What welfare reform did is a measurement question before a verdict question. The statute under examination is Public Law 104-193, the Personal Responsibility and Work Opportunity Reconciliation Act of 1996, which ended Aid to Families with Dependent Children and replaced an individual entitlement to cash assistance with the Temporary Assistance for Needy Families block grant. State programs under the block grant operated from 1997, and each state set benefit levels, eligibility rules, sanction rules, and time limits inside a federal funding structure that paid a fixed amount regardless of caseload movement. The treatment under examination therefore had no single start date and no single national form, because the federal government had approved state waiver experiments before enactment, and several waiver programs had tested work requirements, earnings disregards, and time limits under the predecessor program during the middle 1990s.

The provisions that produced the record are set out in the guide to the welfare reform provisions and the TANF block grant, and the distinction that organizes the evidence is structural. Under the predecessor program, federal spending rose with eligible enrollment, and a family that met the governing tests held a claim that a state had to honor. Under the block grant, federal spending held at a fixed nominal amount, no individual entitlement survived, and a state gained discretion to narrow entry, reduce grants, divert applicants, and redirect funds toward child care, work programs, and administration. Every finding in the sections that follow traces back to that substitution, to the labor market that received the families affected, or to the tax and benefit programs that operated beside cash assistance.

Five findings organize the evidence. The first finding concerns caseloads: families receiving cash assistance fell by roughly three quarters between the middle 1990s peak and 2018, in the Congressional Research Service series based on Department of Health and Human Services data, and the direction of that movement drew no dispute among named researchers. The second finding concerns employment: employment among single mothers rose sharply during the 1990s, and researchers agreed on the direction while assigning sharply different shares to the statute, to the earned income tax credit expansion of 1993, and to the strongest labor market in a generation. The third finding concerns poverty: the official child poverty rate fell through 2000 and rose afterward, while income based measures recorded an increase in deep poverty among families with children that consumption based measures did not confirm. The fourth finding concerns coverage: the share of families in poverty that received cash assistance fell from about two thirds at enactment to about one fifth by the middle 2010s, in the ratio series published by the Center on Budget and Policy Priorities. The fifth finding concerns responsiveness: during the recession of 2007 to 2009, an entitlement program and a block grant program faced the same shock and answered at very different scales.

Attribution is the difficulty that runs through the first four findings. The waiver experiments, the 1993 tax credit expansion, the 1996 statute, and a falling unemployment rate arrived together, and no national comparison can separate forces that shared a calendar. Researchers answered with variation that the calendar lacked: differences in waiver timing among states, differences in credit amounts by family size, random assignment inside demonstration programs, and comparisons between single mothers and single women without children. Rebecca Blank surveyed the resulting literature in the Journal of Economic Literature in 2002 and described reinforcing forces rather than a single dominant cause. Bruce Meyer and Dan Rosenbaum assigned the largest share of the employment increase to tax changes, while Jeffrey Grogger isolated a smaller and bounded effect for time limits, and the distance between those estimates is a finding in itself.

The recession settled one question. The downturn of 2007 to 2009 gave a clean comparison between an entitlement and a block grant facing the same shock, because unemployment doubled, eligibility for food assistance expanded under national standards, and the block grant amount did not move. The entitlement expanded with need while the block grant did not: food assistance caseloads rose by 45 percent between December 2007 and December 2009 in the state collected series published by the Center on Budget and Policy Priorities, while cash assistance caseloads rose by 13 percent in the same series. Congress answered the gap with a temporary emergency fund for fiscal 2009 and fiscal 2010, and the need for a separate enactment confirmed the structural point. A financing rule, rather than a judgment about any individual family, produced the difference in response.

The sections that follow state each finding with the series, the researchers, and the identification problem that belong to that finding, and the closing section restates each finding with a confidence level. Two verdicts on the statute, a success reading associated with Ron Haskins and a failure reading associated with Peter Edelman, Kathryn Edin, and H. Luke Shaefer, receive full statement in the final body section, because a verdict can be weighed only after the movements underneath the verdict have been separated from the explanations offered for those movements. Evidence in every section comes from the record available by October 15, 2019, and each figure carries the name of the institution or the researchers who produced the figure.

The five-finding evidence table

Finding	Direction	Period	Principal researchers	Identification problem	Rating
Caseloads	Fell by roughly three quarters	1994 to 2018	Congressional Research Service and Grogger, Haider and Klerman	Decline began before enactment under waivers and a strengthening labor market	Settled
Employment	Rose sharply for single mothers	1993 to 2000	Meyer and Rosenbaum, Grogger and Blank	Statute, tax credit expansion and tight labor market arrived together	Settled as to direction, contested as to attribution
Poverty and deep poverty	Official child poverty fell then rose, deep poverty rose on income measures	1993 to 2011	Trisi and Sherman, Edin and Shaefer, Meyer and Sullivan	Income reporting error and exclusion of noncash benefits and tax credits	Contested
Coverage	Fell from 68 per 100 families in poverty to 23 per 100	1996 to 2015	Center on Budget and Policy Priorities	Ratio measures reach against poverty rather than receipt among state eligible families	Settled
Recession response	Food assistance rose 45 percent while cash assistance rose 13 percent	December 2007 to December 2009	Pavetti, Trisi and Schott, Bitler and Hoynes	State discretion varied inside a common national shock	Settled

The Caseload Decline

Any account of what welfare reform did has to begin by deciding what was measured, from which starting point, and against which alternative. The statute at the center of the inquiry is Public Law 104-193, the Personal Responsibility and Work Opportunity Reconciliation Act of 1996. That law ended Aid to Families with Dependent Children and replaced that program with Temporary Assistance for Needy Families, a fixed block grant under which states designed and operated state programs of cash assistance for needy families with children. The object counted after 1996 was therefore not a single national program with uniform rules. The object was a set of state programs, sharing a federal funding stream and federal requirements, but differing in benefit levels, eligibility rules, work rules, sanction rules, and time limits, as the Congressional Research Service describes in its legislative history of the Temporary Assistance for Needy Families block grant. The structural contrast between the replaced program and the successor program is set out in a direct comparison of the AFDC and TANF welfare models.

The timing creates the first complication. Enactment took place in 1996, and state Temporary Assistance for Needy Families programs operated from 1997, so a simple account treats 1996 as the last year of the old order and 1997 as the first year of the new order. The boundary, however, was not clean. Before enactment, the federal government had already allowed states to test many devices that the 1996 law later generalized. The Congressional Research Service, in Temporary Assistance for Needy Families: Welfare Waivers, records that between January 1993 and August 1996 a total of 83 waiver applications from 43 states and the District of Columbia were approved under the waiver authority that applied to Aid to Families with Dependent Children. Some demonstrations were small, confined to selected counties or narrow groups, and some were statewide programs of state designed cash assistance and work activity.

The content of the waivers deepens the problem. According to the Congressional Research Service account of the waiver period, state waiver programs tested disregarding a greater share of earnings in determining eligibility and benefits than federal law allowed, permitting greater savings while retaining eligibility, expanding work requirements to more mothers receiving assistance than federal law required, and imposing time limits in some states. The policy treatment therefore did not begin on a single date. In a state with a statewide waiver in operation in 1994 or 1995, a family entering assistance faced rules resembling the rules of the later state program. A comparison that treats every case recorded in 1995 as untreated, and every case recorded in 1997 as treated, misclassifies part of the earlier caseload and assigns to the statute alone effects that belonged in part to state action taken before the statute existed.

The caseload record confirms that movement predated enactment. In the Congressional Research Service series based on data from the United States Department of Health and Human Services, recipients, meaning individual adults and children on whose behalf assistance was paid, averaged 14.161 million in calendar year 1994 and 12.321 million in calendar year 1996. Families, meaning assistance cases rather than individual persons, averaged 5.033 million in calendar year 1994 and 4.434 million in calendar year 1996, in the same series. The peak therefore sits in 1994, and a decline of roughly one eighth in the recipient count occurred between 1994 and 1996, before any state Temporary Assistance for Needy Families program had begun operation. That early decline does not show that the statute had no effect. The early decline shows that the starting point changes the size of the effect attributed to the statute, and that economic conditions, waiver policies, and administrative practice were already moving the count.

A second decision concerns the unit of the count. Families and recipients are not interchangeable measures. A family in the administrative series is a case, a household unit approved for assistance, and a recipient is a person within such a case, most often a child. Because an average assistance family contained more than one recipient, the recipient count was larger than the family count by a factor approaching three in the mid 1990s, and the two counts did not have to move at identical rates. A statement that the caseload fell by a given amount carries no meaning until the measure is named, and a statement that mixes a family figure for one year with a recipient figure for another year manufactures a movement that the underlying series does not contain.

A third decision concerns the period covered by a figure. Some figures are monthly counts, some are annual averages of monthly counts, and some are estimates of distinct persons who received assistance at any point during a year, a measure necessarily larger than an annual average because families enter and leave during a year. The Congressional Research Service illustrates the gap in work using the TRIM3 microsimulation model, summarized in Temporary Assistance for Needy Families: Size of the Population Eligible for and Receiving Cash Assistance. In that analysis, an estimated 17.6 million persons received cash assistance at some time during 1995, whereas the administrative series based on Department of Health and Human Services data gives an annual average of 13.418 million recipients for 1995. The difference is a difference of definition, not a disagreement about underlying events.

With those cautions in place, the central fact can be stated with a precision that disputes in the field rarely disturb. The number of families receiving cash assistance, and the number of recipients within those families, fell by roughly three quarters between the mid 1990s peak and the late 2010s. The decline was large in absolute and proportional terms, sustained across two decades with one interruption during and after the recession of 2007 to 2009, and visible in both measures. Disagreement in the literature concentrates on why the count fell, on which policies and economic conditions produced which shares of the fall, and on what happened to families who left or never entered, and disagreement concentrates far less on whether the count fell. That combination, a firm count and a contested explanation, is the reason the caseload decline functions as the cleanest and least disputed fact in the field.

How far did caseloads actually fall?

In the Congressional Research Service series based on Department of Health and Human Services data, recipients fell from 14.161 million in 1994 to 3.150 million in 2018, a decline of about 78 percent, while families fell from 5.033 million in 1994 to 1.196 million in 2018.

The steepest segment ran from 1996 to 2001. In the Congressional Research Service series based on Department of Health and Human Services data, recipients averaged 6.143 million in 2000 and 5.717 million in 2001, after averaging 12.321 million recipients in 1996, so the recipient count fell by more than half between 1996 and 2000. Families averaged 2.303 million in 2000 and 2.192 million in 2001, after averaging 4.434 million families in 1996, in the same series. After 2001 the direction persisted and the pace changed. The Congressional Research Service narrative in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions describes the caseload as falling rapidly in the late 1990s, leveling off in 2001, and beginning another decline in 2004 at a slower pace. The annual figures match that description, with recipients averaging 4.005 million in 2008 and families averaging 1.701 million in 2008, in the same Congressional Research Service series.

The recession of 2007 to 2009 interrupted the decline without reversing the larger movement. In the monthly series described by the Congressional Research Service in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions, families receiving assistance rose from 1.7 million families in August 2008 to a peak close to 2.0 million families in December 2010. Measured against the fall from the 1994 peak, that increase was modest, and the modesty became a point of dispute, because a caseload that responds only slightly to a severe recession behaves differently from a caseload governed mainly by need. After the recession era peak, decline resumed. In the Congressional Research Service series based on Department of Health and Human Services data, recipients averaged 3.516 million in 2017 and 3.150 million in 2018, and families averaged 1.358 million in 2017 and 1.196 million in 2018. The Congressional Research Service, in The Temporary Assistance for Needy Families Block Grant, an In Focus summary based on Department of Health and Human Services data, reports that 1.2 million families received assistance in September 2018, against a peak of 5.1 million families in March 1994, and that the September 2018 figure was the lowest monthly family count since February 1967.

The meaning of the decline depends on a distinction that headline summaries often collapse, the distinction between a count of cases and a measure of need. A caseload is an administrative count. The count records the outcome of applications, eligibility determinations, compliance decisions, and case closures inside state offices, under state rules, during a defined period. A caseload is not a poverty measure, not an income measure, not an employment measure, and not a measure of material hardship. A family can leave the caseload with earnings above the poverty threshold, with earnings below that threshold, with no earnings, with support from relatives, or with benefits from another program, and the series records each departure in the same way, as one fewer case. A family can also remain outside the caseload while meeting every financial test for assistance, and the series records that family not at all. The decline therefore establishes that far fewer families received cash assistance at the end of the period than at the beginning, and the decline establishes nothing by itself about whether families outside the program had greater income, steadier work, or lower hardship.

