T H E   W H I T E   H O U S E

Technical Report: The Effects of Welfare Policy and the Economic Expansion on Welfare Caseloads

Help Site Map Text Only


Technical Report:

The Effects of Welfare Policy and the Economic Expansion on Welfare Caseloads: An Update

August 3, 1999

A Report by the Council of Economic Advisers


This study could not have been completed without the generous assistance of the Department of Health and Human Services in providing data and program information.

The Effects of Welfare Policy and the
Economic Expansion on Welfare Caseloads: An Update

Executive Summary

This study investigates the causes behind recent changes in welfare caseloads, updating a 1997 CEA report of caseload change.

  • The fall in welfare caseloads has been unprecedented, wide-spread, and continuous, and employment of welfare recipients has increased. 14.1 million people received welfare in January 1993, and this number had fallen to 7.3 million by March 1999, according to estimates released today (August 3, 1999). In 31 states the caseload is less than half of what it was when President Clinton took office, and all states have experienced double-digit percentage declines. For 22 states, the percent drop during 1998 was larger than during 1997 (from January to December). Previous analyses by the Department of Health and Human Services show that the percentage of welfare recipients working tripled between 1992 and 1997, and an estimated 1.5 million adults who were on welfare in 1997 were working in 1998.

  • The 1996 legislation has been a key contributor to the recent declines. PRWORA produced a dramatic change in welfare policy: work and self-sufficiency became a primary goal; state and local governments were given much greater control of their programs; and states experimented with a host of program designs. The evidence suggests that these changes caused a large drop in welfare participation, a drop that is independent of the effects of the strong labor market. The estimates imply that TANF has accounted for roughly one-third of the reduction from 1996 to 1998, the last year of data analyzed in this study. In the earlier years, 1993-1996, most of the decline was due to the strong labor market, while welfare waivers played a smaller yet important role.

  • The strong labor market has made work opportunities relatively more attractive, drawing people off welfare and into jobs. The unemployment rate has not declined as much in the post-TANF period as it did in the 1993-96 waiver period. As a result, the share of the decline in the caseload that is attributable to improvements in the labor market was much higher in the 1993-96 period (roughly 26 to 36 percent) than in the 1996-98 period (8 to 10 percent).

  • Past increases in the minimum wage have made work more attractive and, as a result, caused welfare participation to decline. The estimates imply that about 10 percent of the caseload decline was due to increases in state and federal minimum wages.

  • The specific program design adopted by a state can affect its caseload declines. The study examines the effects of a number of specific policies, including family caps, earnings disregards, time limits, work exemptions, and work sanctions on the size of the caseload.

The large sustained declines in caseloads provide one piece of evidence about the effectiveness of welfare reform efforts. However, there are multiple indicators of the impact of welfare reform, including changes in work and earnings among welfare leavers, in marriage rates and out-of-wedlock pregnancies, and in poverty rates. The Clinton Administration is collecting and tracking information on all of these measures in order to fully assess the impact of welfare reform.

The Effects of Welfare Policy and the
Economic Expansion on Welfare Caseloads: An Update

Introduction

The number of people receiving welfare has been declining at record rates. After peaking in March 1994, welfare caseloads have dropped by 48 percent through March 1999; At that time, just 7.3 million people representing 2.7 percent of the population were receiving welfare. Not since 1967 has such a small share of the population relied on welfare.

Not only have the declines been large, they have been widespread and continuous (Table 1). Between 1993 and 1998 (this report examines caseload changes through December 1998), all 50 states and the District of Columbia experienced double digit percent reductions in welfare participation, and in most states the declines were unprecedented. Thirty-seven states have experienced drops of at least one-third, and in 23 states the number of participants is less than half of what it was in 1993. And although a substantial share of the reduction occurred between 1994 and 1996, in many states the largest declines have occurred more recently. In fact, in 22 states the percentage decline in 1998 was greater than it was in 1997 (from January to December). And in almost all states (45) caseloads were still declining during the final months of 1998.

Two primary factors have been posited to explain the recent caseload changes: the strong labor market, and changes in welfare policy. The nation is in the midst of the longest peacetime expansion in its history, with low unemployment and rising wages. Moreover, gains in employment and wages have been experienced by groups who have typically had high rates of welfare use. Expanding labor market opportunities have made work more attractive to potential welfare participants, reducing their need for public transfers.

While the labor market has improved since 1992, there have been substantial changes in welfare policies throughout the past decade. In the early 1990s a growing number of states requested waivers from the traditional welfare program, Aid to Families with Dependent Children (AFDC), allowing them to experiment with alternative policies such as time limits, family caps, work requirements, and

