May 9, 1997
A Report by the
|ln Rst = Ust 1 + Wst 2 + ln Bst 3 + s + t + st||(1)|
|ln Rst = Ust 1 + Wst 2 + ln Bst 3 + s + t + trend*s + st||(2)|
where R represents the share of the population receiving AFDC, U is the unemployment rate, W is an indicator variable for welfare waiver status, B represents real maximum AFDC benefits in 1996 dollars for a three-person family, s indexes states, t indexes time, s and t represent state and year fixed effects, and represents a residual. Year fixed effects capture time-varying factors that affect all states in a given year. Such factors might include changes in welfare policy (like OBRA 1981), other changes in policies targeted to low-income individuals (like the Earned Income Tax Credit), or changes in national attitudes regarding welfare receipt that may have been linked to the welfare reform debate.14 This approach incorporates the contribution of factors like these, although we cannot specifically identify the effects of each one on the rate of welfare receipt. Similarly, state fixed effects control for time-invariant differences across states, such as differences in industrial composition that may affect less-skilled workers or attitudes towards welfare recipients.
As shown earlier, it is also possible that changes may be occurring over time in otherwise unmeasured factors that differ across states, particularly demographic characteristics like the share of female-headed households. Unfortunately, published data on detailed demographic characteristics such as these are unavailable at the state level each year. Such differences could be fully accounted for by including the interaction of state and year fixed effects, but a model including these interactions is under-identified. As an alternative, we include a state-specific time trend. If the rate of increase in, say, female-headed households in a state is constant, this approach will control for these changes and provide an unbiased estimate of the effects of waivers and economic conditions on the welfare rolls.15 The effects of such changes, however, cannot be separately identified.
Figure 4 presents a comparison of Florida and Georgia that is intended to provide some intuition for the statistical methodology and the manner in which the effects of economic activity are estimated separately from other potential confounding factors. It should not be considered a rigorous test. The figure plots the difference between the two states in unemployment rates between 1984 and 1996 and in the share of the population receiving AFDC over the same period. Taking the difference between the two states in each year controls for any differences that affect both states simultaneously. Because neither state received a waiver until late in the 1996 fiscal year, the difference in trends through virtually all of this time period are unaffected by differences in waiver provisions or their effectiveness.
Throughout most of the expansion of the middle to late 1980s, unemployment in Georgia had been somewhat higher than in Florida. Over this period, a steady difference in the rate of AFDC recipiency is also apparent. This difference may be attributed to differences in the two states' welfare systems that do not change over time, attitudes towards welfare receipt and the like that are controlled for in the analysis conducted here. When the 1990-91 recession hit, unemployment in Florida rose considerably relative to that in Georgia, and the difference has been slow to recede. Subsequently, AFDC receipt shows an increase in Florida relative to Georgia. It is important to note that a delay in this response is apparent as Florida's AFDC caseload did not begin to rise relative to Georgia's until 1991 or 1992. This timing of the response in the rate of AFDC receipt to changes in unemployment (and waivers) will be examined more carefully in the empirical analysis below.
Table 2 presents estimates from different statistical specifications based on the regression models represented by equations (1) and (2). In column 1, the model does not include state-specific linear time trends and provides a baseline set of estimates to identify the effect of including these trends. In this model, the unemployment rate is shown to have a substantial effect on the rate of welfare receipt; a one percentage point increase in the unemployment rate increases the rate of welfare receipt by almost 5 percent.16 States that were granted any major, statewide waiver had almost a 10 percent fall in the share of the population receiving welfare, based on estimates in this model. Finally, benefit generosity is shown to be significantly positively related to AFDC receipt; the share of the population receiving benefits increases by 3.2 percent for every 10 percent increase in maximum monthly benefit payments.
Column 2 presents estimates of the same specification except that state-specific linear trends are included. Omitting these trends will introduce bias if they are correlated with the rate of welfare recipiency and any of the other explanatory variables. Estimates presented here indicate that these conditions are present. As illustrated in Table 1, trends in factors like female-headed households and poverty rates across states are correlated with waiver status, and ignoring these trends biases the estimated effect of waivers upwards. The estimated effect of introducing a major, statewide waiver falls from 9.4 percent in column 1 to 5.8 percent in column 2. The estimated responsiveness of welfare receipt to unemployment is also smaller in this specification.