The eligibility record makes that gap concrete. The Congressional Research Service, in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions, states that most of the decline after 1994 resulted from a reduction in the share of eligible families receiving benefits, rather than from a reduction in the number of families meeting state definitions of a needy family. The same summary gives that share, on the family measure, as an estimated 82 percent in 1995 and an estimated 27 percent in 2015. Eligibility therefore did not contract at the rate at which receipt contracted, and the decline was, in large part, a decline in take up among eligible families. Take up can fall because earnings made assistance unnecessary, because application procedures made entry difficult, or because a small benefit was judged not worth the compliance burden attached to that benefit. The count registers the combined result as absence from the rolls, and the count alone cannot apportion absence among those paths.

The arithmetic of a caseload also matters, because a caseload is a stock produced by flows. The stock falls when exits exceed entries, when entries fall, or when both movements occur together. Jeffrey Grogger, Steven Haider, and Jacob Klerman, in Why Did the Welfare Rolls Fall during the 1990s, published in the American Economic Review in 2003, examined that flow arithmetic for the 1990s decline. The Congressional Research Service summary of the research literature in its report on the population eligible for and receiving cash assistance cites the Grogger, Haider, and Klerman study for the finding that the reduction resulted from both faster exits from state assistance programs and fewer entries into state assistance programs. The dual source has a practical consequence, because a study that follows families after exit cannot observe families who were diverted or deterred before entry, and leaver evidence alone cannot describe a decline driven in part by non entry.

Exit itself is a bundle of unlike events. A case could close because earnings rose enough to end eligibility, because a time limit was reached, because a sanction for noncompliance reduced or terminated the grant, because a family moved or failed to complete a redetermination, or because a recipient asked for closure. The Brookings Institution, in Welfare Reform and Devolution: Looking Back and Forward, observed that in some states the use of sanctions accounted for a major part of the caseload decline in that state. A sanction closure and an earnings closure both subtract one case, and neither closure, recorded alone, records whether earnings followed after closure. Time limits operated on a delay. Federal law limited federally funded aid to families with an adult recipient to five years, as the Congressional Research Service states in The Temporary Assistance for Needy Families Block Grant, its In Focus summary. A limit enacted in 1996 could not produce terminations for exhaustion of the federal limit until several years after state programs began, although states could impose shorter limits and some waiver programs had started time limit clocks before 1996, so the immediate fall between 1996 and 2000 requires explanation from other sources, including work requirements, sanctions, diversion, the labor market, and the earned income tax credit.

Diversion operated before entry, and diverted families leave no case record of the ordinary kind. Formal diversion included lump sum payments to meet an immediate need in place of an ongoing case, and requirements that applicants conduct and document a job search before approval. Informal diversion included application experiences that led an applicant to withdraw before a determination. Each form reduces the entry flow, and a reduced entry flow lowers the stock even if exit behavior does not change. The Brookings Institution review named Welfare Reform and Devolution: Looking Back and Forward records the message received by states in the earliest stage of implementation as the importance of caseload reduction, and records that states responded to that message. Federal rules reinforced attention to caseload size through the caseload reduction credit, under which a state could lower the work participation percentage required of that state by reducing the caseload, as the Congressional Research Service describes in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions.

Composition and state variation add further cautions. The Congressional Research Service, in tabulations of the Temporary Assistance for Needy Families national data files reported in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions, gives the share of assistance families that was child only in fiscal year 2017 as 38 percent, with child only defined as families in which no adult recipient received benefits in the adult capacity. The reform instruments aimed at adults had limited direct application to a large fraction of such a caseload. States also set most rules that applied to recipient families, and the Congressional Research Service, in The Temporary Assistance for Needy Families Block Grant, gives the maximum monthly benefit for a single parent family of three in July 2017 as ranging from 170 dollars in Mississippi to 1,021 dollars in New Hampshire, so the national decline is an average of state declines that differed in size and mechanism.

Attribution was contested from the earliest studies. The Council of Economic Advisers, in a 1997 analysis of state caseload data for 1976 to 1996, concluded in preferred estimates summarized by the Center on Budget and Policy Priorities in The Determinants of Welfare Caseload Decline, issued in 1999, that economic growth accounted for 44 percent of the decline in Aid to Families with Dependent Children caseloads from 1993 to 1996 and that welfare waivers accounted for 31 percent of that decline. Geoffrey Wallace and Rebecca Blank, in a 1999 analysis summarized in the same review, found that the decline in unemployment rates explained about 8 percent to 19 percent of the caseload decline after 1994. Rebecca Blank surveyed the literature in Evaluating Welfare Reform in the United States, published in the Journal of Economic Literature in 2002. The spread reflects sensitivity to model choice, to the years included, and to whether waiver adoption is treated as a policy variable or as a marker for broader state changes. The fact of the decline can be accepted at full strength, while any precise division of credit rests on assumptions that the count cannot settle.

The earned income tax credit belongs in any attribution that touches employment. The Omnibus Budget Reconciliation Act of 1993 expanded the credit, with the expansion phased in during the middle 1990s, the same years in which waiver programs spread and state programs began. The credit raised the return to paid work for low income families with children and paid nothing to a parent without earnings, so the credit pushed in the same direction as work requirements and time limits. Bruce Meyer and Dan Rosenbaum, in Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers, published in the Quarterly Journal of Economics in 2001, examined tax changes, benefit cuts, and waivers as explanations for the increase in work by single mothers. In a summary given in a Federal Reserve Bank of Chicago review of the literature, tax changes centered on the credit accounted for about 60 percent of the 12 percentage point increase in employment rates of single mothers between 1984 and 1996, and welfare reform accounted for about 15 percent, while for 1992 to 1996 the credit accounted for about 35 percent and welfare reform accounted for about 20 percent. Employment gains therefore cannot be assigned to welfare reform alone, because a large tax change rewarding work operated at the same time, in the same population, with the same predicted direction of effect.

Counting rules at the edges of the series explain small differences between published totals. The Congressional Research Service notes, in connection with tabulations in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions, that the caseload includes families receiving assistance in state funded programs counted toward the maintenance of effort requirement. The Administration for Children and Families, in releasing 2018 caseload data, states that in an average month in calendar year 2018 about 1.2 million families received Temporary Assistance for Needy Families or separate state program cash assistance, representing approximately 3.2 million recipients. A figure without a named series cannot be checked against another figure for the same year.

The caseload finding, stated with its limits, is a finding about receipt. Families fell by roughly three quarters on the family measure, from 5.033 million families in 1994 to 1.196 million families in 2018, and recipients fell by roughly three quarters on the recipient measure, from 14.161 million recipients in 1994 to 3.150 million recipients in 2018, in the Congressional Research Service series based on data from the United States Department of Health and Human Services. The fall began before enactment of Public Law 104-193, accelerated after state programs began operation in 1997, slowed after 2001, rose modestly during and after the recession of 2007 to 2009, and resumed after 2010. What the fall meant for income, poverty, employment, and hardship among families inside and outside the count is a separate question, requiring separate evidence, and the sections that follow take up that evidence on its own terms.

Employment and the Attribution Problem

Employment among single mothers rose sharply during the 1990s, and argument in the research literature concerned explanation rather than existence. Bruce D. Meyer and Dan T. Rosenbaum, in “Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers,” published in The Quarterly Journal of Economics in 2001, measured the change in two named Current Population Survey series, the Outgoing Rotation Group files and the March files. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, reported that weekly employment among single mothers increased by about six percentage points between 1984 and 1996, while annual employment increased by nearly nine percentage points. The employment rise unfolded against the strongest labor market in a generation and alongside the expansion of the earned income tax credit enacted in 1993, and neither condition can be separated from the record by timing alone.

The labor market condition was exceptional. The Bureau of Labor Statistics, in State and Regional Unemployment, 2000 Annual Averages, published in 2001, reported that the national unemployment rate averaged 4.2 percent in 1999 and 4.0 percent in 2000. The Bureau of Labor Statistics, in “The Employment Situation: April 2000,” published in 2000, reported a seasonally adjusted rate of 3.9 percent in April 2000, the first rate below 4.0 percent since January 1970. The tax condition changed on a parallel track. The Congressional Research Service, in The Earned Income Tax Credit (EITC): A Brief Legislative History, report R44825, published in 2017, recorded a maximum credit for a family with two children of 1,384 dollars in 1992, against 1,324 dollars for a family with one child. The Congressional Research Service, in report R44825, published in 2017, recorded that the expansion enacted in the Omnibus Budget Reconciliation Act of 1993 raised the maximum for a family with two children to 2,528 dollars by 1994, with the schedule reaching the enacted level by 1996. The legislative path of that credit is traced in the account of the earned income tax credit law.

Was it welfare reform or the economy?

Employment rose under all three forces at once. The 1996 statute, state waivers, and time limits pushed toward work, the 1993 earned income tax credit expansion paid for work, and the tight late 1990s labor market supplied jobs, so no single cause accounts for the full increase in the evidence reviewed through 2019.

The contest persisted because the treatments arrived together. State waivers operated before the federal statute, the earned income tax credit expansion enacted in 1993 phased in between 1994 and 1996, the Personal Responsibility and Work Opportunity Reconciliation Act was enacted in 1996, and state Temporary Assistance for Needy Families programs began operation in 1997. Unemployment fell across the span from 1993 to 2000 in the annual averages published by the Bureau of Labor Statistics in 2001. A national before and after comparison around 1996 therefore bundles waiver policy already in place, a federal statute that generalized waiver devices and added a federal time limit, a tax credit that raised the return to entering work, and a labor market that absorbed workers with limited experience at unusual rates. Rebecca M. Blank, in “Evaluating Welfare Reform in the United States,” published in the Journal of Economic Literature in 2002, treated that bundling as a central problem in a survey of the econometric and experimental literature on caseload, labor force, poverty, income, and family formation changes.

Researchers attempted separation through variation that the national calendar lacked. States adopted major statewide waivers on different dates before 1996, implemented Temporary Assistance for Needy Families programs on different schedules after 1996, and set different benefit levels, earnings disregards, sanction rules, and time limits. Family structure supplied a second source of variation, because the earned income tax credit expansion enacted in 1993 paid different amounts by number of children, in the schedule recorded by the Congressional Research Service in report R44825, published in 2017, while welfare rules applied to families receiving or applying for cash assistance. Experimental assignment in waiver demonstrations supplied a third source, because eligible families entered a reform program or a control program by lottery.

Comparison groups carried much of the identification burden, and the choice of group moved the answers. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, built the central comparison around single women without children, because tax and welfare law treated single mothers and single women without children differently, so a change in the gap between the groups could reveal policy effects that a change in the level for single mothers alone could not. The design held a weakness, because single women without children differed in education and earnings capacity and responded to the strongest labor market in a generation through different occupations, so parallel movement had to be assumed for the counterfactual years. Mothers with two or more children gained more from the 1993 earned income tax credit expansion than mothers with one child, in the Congressional Research Service schedule in report R44825, published in 2017, and a contrast between those groups isolated the credit while entangling family size with welfare eligibility.

State waiver timing supplied the variation most favorable to a welfare reform reading, and the caseload literature showed the fragility of that variation. The Council of Economic Advisers, in Explaining the Decline in Welfare Receipt, 1993 to 1996, published in 1997, used annual state panel data and attributed about 44 percent of the decline in welfare receipt between 1993 and 1996 to economic growth and about 31 percent to state welfare waivers. The Council of Economic Advisers, in the Economic Report of the President, published in 1999, attributed about 20 percent of the decline in caseloads between 1994 and 1998 to economic conditions. Those figures measured caseloads rather than employment, and the distance between the 1997 and 1999 attributions showed how a share could move when the period, the controls, or the waiver coding changed. Blank, in the Journal of Economic Literature in 2002, reported a related caution, namely that a waiver indicator could stand for a bundle of administrative and political changes inside an adopting state rather than for a single rule.