Table 1. Changes in the Number of Recipients in Each State

Number of recipients

Percentage Change From

State

1993

1998

'93 to '96

'96 to '98

'93 to '98

Alabama

138,465

54,635

-26

-46

-61

Alaska

37,078

29,582

-1

-19

-20

Arizona

199,153

102,511

-16

-39

-49

Arkansas

71,989

32,633

-21

-43

-55

California

2,511,293

1,998,618

3

-23

-20

Colorado

122,890

50,746

-22

-47

-59

Connecticut

162,481

117,777

-2

-26

-28

Delaware

27,736

15,820

-16

-32

-43

DC

69,549

54,856

0

-21

-21

Florida

691,053

261,581

-22

-52

-62

Georgia

398,077

185,052

-15

-45

-54

Hawaii

57,336

46,724

16

-30

-19

Idaho

21,877

3,867

1

-83

-82

Illinois

694,050

476,576

-7

-26

-31

Indiana

215,367

111,176

-35

-21

-48

Iowa

102,438

65,665

-16

-24

-36

Kansas

88,363

34,536

-26

-47

-61

Kentucky

220,766

119,360

-22

-31

-46

Louisiana

259,762

124,800

-12

-46

-52

Maine

66,914

39,423

-18

-28

-41

Maryland

219,998

116,456

-11

-40

-47

Massachusetts

321,219

167,043

-28

-27

-48

Michigan

689,139

332,240

-26

-35

-52

Minnesota

192,173

143,685

-12

-15

-25

Mississippi

168,924

52,523

-26

-58

-69

Missouri

262,382

147,105

-14

-35

-44

Montana

34,875

19,540

-13

-35

-44

Nebraska

47,840

36,665

-20

-4

-23

Nevada

36,009

25,472

-2

-28

-29

New Hampshire

29,797

15,409

-22

-34

-48

New Jersey

345,370

196,947

-19

-30

-43

New Mexico

97,246

74,170

2

-25

-24

New York

1,215,526

886,746

-5

-23

-27

North Carolina

335,620

169,144

-20

-37

-50

North Dakota

18,215

8,541

-28

-35

-53

Ohio

712,277

340,179

-24

-37

-52

Oklahoma

135,762

61,191

-27

-38

-55

Oregon

117,852

46,001

-31

-43

-61

Pennsylvania

610,531

360,009

-14

-32

-41

Rhode Island

62,187

54,150

-8

-6

-13

South Carolina

146,280

60,110

-22

-48

-59

South Dakota

19,913

9,653

-21

-39

-52

Tennessee

310,486

149,089

-20

-40

-52

Texas

784,816

370,857

-16

-44

-53

Utah

52,144

28,258

-25

-28

-46

Vermont

28,301

19,643

-12

-21

-31

Virginia

194,765

99,053

-20

-36

-49

Washington

289,965

202,573

-6

-25

-30

West Virginia

118,113

38,638

-25

-56

-67

Wisconsin

235,247

40,167

-33

-75

-83

Wyoming

17,859

2,471

-32

-80

-86

Total

14,007,468

8,199,666

-13

-33

-41

Data are the average monthly caseloads for the calendar year.

a variety of other options. During the Clinton Administration (from the beginning of 1993 to 1996), 43 states received welfare waivers, more than any previous Administration. At the federal level, welfare policy was changed dramatically by the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA), which replaced the AFDC program with the Temporary Assistance for Needy Families (TANF) block grant. Under PRWORA, welfare became more work-focused and time-limited: with few exceptions, federal welfare assistance is strongly linked to the recipient's efforts to find a job. In most cases, adults cannot receive federal aid for more than a total of 5 years during their lifetime, and some states have chosen to set shorter time limits. PRWORA also shifted primary responsibility for welfare program design and management to States and localities.

In 1997, the Council of Economic Advisers issued a report using 1976 to 1996 data that examined the reasons for the decline in caseloads between 1993 and 1996. That study found that roughly 45 percent of the decline was accounted for by improved labor market conditions, about 30 percent was due to welfare waivers, and the remaining 25 percent was explained by other factors. Several subsequent studies were conducted that examined changes in welfare caseloads during this and earlier periods (Bartik and Eberts, 1998; Blank, 1997; Figlio and Ziliak, 1998; Levine and Whitmore, 1998; Moffitt, 1999; Stapelton, 1998; Wallace and Blank, 1998; Ziliak, Figlio, Davis, and Connolly, 1997).

Since 1996 caseloads have continued to fall, the labor market has grown even stronger, and welfare policy has been fundamentally changed, making it important to update the earlier report. This study extends the earlier study on several dimensions. Most importantly, the effects of TANF are assessed by analyzing data through 1998. In addition, the study provides more recent evidence of the effects of labor market conditions on changes in caseloads, and the study examines whether increases in the minimum wage also played a role.

The large sustained declines in caseloads provide one piece of evidence about the effectiveness of welfare reform efforts. However, there are multiple indicators of the impact of welfare reform, including changes in work and earnings among welfare leavers, in marriage rates and out-of-wedlock pregnancies, and in poverty rates. The Clinton Administration is collecting and tracking information on all of these measures in order to fully assess the impact of welfare reform.

Factors Affecting Caseload Trends
Economic Conditions

Caseloads normally fluctuate with the business cycle, rising in periods of high unemployment and declining when unemployment falls. Chart 1 illustrates this relationship between labor market opportunities and welfare participation (i.e., the number of welfare recipients divided by the total population) over the past three decades. When unemployment increased in the early 1970s, so too did welfare participation. The increase in welfare participation in the late 1980s and early 1990s, as well as the decline that began in 1994, also correspond with changes in employment opportunities during these periods. However, the trend in welfare participation does not always match that in unemployment, most notably when other important changes are taking place, including changes in family structure and welfare policies. Indeed, increases in welfare participation during the recession of the early 1980s were truncated by eligibility restrictions that were part of President Reagan's welfare reform efforts in 1982. Over the entire 1980s the simple correlation between unemployment and welfare participation was much lower (0.23) than in the 1970s (0.41) or the 1990s (0.78).

Economic conditions vary across states as well as over time. Chart 2 displays a scatterplot of the unemployment rate versus the welfare participation rate for each state and the District of Columbia in 1994, when participation was near its peak. (California and New York are highlighted because they are home to roughly one-third of the nation's welfare recipients, and DC is highlighted because it is an outlier on this Chart.) This relationship is quite strong, with a simple correlation of 0.65. While this correlation suggests a strong role for economic factors, it is likely to over-state the true role of economic factors. Fixed characteristics of states that cause them to have high unemployment rates may also lead them to high welfare participation. These characteristics include the state's age distribution, educational level, metropolitan/rural population shares, and racial and ethnic composition. While these factors may change over time, such change occurs more slowly than changes in policy or economic conditions. One way to abstract from these factors is to examine changes over time within states, which is the approach employed in the econometric models below. Chart 3 displays the simple relationship between the change in the unemployment rate and the change in the welfare participation rate in each state between 1994 and 1998 to illustrate the potential importance of these fixed characteristics. The chart demonstrates that once state fixed effects are removed by examining changes in these variables, the relationship is not nearly as strong as the simple cross-sectional one, with a correlation of 0.17.

Federal and State Policies

Welfare Waivers. Since 1962, the Secretary of Health and Human Services has had the authority to waive federal welfare requirements if a state proposed experimental or pilot programs that furthered the goals of AFDC. Although there were a few waivers granted in the early 1980s, it was not until the early to mid-1990s that major, state-wide waivers became widespread.

