDC (5.45) (-3.13)

References

Bane, M. and D. Ellwood, The Dynamics of Dependence: The Routes to Self-Sufficiency. Washington, DC(1983): Department of Health and Human Services.

Blank, E. and P. Ruggles, "Short-Term Recidivism Among Public Assistance Recipients", American Economic Review 84 (1994): 49- 53.

Boskin, M. and F. Nold, "A Markov Model of Turnover in Aid to Families with Dependent Children", Journal of Human Resources 10 (1975): 467-81.

Ermisch, J. and R. Wright, "Welfare Benefits and Lone Parents' Employment in Great Britain", Journal of Human Resources 26 (1991): 424-56.

Gardner, J. and D. Herz," Working Poor in 1990", Monthly Labor Review (December, 1992): 20-27.

Harrison, B., "How American Households Mix Work and Welfare", Challenge 21 (1978): 49-54.

-----, Education, Training and the Urban Ghetto. Baltimore: (John Hopkins University, 1972).

Hutchens, R., "Entry and Exit Transitions in Government Transfer Program: The Case of Aid to Families with Dependent Children", Journal of Human Resources 16 (1981): 317-37.

Mead , L., Beyond Entitlement: The Social Obligations of Citizenship. New York: (Free Press, 1986).

Moffitt, R., "An Economic Model of Welfare Stigma", American Economic Review 73 (1983): 1023-1035.

Murray, C., Losing Ground: American Social Policy 1950-1980. New York: (Basic Books, 1984).

Nightingale, D. and L. Burbridge, The Status of Work-Welfare Programs in 1986: Implications for Welfare Reform. Washington, DC (1987): Urban Institute.

Nord, S. and R. Sheets, "The Relation of AFDC Payments and Employment Structure on the Labour Force participation and underemployment Rates of Single Mothers", Applied Economics 22 (1990): 187-99.

Phelps, E., "Low-Wage Employment Subsidies Versus the Welfare State", American Economic Review 84 (1994): 54-58.

Plotnik, R.," Turnover in the AFDC Population: An Event History Analysis", Journal of Human Resources 18 (1983): 65-81.

Schiller, B. and N. Brasher, "Effects of Workfare on AFDC Caseloads", Contemporary Policy Issues 11 (1993): 39-48.

Snower, D., "Converting Unemployment Benefits into Employment Subsides", American Economic Review 84 (1994): 65-70.

Solow. R., et al., Work and Welfare. Princeton, NJ: (Princeton University, 1998).

Weeks, G., Leaving Public Assistance in Washington State. Evergreen State College (April, 1991): Washington State Institute for Public Policy.

COPYRIGHT 2000 American Journal of Economics and Sociology, Inc.
COPYRIGHT 2000 Gale Group