The time limit mechanism received the sharpest theoretical statement in the literature, and the statement generated a testable prediction. A time limit converts cash assistance into a finite stock of eligible months. A recipient who expects future need, faces uncertain earnings, and cannot borrow freely gains an incentive to conserve eligible months by leaving assistance earlier, entering work earlier, or avoiding entry, before any limit binds. Jeffrey Grogger and Charles Michalopoulos, in “Welfare Dynamics under Time Limits,” published in the Journal of Political Economy in 2003, derived the prediction that conservation would operate most strongly among families with the youngest children, because a young child lengthened the horizon over which conserved months held value. Grogger and Michalopoulos, in the Journal of Political Economy in 2003, tested the prediction in the Family Transition Program in Escambia County, Florida, a randomized demonstration in which the treatment group faced limits of 24 months in a 60 month period for most recipients and 36 months in a 72 month period for a more disadvantaged group, alongside a financial work incentive, while the control group faced no limit. The estimated pattern matched the predicted age gradient. Grogger and Michalopoulos, in the Journal of Political Economy in 2003, estimated that the time limit in that program, absent other reforms that raised welfare use, would have reduced welfare receipt by 16 percent. The experiment took place in the tight labor market of the middle 1990s and after the 1993 earned income tax credit expansion.

Jeffrey Grogger carried the time limit question to national survey data, and the national estimates assigned time limits a meaningful and bounded share. Grogger, in “The Effects of Time Limits, the EITC, and Other Policy Changes on Welfare Use, Work, and Income among Female-Headed Families,” published in The Review of Economics and Statistics in 2003, used March Current Population Survey data and the age based predictions developed with Michalopoulos to separate time limits from the remaining waiver and Temporary Assistance for Needy Families bundle and from the earned income tax credit. Grogger, in The Review of Economics and Statistics in 2003, estimated that time limits accounted for about one eighth of the decline in welfare use since 1993 and about 7 percent of the rise in employment since 1993, with no significant effect on earnings or income. The same analysis assigned important effects to the collective remainder of welfare reforms and identified the earned income tax credit as a particularly important contributor to the decrease in welfare use and to the increase in employment, labor supply, and earnings.

The full strength case for the welfare bundle rested on more than a single coefficient. Anticipatory response meant that an employment effect could precede any termination, so a count of terminated families understated the reach of the policy in the framework published by Grogger and Michalopoulos in the Journal of Political Economy in 2003. Work requirements, sanctions, diversion at application, and earnings disregards operated alongside limits inside state programs, and a design that credited only the limit would assign the bundle effect to whichever component a specification isolated. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, reported that one or two year leads of time limit and termination variables carried small and statistically insignificant coefficients, a result interpreted in that article as evidence that the provisions themselves, rather than publicity or administrator attitudes alone, produced the associated employment increases. Movement also followed waiver adoption in several state panel designs reviewed by Blank in the Journal of Economic Literature in 2002.

Meyer and Rosenbaum produced the decomposition most favorable to the earned income tax credit, built to compare policies inside a single sample, period, and method. In the main specifications reported by Meyer and Rosenbaum in The Quarterly Journal of Economics in 2001, the earned income tax credit explained 62 percent of the increase in weekly employment of single mothers relative to single women without children between 1984 and 1996 and 61 percent of the increase in annual employment. For the shorter span between 1992 and 1996, Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, assigned the credit 27 percent of the increase in weekly employment and 35 percent of the increase in annual employment, and assigned waivers about 14 percent to 20 percent across that span. The shares described the closing of a gap between two groups, measured in a tight labor market and across the years of the 1993 earned income tax credit expansion, rather than the whole rise in the level of employment among single mothers.

That relative construction explained much of the distance between the Meyer and Rosenbaum shares and the Grogger shares. A factor that raised employment for single mothers and for single women without children alike dropped out of a relative decomposition, because the gap recorded no movement from a common force. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, handled the labor market outside the headline shares and estimated that changes in state unemployment rates raised the absolute level of employment of single mothers by 2.0 percentage points during a typical week and by 1.4 percentage points during a year between 1984 and 1996. Those amounts equaled 28.4 percent of the relative increase in weekly employment and 12.2 percent of the relative increase in annual employment in the calculations reported by Meyer and Rosenbaum in The Quarterly Journal of Economics in 2001. A comparison of the Meyer and Rosenbaum tax share with the Grogger time limit share therefore compares a share of a relative increase over 1984 to 1996 with a share of an absolute increase after 1993, in different samples and under different assumptions. Blank, in the Journal of Economic Literature in 2002, presented the divergence as a consequence of data, period, and method.

An earlier tax study supplied the foundation for the Meyer and Rosenbaum decomposition. Nada Eissa and Jeffrey B. Liebman, in “Labor Supply Response to the Earned Income Tax Credit,” published in The Quarterly Journal of Economics in 1996, compared single women with children and single women without children around the expansion enacted in the Tax Reform Act of 1986. Eissa and Liebman, in The Quarterly Journal of Economics in 1996, estimated an increase in labor force participation for single mothers relative to single women without children of up to 2.8 percentage points attributable to that expansion, with a larger response among women with limited education. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, reported that tax changes between 1984 and 1996 raised weekly employment by 4.4 percentage points and annual employment by 7.2 percentage points in the main specifications, with about 40 percent of that change concentrated between 1992 and 1996. The gradient by family size strengthened the tax reading, because the 1993 expansion raised the maximum credit for families with two or more children far more than for families with one child, in the Congressional Research Service schedule in report R44825, published in 2017, and employment rose most among groups with the largest scheduled gains in the Meyer and Rosenbaum analysis published in 2001.

The full strength case for the earned income tax credit rested on mechanism as well as on shares. The credit paid only to a worker, rose with earnings across an initial range, and reached a single mother at the margin between no work and some work, so the schedule raised the payoff to entering employment without contact with a welfare office, compliance monitoring, or a sanction. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, concluded that financial incentives exerted substantial effects on employment decisions and that the earned income tax credit and other tax changes played the dominant role in the employment increase between 1984 and 1996, a conclusion connected in that article to a preference for rewarding work over punitive measures such as time limits and work requirements. Grogger, in The Review of Economics and Statistics in 2003, reached a compatible emphasis from a design built to elevate time limits, describing the earned income tax credit on page 408 of that article as a candidate for the single most important policy measure behind the decrease in welfare use and the rise in work and earnings among female headed families.

The full strength case for the economy rested on demand. Employment among single mothers rose before the federal statute of 1996, during the waiver years and during the phase in of the 1993 earned income tax credit expansion, in the Meyer and Rosenbaum series published in The Quarterly Journal of Economics in 2001, and the rise accelerated as unemployment fell toward 4.0 percent in the Bureau of Labor Statistics annual averages published in 2001. A tight labor market raised the probability that a job search ended in an offer, raised hours offered to workers with limited experience, and drew employers toward applicants passed over under slacker conditions. The Council of Economic Advisers, in Explaining the Decline in Welfare Receipt, 1993 to 1996, published in 1997, assigned the largest single share of the caseload decline in that span to economic growth, at about 44 percent. Under the economy reading, welfare reform and the earned income tax credit expansion enacted in 1993 changed incentives on the supply side, while the strongest labor market in a generation supplied the vacancies that converted changed incentives into recorded employment.

Blank offered the synthesis that organized the field without closing the contest. Blank, in “Fighting Poverty: Lessons from United States History,” published in the Journal of Economic Perspectives in 2000, concluded that the changes of the 1990s were causally related to the strong macroeconomy, to welfare reform efforts, and to other policy changes, with the earned income tax credit and minimum wage expansions named in that conclusion, and that the forces reinforced one another in producing large behavioral changes. Blank, in the Journal of Economic Literature in 2002, extended the synthesis across the econometric and experimental record and reported a growing body of evidence that the policy changes of the 1990s had influenced behavior and well being in a variety of ways, with a particular set of new style welfare programs showing increased work and earnings and reduced poverty in the studies under review. Employment effects, under the synthesis in Blank, in the Journal of Economic Perspectives in 2000 and in the Journal of Economic Literature in 2002, were joint products of the three forces, and disagreement concerned shares under interaction rather than inclusion.

Interaction defeated additive accounting in principle, and the literature acknowledged the defeat while publishing shares. An earned income tax credit expansion pays nothing to a person without earnings, so the employment effect of the 1993 expansion depended on vacancies that the strongest labor market in a generation supplied. A work requirement or a time limit moves a recipient toward a job search, and the yield of that search depended on the same vacancies and on a tax schedule that made a low wage job pay more after the 1993 expansion. A decomposition assigning fixed percentages to the statute, the credit, and the economy therefore imposed separability on a process that Blank, in the Journal of Economic Perspectives in 2000, described as reinforcement. A specification assigning family size variation to the credit credited the credit with variation that a specification allowing welfare effects to vary by family size could assign to welfare policy, because benefit amounts and eligibility also varied by family size.

Measurement choices added a second layer of uncertainty. Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, noted a concern surrounding the redesign of the Current Population Survey questionnaire that began in January 1994, in the middle of the span of rapid employment growth, and a redesign altering reported work or program participation could shift a measured trend independently of behavior. Participation and hours also diverged, because Eissa and Liebman, in The Quarterly Journal of Economics in 1996, and Meyer and Rosenbaum, in The Quarterly Journal of Economics in 2001, each found the response concentrated in the decision to work rather than in hours among workers. A participation gain could coexist with limited movement in earnings, as in the Grogger estimates published in The Review of Economics and Statistics in 2003, where time limits showed no significant effect on earnings or income.

By 2019, the unsettled core of the fight had a defined shape. Agreement extended to direction: the welfare bundle raised employment, the earned income tax credit expansion enacted in 1993 raised employment, and the strongest labor market in a generation raised employment, in the synthesis offered by Blank in the Journal of Economic Literature in 2002 and in the primary estimates published by Grogger in The Review of Economics and Statistics in 2003 and by Meyer and Rosenbaum in The Quarterly Journal of Economics in 2001. Agreement failed on magnitude and on transportability. The share assigned to the credit ranged from about one quarter to about three fifths of a measured increase depending on span, outcome definition, and whether the increase was stated in absolute or relative terms, in the Meyer and Rosenbaum decomposition published in The Quarterly Journal of Economics in 2001. The share assigned to time limits stood at about 7 percent of the employment rise after 1993 in the Grogger estimates published in The Review of Economics and Statistics in 2003, while the collective welfare bundle received larger and less stable shares across the studies reviewed by Blank in the Journal of Economic Literature in 2002. No study in the record reviewed through 2019 observed the 1996 statute under a slack labor market with an unexpanded credit, or observed the 1993 expansion under slack demand at the scale of the late 1990s. The counterfactual path under one force without the others therefore rested on model assumptions rather than on an observed American experience, and employment effects remained effects estimated jointly in a tight labor market and after the 1993 earned income tax credit expansion.

Poverty and the Deep Poverty Dispute

The poverty record after the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 contained two findings that pointed in different directions. On the official poverty measure published by the Census Bureau, poverty among children fell sharply during the second half of the 1990s, reached a low point at the end of that decade, and then rose during the following decade. On several income based measures, the share of families with children living far below the poverty line, in the condition researchers called deep poverty, increased during years in which the overall child poverty rate first fell. The official rate counted every person below a single threshold in the same way, regardless of distance from that threshold. The dispute examined in this section arose because researchers disagreed over whether the second finding described a real change in living conditions or an artifact produced by the instruments used to measure income at the bottom.

The official measure provided the common starting point, and its definition mattered for everything that followed. The Census Bureau calculated the official rate from pretax cash income reported in the Current Population Survey Annual Social and Economic Supplement, comparing that income against thresholds that varied by family size and composition and that rose with inflation. The measure counted earnings, cash welfare benefits, Social Security benefits, unemployment insurance, and other cash income. The measure did not count the Earned Income Tax Credit or other tax credits, and the measure did not count noncash benefits such as food assistance under the Supplemental Nutrition Assistance Program, housing subsidies, or school meals. The trend in the official rate therefore described movement in pretax cash income relative to the thresholds rather than movement in total resources.

Within that series, the path of child poverty had a distinct shape. The Center on Budget and Policy Priorities, summarizing the Census Bureau report for 2000, stated that the poverty rate among children fell from 22.7 percent in 1993, the peak year in that series, to 16.2 percent in 2000, the lowest level recorded in the series since 1979. The Assistant Secretary for Planning and Evaluation of the Department of Health and Human Services, summarizing the Current Population Survey data released in 2011, reported the same 16.2 percent rate for 2000 and then traced the reversal, with the child poverty rate at 17.8 percent in 2004, 18.0 percent in 2007, 20.7 percent in 2009, and 22.0 percent in 2010. The 2010 figure represented an increase of 5.8 percentage points from the 2000 figure, and the summary described the 2010 rate as the highest rate since 1993. The complete Census Bureau series therefore contained both a large reduction from 1993 to 2000 and a large increase from 2000 to 2010, in sequence.