These waivers varied substantially across states, and in many cases they differed greatly from the rules under AFDC. Some waivers increased the amount of earnings recipients were allowed to keep and still be eligible for welfare. Other waivers expanded work requirements to a larger number of recipients, established limits on the length of time recipients could remain on aid, permitted states to sanction participants who failed to meet work requirements, or allowed states to eliminate benefit increases to families who conceived and gave birth to children while on welfare (the so-called "family cap"). Given the widespread use of waivers and the degree to which these policies differed from traditional AFDC policy, there is substantial reason to believe that waivers contributed to changes in welfare caseloads.

PRWORA. In August of 1996, President Clinton signed the Personal Responsibility and Work Opportunity Reconciliation Act into law, dramatically changing federal welfare policy. PRWORA was designed to emphasize self-sufficiency and employment in place of welfare dependency, and it gave states greater flexibility to design and implement programs to achieve these goals. Benefits are time-limited; adults usually cannot receive federal aid for more than 5 years during their lifetime, and some states have chosen to set shorter time limits. Most recipients must also participate in a work activity within two years to continue receiving aid.

Under the TANF block grant established by PRWORA, federal assistance consists of an annual fixed transfer to each state equal to the amount of federal transfers the state received in fiscal year 1994, 1995, or the average of 1992-4, whichever was higher. In addition, most of the authority to design welfare programs was passed along to the states, who are required to have half of all recipients working by 2002 (40 percent by 2000). As a result, there are now substantial differences in how welfare programs operate across the nation. Some states increase benefits to welfare families who have additional children, while others do not. Some states stop payment of benefits to the entire family at the first instance of their failure to meet work activity requirements, while other states never sanction more than the adult. And some states allow welfare recipients to keep a substantial portion of their labor market earnings without reducing their welfare payments, while others do not.

AFDC/TANF Benefit Levels. States have long set their own level of maximum monthly benefit payments, with variation by family size and composition. All else equal, higher benefit levels are expected to increase the number of participants. Over the period of this study, the inflation-adjusted level of welfare benefits fell in almost all states. In some cases the state explicitly changed benefits, but in most states benefit levels were fixed and eroded over time with inflation.

Minimum Wage. The real value of the federal minimum wage decreased substantially between 1976 and 1989. A $0.45 legislated increase in 1990, followed by a $0.45 increase in 1991, offset some of this long-run decline, but by 1995 the real minimum wage ($4.55) was nearly as low as it was in 1989. The minimum was then legislatively raised by $0.50 in 1996 and an additional $0.40 in 1997. During the period analyzed in this study, 1976-1998, several states established minimum wage levels that were above the federal minimum that prevailed at that time.

A higher minimum wage can make work more attractive, giving welfare participants a greater incentive to enter the workforce and leave public assistance. On the negative side, if a higher minimum wage reduces employment of low-skilled workers, some people may lose their jobs and enter welfare. At the same time, an increase in the minimum wage may lead employers to substitute away from teenagers (a relatively large share of whom work for the minimum wage) and towards older welfare workers (who are perhaps not as likely to work at the minimum wage, but more likely to be working just above the minimum than teenagers). The evidence on the disemployment effects of the minimum wage is mixed. Some studies have found that a 10 percent increase in the minimum wage causes a 1 to 2 percent decline in employment (e.g., Neumark and Wascher, 1992; Neumark and Wascher, 1994; or the estimates surveyed by Brown et al., 1982), while other studies have found no disemployment effects (e.g., Katz and Krueger, 1992; Card, 1992a; Card, 1992b; Card, Katz, and Krueger, 1994; Bernstein and Schmitt, 1998; Card and Krueger, 1998). Two recent studies have examined the effects of minimum wages on welfare caseloads, with one finding a negative effect over the 1990-91 period (Turner, 1999) and the other finding a positive effect over the 1983-96 period (Page, Spetz, Millar, 1999).

There are a variety of other factors that may affect caseloads, including the Earned Income Tax Credit, the availability of child care, transportation, and Medicaid coverage, family structure, and out-of-wedlock births. Although our models do not directly examine these factors, our approach controls for them indirectly, as described in the next section.

Econometric Specification

Two approaches are implemented to estimate the effects of policy and economic conditions over the period 1976-1998. Both approaches utilize the same dependent variable, use the unemployment rate to capture the effects of labor market conditions, and specify the minimum wage and welfare benefit levels in identical ways. The difference between the two models is the specification of the remaining welfare policy variables. The first model uses two simple 0/1 indicator variables: one to capture the period during which a major waiver was in effect in each state, and one to capture the period during which TANF was in effect in each state. Specifically, Model (1) is:

The variables are defined for state s in calendar year t as follows:

R: the ratio of the number of recipients to the population under 65 years of age (the number of recipients is obtained from administrative reports on AFDC/TANF); the model estimates the natural log of this ratio.

Waiver: an indicator variable that takes the value of one if the state had a major waiver in effect; the indicator is turned off when TANF is implemented in the state.

TANF: an indicator variable that takes the value of one if TANF was in effect in the given state (the TANF implementation date varied across states, as discussed below).

Benefits: the maximum monthly benefit for a family of three on AFDC/TANF.

MinWage: the value of the state-specific minimum wage expressed as a monthly amount (to make comparable with the benefits variable) assuming employment for 30 hours per week for 4.33 weeks. (In most cases, this is the federal minimum wage.)

Unemployment: the unemployment rate (current, lagged one year, lagged two years).

γs: state fixed effects

γt: year fixed effects

trend*γs: linear state-specific time trends

All dollar values are expressed in 1998 dollars using the CPI-U-X1.

The second approach examines the effects of specific welfare policies, regardless of whether the policy was implemented under waivers or TANF. That is:

In Model (2), Xst represents a vector of variables that describe specific policies that are in effect in state s in year t. There are a variety of policies that could be analyzed. The five policies that were examined were chosen because, a priori, they were expected to significantly influence participation and they could be quantified based on available sources. The five policies are:

  1. Termination or work requirement time limits are represented by an indicator variable for whether the state either terminates eligibility, reduces benefits, or requires participants to work (not just participate in a "work activity") after a given duration on aid. The date that participants first began to reach the time limit was used as the date that this policy came into effect. (These time limits had become binding in too few states for us to examine the distinct effects of each of these three policies.)

  2. A second indicator variable takes the value of 1 (0 otherwise) if the state has a family cap, that is, the state does not increase benefits for participants who give birth to or conceive a child while on aid.