One surprising finding in this specification is that more generous benefits are estimated to reduce the welfare rolls, although this effect is not significantly different from zero.17 This finding is counterintuitive and is the result of the statistical procedure that has absorbed a significant share of the variability in the data. In a model with state and year fixed effects and state-specific linear trends, the only type of variation that can provide statistical identification are those resulting from sharp changes within a state over time in the respective variables. Changes like this are exactly what are observed in variables like unemployment and, particularly, in indicator variables like those representing waiver status. AFDC benefits generally exhibit little of this sort of behavior; typically benefit increases are small and benefit cuts largely occur as inflation slowly erodes the purchasing power of the benefit. Therefore, with little variation left to identify the effect of changes in AFDC benefits, the estimated effect becomes less robust. This becomes clear in the subsequent model specifications reported in this table where an increase in AFDC benefits is estimated to increase welfare receipt, although some of these effects are only marginally statistically significant. In essence, these results indicate that the methodology employed here is not a particularly powerful one to determine the effects of the generosity of AFDC benefits on the level of welfare receipt.
Estimates in column 3 are obtained from a model that includes a one-year lagged measure of the unemployment rate within a state, providing a more flexible specification of the timing of the response in welfare receipt to economic conditions. Lagged unemployment may be related to welfare receipt if, for instance, the onset of a recession leads those low-income workers who lose their jobs to spend some time looking for a new one while drawing down their limited assets before applying for welfare. As a recession ends, these typically less-skilled workers may be the last ones hired. Evidence appears to support this intuition, as lagged unemployment is strongly related to the share of the population receiving welfare. To interpret these findings, consider a 1 percentage point increase in the unemployment rate that lasts for two years. In the second year, the share of the population receiving welfare will be 4 percent larger (because the coefficients on the two unemployment measures are summed). States awarded a major statewide waiver are estimated to experience a 5.2 percent decline in welfare recipiency in this model.
So far, waivers have been aggregated into a simple indicator variable that measures whether any waiver had been approved. Column 4 presents estimates of the effects of each of the six major types of waivers studied in this analysis on the rate of welfare receipt. In this model, the only type of waiver that significantly affects the extent of welfare receipt is JOBS sanctions.18 This type of waiver is estimated to reduce the share of the population receiving welfare benefits by almost 10 percent.19 Disaggregation of the waiver categories did not substantially change the estimated impact of an increase in unemployment.
One potential shortcoming of the model presented in column 4 is that many waivers include several of the different types all at once, limiting the ability of the statistical analysis to separately identify their effects. Column 5 presents estimates of a more parsimonious model that includes whether the state received any major statewide waiver and whether that waiver included JOBS sanctions. In this specification as well, no other type of waiver is shown to have a significant effect on welfare receipt besides JOBS sanctions. Again, the responsiveness of the welfare rolls to the business cycle is relatively unaffected by the changes in waiver specification. The analysis reported so far has restricted the effect of waivers to be observed no sooner than the time the waiver was approved. This restriction does not allow for the possibility that the waiver application process, the publicity surrounding it, and potential changes in case workers' behavior and attitudes may provide a signal to potential recipients that the environment in which the welfare system operates is about to change. It may lead some individuals contemplating applying for benefits to find other sources of income support, whether from work or elsewhere. This possibility is considered in column 6, where the presence of any statewide waiver and those including a sanction provision are included in the model at the time the waiver was approved and, in separate variables, a year before the waiver was approved (a "lead").
Estimates of models including leads of the waiver measures are reported in Column 6 of Table 2. The "threat effect" of applying for a waiver does appear to reduce the number of individuals who receive benefits the year before the waiver is approved; the share of the population receiving welfare is estimated to fall by 6.3 percent in that year. In the following year no additional reduction is observed. On the other hand, the effect of waivers that include JOBS sanctions is not observed until the year such a waiver is approved.