Interpretation of the first movement divided along familiar lines. Analysts who emphasized the late 1990s decline pointed to strong employment growth, rising earnings among single mothers, the expanded Earned Income Tax Credit enacted in 1993, state welfare waivers that preceded the federal law, and the work requirements and time limits of the 1996 law itself, The Congressional Research Service, covering female headed families with children from 1987 to 2013, recorded that cash welfare receipt declined substantially among single mothers after 1996 while work increased, and that the change began before the 1996 law. Analysts who emphasized the rise after 2000 pointed to two recessions, in 2001 and in 2007 to 2009, and to the reduced reach of cash assistance, which meant that a job loss translated into a larger income loss for a family that received no cash benefit than for a family on the earlier Aid to Families with Dependent Children program. The official series alone could not separate those explanations, since the series recorded outcomes rather than causes.

Deep poverty had a precise definition, distinct from both poverty and the extreme poverty threshold used by Kathryn Edin and H. Luke Shaefer. A person lived in deep poverty, in the standard definition used by the Center on Budget and Policy Priorities and by Yonatan Ben Shalom, Robert Moffitt, and John Karl Scholz, when family income fell below half of the official poverty threshold for a family of that size and composition. Danilo Trisi and Arloc Sherman of the Center on Budget and Policy Priorities, in the 2014 analysis titled Deep Poverty Among Children Worsened in Welfare Law First Decade, used a comprehensive income measure that counted cash income, noncash benefits, and tax credits and that corrected for the underreporting of safety net benefits in Census Bureau data. Trisi and Sherman estimated that the share of children in deep poverty rose from 2.1 percent in 1995 to 3.0 percent in 2005, and that the number of children in deep poverty climbed from 1.5 million to 2.2 million across the same ten years. The endpoints held labor market conditions roughly constant, with unemployment at 5.6 percent in 1995 and 5.1 percent in 2005.

The increase appeared with particular clarity among the families most directly subject to the 1996 law. A Center on Budget and Policy Priorities analysis of the same 1995 to 2005 comparison reported that deep poverty rose from 5.4 percent to 7.4 percent among children in single mother families, corresponding to 300,000 additional children in deep poverty in 2005. Trisi and Sherman, in a separate 2016 analysis titled Incomes Fell for Poorest Children of Single Mothers in Welfare Law First Decade, found that average family income for the poorest tenth of children in single mother families fell by nearly 2,400 dollars, or 18 percent, between 1995 and 2005, while income rose for every other group of children of single mothers in the same analysis. Trisi and Sherman attributed the decline for the poorest tenth to a fall of about 2,800 dollars in cash assistance through Temporary Assistance for Needy Families. Related measures pointed in the same direction. Pamela Loprest and Austin Nichols of the Urban Institute defined disconnected single mothers as low income single mothers with no earnings, no cash benefit from Temporary Assistance for Needy Families or Supplemental Security Income, and no enrollment in school, and estimated, in the summary given in later literature, that one in eight low income single mothers met that definition in 1996, while one in five met that definition during the years from 2004 to 2008.

Did welfare reform increase deep poverty?

On cash income measures from household surveys, deep poverty among families with children rose after 1996, and the sharpest increases appeared among single mother families. On consumption measures and on income measures corrected with administrative records, the same increase became smaller, disappeared, or reversed. Researchers disputed the measure, not only the trend.

The extreme poverty claim entered the debate through a specific measure, a specific survey, and a specific pair of researchers. Kathryn Edin, a sociologist, and H. Luke Shaefer, a researcher on poverty measurement, published the article titled Rising Extreme Poverty in the United States and the Response of Federal Means Tested Transfer Programs in Social Service Review in 2013. Shaefer and Edin used the Survey of Income and Program Participation, the Census Bureau survey that interviewed households repeatedly and recorded income by month and by source. The measure in the 2013 article was cash income in the Survey of Income and Program Participation, expressed per person per day, and the threshold was 2.00 dollars per person per day. Shaefer and Edin estimated that the number of households with children living on cash income of 2.00 dollars or less per person per day in a given month increased from about 636,000 in 1996 to about 1.65 million in mid 2011, a growth of 159.1 percent, and that about 3.55 million children lived in such households in mid 2011. Edin and Shaefer then presented the same research program in the book titled $2.00 a Day, Living on Almost Nothing in America, published in 2015. The book reported the headline figures in rounder form, stating that 1.5 million American households lived on 2.00 dollars per person per day, including about 3 million children. The book combined the survey estimates with fieldwork in Chicago, Cleveland, Johnson City in Tennessee, and the Mississippi Delta.

Three features of the Edin and Shaefer argument gave the claim weight beyond the headline threshold. The first feature was an internal adjustment that the authors themselves performed. When Shaefer and Edin counted the Supplemental Nutrition Assistance Program as income in the 2013 article, the number of households in extreme poverty rose by 80.4 percent between 1996 and 2011, from roughly 475,000 to 857,000, rather than by 159.1 percent on cash income alone. When Shaefer and Edin added refundable tax credits and housing subsidies as well as food assistance, the increase fell to about 50 percent, from 409,000 to 613,000 households, including 1.17 million children in 2011. The direction of the trend survived each adjustment, while the magnitude changed substantially. The second feature was a counterfactual calculation. Edin and Shaefer reported in the 2016 paper titled What Is the Evidence of Worsening Conditions among America Poorest Families with Children that Aid to Families with Dependent Children had lifted more than a million households above the 2.00 dollar line in 1996, and that removing that assistance from the calculation eliminated the increase in extreme poverty across the period. The third feature was corroboration from administrative records. Shaefer, Edin, and Elizabeth Talbert, in the 2015 article titled Understanding the Dynamics of $2 a Day Poverty in the United States in the Russell Sage Foundation Journal of the Social Sciences, and Edin and Shaefer in the 2016 paper, cited counts from Supplemental Nutrition Assistance Program administrative data of households with children that received food assistance and reported no other source of cash income. Those counts rose from 316,000 households in 1996 to 599,000 in 2005 and to 1.3 million in 2013.

The opposing argument, developed most fully by Bruce Meyer of the University of Chicago and James Sullivan of the University of Notre Dame, held that the instruments rather than the underlying conditions had changed. In articles in the Journal of Human Resources in 2003 and the Canadian Journal of Economics in 2011, Meyer and Sullivan reported that consumption based classifications identified a more disadvantaged group on other indicators than income based classifications identified. In Identifying the Disadvantaged, Official Poverty, Consumption Poverty, and the New Supplemental Poverty Measure, published in the Journal of Economic Perspectives in 2012, Meyer and Sullivan argued that the official measure omitted tax credits and noncash benefits and contained substantial reporting error. In Winning the War, Poverty from the Great Society to the Great Recession, published in Brookings Papers on Economic Activity in 2012, Meyer and Sullivan reported that a consumption based poverty rate, with thresholds adjusted for bias in the price index used to update the thresholds, fell by 26.4 percentage points between 1960 and 2010, with 8.5 percentage points of that decline occurring after 1980, and that the consumption based measure showed considerably greater improvement for single parent families than income based measures showed. On the consumption approach, poverty among the families at the center of the welfare debate had fallen substantially across the long period that included welfare reform.

The measurement objection had several distinct components. The first component concerned omitted resources. The official measure and the cash only extreme poverty measure both excluded the largest expansions in assistance after 1996. Meyer and Sullivan observed, in the work summarized in the 2019 American Enterprise Institute working paper titled The Use and Misuse of Income Data, that food assistance payments more than doubled between 1996 and 2011 and that Earned Income Tax Credit benefits increased by approximately half across the same period, with both increases transferring more dollars than cash assistance lost through Temporary Assistance for Needy Families. Edin and Shaefer argued that food assistance could not pay rent or a utility bill and that a household with food assistance and almost no cash faced constraints that an income total treating a dollar of food assistance as equivalent to a dollar of cash would conceal. Meyer and Sullivan argued that in kind resources freed cash for other uses and that excluding those resources understated living standards.

The second component concerned reporting error within the amounts that surveys did attempt to record. Bruce Meyer, Wallace Mok, and James Sullivan, in Household Surveys in Crisis, published in the Journal of Economic Perspectives in 2015, compared survey reports against administrative totals for nine large transfer programs and reported a sharp rise in downward bias across the preceding three decades. Meyer, Mok, and Sullivan reported that, in the final years covered by that study, more than half of welfare dollars and nearly half of food stamp dollars went unrecorded in several major surveys, and that the error took the form principally of recipients recorded as nonrecipients rather than of recipients reporting amounts modestly below the amounts received. Error of that form had a concentrated effect at the bottom, because a recipient household recorded as receiving nothing fell to the lowest income category in the survey even though program records showed a benefit. Growing bias also produced a measured increase in zero income households even under a constant true distribution.

Administrative linkage supplied the most direct test of the reporting objection, because linkage replaced a survey answer with a program record for the same household. Bruce Meyer and Nikolas Mittag, in Using Linked Survey and Administrative Data to Better Measure Income, published in the American Economic Journal, Applied Economics, in 2019, linked administrative records for four transfer programs in New York State to responses in the Current Population Survey. The four programs were food assistance, Temporary Assistance for Needy Families, General Assistance, and subsidized housing. Meyer and Mittag reported that program receipt went unrecorded in the survey for more than one third of housing assistance recipients, more than 40 percent of food stamp recipients, and more than 60 percent of recipients of Temporary Assistance for Needy Families and General Assistance, Using the combined data rather than the survey alone nearly doubled the estimated poverty reducing effect of the four programs together and tripled the estimated effect of housing assistance. Correction of the same errors often reduced the share of single mothers classified as falling through the safety net, defined through absence of earnings and absence of program receipt, by one half or more. The linkage evidence came from one state and from the Current Population Survey rather than from the Survey of Income and Program Participation, so the magnitudes did not transfer mechanically to the Edin and Shaefer estimates.

The third component concerned the divergence between income and consumption as concepts, apart from error in either series. Income recorded in a survey described a flow during a reference month or year, and a household with a temporary job loss, a gap between jobs, or irregular earnings could record near zero income for that period while maintaining consumption through savings, borrowing, credit, help from relatives, or durable goods and housing acquired earlier. Consumption in the Consumer Expenditure Survey smoothed across short income fluctuations. Meyer and Sullivan argued, in the 2012 Journal of Economic Perspectives article, that the most disadvantaged families appeared to report consumption more accurately than income, and that consumption reflected savings use, ownership of durable goods, and access to credit in addition to program use. Scott Winship of the Manhattan Institute, in the 2016 report titled Poverty After Welfare Reform, reexamined extreme poverty with adjustments to income data, including the addition of food assistance and other noncash benefits and tax credits, and reported lower levels than Edin and Shaefer reported, with a smaller absolute increase. Winship reported, for the comparison with food assistance counted as income, an increase in extreme poverty among children from 1.4 percent in 1996 to 1.6 percent in 2012 in the Edin and Shaefer framework, against an increase from 0.7 percent to 1.0 percent in the adjusted estimates in the 2016 report.

Shaefer and coauthors answered each component. H. Luke Shaefer and Joshua Rivera, in the 2018 working paper titled Comparing Trends in Poverty and Material Hardship Over the Past Two Decades, compared annual poverty rates from the official measure, the Supplemental Poverty Measure, and the Meyer and Sullivan consumption measure against measures of material hardship and labor market outcomes. Shaefer and Rivera argued that short term changes in income poverty tracked changes in other measures of well being more closely than short term changes in consumption poverty tracked those measures, that long run trends in income poverty aligned more closely with long run trends in other well being measures than consumption trends did, and that the level of poverty implied by consumption measures was unreasonably low beside other indicators of well being. Shaefer and Rivera also questioned the price index adjustment in the consumption series. Meyer and Sullivan, in the 2018 response titled Levels and Changes in Income Poverty, Consumption Poverty, and Material Well Being, answered that the central correlations in Shaefer and Rivera suffered from spurious correlation, that corrections for that problem reversed the central findings, that Shaefer and Rivera relied on selected pieces of evidence against a broader literature, and that a comprehensive examination supported the conclusion that consumption based measures reflected the circumstances of households with few resources more accurately, both at a point in time and over time.

Edin and Shaefer also addressed the reporting objection through arguments specific to the Survey of Income and Program Participation. Edin and Shaefer noted that the 2016 paper opened with the statement that no data source was perfect for measuring the conditions faced by the poorest families with children, and the paper then assembled the survey estimates, the food assistance administrative counts, school reports of homeless children, and data on the use of charitable emergency food programs, together with fieldwork from four sites. Edin and Shaefer argued that the sources moved together, and that a reporting artifact confined to one survey would not explain parallel movement in program records and school counts. Edin and Shaefer further observed that the survey estimates may have understated extreme poverty in the later years, because survey estimates fell behind administrative counts after 2012 and because households experiencing homelessness were difficult for a household survey to reach. Meyer and coauthors, in the research program summarized in the 2019 working paper on income data and extreme poverty, answered that the Survey of Income and Program Participation also contained substantial holes, citing evidence that benefit dollars went unreported in that survey as well as in the Current Population Survey and that the share of dollars missed for several programs had risen after 2000.