  3. Work exemptions are represented by three indicator variables based on the state's policy toward families with young children: the first takes the value of 1 if the state exempts mothers with a child as old as 6 months to 3 years, 0 otherwise; a second indicator takes the value of 1 if the exemption applies to mothers with a child newly born to 6 months old (and not older), 0 otherwise; and a third takes the value of 1 if the state allows no exemptions based on the age of the mother's children, 0 otherwise. Years in which a state has a traditional AFDC/JOBS exemption policy serves as the reference group. These four groups are mutually exclusive.

  4. A set of three indicator variables capture the aggressiveness of work sanction policies. One indicator represents states that impose full family sanctions with the first offense ("full/full"), a second indicator represents states that impose full family sanctions only after repeated offenses ("partial/full"), and a third indicator represents states whose maximum sanction is a partial family sanction ("partial/partial"). States that impose no sanction or some lesser sanction, which was the case under traditional AFDC, serve as the reference group.

  5. The aggressiveness of disregarding earned income is represented by the amount of earnings disregard if a welfare recipient earns $750 per month (in 1998 dollars). When the disregard formula varies with duration on welfare, the disregard applicable for the longest duration (typically more than 3 months) is assumed.

The "policy oriented" approach used in Model (2) has the advantage of being able to identify the specific policies that influence caseloads. However, there a number of TANF policies and practices that may affect participation that could not be captured in Model 2 because of data limitations, such as diversion policies, work requirements and targets, and welfare office culture. The simple indicator-variable approach used in Model 1 is more effective in capturing the total effect of waiver and TANF policies.

State, year, and state-specific time trends are included to capture unobserved factors, such as family structure and other policies, that may be correlated with the observed variables. Most policies were not in effect the entire calendar year that they were implemented. In these cases, fractional values are used corresponding to the share of the calendar year that the policy was in effect. The model is estimated with weighted least squares, where the weight is the population under 65 in state s in year t. The standard errors of the coefficient estimates are corrected for general forms of heteroscedasticity.

Before discussing the results, it should be acknowledged that a maintained assumption in this study is that welfare policies are exogenous to welfare participation (after controlling for the factors in the models described above). All previous studies have also made this assumption. Endogenous policy is probably more likely to affect the estimates of Model 2. While most states received waivers, and every state has implemented TANF, the specific types of policies vary considerably. For example, states whose caseloads were increasing (or not decreasing as much as desired), may have adopted relatively harsh policies.

Data

Using annual calendar year data from 1976 to 1998 on all states and the District of Columbia, the analysis is based on 1,173 observations. Most of the data used in the analysis come from well-known sources, with a few exceptions (described below). The federal and state minimum wage data were obtained from the Wage and Hours Division of the Bureau of Labor Statistics.

Welfare Waivers

The data that are unique to this study are the waiver implementation dates and TANF policies. These policies are difficult to categorize and measure, and the pace and intensity of their implementation typically vary across and within states. Experts from the Department of Health and Human Services as well as non-government research institutions were consulted to characterize these policies as fully as possible. Specifically, information on waivers was obtained from the Department of Health and Human Services. Most waivers permitted simultaneous implementation of various provisions. For example, the California Work Pays Demonstration increased the AFDC resource limit for recipients to $2,000, increased the excludable equity value for a vehicle to $4,500, allowed recipients to place up to $5,000 in restricted accounts which did not count against the resource limit and which may only be withdrawn for certain uses, and (among other things) required pregnant or parenting teens (under 19) who did not possess a high school diploma or equivalent to participate in CalLEARN.

Like the 1997 CEA study, this report focuses on six "major" types of waivers that received approval to be implemented state-wide: termination time limits, work requirement time limits, family caps, JOBS exemptions, JOBS sanctions, and the earnings disregard. Each of these policies was discussed in detail in the appendix to the 1997 CEA Technical Report.

Some of the waivers that were approved for state-wide implementation were initially implemented state-wide, some were implemented in selected areas of the state, while still others began in small regions of the state but were eventually phased-in state-wide. Information on the pace of implementation is not available for all states. Therefore, the date that is used to signal implementation is the date that the waiver began to be implemented. The earliest dates that these waivers were approved and implemented in each state are listed in Table A1.

PRWORA & TANF

PRWORA was signed into law in August of 1996, but a given state could not begin its TANF-funded program until that state submitted its TANF plan and it was certified as complete by the federal government. Beginning on the date the state formally implemented its TANF plan, the state could begin to draw down federal funds and was subject to all of the requirements and restrictions in TANF. The earliest official implementation date was September 1996 and the latest was July 1997,

when all states were required to begin operating under TANF. The date that the state formally implemented its TANF plan is the date that is used to construct the TANF indicator variable in Model (1). However, in some states the initial plan was simply a placeholder, designed to allow the state to begin to draw down its TANF block grant, and some state policies were not changed until a later date. Therefore, the actual implementation date may differ from the official date. In particular, in five states (California, Mississippi, New Jersey, New York, and Wisconsin) specific information was available indicating that the policies most associated with TANF time limits, work requirements, sanctions, etc. were not implemented until a later date; in these cases, the later date was used to construct the TANF indicator. Table A1 reports the official and actual TANF implementation dates for each state.

To specify Model (2) the policies that were in effect in each state in each year were determined. To construct indicator variables for the existence of a termination or work requirement time limit and a family cap, we used the date that the relevant waiver was implemented (for time limits, the date that participants began to hit the limit) and assumed that the waiver continued to be in effect until (at least) TANF was implemented in that state (i.e., the date listed in Table A1).

For the TANF period, we use information on state TANF plans as of October 1997 (Gallagher et al., 1998) along with the date the current policy (as of October 1997) was implemented to determine which policies were in effect in each state in each year. It is assumed that the policies in place in October 1997 were not changed by December 1998, which is the end of our sample period. If a policy was implemented and rescinded between the date that TANF was implemented and October 1997, we would not capture this policy change. However, the earliest TANF implementation was October 1996, just one year prior to our TANF information, and many states implemented TANF in the first 6 months of 1997. Therefore, it is unlikely that a policy was both implemented and rescinded within such a short period.