One alternative to a causal interpretation of these findings is that those states which implemented waivers were among the ones that experienced the most dramatic run-up in their welfare rolls in the late 1980s and early 1990s. This trend may have inspired the waiver request and mean reversion may be responsible for the subsequent decline in the rate of welfare receipt relative to other states. Tests of this hypothesis, however, indicate that waiver states did not experience a larger-than-average increase in their welfare rolls between 1989 and 1993. In fact, little relationship across states is apparent between the 1989-1993 increase and the 1993-96 decline.
The results reported in Table 2 can be used to estimate the share of the reduction in welfare receipt between 1993 and 1996 that can be attributed to economic growth and federal welfare waivers granted to states. The product of the estimated parameters for, say, unemployment and its lag and the respective changes in unemployment in each state between 1993 and 1996 provides an estimate of the predicted change in welfare recipiency over the period based solely on changes in unemployment. The ratio of the predicted change to the actual change indicates the share of the reduction attributed to unemployment. An analogous exercise can be conducted to estimate the extent to which waivers contributed to the decline in the welfare rolls. Other unidentified factors would be responsible for the difference remaining after accounting for these two effects.20
Table 3 presents the results of this exercise for several of the statistical specifications reported in Table 2. The results indicate that the decline in unemployment that continued through the economic expansion contributed about 44 percent towards the decline in welfare recipiency in models that included both contemporaneous and lagged unemployment.21 Waivers accounted for roughly 15 to 20 percent of the decline in models that ignore the potential effects of an impending waiver grant. Once these effects are included (Column 6 of Table 2), estimates indicate that waivers can explain 31 percent of the decline in the share of the population receiving welfare. In this model, other unidentified factors explain an additional 25 percent.
A similar exercise could be conducted for the 1989-1993 period that saw a tremendous increase in the rate of welfare receipt. As discussed earlier, the magnitude of the increase is somewhat surprising given the relatively mild recession in the period. The estimates provided here reinforce the mystery; changes in unemployment can only explain about 30 percent of the rise in welfare rolls. Waivers were relatively new by 1993 and are found to have very little impact on the share of the population receiving welfare; in fact, they are expected to lead to a small decline. That leaves roughly 70 percent of the rise unexplained by this statistical analysis. Other forces that are more difficult to quantify must have been changing over this period, contributing to the increase.
The findings presented in this paper indicate that a robust economy and federal waivers allowing states to experiment with new welfare policies have each made large contributions towards reducing the rate of welfare receipt. The estimates provided here suggest that over 40 percent of the decline in welfare receipt between 1993 and 1996 may be attributed to the falling unemployment rate and almost one-third can be attributed to the waivers. Other factors that are not identified in this
analysis are responsible for the remainder. The methodology employed in this analysis poses two problems in interpreting these results. First, it is possible that the estimated effect of waivers on AFDC receipt may be capturing the tendency for states with shrinking welfare rolls to be the ones most willing to experiment with waiver policies.22 Another shortcoming of this research is that it cannot determine the outcomes for those individuals who otherwise would have collected benefits had waivers not been granted. Additional research that can determine how individuals fare under the alternative waiver provisions, rather than an aggregate analysis examining the share of the population receiving welfare, is clearly desirable to help address this issue.
Congressional Budget Office. Forecasting AFDC Caseloads, with an Emphasis on Economic Factors. Washington, DC. July 1993.
Council of Economic Advisers. Economic Report of the President. Washington, DC: Government Printing Office. February 1997.
Gabe, Thomas. Demographic Trends Affecting Aid to Families with Dependent Children (AFDC) Caseload Growth. Congressional Research Service. December 9, 1992.
Hoynes, Hilary Williamson. "Local Labor Markets and Welfare Spells: Do Demand Conditions Matter?" National Bureau of Economic Research, working paper 5643, June 1996.
Moffitt, Robert. "Historical Growth in Participation in Aid to Families with Dependent Children: Was There a Structural Shift?" Journal of Post Keynesian Economics. Spring 1987. pp. 347-363.
Moffitt, Robert A. "The Effect of Employment and Training Programs on Entry and Exit from the Welfare Caseload." Journal of Policy Analysis and Management. Vol. 15, No. 1 (1996). pp. 32-50.
Pavetti, LaDonna A. and Amy-Ellen Duke. Increasing Participation in Work and Work-Related Activities: Lessons from Five State Welfare Reform Demonstration Projects. The Urban Institute: Washington, DC. September 1995.