The concessions recorded in the literature by October 2019 defined the ground that the dispute had actually settled. Edin and Shaefer conceded, through the adjusted series in the 2013 article itself, that counting food assistance, refundable tax credits, and housing subsidies cut the measured level of extreme poverty sharply and reduced the measured increase from 159.1 percent on cash income alone to about 50 percent on the fullest adjustment presented, so the headline cash figure did not describe total resources. Meyer and Sullivan conceded, in Winning the War in 2012, that measurement error in income likely explained some of the most noticeable differences between changes in income poverty and changes in consumption poverty, and Meyer and Sullivan described saving and dissaving as playing a modest role for most demographic groups. Meyer and coauthors also credited Shaefer and Edin with finding that food assistance, tax credits, and housing subsidies together cut the pretax cash extreme poverty rate for households with children by 63 percent in 2011. The record available by October 15, 2019 therefore supported two statements at once. Cash income at the bottom, as recorded in household surveys and corroborated in part by food assistance administrative counts, deteriorated for an identifiable group of families with children after 1996, concentrated among single mother families without earnings. Total resources and consumption for a broader population of poor families, as recorded in consumption data and in income data corrected through administrative linkage, showed improvement or stability across overlapping periods.

Coverage and the Recession Test

Two findings organize the evidence on reach. The first is a long trend in coverage, measured by the relationship between families receiving cash assistance and families with children living in poverty. The second is a short and severe test of responsiveness, supplied by the recession that the National Bureau of Economic Research dated from December 2007 to June 2009. During that downturn, the Supplemental Nutrition Assistance Program, known as SNAP and called food stamps before an October 2008 renaming, retained an entitlement structure under which federal benefit funding rose with eligible enrollment. Temporary Assistance for Needy Families, known as TANF, operated through a fixed block grant under which federal funding did not rise automatically with enrollment. A divergent response to a single downturn isolates the role of funding structure more closely than a long trend alone, because the economic shock was common to both programs while the financing rule was not.

The coverage measure used in this section is the TANF-to-poverty ratio published by the Center on Budget and Policy Priorities, built from poverty data in the Current Population Survey and from caseload data reported by the Department of Health and Human Services, with caseload data collected by the Center from state agencies after September 2006. The ratio counts the yearly average number of families receiving cash assistance under Aid to Families with Dependent Children, known as AFDC, or under TANF, and divides that count by the number of families with children in poverty. In 1996, the final year of AFDC, the Center counted 4,380,430 families receiving assistance and 6,400,950 families with children in poverty, a ratio of 68 per 100. The same series recorded 31 per 100 in 2006 and 27 per 100 in 2010. In 2015 the Center counted 1,510,211 families receiving TANF cash assistance and 6,477,753 families with children in poverty, a ratio of 23 per 100, and reported that the ratio had reached that level in 2014 and had remained there in 2015. The same series recorded 23 per 100 in 2016, 23 per 100 in 2017, and 22 per 100 in 2018. The movement from 68 per 100 in 1996 to 23 per 100 in 2015 is the figure used in Article 231 of this series. In 2015 the count of families in poverty was larger than the 1996 count by roughly 77,000 families, while the count receiving cash assistance was smaller by roughly 2.87 million families.

That construction explains why the ratio measures coverage independently of the poverty rate itself. A poverty count describes the size of the population in need, and it says nothing on its own about how many of those families received a cash grant. The ratio places receipt in the numerator and need in the denominator, so a change in poverty alone does not move the measure. If families in poverty doubled and families receiving assistance doubled in proportion, the ratio would remain unchanged, and the same arithmetic holds in reverse during a contraction in poverty. The ratio falls only when receipt grows more slowly than poverty, shrinks while poverty grows, or shrinks faster than poverty shrinks. An illustration published by the Center in 2012 compared Colorado and Connecticut between the mid 1990s and the 2009 to 2010 period. Caseloads fell by 67 percent in Colorado and 68 percent in Connecticut on the Center figures, yet families with children in poverty fell by 35 percent in Connecticut and rose by 89 percent in Colorado, so the Connecticut ratio declined from 86 per 100 to 42 per 100 while the Colorado ratio declined from 72 per 100 to 13 per 100. Similar caseload behavior produced very different coverage outcomes once the denominator was taken into account.

The ratio has limits. The denominator counts families in poverty whether or not a family met a state income test, asset test, work condition, or time limit, so the ratio measures reach against poverty rather than receipt among families a state deemed eligible. The numerator counts families receiving cash assistance, and excludes families that received only a TANF funded service, a child care subsidy, a benefit classified as nonassistance, or a diversion payment. The ratio also says nothing about grant adequacy. Within those limits, the Center series applied the same numerator and denominator concepts from AFDC through TANF, so cash assistance that reached about two thirds of families in poverty at enactment reached about one fifth by the end of the period covered here. Geography widened as coverage contracted. The Center reported that in 2006 only two states, Idaho and Wyoming, had ratios of 10 per 100 or less, while in 2015 fourteen states were at or below that level. That dispersion followed from state control over income thresholds, benefit levels, application procedures, sanction policies, and any time limit shorter than the federal sixty month limit, as described in Article 235 of this series.

The first mechanism behind the fall was the fixed block grant. The Congressional Research Service recorded that the basic TANF block grant amount established in the 1996 law totaled 16.6 billion dollars for the fifty states, the District of Columbia, the territories, and the tribes, that the amount remained the same from fiscal 1997 through fiscal 2016, and that it was not adjusted for inflation, caseload size, or changes in the poverty population. The Research Service calculated that the real value of the block grant declined by 33.1 percent over the period in which the nominal amount was unchanged, and that the fiscal 2018 block grant stood 36 percent below its fiscal 1997 value in real terms. Each state share rested on AFDC spending in the mid 1990s rather than on the state share of poor children in any later year. Under AFDC, federal spending had been open ended matching spending, so serving an additional eligible family drew additional federal dollars. Under TANF, serving an additional family drew no additional basic grant dollars, and the cost came from a fixed pool that also funded child care, work programs, and administration. Sustained expansion therefore required state funds above the maintenance of effort level or reductions elsewhere inside the same envelope. The grant mechanics behind that arithmetic are detailed in the analysis of the TANF block grant structure.

State discretion over spending supplied the second mechanism, because the fixed pool was fungible across allowable purposes and only a portion reached families as monthly cash. The Department of Health and Human Services reported that in fiscal 2015 combined federal TANF and state maintenance of effort expenditures and transfers totaled 31.7 billion dollars, that about 25 percent, or 7.7 billion dollars, was spent on basic cash assistance, that less than 7 percent, or 2.1 billion dollars, was spent on work, education, and training activities, and that 17 percent, or 5.4 billion dollars, was used for child care, including transfers to the Child Care and Development Fund. The Center, analyzing the same reports for 2015, found that about half of overall funds went to basic assistance, child care, and work related programs or supports combined. Child care and transportation funds can determine whether a parent can accept a job, yet every dollar assigned to another allowable purpose was unavailable for a monthly grant. Federal law imposed no minimum benefit level and no minimum coverage level, so the ratio could fall without any state violating a federal coverage standard, because no such standard existed.

Diversion and sanction practice supplied the third mechanism, at the point where a family met the program. A Congressional Research Service report on diversion defined diversion as a payment, program, or activity intended to divert applicants from completing the TANF application process and thereby becoming eligible for monthly assistance, and described lump sum payments, vendor payments, supportive services, and resource referral as forms of diversion, with applicant job search used by some states as a diversion activity. The Research Service summary of reported state data counted twenty eight states and the District of Columbia providing lump sum diversion payments, generally equal to several months of benefits, offered in lieu of extended cash benefits, with acceptance ordinarily making a family ineligible for TANF for a specified period. The same summary counted sixteen states and the District of Columbia requiring applicant job search before approval. Under the regulations described in that report, nonrecurrent short term benefits for no more than four months in a crisis did not count as TANF assistance, so a diverted family did not enter the caseload or the numerator of the ratio. Sanctions operated after a case opened. The Congressional Research Service primer on TANF recorded that states were required to sanction a family with a member who refused without good cause to engage in work, that states determined the sanction amount, and that states could reduce benefits or terminate benefits for the whole family under a full family sanction. A terminated case left the caseload in the same statistical sense as a case closed for earnings.

The caseload reduction credit supplied the fourth mechanism, operating on state incentives rather than on individual families. As the Congressional Research Service explained the provision, the statute set a work participation standard of 50 percent for the overall caseload and 90 percent for the two parent portion, and the credit reduced the standard a state had to meet by one percentage point for each percentage point decline in the state caseload. Rebecca Blank and Ron Haskins, writing for the Brookings Institution, described the credit as enacted in those terms, designed to reward states that helped recipients leave the rolls, and recorded that in 2002 the credit completely offset the work participation requirement for twenty one states, that only eleven states had to meet a rate greater than 10 percent, and that states on average had to meet a rate of 6 percent after the credit. A Congressional Research Service overview recorded that in fiscal 2004 nineteen states had caseload declines of 50 percent or more since fiscal 1995, reducing the effective standard for those states to zero. The credit as enacted measured decline against the caseload rather than against employment outcomes for families who left, so a decline produced by diversion, sanction, a time limit, a tighter eligibility rule, or earnings at exit all reduced the standard a state faced, subject to statutory adjustments described in guidance from the Administration for Children and Families. No countervailing federal standard rewarded a state for maintaining coverage among families in poverty.

These mechanisms altered the default rather than mandating restriction, and several states widened entry at particular points, as the recession evidence below shows. Under AFDC the default fiscal response to a larger eligible caseload was larger federal spending, against the background of an individual entitlement that eligible families could enforce. Under TANF the default response was no change in federal spending, expansion required money from inside a fixed envelope or added state money, and no individual entitlement to cash assistance existed. The financing rule helps explain why the ratio could keep falling when poverty was falling. Between 2010 and 2015, on the Center series, families with children in poverty declined from 7,263,610 to 6,477,753, while families receiving assistance declined faster, from 1,984,080 to 1,510,211.

Did welfare reform work during the recession?

Between December 2007 and December 2009, TANF caseloads rose 13 percent in the Center on Budget and Policy Priorities state collected series, while SNAP caseloads rose 45 percent and the number of unemployed persons doubled, so cash assistance responded weakly compared with food assistance and with need.

The comparison begins with the entitlement that welfare reform left intact. SNAP benefits were federally funded and payable to households meeting national eligibility standards, so enrollment could expand without a prior appropriation for a larger caseload and without a state decision to appropriate additional benefit dollars, although states administered the program and bore part of the administrative cost. The Department of Agriculture Food and Nutrition Service series on average monthly participation, reproduced in a Bureau of Labor Statistics analysis of SNAP participation through the recession, recorded 26,316,000 participants in fiscal 2007, 28,223,000 in fiscal 2008, 33,490,000 in fiscal 2009, and 40,302,000 in fiscal 2010, an increase of about 14 million participants, or about 53 percent, from fiscal 2007 to fiscal 2010. Congress supplemented that response in the American Recovery and Reinvestment Act of 2009, which the Congressional Research Service recorded as raising maximum monthly SNAP benefits by 13.6 percent in fiscal 2009. The participation rise rested on the underlying entitlement, because a household that became eligible through lost earnings could enroll and draw a federally funded benefit without displacing another household from a capped pool. The program behind that response has a separate legislative history, traced in the history of the Food Stamp Act and the SNAP program.

TANF faced the same shock with the opposite financing rule. The Department of Health and Human Services, in the Ninth Report to Congress on TANF, recorded that caseloads declined slightly throughout fiscal 2007, continued to decline from December 2007 through July 2008, and then began to rise, with close to 199,000 families added from the low in July 2008 through September 2009, an increase of 12 percent from that low. The Center analysis by LaDonna Pavetti, Danilo Trisi, and Liz Schott, using caseload data collected directly from the states, estimated that TANF caseloads increased by 13 percent between December 2007 and December 2009, while SNAP caseloads grew by 45 percent and the number of unemployed persons doubled. The Congressional Research Service recorded that the assistance caseload rose from 1.7 million families in August 2008 to a peak close to 2.0 million families in December 2010. Each measure shows growth, and none shows growth on the scale of the change in need or in food assistance. The observed TANF growth represented state decisions to spend a larger share of a fixed resource on basic assistance, to draw on accumulated balances, or to use the separate emergency funding described below.