Results

Table 2 contains the estimates of Models 1 and 2. The table also reports a version of each of these models that excludes state-specific time trends. The rationale for including these trends is to control for unobserved changes over time that are specific to each state. For example, if there is a long-run increase in female-headed households, and the rate of this increase varies between states, other variables in the models may be biased if this factor is not controlled. On the other hand, some of the interesting and important variation for identifying effects of some of the variables of interest may be reduced substantially by the inclusion of these trends, making it difficult to identify their effects. For example, cash benefit levels follow a long-run trend in some states, and including the state-specific trends leaves much less variation in benefits to identify its effects. Therefore, estimates with (Models 1 and 2) and without (Models 1A and 2A) the state-specific trends are reported.

Estimates from Model 1

Waivers had a large and precisely estimated effect on welfare participation (Table 2). The estimates in Models 1 and 1A imply that states that implemented a major waiver experienced a decline in participation that was 8 to 9 percent greater than other states. The implementation of TANF is associated with a decline in participation of 18 percent, roughly double the size of the effect of waivers.

All other statistically significant estimates in Models 1 and 1A alter participation in the expected direction. Higher cash welfare benefits raise participation. The estimates in Model 1 imply that a $50 increase in the monthly benefit above its 1998 average monthly value would increase participation by 1.8 percent. For the reasons described above, the estimates from Model 1A, which exclude the state-specific linear trends, are much larger and imply that the same $50 increase would lead to a 6.2 percent increase in participation.

Increases in the minimum wage are found to decrease welfare participation. In particular, consider an increase in the minimum wage by $0.50. If this increase were on top of the average minimum that existed in 1998, monthly earnings at the minimum wage (evaluated at 30 hours per week, full month) would increase by $65. This rise would translate into a decline in welfare participation of roughly 3.7 to 5.9 percent.

Tight labor markets, as measured by the unemployment rate, reduce welfare participation. The models demonstrate the lagged nature of the unemployment effects. In fact, the largest effects are for unemployment lagged two years. Model 1 implies that a one percentage point decrease in the unemployment rate that persists for three years is associated with a 5.41 percent (4.27+1.50-0.36) decline in welfare participation. The estimates are substantially higher if state-specific time trends are not included in the model.

Estimates from Model 2

The effects of cash benefits, minimum wages, and the unemployment rate estimated for Models 2 and 2A are similar to those estimated in Models 1 and 1A, respectively. The welfare reform policy variables included in Models 2 and 2A show mixed results. The coefficient on the time-limit indicator variable is negative, as expected, but it is not precisely estimated. It is important to note that all participants who have hit time limits by the end of 1998 were doing so under a waiver policy. And because only a small number of states had time limit waivers, a relatively small number of participants had hit a time limit. Therefore, it is not surprising that, through 1998, time limits had not significantly altered national caseloads.

As expected, a higher earnings disregard raises participation (at least in the short-run), but this effect is relatively small. The estimates suggest that an increase in the disregard equivalent to $50 on a monthly basis is associated with less than a 1 percent increase in participation. Family caps do not have the expected negative effect; in fact, they are positive and precisely estimated. Similarly, looking across Model 2 and 2A, it appears that work exemption policies based on the age of the youngest child do not play a substantial role in determining caseloads. In fact, the one significant effect is of unexpected sign.

Not surprisingly, policies that sanction recipients who do not go to work are associated with large declines in welfare participation. The effects of the work sanction policies may be due to the fact that impending sanctions cause welfare recipients (or potential recipients) to accelerate their job search and find employment, or the effect may be due to the fact that recipients did not find a job and were sanctioned. States with full family sanctions on the first violation of work requirements have much lower caseloads than other states. States whose most severe work sanction policy is a partial reduction in benefits also have lower participation, but not nearly as low as the rates for states with full family sanctions. As with all policies examined in the model, the effects of these sanctioning policies on the caseload may be distinct from their effects on other important factors, such as child health and development, illegitimacy, education, poverty, and work participation.

Relative Contribution of Each Factor

1993-96 Welfare Waiver Period. Table 3 provides estimates of the relative contribution of each factor to the change in welfare participation during two periods: 1993-1996 (the waiver period under the Clinton Administration) and 1996-98 (the TANF period). Specifically, the change in the national average of each variable (obtained by weighting by the state population under 65) is multiplied by its respective coefficient estimate to determine the change induced by that factor. The ratio of the share

Table 3.

Percentage of Change in

Participation Attributable to Each Factor

(Based on Estimates of Models 1 and 1A in Table 2)

Based on Model 1

Based on Model 1A

Factor

1993-96

1996-98

1993-96

1996-98

Welfare waivers

14.6%

12.4%

TANF

36.2%

34.8%

Decline in unemployment

26.4%

7.8%

35.6%

10.4%

Increased minimum wage

-9.7%

9.6%

-15.6%

15.5%

Lower cash benefits

6.3%

1.4%

21.7%

4.7%

Other

62.4%

45.0%

45.9%

34.5%

of this change to the total change in participation during this period is reported in Table 3. For example, 22 percent of the population under 65 lived in states with major waivers in place in 1993.

By 1996, this share increased to 53 percent. Multiplying the change in the share living under waivers (0.53-0.22=0.31) by the respective coefficient estimate in Model 1 (-9.40), it is found that the expansion of waivers led to a 2.91 percent decline in participation during this period. Participation in total dropped by about 20 percent between 1993 and 1996, which implies that roughly 14 percent of the decline can be attributed to the increase in waivers.

While waivers accounted for about 14 percent of the decline from 1993-96 according to Model 1, the lower unemployment rate was responsible for 26 to 36 percent of the decline (depending on the model). Cash benefits declined by about 8 percent from 1993 to 1996, which led to a decline in participation. The actual amount of the decline that can be attributed to the benefit reduction differs substantially between the two models; 6 percent for Model 1 and 22 percent for Model 1A. The real value of the minimum wage fell between 1993 and 1996 (the increase in 1996 was in October, so it was not effective most of the year), which is why the minimum wage explains a negative share of the caseload decline; the caseload would have increased between 1993 and 1996 if the only change that had occurred were the decline in the real minimum wage.