Yelowitz, Aaron S. "The Medicaid Notch, Labor Supply, and Welfare Participation: Evidence from Eligibility Expansions." Quarterly Journal of Ecomics. November 1995. pp. 909-939.
Most waivers awarded to states include a multitude of provisions that vary in the degree of their implications. Some affect the entire caseload while others affect a very small segment, like those that were introduced in pilot sites, such as a few counties. Some contain generally standard provisions while others are more complicated and require some judgement in categorizing them. In this paper, six major types of waivers that were implemented in most, if not all, of the state are considered. This appendix will provide some background regarding each of these different types of waivers, and how they have been coded for this analysis.
Termination and Work-Requirement Time Limits. Under AFDC, families were entitled to receive benefits as long as they met the eligibility requirements; states could only impose a time limit on the duration of benefit receipt if they were granted a waiver. Several states received such a waiver to implement to two main types of time limits. Termination time limits result in the loss of benefits for the entire family or just for the adult members, depending on the individual state's plan. While most states set a limit of 24 months or so for all recipients, other states had variable time limits. For example, Iowa's plan called for recipients to develop a self-sufficiency plan that included individually-based time limits, and Texas limited benefits to 12, 24, or 36 months depending on the recipient's education and work experience. Illinois provides an example of a state that contained this type of waiver provision but that is not coded as such here because it applied to a small fraction of the recipients (those with no children under age 13).
Work-requirement time limit waivers continue to provide benefits to adult recipients who reach the time limit as long as they comply with mandatory work requirements. For example, Massachusetts requires recipients unemployed after 60 days of AFDC receipt to do community service and job search to earn a cash "subsidy." California requires individuals who received AFDC for 22 of the previous 24 months to participate in a community service program for 100 hours per month. New Hampshire alternates 26 weeks each of job search and work-related activities for recipients. West Virginia's plan only requires participation in its work experience program by one parent in two-parent AFDC-UP cases, which are a small share of the total caseload, so it is not coded as a work-requirement time limit.
Some time limit waivers contain more complicated provisions that make them difficult to code. For instance, Delaware requires "employable" adults to participate in a pay-for-performance work experience program after receiving benefits for 24 months; after 24 months of program participation, the family completely loses cash benefits. Time limits with provisions such as this have been coded as containing both termination and work requirement provisions. Washington's plan is a grant-reduction time limit, subtracting 10 percent of the benefit for those who have received benefits for 48 of 60 months, then 10 percent for every 12 months thereafter. Because the time frame before a significant reduction in benefits could occur is so long, no time limit is coded for Washington.
Family Caps. Under AFDC, a family's benefit level depended upon its size, so if a recipient had a baby the grant amount rose. Family cap waivers allowed states to eliminate or reduce the increase in benefits when an additional child was born. A few states, like South Carolina, provide vouchers for goods and services worth up to the amount of the denied benefit increase. Others allow child support collected for the additional child to be excluded from AFDC income calculation. All family cap waivers except New Jersey's exempt children conceived as a result of rape or incest from the family cap. Several states, such as Wisconsin, Massachusetts and Illinois, specify that a child born or conceived after a family no longer receives AFDC can be denied benefits if the family returns to AFDC.
JOBS Exemptions. The Job Opportunities and Basic Skills Training Program (JOBS), part of the 1988 Family Support Act, provides education, training and work experience activities to AFDC recipients who did not fall into one of the exemption categories. The exemption categories were rather large, however. For instance, parents with children under age 3 were exempt and those with children under age 6 could only be required to participate if the state guaranteed child care. Some states requested a waiver to narrow the exemption criteria. The most commonly requested waiver required parents with young children (sometimes as young as 12 weeks) to participate in JOBS. Other waivers allowed teen parents attending school and people working 30 hours a week to be considered as JOBS participants. Hawaii had a JOBS waiver approved for a pilot site in Oahu, where a large share of the state's population lives, so it was coded as statewide.