State variation confirms that discretion, rather than a uniform national response, governed the outcome. Pavetti, Trisi, and Schott reported that in twenty two states TANF caseloads responded very little or not at all between December 2007 and December 2009, with sixteen states recording increases of less than 10 percent and six states recording declines. The appendix table in the same analysis recorded Colorado rising by 48.3 percent, from 9,226 families in December 2007 to 13,681 in December 2009, New Mexico rising by 43.0 percent, from 14,060 to 20,103 families, and New Hampshire rising by 41.9 percent, from 4,584 to 6,505 families. The same table recorded California rising by 20.4 percent, from 466,853 to 561,909 families, and Florida rising by 25.7 percent, from 48,608 to 61,097 families. At the other end, North Dakota fell by 17.8 percent, from 2,423 to 1,991 families, Indiana fell by 9.7 percent, from 40,985 to 36,989 families, and Georgia fell by 5.6 percent, from 22,719 to 21,444 families. Legal Momentum, compiling figures published by the Department of Health and Human Services, found that caseloads were lower in twenty two states in March 2009 than in December 2007, and that the increase in California accounted for almost three quarters of the national increase in recipients over those sixteen months, leaving an aggregate increase of 1.8 percent across the other states and the District of Columbia. Unemployment rose in every state, so the divergence reflected eligibility rules, benefit policy, reserves, application practice, and state budget conditions inside the discretion the block grant conferred.

The TANF Emergency Contingency Fund was the partial exception, and its design demonstrates the rule it interrupted. The regular statute contained a contingency fund under section 403(b) of the Social Security Act, and guidance from the Administration for Children and Families stated that the new emergency fund was not to be confused with that existing fund. Congress created the Emergency Contingency Fund in section 403(c), as added by the American Recovery and Reinvestment Act of 2009, signed on February 17, 2009. The Congressional Research Service legislative history recorded that the fund made 5 billion dollars available for fiscal 2009 and fiscal 2010 and reimbursed states, territories, and tribes for 80 percent of increased costs of basic assistance, nonrecurrent short term benefits, and subsidized employment. Reimbursement for basic assistance depended on an increase in the caseload, so the fund restored, for two fiscal years and three categories, a federal contribution toward growth that the basic grant did not provide. The Department of Health and Human Services reported that by the end of fiscal 2010 the full 5 billion dollars had been obligated to forty nine states, with Wyoming not submitting an application, together with the District of Columbia, Puerto Rico, the Virgin Islands, and twenty four tribes. The same report allocated 1.6 billion dollars to basic assistance, over 2 billion dollars to nonrecurrent short term benefits, and over 1.3 billion dollars to subsidized employment, and cited a Center survey of state officials indicating that states placed approximately 260,000 unemployed individuals in subsidized jobs funded in whole or in part by the fund, including over 138,000 youth. A Congressional Research Service report on the fund recorded that it expired on September 30, 2010, with all 5 billion dollars awarded by that date. After expiration, financing of further caseload growth reverted to the fixed basic grant and state funds.

Interpretation of the recession record divided along the standard applied. The Center concluded from the comparison among TANF growth, SNAP growth, and the doubling of unemployment that TANF had been only modestly responsive to the downturn, and argued in a separate analysis that the weakness illustrated a defect in the block grant structure as a model for other programs. Ron Haskins, Vicky Albert, and Kimberly Howard, in a Brookings Institution report titled The Responsiveness of the Temporary Assistance for Needy Families Program during the Great Recession, examined TANF rolls against AFDC rolls in earlier recessions, state relationships between TANF rolls and unemployment, and receipt data for single mothers in the 2001 and 2007 recessions, and drew on interviews with forty four state TANF directors. The Brookings authors concluded that the American system of balancing work requirements and welfare benefits worked fairly well during the most severe recession since the depression of the 1930s. The Center measured TANF against contemporaneous need and a contemporaneous entitlement, and on those measures the gap was large. The Brookings report measured TANF against historical cash assistance behavior and the combined response of several programs, and on those measures a modest rise accompanied by other supports could be recorded as a functioning system. Both assessments rested on the same underlying caseload figures.

Read together, the trend and the test describe one structural fact from two directions. The ratio series shows coverage falling from 68 per 100 in 1996 to 23 per 100 in 2015, through expansions and through a recession, with no year in that span recording a return toward the 1996 level. The recession series shows why coverage was unlikely to recover automatically when need was greatest, because between December 2007 and December 2009 the entitlement in the comparison grew by 45 percent on the Center caseload measure and by about 53 percent between fiscal 2007 and fiscal 2010 on the Department of Agriculture participation measure, while the block grant program grew by 13 percent on the Center state collected measure. A fixed grant does not expand when the economy contracts, and the one episode in which federal TANF funding did expand in response to rising caseloads occurred because Congress enacted a temporary 5 billion dollar fund for fiscal 2009 and fiscal 2010, with an 80 percent reimbursement rate and a statutory expiration date. That sequence, documented by the Congressional Research Service, the Department of Health and Human Services, the Department of Agriculture, and the Center on Budget and Policy Priorities, is the empirical content of the claim that a block grant does not function as an automatic stabilizer.

States, Waivers and the Wider Safety Net

The national figures for caseload decline, employment, and poverty described a single country, but the system that produced those figures was never a single program. Before 1996, states had already tested many of the devices that the federal law later adopted, under conditions that limited what the tests could prove. After 1997, the block grant gave states wide discretion over benefit levels, time limits, sanctions, and spending, and states exercised that discretion in sharply different ways. During the same years, the rest of the safety net moved in the opposite direction from cash assistance, as food assistance, refundable tax credits, health coverage, and child care support expanded for low income working families. An account measured by the cash caseload alone therefore missed a broader set of supports that grew as cash assistance contracted, and a population of families with neither earnings nor cash assistance that fell outside both developments.

The waiver era supplied the prehistory for that variation. Under Section 1115 of the Social Security Act, the federal government could waive provisions of Aid to Families with Dependent Children for state demonstration projects, and state requests multiplied in the late 1980s and early 1990s. The Congressional Research Service history of Temporary Assistance for Needy Families waivers recorded that between January 1993 and August 1996, a total of 83 waiver applications from 43 states and the District of Columbia received approval. Projects ranged from small demonstrations in selected counties to statewide redesigns. Some disregarded a larger share of earnings than federal law allowed, or permitted larger savings. Others expanded work requirements to mothers whom federal law had exempted, particularly mothers with very young children. A further group tested time limits, family caps that denied a benefit increase for a child born to a family receiving assistance, and sanctions that removed the entire grant rather than the adult portion alone.

Time limits illustrated how much the waiver version of a policy could differ from the version enacted in 1996. The review prepared by the Lewin Group and MDRC for the Administration for Children and Families recorded that the Department of Health and Human Services began granting time limit waivers in 1993 and 1994, and that by August 1996 a total of 32 states had received such waivers. Many early limits did not operate statewide, many did not terminate the entire grant, and many were periodic rather than lifetime limits, allowing a defined number of months within a longer calendar period or requiring a spell off assistance before eligibility could resume. The 1996 law removed much of that scaffolding, barred the use of federal funds after sixty months subject to a hardship exception for up to 20 percent of the caseload, and left states to decide how closely state policy would track the federal ceiling.

Two waiver evaluations received particular attention because each used random assignment, the design that allowed evaluators to attribute differences in outcomes to the program rather than to differences in the families enrolled. Florida operated the Family Transition Program from 1994 to 1999 in Escambia County, under a contract between the Florida Department of Children and Families and MDRC, with funding from the Department of Health and Human Services and the Ford Foundation. The MDRC final report described a limit of 24 months of receipt in any 60 month period for most recipients and 36 months in any 72 month period for recipients judged least ready for employment, combined with financial incentives, mandates, and an unusually rich array of services. MDRC interim findings showed a staged pattern. During the first two years of follow up, before participants could have reached the limit, the program increased employment and earnings, did not affect the rate of welfare receipt, and raised combined income from earnings and public assistance. For the earliest enrollees, the pattern changed after the two year point, as small numbers reached the limit and lost benefits, and the program began to generate reductions in welfare receipt and increases in the share who worked without cash assistance.

Minnesota tested a different combination. The Minnesota Family Investment Program began in April 1994 in seven Minnesota counties and paired a more generous earnings disregard, which allowed families to keep more of the grant as earnings rose and to remain eligible until income reached a level above the poverty line, with mandatory employment services for longer term recipients and payment of child care costs. MDRC evaluated the program through random assignment of families between April 1994 and March 1996 to either the new program or the Aid to Families with Dependent Children system. The MDRC final report, subtitled Reforming Welfare and Rewarding Work, found the strongest effects among single parent families that had received assistance for a long period before entry. For that group the program increased employment, earnings, and total income, and reduced measured poverty, while also increasing receipt of welfare, because the disregard allowed working parents to combine earnings with a partial grant rather than leaving assistance at the first job. The MDRC six year follow up, published in July 2005 under the title Turning Welfare into a Work Support, found that effects on earnings and income for most parents had faded by the sixth year, that effects persisted for a subgroup of the most disadvantaged single parents, and that children of single parent long term recipients performed better in school than children in the control group.

The waiver record could show that specific packages changed behavior under demonstration conditions, and random assignment gave those findings a credibility that ordinary caseload comparisons lacked. The record could not show how the same devices would perform as permanent statewide policy for an entire caseload. Demonstrations operated in selected counties, enrolled defined cohorts, carried extra services and attention, and ran during the strong labor market of the middle and late 1990s, so gains measured in one county during an expansion did not predict outcomes for a statewide program during a recession. Waiver limits also differed in kind from the federal limit, since periodic limits, reduction limits that removed only the adult portion, and extension guarantees softened the consequence that a termination limit imposed. The Minnesota result depended on spending more per family and on allowing welfare receipt to rise alongside employment, an outcome that a fixed block grant and a federal work participation standard later penalized rather than rewarded. The Florida result showed employment gains before any family reached the limit, which suggested that services, mandates, and incentives did much of the measured work, and left open how much of the later caseload effect came from benefit cancellation itself.

Did states run the new system the same way?

No. Federal law set a sixty month ceiling on federally funded assistance and broad work expectations, but states set benefit levels, chose shorter limits or longer support through state funds, defined sanctions, and allocated block grant dollars among cash, child care, and other uses, producing markedly different programs under one federal name.

Benefit levels provided the clearest contrast. The Congressional Research Service report titled The Temporary Assistance for Needy Families Block Grant, Responses to Frequently Asked Questions reported the July 2017 range for a family of three documented in the caseload section, with only New Hampshire, at 60 percent of the federal poverty guidelines, holding a maximum amount for a family of that size above 50 percent of poverty level income. Because most states left nominal grants unchanged for long stretches, inflation reduced grant value without a recorded vote, and the real value of the federal money fell at the same time. The Congressional Research Service table of block grant funding in nominal and constant dollars, computed with the Consumer Price Index for All Urban Consumers from the Bureau of Labor Statistics, showed the state family assistance grant at 16.567 billion dollars in fiscal 1997 and 16.512 billion dollars in fiscal 2018 in nominal terms, and at 10.564 billion dollars in constant 1997 dollars in fiscal 2018, a cumulative decline of 36.2 percent in purchasing power between fiscal 1997 and fiscal 2018.

Time limit choices diverged in both directions from the federal sixty month standard. The Lewin Group and MDRC review for the Administration for Children and Families categorized state policies by the most stringent limit in each state and reported that about 24 percent of adult headed families receiving assistance lived in states with a sixty month termination limit, about 47 percent lived in states with a sixty month limit that did not routinely terminate all assistance, and about 22 percent lived in states with a shorter termination limit. New York used state and local funds to move families that reached the limit into a state and locally funded safety net program, and California removed the adult from the assistance unit at the limit while continuing assistance for the needs of the children, according to the state policy descriptions in the same review. The Center on Budget and Policy Priorities report on the expiration of Temporary Assistance for Needy Families supplemental grants documented the opposite choice in Arizona, which reduced the lifetime limit from 60 months to 36 months and then to 24 months, described in that report as one of the shortest lifetime limits in the country, while scaling back child care subsidy funding in a change estimated to affect 13,000 children.