TANF Period: 1996-98. Welfare participation declined by roughly 33 percent between 1996 and 1998, and TANF was a major contributing factor. Roughly one-third of the decline is due to TANF. Economic factors are still important in drawing people off welfare, but since the unemployment rate has declined relatively little since 1996, it accounts for just 8 to 10 percent of the decline in participation over this period. Higher minimum wages accounted for about 10 percent of the drop in participation, and reductions in cash benefits accounted for an additional 1 to 5 percent decline. The remaining share is unexplained and may be due to other changes in policy, practice, or behavior.

Alternative Specifications

Several alternative specifications were estimated to examine the robustness of the findings, and some of these results are reported in Table 4. All of the models in Table 4 include state-specific time trends, and the estimates from Model 1 of Table 2 ("Baseline") are listed for comparison.

It has been argued that analyses of waiver policies should not utilize population weights (Martini and Wiseman, 1997). Comparison 1 demonstrates that the effects of waivers, TANF, cash benefits, and the unemployment rate are not very sensitive to whether weighting is used. However, the effects of the minimum wage are substantially larger when the weights are not used.

Quite often it is said that welfare reform would not have been as effective in reducing caseloads if it had not been for the strength of the labor market. This hypothesis is tested in Comparison 2 by

Table 4.

Alternative Specifications of Model 1

(Coefficient estimates are multiplied by 100)

Comparison 1

Comparison 2

Comparison 3

Comparison 4

Comparison 5

Baseline

Without

Policy & Economy

Changing Economic Effects

With Leads of

Population as an

Model 1

Population Weights

Interactions

Model A

Model B

TANF and Waivers

Explanatory Variable

Beta

t-stat

Beta

t-stat

Beta

t-stat

Beta

t-stat

Beta

t-stat

Beta

t-stat

Beta

t-stat

Any waiver

-9.40

2.90

-7.34

2.95

-1.90

0.21

-8.86

2.42

-9.34

2.54

-5.53

1.82

-8.29

3.01

Any waiver, lead

-6.84

2.39

TANF

-18.84

4.37

-18.04

2.38

-46.23

2.77

-21.28

4.23

-22.07

4.14

-15.19

3.20

-15.94

3.94

TANF, lead

-4.84

1.19

Log max. monthly benefit

14.98

1.93

20.92

3.34

-5.44

0.78

-6.99

0.87

-6.10

0.75

14.91

1.95

29.06

4.27

Log monthly min. wage

-39.59

4.02

-67.31

4.01

-53.00

3.73

-51.59

3.81

-47.44

3.44

-40.28

4.26

-15.14

1.48

Unemployment rate

Current

-0.36

0.74

0.63

1.36

3.21

8.51

3.17

8.80

-0.26

0.54

0.74

1.70

One lag

1.50

2.40

1.80

3.23

1.51

2.44

1.25

2.31

Two lags

4.27

8.92

3.66

8.12

4.17

8.78

2.68

6.04

Current*1976-80

1.48

1.93

Current*1981-86

3.20

7.97

Current*1987-92

3.87

6.03

Current*1993-98

4.37

3.54

Waiver*Current

-1.01

0.63

TANF*Current

5.32

1.57

Log(Population under 65)

-136.77

4.62

All models include state effects, year effects, and state-specific time trends. Estimates use the population under 65 as weights and robust calculation of standard errors, except in Comparison 1 where the weights are not used.

interacting the unemployment rate with the waiver indicator and with the TANF indicator.

Although the precision of the estimate of the interaction between TANF and the unemployment rate

is slightly below standard levels for determining statistical significance (with a p-value of 0.12), the coefficient estimate implies that TANF policy is more effective when unemployment is low. For

example, after adjusting for other factors, TANF is estimated to reduce participation by 14.8 percent if the unemployment rate were 5.9 (as it was in California when it implemented TANF in 1998) and by 20.2 percent if the unemployment rate were 4.9 (as it was in Michigan when it implemented TANF in 1996).

It has been argued that the effects of waivers may be accounted for by an increase in the sensitivity of the caseload to labor market conditions in the 1990s (Moffitt, 1999). For this argument to hold, economic conditions must be correlated with waivers, the caseload must have become more sensitive to the unemployment rate over time, and the model must not have allowed the effects of the economic factors to change over time. Comparison 3 (Model B) tests this hypothesis by allowing the effects of the unemployment rate to differ between four periods: 1976-80, 1981-86, 1987-92, and 1993-98. (While Model B allows the effects of unemployment to vary across time, it does not include lagged unemployment effects. Therefore, the baseline model, which does not incorporate time-varying unemployment effects, is re-estimated with no lags in unemployment so that proper comparisons can be made. This specification appears as Model A in Comparison 3.) Indeed, the caseload has become more sensitive over the past two decades. A one percentage point increase in unemployment led to an increase in welfare participation of 1.5 percent in the 1976-80 period, 3.2 percent in the 1981-86 period, 3.9 percent from 1987-92, and 4.4 percent since 1993. (The 1976-80 period is statistically significantly different from each of the other three periods, but the three latter periods are not statistically significantly different from each other.) This rise may be due to the fact that most of the changes to AFDC introduced by waivers and TANF emphasize employment. This also suggests that the estimates of the contribution of the unemployment rate reported in Table 4 may be a lower bound. Most importantly for this study, however, the effects of waivers and TANF are robust to this specification, changing very little from the baseline model.

Comparison 4 permits "lead" effects of TANF and waivers. The 1997 CEA study argued that welfare policies may begin to have an effect on behavior in the year leading up to their enactment because of the heightened awareness generated by the debate surrounding their passage. Indeed, the 1997 study found that state caseloads were declining significantly in the year prior to receiving approval for a waiver. The estimates with the data through 1998 and incorporating TANF imply a fairly large and statistically significant association between welfare participation and the one-year lead of waivers; the lead of TANF is not significant. However, it is difficult to interpret these estimates. While a true causal interpretation is plausible, an alternative interpretation is that the leads are picking up unobserved differences across states or within states across time. For example, perhaps states with recently declining caseloads (or caseloads declining more -- or increasing less than expected) had slack resources and manpower to design and submit a waiver. In this case, waivers themselves may not be causing the decline. For this reason, the estimates without the leads are emphasized.