JOBS Sanctions. Some states found that the sanctions for non-compliance with JOBS were not strong enough to motivate unwilling participants; they requested and were granted waivers to impose harsher sanctions. Twenty-two of the states were allowed to impose full-family sanctions (such as suspension of the entire family's AFDC grant) after a continued period of non-compliance. Other states requested tougher sanctions imposed upon the recipient only, leaving the children on the welfare rolls regardless of the parent's behavior. An informal survey of state welfare agencies conducted by the Council of Economic Advisers indicates that the use of sanctions has varied considerably across states. Some states have been very aggressive, sanctioning large numbers of recipients while others have sanctioned few, if any. For example, over the 1996 fiscal year Missouri reported sanctioning an average of 3,100 people per month, including sanctions of different severity levels. Massachusetts terminated benefits for 1,200 families in 1996 for failure to comply with training/work requirements. On the other hand, Georgia sanctioned few recipients in 1996.
Earnings Disregard. Without a waiver, individuals are allowed to
keep $30 plus one-third of all additional earnings for the first three months
of benefit receipt (the "standard AFDC disregard"). After that almost every
dollar of earnings results in a dollar reduction in benefits. Some states
received statewide waivers to improve the economic incentives for recipients to
work by increasing earned income disregards. The changes ranged from removing
the time limit on the standard AFDC disregard to disregarding all earned income
up to the poverty line.
|Approval Dates of Major Statewide Welfare Waivers in the Bush and Clinton Administrations|
|family cap||JOBS||Earnings Disregard||Sanctions|
|California||10/29/92, 9/11/95, 8/19/96||9/11/95||8/19/96||10/29/92|
|Connecticut||8/29/94, 12/18/95||12/18/95||12/18/95||8/29/94, 12/18/95||8/29/94||8/29/94|
|Illinois||11/23/93, 9/30/95, 6/26/96||9/30/95||9/30/95||11/23/93||6/26/96|
|Iowa||8/13/93, 4/11/96||8/13/93||8/13/93, 4/11/96||8/13/93||8/13/93|
|Oregon||7/15/92, 3/28/96||3/28/96||7/15/92, 3/28/96||3/28/96|
|Table 1: State Characteristics Over Time, by Welfare Waiver Status|
|Characteristic||States without Major
|States with Major
|Short-Term Changes, 1993-1996|
|% of population receiving AFDC||5.3||4.7||5.5||4.7|
|max AFDC benefit (3 person
family, 1996 dollars)
|Long-Term Changes, 1980-1990|
|% of Families Headed
|Table 2: Effect of Economic Activity and Federal
Welfare Waivers on Rate of AFDC Recipiency
(coefficients multiplied by 100, standard errors in parentheses)
|log of maximum
|termination time limits||-6.37
|lead of any statewide waiver||-6.28
|lead of JOBS sanction waiver||-1.50
|state fixed effects||x||x||x||x||x||x|
|year fixed effects||x||x||x||x||x||x|
|Note: The dependent variable is the share of the population receiving welfare, measured in natural logs.|
|Table 3: Percentage of Change in Welfare
Attributable to Different Factors
Standard Errors in Parentheses)
|Based on Results in Table 2, Column:|
|change in unemployment||31.3
|welfare waiver approval||14.9
|change in unemployment||23.9
2 The statistical analysis presented here uses data on the average monthly share of the population receiving welfare in a fiscal year. Between the 1993 and 1996 fiscal years (October 1, 1992 to September 30, 1996), the average monthly share of the population receiving welfare fell from 5.4 percent to 4.7 percent.
6 Moffitt (1996) has argued that the JOBS program (and, by implication, an extension of the JOBS program) may provide incentives for some to participate in welfare programs so that they can receive the potential benefits of these policies and could lead to an increase in the caseload.
7 It is also possible that expanded Medicaid eligibility may have increased AFDC participation. As more people come into contact with the social welfare system through Medicaid, they may find that they are eligible for AFDC benefits as well.
8 This analysis does control for some of the recent changes in Medicaid eligibility that have occurred at the national level even though their effects cannot be separately identified from other factors that affect all states in a given year.
9 All AFDC recipients area counted here, including those in two-parent families who receive AFDC-UP. Those in the latter category are probably more responsive to business cycle conditions because constraints facing single-parents, like finding affordable day care for their children while they work, are smaller in two-parent families. Therefore, they are more able to work when jobs are available. Still, AFDC-UP families represent a very small part of the total AFDC caseload and including them in this analysis should have minimal effects on the estimated parameters.