Sanction policy diverged as widely, and sanction exposure reached more families than time limit exposure during the early years. Federal law required states to reduce or terminate benefits when a recipient refused without good cause to comply with work requirements, but states determined the amount, speed, and duration of the penalty. The Brookings Institution analysis titled Sanctions and Welfare Reform, prepared for the 2002 reauthorization debate, reported counts from the State Policy Documentation Project operated by the Center for Law and Social Policy showing that 36 states used full family sanctions that terminated the entire grant, that 18 of those states imposed a full family sanction at the first instance of noncompliance, and that in the other 18 states partial sanctions could escalate to full family sanctions after repeated or continued noncompliance, while California, New York, and Texas, three states with large caseloads, did not use full family sanctions. The same analysis summarized a General Accounting Office study from 2000 estimating that 136,000 families, or 5 percent of the national caseload, received reduced benefits or no benefits because of sanctions in a typical month in 1998.

Spending choices completed the divergence, because the block grant allowed states to move money away from monthly cash grants toward other allowable purposes. The Center on Budget and Policy Priorities analysis of Department of Health and Human Services financial data, titled How States Have Spent Federal and State Funds Under the Temporary Assistance for Needy Families Block Grant, reported that states spent 70 percent of combined federal and state funds on basic assistance in fiscal 1997 and 58 percent in fiscal 1998, and that the share fell to 28 percent in fiscal 2008 and fiscal 2009, 30 percent in fiscal 2010, and 29 percent in fiscal 2011. The success reading of state flexibility treated that reallocation as the intended operation of a block grant, with states tailoring resources to local labor markets and investing in work supports rather than unconditional grants. The failure reading, advanced by analysts at the Center on Budget and Policy Priorities in the spending analysis cited in this section, treated the same reallocation as a weakening of the cash safety net, because basic assistance was the only component that put unrestricted cash in the hands of families with no earnings.

The wider safety net moved in the opposite direction from cash assistance, and that movement changed the meaning of the caseload decline. The 1996 law separated Medicaid eligibility from cash assistance, so that loss of a cash grant no longer automatically ended health coverage, and the State Children Health Insurance Program, enacted in 1997, extended coverage to children in low income families with incomes above Medicaid limits. The Center on Budget and Policy Priorities summary of the post reform safety net observed that virtually all low income children held eligibility for Medicaid or the State Children Health Insurance Program in the years covered by that summary, and that only a small fraction of children covered through those programs also received cash assistance, because most covered children lived in low income working families. Food assistance followed a different trajectory after an initial decline. The Economic Research Service of the Department of Agriculture reported in the 2012 study titled Alleviating Poverty in the United States, The Critical Role of Supplemental Nutrition Assistance Program Benefits that the average monthly caseload of the Supplemental Nutrition Assistance Program rose from 17.2 million individuals in 2000 to 44.7 million individuals in 2011, with total benefits of almost 72 billion dollars in 2011.

The earned income tax credit supplied the largest single work based support in the expanded net. The Congressional Research Service analysis of the credit recorded that the Omnibus Budget Reconciliation Act of 1993 greatly expanded the credit, restructured the credit by family size, extended eligibility to workers without qualifying children for the first time, and produced, after full phase in, a maximum credit of 2,152 dollars for a taxpayer with one child and 3,556 dollars for a taxpayer with two or more children in 1996. Economists Nada Eissa of Georgetown University and Hilary Hoynes of the University of California, Davis, as quoted in the Center on Budget and Policy Priorities analysis of the empirical literature, concluded that the credit had been especially successful at encouraging employment of single parents, particularly mothers. Economist Jeffrey Grogger of the University of Chicago, in the study of policy changes between 1993 and 1999 cited in the same analysis, estimated that the credit contributed about as much to the decline in receipt of cash assistance among female headed families during those years as time limits and other welfare reforms combined. Child care support also expanded, as the 1996 law consolidated funding into the Child Care and Development Fund and allowed states to transfer block grant funds to child care. The combined effect was that total support for low income families with earnings rose as support for families without earnings fell, a rebalancing that the success reading presented as a coherent work based system and that the failure reading presented as a system that assisted families after employment began and assisted families without employment far less.

The disconnected population sat at the point where the contracted cash system and the expanded work based system failed to meet. Researchers used the term disconnected for low income parents with little or no connection to either the labor market or cash public assistance, and the Administration for Children and Families research summary titled Disconnected Families and Temporary Assistance for Needy Families, published on November 15, 2011, recorded that national and state studies had identified a significant minority of former recipients who did not leave welfare with employment, and that take up rates for cash assistance, defined as the percentage of eligible mothers who received benefits, had fallen. Rebecca Blank of the University of Michigan and Brian Kovak, in the 2007 analysis cited in the Labor and Employment Relations Association research volume chapter on single mothers, estimated that low income single mothers who were neither working nor receiving welfare comprised up to one fourth of all low income single mothers. Rebecca Blank, in the Brookings Institution paper titled From Welfare to Work, What the Evidence Shows, published in January 2002, summarized leaver studies finding that between 25 percent and 40 percent of women who had left welfare were not working at the time of follow up in the studies reviewed.

The characteristics of disconnected families explained part of the policy difficulty. Rebecca Blank and Brian Kovak, in the 2008 summary of the disconnected research reported by the American Public Human Services Association, stated that compared with women who left welfare for work, disconnected mothers had less education, younger children, higher rates of mental and physical health problems, higher rates of substance abuse, and a greater history of domestic violence. LaDonna Pavetti of the Center on Budget and Policy Priorities, in testimony summarizing an Urban Institute study of data from 2005 and 2006, reported that 27 percent of cash assistance recipients faced significant work limitations, defined as physical, mental, or emotional problems that prevented employment or limited the kind or amount of work, compared with 5 percent of all adults and 6 percent of low income single mothers in the general population measures used in that study, and that 14 percent of recipients lived with a family member with a disability. Because food assistance and health coverage reached many disconnected families even when cash assistance and earnings did not, disconnection from cash and work did not always mean disconnection from every program.

State variation, the waiver record, the wider safety net, and the disconnected population together reframed the national averages. The waiver experiments demonstrated that earnings disregards combined with mandates could raise employment and income for defined groups under funded demonstration conditions, and that time limit cancellations arrived late in the follow up periods and affected small numbers during the evaluation windows. The block grant then produced fifty distinct settlements on benefits, limits, sanctions, and spending, financed by a grant that lost 36.2 percent of purchasing power between fiscal 1997 and fiscal 2018 according to the Congressional Research Service constant dollar series. The disconnected estimates meant that a substantial minority of low income single mothers, up to one fourth in the estimate by Rebecca Blank and Brian Kovak for 2007, stood outside both halves of that settlement, without earnings to trigger work based supports and without a cash grant to replace earnings.

Two Verdicts and the Longer Run

The success reading at full strength

Ron Haskins, a Republican staff director during the drafting of the 1996 law and later a senior fellow at the Brookings Institution, presented the record as a statute meeting the purposes written into the statute. The Congressional Research Service, drawing on caseload data from the United States Department of Health and Human Services, recorded a peak of 5.1 million families receiving Aid to Families with Dependent Children in March 1994, 4.4 million families in August 1996, the month of enactment, and 2.4 million families in December 1999. Haskins and the congressional architects named in the legislative debates treated a decline of that size, sustained through the end of the 1990s, as evidence that work expectations, time limits, and state administration produced a different result from an entitlement to cash aid.

Employment supplied the second leg of the account given by Haskins. In congressional testimony, Haskins reported that the share of single mothers in employment reached nearly 75 percent by 2000, an increase of more than one fifth from the level in 1995 and the highest level in the series cited for 1995 to 2000. Robert Moffitt of Johns Hopkins University, summarizing a Department of Health and Human Services review of state leaver studies for the Brookings Institution in 2002, reported an employment rate among welfare leavers of approximately 60 percent shortly after exit and a finding that about three quarters of leavers worked at some point during the first year after leaving the rolls, based on studies covering exits in the late 1990s. Moffitt also reported that leavers who worked generally worked full time in the studies covered.

Earnings formed the third leg, and Haskins stated the wage level plainly. In testimony summarized by the Brookings Institution, Haskins reported that most jobs taken by mothers leaving or avoiding welfare paid low wages, around 8 dollars per hour in 2000, in the studies and federal data cited for that year. Moffitt reported a comparable range of 7 dollars to 8 dollars per hour, with earnings of about 3,000 dollars per quarter for leavers who worked, in the department review. Haskins argued that earnings did not stand alone because Congress had expanded the Earned Income Tax Credit, child care assistance, food assistance, and Medicaid. In the Census Bureau series cited by Haskins for female headed families in the bottom fifth of the income distribution, in constant 2001 dollars for 1993 to 2001, earnings and the Earned Income Tax Credit rose as welfare income fell, and total income rose by around 20 percent across the period shown.

Child poverty supplied the fourth leg. The Census Bureau official poverty measure for children, as tabulated in the Current Population Survey historical series and summarized by Child Trends, fell from 23 percent in 1993 to approximately 16 percent in 2000. Haskins reported that poverty among Black children and among children in female headed families each reached the lowest level in the recorded series by the end of the 1990s, in the Census Bureau series cited in the testimony. The success reading, as Haskins framed the reading, rested on four movements in a single short period, with caseloads falling, employment rising, earnings rising for the lowest income group of single mother families, and child poverty falling, each measured in a named federal series for 1993 to 2001.

Rebecca Blank of the University of Michigan gave a decomposition that supported sustained gains while distributing causes across more than one force. In a 2000 lecture in the Journal of Economic Perspectives, Blank concluded that the strong macroeconomy, welfare reform, and other policy changes, especially expansions of the Earned Income Tax Credit and increases in the minimum wage, reinforced each other and produced very large behavioral changes during the 1990s. In a 2004 Brookings Papers on Economic Activity study, Blank recorded a fall in welfare participation among single mothers from 25 percent in 1996 to 9 percent in the later year covered and a rise in the share in work from 74 percent in 1996 to 79 percent in that later year, in the survey series used. Robert Schoeni and Rebecca Blank, in a 2000 study, found that unmarried women with dependent children who had left school without a diploma raised employment by 2.0 percentage points to 4.5 percentage points in waiver states relative to states without waivers. Jeffrey Grogger, in an analysis of Survey of Income and Program Participation data for 1993 to 1999, assigned 12 percent of the decline in welfare participation to reform, 10 percent to the Earned Income Tax Credit, and 5 percent to the unemployment rate.

The failure reading at full strength

Peter Edelman, an assistant secretary in the Department of Health and Human Services who resigned in 1996 in protest against the signing of the law, presented the opposing reading from enactment forward. In a 1997 article in The Atlantic, Edelman argued that the law would move 2.6 million people, including 1.1 million children, into poverty, in the estimate quoted in later summaries of the article, and predicted increases in malnutrition, crime, infant mortality, family violence, and abuse. In a 2001 letter responding to a commentary by Rebecca Blank and Ron Haskins, Edelman stated that research cited in the letter showed about 40 percent of the more than 2.5 million women who had left welfare after 1996 had neither a job nor cash assistance, a group placed at more than one million women with more than two million children. Edelman argued in later interviews that a block grant left a state under no obligation to aid any particular family and that the recession after 2007 exposed that design choice.

Kathryn Edin of Princeton University and Johns Hopkins University and H. Luke Shaefer of the University of Michigan gave the failure reading a specific poverty claim in a 2015 book based on Survey of Income and Program Participation data and on fieldwork between 2012 and 2014 in Chicago, Cleveland, Appalachia, and the Mississippi Delta. Edin and Shaefer estimated that in early 2011, 1.5 million households with roughly 3 million children lived on cash incomes of no more than 2 dollars per person per day in any given month, a measure adapted from a World Bank threshold for extreme poverty and applied to cash income in that survey. Edin and Shaefer reported that families in that category had more than doubled after 1996, in the same series for 1996 to 2011, and that counting the Earned Income Tax Credit and food assistance reduced the estimate substantially. The fieldwork documented sale of plasma, collection of scrap metal and cans, and reliance on private charity and on kin, and Edin and Shaefer attributed growth in the cash poor population in part to loss of an entitlement to cash aid and to barriers, sanctions, and time limits.