The final alternative specification, Comparison 5, uses a less restrictive functional form by using the population variable as an explanatory variable instead of using it as the denominator in the dependent variable. In this model the dependent variable is simply the natural log of the number of recipients. The results are fairly stable to this specification change. However, the coefficient estimate on the minimum wage, while still negative, is reduced, and it has a p-value of 0.14.

CONCLUSIONS

There has been an unprecedented decline in welfare caseloads. The drop has occurred in every state in the nation, and it has persisted now for almost 5 years. In the earlier years, from 1993 to 1996, most of the decline was due to the strong labor market and welfare waivers. The declines in the more recent period, from 1996 to 1998, have been very large, and the single most important factor that can be identified is the implementation of TANF. PRWORA produced a dramatic change in welfare policy: work and self-sufficiency became a primary goal; state and local governments were given much greater control of the programs they ran; and states experimented with a host of program design changes. The evidence suggests that these changes have caused a large drop in welfare participation, a drop that is independent of the effects of the strong labor market during this period. The estimates imply that TANF alone has accounted for roughly one-third of the reduction from 1996-98.

The strong labor market has made work opportunities relatively more attractive, drawing people off welfare and into jobs. In fact, the size of the caseload has become more sensitive to labor market changes in recent periods. However, the unemployment rate has not declined as much in the post-TANF period (1996-98) as it did in the 1993-96 waiver period. As a result, the share of the decline in the caseload that is attributable to improvements in the labor market was much larger in the 1993-96 period (roughly 26 to 36 percent) than in the 1996-98 period (8 to 10 percent).

While this study helps to explain the post-TANF changes in welfare participation, there is much about welfare participation that is unknown. In most models that were estimated, a large share of the variation over time could not be explained. The variation across states in welfare policy and management has increased as a result of TANF, and the research community will struggle to keep abreast of these changes. Merely documenting the changes, let alone understanding their effects on caseloads, work, self-sufficiency, child well-being and the like, is a major challenge.

Appendix A

Comparison with the 1997 CEA Study

A replication of the estimates reported in the 1997 CEA study is provided in Table A2. There are five reasons why the "old" estimates may differ from the "new" estimates:

1. different time periods of analysis

2. different variables included in the models

3. use of approval vs implementation date of waivers

4. use of calendar vs fiscal year data

5. use of population under 65 instead of all population in calculating participation rates.

All models in Table A2 analyze the 1976-1996 period and include the same explanatory variables. Comparison between the "old CEA" estimates and the estimates in Model I of Table A2 shows that the effects of waivers are larger when calendar year data is used instead of fiscal year data. This finding is not surprising because the caseload continued to decline at the end of 1996, and some of this decline is attributed to waivers in Model I. Table A2 also demonstrates that the effects of waivers is somewhat smaller when the implementation date (Model IV) is used instead of the approval date (Model II). Use of the population under 65 (Model IV) instead of the total population (Model III) in the denominator of the recipiency rate alters the results very little.

Although the use of the implementation date instead of the approval date and a different population control does not alter our results substantially, two other choices do. First, we include a second lag of the unemployment rate in our models in the current study (Table 2). The effect of the second lag is quite large and precisely estimated. It turns out that the inclusion of the second lag explains an important difference in the reported results between the two studies. With only one lag in unemployment, the 1997 study found that unemployment could explain 45 percent of the change in

participation from 1993-96. (See Table 3, column labeled (3), in the 1997 report.) Using the 1976-

1998 data, and the full specification reported as Model 1 in Table 2 but without the second lag in unemployment, we find results that are almost identical to those reported in the 1997 study: unemployment explains 42 percent of the change between 1993 and 1996. But with the second lag included, the share explained by unemployment falls to 26 percent. Therefore, the specification of the lag structure does alter the results from the simulations. However, the effects of waivers change very little with the specification of the lag structure of unemployment: the share explained by waivers between 1993-96 based on Model 1 in Table 2 is approximately 15 percent with either one or two lags.

The studies also differ in their findings regarding the importance of waivers. However, the primary difference is not due to different estimates within the same specification, but in the choice of which specification to emphasize. The 1997 study emphasized results from a specification that included a lead value of the waiver variables (model 6 in Table 2 of the 1997 report) while the current study emphasizes models that exclude the leads (model 3 in Table 2 of the 1997 report). As described in the 1997 technical report, "... it may be 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 (p. 15)." While this is a plausible scenario, an alternative interpretation is that the leads are picking up unobserved differences across states or within states across time. For example, perhaps states with recently declining caseloads (or caseloads declining more -- or increasing less than expected) had slack resources and manpower to design and submit a waiver. For this reason, the current study uses the simple contemporaneous value for waivers and TANF.

Excluding the leads does not change the estimates of the effect of unemployment rates. However, the waiver effects are substantially smaller without the leads. As reported in Table 3 of the 1997 study, the share of the 1993-96 change explained by waivers falls from 31 percent if the leads are included to 13 percent if the leads are not included. The 13 percent estimate in the 1997 study is comparable to the estimate of 14.6 percent in Table 3 of the current study.

Other than these differences, the updated study is quite consistent with the earlier report. Most importantly, strong labor markets, as measured by the unemployment rate, and welfare waivers played important roles in explaining the declines from 1993-96. The new study builds on the 1997 report and finds that TANF has had an even more profound effect on participation than did waivers.