10 The difference in the average reduction across waiver and nonwaiver states is not statistically significant. The power of this test, however, is very weak in that waiver states may have had a waiver in effect for a very small part of this three year period. In addition, the normal variation across states in the share of the population receiving welfare swamps any variation across the groups of states over time. The regression analysis reported below adjusts for these problems and results from model specifications that mimic this simple "difference-in-difference" test statistic indicate that the reduction in waiver states is significantly larger than that in nonwaiver states.
11 This analysis uses the unemployment rate in each state and fiscal year. Because state level unemployment data have only been available since 1976, the 1976 fiscal year unemployment rate is measured just for the last three quarters (January through September) of that fiscal year. Other measures of unemployment may be more appropriate for this analysis. For instance, a measure of unemployment for younger women may better represent the labor market opportunities of potential welfare recipients. This measure may be somewhat endogenous, however, because changes that affect the labor supply of welfare recipients will to some extent, also affect the unemployment rate of younger women. Therefore, one might want to use the prime-age male unemployment rate because it does not suffer from this sort of endogeneity. Unfortunately, neither of these alternative measures is available on a state/year basis.
12 Another measure of welfare receipt that could be used as the dependent variable for this analysis is the number of families, or cases, receiving benefits. Patterns in the welfare caseload over time may differ across states as the number of child-only cases has proliferated at differential rates. All of the models reported below have also been estimated using the log of other welfare caseload as the dependent variable and mainly find similar results. The main difference is that JOBS sanctions apparently have a larger effect on recipients than on cases. This is consistent with the fact that many of these waivers only sanction the parent and maintain benefits for the children so that the case remains open even though the number of recipients fell.
14 Previous studies of the welfare caseload that use national time series data (CBO, 1993) have difficulty controlling for this type of pattern in the data. The results presented in Moffitt (1987) imply that it is important to control for such "structural shifts."
15 If differences across states over time are nonlinear they will not be captured by these trends and, if these differences are correlated with waiver awards, the estimated effect of waivers on the rate of welfare receipt will be biased. Although few candidates for such changes are readily apparent, one possibility may be the growth in income inequality since the late 1970s, documented in the Economic Report of the President (1997). Blank and Card (1993) show that the rate of growth in inequality has not been constant and has varied across regions of the country; if these differences occur across states and are correlated with waiver policies they may introduce a bias in the results reported here. Future research should investigate this possibility in more detail.
16 Additional measures of cyclical activity besides the unemployment rate may have a significant effect on welfare receipt. Preliminary estimates using the rate of employment growth within states over time, however, added no additional explanatory power in models that also included lags of the unemployment rate.
17 It is possible that this result is driven by a sort of policy endogeneity where sharp changes cuts in benefit levels occur in response to swelling welfare rolls, providing a negative relationship between these variables. Benefit cuts in California in the early 1990s that occurred as caseloads were rising in that state may be an example of this endogeneity.
20 Simply subtracting the sum of the two effects from 100 only indicates the contribution of other factors if no interaction between changes in unemployment and waiver policy on welfare receipt occurs. It may be the case, for example, that waiver policies are more effective in states with low unemployment rates. Models that incorporated this possibility were also estimated but the results indicated that the interaction between unemployment and waivers was not statistically significantly different from zero at conventional significance levels.
21 Based on estimates from a model of the duration of welfare spells and permanent changes in labor market conditions, Hoynes (1996) estimates that a typical economic expansion would result in an 8 to 10 percent reduction in the welfare caseload. This estimate is somewhat higher than the findings presented here and the difference is consistent with the fact that the current expansion is ongoing and, therefore, does not represent a permanent change in labor market conditions.
22 One might expect states with difficulties in holding down their welfare rolls to experiment with approaches to achieve that end. This sort of policy endogeneity would bias the results towards finding a positive relationship between waivers and the rate of welfare receipt.
President and First Lady | Vice President and Mrs. Gore
Record of Progress | The Briefing Room
Gateway to Government | Contacting the White House
White House for Kids | White House History
White House Tours | Help | Text Only