Table A1. Dates of TANF Implementation and Major Welfare Waivers

Date of First Major Waiver

TANF Implementation

Approval

Implementation

Official

Actual, if Different from Official

Alabama

November-96

Alaska

July-97

Arizona

May-95

November-95

October-96

Arkansas

April-94

July-94

July-97

California

October-92

December-92

November-96

January-98

Colorado

July-97

Connecticut

August-94

January-96

October-96

Delaware

May-95

October-95

March-97

DC

March-97

Florida

June-96

October-96

Georgia

November-93

January-94

January-97

Hawaii

June-94

February-97

July-97

Idaho

August-96

July-97

Illinois

November-93

November-93

July-97

Indiana

December-94

May-95

October-96

Iowa

August-93

October-93

January-97

Kansas

October-96

Kentucky

October-96

Louisiana

January-97

Maine

June-96

November-96

Maryland

August-95

March-96

December-96

Massachusetts

August-95

November-95

September-96

Michigan

August-92

October-92

September-96

Minnesota

July-97

Mississippi

September-95

October-95

October-96

July-97

Missouri

April-95

June-95

December-96

Montana

April-95

February-96

February-97

Nebraska

February-95

October-95

December-96

Nevada

December-96

New Hampshire

June-96

October-96

New Jersey

July-92

October-92

February-97

July-97

New Mexico

July-97

New York

December-96

November-97

North Carolina

February-96

July-96

January-97

North Dakota

July-97

Ohio

March-96

July-96

October-96

Oklahoma

October-96

Oregon

July-92

February-93

October-96

Pennsylvania

March-97

Rhode Island

May-97

South Carolina

May-96

October-96

South Dakota

March-94

June-94

December-96

Tennessee

July-96

September-96

October-96

Texas

March-96

June-96

November-96

Utah

October-92

January-93

October-96

Vermont

April-93

July-94

September-96

Virginia

July-95

July-95

February-97

Washington

September-95

January-96

January-97

West Virginia

January-97

Wisconsin

June-94

January-96

September-96

September-97

Wyoming

January-97

*New Mexico implemented its TANF program in July 1997. It was found unconstitutional in September 1997.

A revised TANF program was implemented in April 1998.

 

 

 

 

 

References

Bartik, Timothy J., and Randall W. Eberts (1998). "Examining the Effect of Industry Trends and Structure on Welfare Caseloads," mimeograph, November, W.E. Upjohn Institute for Employment Research.

Bernstein, Jared, and John Schmitt (1998). Making Work Pay: The Impact of the 1996-97 Minimum Wage Increase. Economic Policy Institute, Washington, D.C.

Blank, Rebecca M. (1997). "What Causes Public Assistance Caseloads to Grow?" National Bureau of Economic Research Working Paper No. 6343.

Blank, Rebecca M., and Patricia Ruggles (1996). "When Do Women Use AFDC and Food Stamps? The Dynamics of Eligibility vs. Participation," Journal of Human Resources, 31(1): 57-89.

Brown, Charles, Curtis Gilroy, and Andrew Kohen. (1982). "The Effects of the Minimum Wage on Employment and Unemployment," Journal of Economic Literature, 20(June): 487-528.

Card, David (1992a). "Using Regional Variation in Wages to Measure the Effects f the Federal Minimum Wage," Industrial and Labor Relations Review, 46(1): 22-37

Card, David (1992b). "Do Minimum Wages Reduce Employment? A Case Study of California, 1987-1989," Industrial and Labor Relations Review, 46(1): 38-54

Card, David, Lawrence F. Katz, and Alan B. Krueger (1994). "Comment on David Neumark and William Wascher, "Employment Effects of Minimum and Subminimum Wages: Panel Data on State Minimum Wage Laws," Industrial and Labor Relations Review, 47(3): 487-496.

Card, David, and Alan B. Krueger (1998). "A Reanalysis of the Effect of the New Jersey Minimum Wage Increase on the Fast-Food Industry with Representative Payroll Data," National Bureau of Economic Research Working Paper 6386.

Council of Economic Advisers (1997). "Explaining the Decline in Welfare Receipt, 1993-1996: Technical Report," Executive Office of the President of the United States.

Figlio, David N., and James P. Ziliak (1998). "Welfare Reform, the Business Cycle, and the Decline in AFDC Caseloads," mimeograph, October, University of Florida.

Gallagher, L. Jerome, Megan Gallagher, Kevin Perese, Susan Schreiber, and Keith Watson (1998). "One Year after Federal Welfare Reform: A Description of State Temporary Assistance for Needy Families (TANF) Decisions as of October 1997," Urban Institute Occasional Paper Number 6.

Katz, Lawrence F., and Alan B. Krueger (1992). "The Effect of the Minimum Wage on the Fast-Food Industry," Industrial and Labor Relations Review, 46(1): 6-21.

Levine, Phillip B. and Diane M. Whitmore (1998). "The Impact of Welfare Reform on the AFDC Caseload," National Tax Association Proceedings - 1997. Washington, DC: National Tax Association, pp. 24-33.

Martini, Alberto, and Michael Wiseman (1997). "Explaining the Recent Decline in Welfare Caseloads: Is the Council of Economic Advisers Right?" mimeograph, July, Urban Institute.

Moffitt, Robert A. (1999). "The Effects of Pre-PRWORA Waivers on AFDC Caseloads and Female Earnings, Income, and Labor Force Behavior," mimeograph, May, Johns Hopkins University.

Neumark, David, and William Wascher (1997). "The New Jersey-Pennsylvania Minimum Wage Experiment: A Re-Evaluation Using Payroll Records," mimeo, March, Michigan State University.

Neumark, David, and William Wascher (1992). "Employment Effects of Minimum and Subminimum Wages: Panel Data on State Minimum Wage Laws," Industrial and Labor Relations Review, 46(1): 55-81.

Neumark, David, and William Wascher (1994). "Employment Effects of Minimum and Subminimum Wages: Reply to Card, Katz, and Krueger," Industrial and Labor Relations Review, 47(3): 497-512.

Page, Marianne E., Joanne Spetz, and Jane Millar (1999). "Does the Minimum Wage Affect Welfare Caseloads?" mimeograph, Public Policy Institute of California.

Scholz, John Karl (1994). "The Earned Income Tax Credit: Participation, Compliance, and Antipoverty Effectiveness," National Tax Journal, 59-81.

Stapelton, David C., Gina Livermore, and Adam Tucker (1998). "Determinants of AFDC Caseload Growth," report by the Lewin Group to the Assistant Secretary for Planning and Evaluation, Department of Health and Human Services. July 1997.

Turner, Mark (1999). "The Effects of Minimum Wages on Welfare Recipiency," mimeo, June, Urban Institute and Johns Hopkins University.

Wallace, Geoffrey, and Rebecca M. Blank (1998). "What Goes Up Must Come Down? Explaining Recent Changes in Public Assistance Caseloads," mimeograph, Northwestern University Department of Economics.

Ziliak, James P., David N. Figlio, Elizabeth E. Davis, and Laura S. Connolly. (1997). "Accounting for the Decline in AFDC Caseloads: Welfare Reform or Economic Growth," Institute for Research on Poverty Discussion Paper No. 1151-97, University of Wisconsin.


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

Privacy Statement