Analysts at the Center on Budget and Policy Priorities gave the failure reading a coverage measure. The center defined a Temporary Assistance for Needy Families to poverty ratio as families receiving cash assistance for every 100 families with children in poverty, using Current Population Survey poverty data and Department of Health and Human Services caseload data and, after September 2006, caseload data collected by the center from state agencies. In that series, coverage fell across the full span documented in the coverage section, reaching 22 families receiving cash assistance for every 100 families with children in poverty in 2018. Center analysts, including Arloc Sherman and Danilo Trisi, linked the decline to strict eligibility barriers, work requirements, sanctions, diversion, and time limits, and the center estimated that a ratio held at the 1996 level in 2017 would have placed an additional 2.5 million families on cash assistance in that year. Trisi found that the share of children below half the poverty line rose between 1995 and 2005 and attributed the rise mainly to loss of cash assistance.

The recession after 2007 supplied a comparison across programs. Marianne Bitler and Hilary Hoynes of the University of California, in a 2010 Brookings Papers on Economic Activity study covering the recession of 2007 to 2009, found that food assistance and other noncash programs had become significantly more responsive to the economic cycle after reform, rising more as unemployment rose, while the study found no evidence that cash welfare had become more responsive and some evidence that cash welfare had become less responsive. Peter Edelman made the contrast in counts in an interview account, stating that food assistance enrollment reached 46 million persons while cash welfare enrollment moved from 3.9 million persons to 4.4 million persons during the downturn described. The Congressional Research Service, based on Department of Health and Human Services data, recorded a cash caseload in December 2008 that stood 65 percent below the caseload in December 1994.

Funding erosion completed the failure reading. The Congressional Research Service constant dollar series documented in the state section recorded a decline of 36.2 percent in the real value of the state family assistance grant between fiscal 1997 and fiscal 2018, across a span in which the grant faced no adjustment for inflation, caseload change, or change in the poverty population during fiscal 1997 through fiscal 2016, and a reduction of 0.33 percent beginning in fiscal 2017 financed related research and technical assistance, according to the service account.

The longer run record

The employment record after 2000 complicated each reading without reversing the direction of the 1990s change. The Legal Momentum analysis titled Single Mothers Since 2000, drawing on Bureau of Labor Statistics data from Employment Characteristics of Families for mothers in families maintained by women with children under age 18, reported a fall in single mothers employed in an average month from 76 percent in 2000 to 73 percent in 2007 and to 68 percent in 2009, and a fall in full time year round employment from 51 percent in 2002 to 49 percent in 2007 and to 45 percent in 2009, in that series. Ron Haskins stated in Brookings Institution testimony that employment among females heading families had declined almost to the prereform level during the deep downturn described. Congressional Research Service summaries noted that the steady increase in employment rates for single mothers ended in 2000 while families receiving assistance continued to decline after 2000, in the series for 1994 to 2019.

Wages and retention bounded the earnings claim. Gregory Acs and Pamela Loprest, in a 2001 synthesis of state leaver studies for the Assistant Secretary for Planning and Evaluation, reported median hourly wages for single parent leavers ranging from 6.50 dollars in the South Carolina study of an October 1998 to March 1999 exit cohort to 9.00 dollars in the Bay Area study of a fourth quarter 1998 cohort, in surveys 6 months to 18 months after exit. Robert Moffitt reported that only a little over one third of leavers who worked held employment for four quarters in a row in the studies reviewed. The Legal Momentum analysis, based on Census Bureau data for 2000 to 2009, reported an increase in single mothers below the poverty level from 33 percent in 2000 to 38.5 percent in 2009, below half the poverty level from 14 percent to 19 percent, in food insecurity from 31 percent to 37 percent, and in use of food pantries from 8.6 percent to 13.7 percent, in the federal series assembled in that analysis.

Deep poverty remained dependent on a measurement choice that named researchers defended on incompatible grounds. Bruce Meyer of the University of Chicago and James Sullivan of the University of Notre Dame argued in studies between 2003 and 2017, and in a 2018 response to H. Luke Shaefer and Gregorio Rivera, that consumption based measures captured the material circumstances of families with few resources more accurately than income based measures, that income went underreported in surveys by the most disadvantaged families, and that consumption poverty fell across decades in the Consumer Expenditure Survey series constructed in the Meyer and Sullivan work. Shaefer and Rivera, in the 2018 paper answered by Meyer and Sullivan, argued that income poverty tracked other indicators of material hardship more closely than consumption poverty in the comparisons presented. Trisi reported in the center account that deep poverty among children rose between 1995 and 2005 and fell during a later period as nutrition assistance and refundable tax credits strengthened, through the years covered before 2019.

Disconnected status gave the post 2000 record a population that neither the employment series nor the caseload series captured alone. Pamela Loprest, in Urban Institute work, and Gregory Acs and Pamela Loprest in joint work, documented mothers with no recorded work and no recorded cash assistance during defined observation periods in the late 1990s and the 2000s, in the survey and administrative series used in those studies. Ron Haskins, in the Brookings Institution account titled TANF: It Worked, Mostly, acknowledged a substantial group of disconnected mothers who did not work steadily and did not receive cash welfare during the recession after 2007, and described a tradeoff between strict work requirements and an increase in disconnected mothers with a high likelihood of poverty. Disagreement remained on size by definition, on duration, and on shares attributable to sanctions, time limits, health barriers, caregiving demands, and labor market conditions.

Evidence on children arrived through studies of programs that predated or paralleled the 1996 law more often than through a direct national evaluation of the law, and MDRC synthesis work made that limitation explicit. Pamela Morris, Aletha Huston, Greg Duncan, Danielle Crosby, and Johannes Bos, in a 2001 MDRC synthesis of random assignment programs, found that programs raising both parental employment and family income through earnings supplements improved school achievement among elementary school age children, in the measures reported. The same synthesis found that programs mandating employment activities, raising employment without raising income, produced few effects on elementary school age children. Lisa Gennetian and coauthors, in an MDRC synthesis of eight studies of sixteen programs for adolescents aged 12 years to 18 years at survey, found that parents in the programs reported worse school performance, a higher rate of grade repetition, and greater use of special educational services for adolescents than control group parents reported, with no overall difference in dropout, suspension, completion, or childbearing. No MDRC synthesis cited in this section claimed a direct estimate of the effect of the 1996 statute alone on a national cohort followed into adulthood by 2019.

Do researchers agree about welfare reform results?

Researchers agreed on several recorded movements and disagreed on causes and measurement. Caseload decline and coverage decline rested on administrative counts. Employment rise rested on survey records. Attribution among reform, expansion, and earnings supplements stayed contested, and deep poverty estimates depended on income reporting and benefit counting choices in surveys.

The caseload finding carried a settled confidence level. Administrative counts from the Department of Health and Human Services, as compiled by the Congressional Research Service for 1994 to 2019 and by the Center on Budget and Policy Priorities for 1996 to 2018, showed a large and sustained fall in families receiving cash assistance, and no named researcher in the record summarized in this section disputed the direction or approximate magnitude of that fall. The employment rise carried a settled confidence level for direction between the mid 1990s and 2000 and a contested confidence level for attribution. Series cited by Haskins, by Blank, and by the Bureau of Labor Statistics based analysis in Single Mothers Since 2000 agreed on a rise to a peak around 2000 and a partial retreat between 2000 and 2009. Attribution among the statute, the Earned Income Tax Credit expansion of 1993, state waivers before 1996, minimum wage increases, and the late 1990s unemployment rate stayed contested, with Grogger assigning a 12 percent share to reform for 1993 to 1999, Meyer and Rosenbaum assigning larger shares to tax changes for 1984 to 1996, and Blank describing reinforcing forces without a single dominant share.

Deep poverty carried a contested confidence level on measurement grounds. Edin and Shaefer reported a doubling of families with cash income at or below 2 dollars per person per day between 1996 and 2011 in Survey of Income and Program Participation cash income, Trisi reported a rise in deep poverty among children between 1995 and 2005 in Census Bureau based analyses, and Meyer and Sullivan reported falling consumption poverty in a Consumer Expenditure Survey based series, with each result following from the resource definition selected. Coverage decline carried a settled confidence level, with the center ratio moving from 68 families for every 100 families with children in poverty in 1996 to 22 for every 100 families in 2018. The recession contrast carried a settled confidence level, with Bitler and Hoynes finding food assistance and noncash programs significantly more responsive to rising unemployment after reform and cash welfare no more responsive, with some evidence of a weaker response, between 2007 and 2009.

The series thesis applied to the 1996 law required two separations. A statute fell to be assessed against the aims stated for the statute, including reduced dependence on cash aid, increased work, and reduced poverty, and each aim pointed to a different series with a different confidence level. A treatment also fell to be separated from the conditions that accompanied the treatment, including an unemployment rate that fell below 4 percent during months in 2000 in Bureau of Labor Statistics data, an Earned Income Tax Credit expansion enacted in 1993, state experiments under waivers before 1996, and a recession between 2007 and 2009 that tested a fixed block grant under conditions opposite to the conditions of enactment. Decomposition work by Blank, by Grogger, and by Meyer and Rosenbaum attempted the second separation with different data and assumptions, and the distance between the resulting shares measured the size of the attribution problem. Settled findings named movements established under multiple definitions. Contested findings named movements that changed with a resource definition, a population definition, or an attribution assumption, in measurement for deep poverty, in shares for employment attribution, and in duration and cause for disconnected status.

Reading the Evidence

The evidence supports five findings at three confidence levels. The caseload finding carries a settled confidence level: families receiving cash assistance fell by roughly three quarters between 1994 and 2018 in the Congressional Research Service series, recipients fell by a comparable share, and the fall began before enactment, accelerated after state programs began in 1997, and resumed after a modest recession era increase. The employment finding carries a settled confidence level for direction and a contested confidence level for attribution: employment among single mothers rose sharply through 2000, and the earned income tax credit expansion of 1993, the welfare bundle of waivers and the 1996 statute, and a labor market with unemployment at 4.0 percent in 2000 each contributed, with shares that moved by sample, period, and method across the estimates published by Meyer and Rosenbaum, by Grogger, and in the synthesis offered by Blank.

The poverty finding divides by measure. The official child poverty rate fell from 22.7 percent in 1993 to 16.2 percent in 2000 and rose to 22.0 percent in 2010 in the Census Bureau series, a path that neither verdict reading can quote in part without distortion. Deep poverty carries a contested confidence level: income based measures corrected for underreporting in the analysis by Trisi and Sherman recorded an increase among children between 1995 and 2005, Edin and Shaefer recorded a large increase in extreme cash poverty between 1996 and 2011, and Meyer and Sullivan recorded falling consumption poverty across overlapping years, with the disagreement turning on survey reporting error and on whether food assistance and tax credits count as resources. The coverage finding carries a settled confidence level: the ratio published by the Center on Budget and Policy Priorities fell from 68 families receiving assistance for every 100 families with children in poverty in 1996 to 23 for every 100 in 2015, and the recession finding carries a settled confidence level as well.

The recession settled one question, and the question was structural. Between December 2007 and December 2009, food assistance caseloads rose by 45 percent and cash assistance caseloads rose by 13 percent in the state collected series published by the Center on Budget and Policy Priorities, while the number of unemployed persons doubled. The entitlement expanded with need while the block grant did not, and the temporary Emergency Contingency Fund that Congress enacted for fiscal 2009 and fiscal 2010 stands as the exception that identifies the rule, because expanded federal support for caseload growth required a new statute with an expiration date. Neither the success reading nor the failure reading changes that comparison, and the comparison changes neither the attribution dispute over employment nor the measurement dispute over deep poverty. Each verdict draws strength from a different finding, and each verdict meets a limit in a finding stated at a different confidence level. A reader assembling the full record behind those verdicts can keep your statute notes, citations, and case chronologies together free on VaultBook while separating settled movements from contested explanations.

Frequently Asked Questions

Q: Did welfare reform reduce caseloads?

Yes, on administrative counts. The Congressional Research Service series based on Department of Health and Human Services data recorded families receiving cash assistance falling from 5.033 million in 1994 to 1.196 million in 2018, and recipients falling from 14.161 million to 3.150 million across the same span. The decline began before enactment of Public Law 104-193, since recipients had already fallen to 12.321 million by 1996, and the steepest segment ran from 1996 to 2001. Jeffrey Grogger, Steven Haider, and Jacob Klerman found that the 1990s decline came from both faster exits and fewer entries. The count measures receipt rather than need, income, or hardship among families outside the rolls.

Q: Did welfare reform increase employment for single mothers?

Employment among single mothers rose sharply during the 1990s, and researchers agreed on the direction while disputing the shares. Bruce Meyer and Dan Rosenbaum, using Current Population Survey data, assigned the largest share of the increase between 1984 and 1996 to tax changes centered on the earned income tax credit expansion of 1993. Jeffrey Grogger estimated that time limits accounted for about 7 percent of the rise in employment after 1993 and assigned important effects to the remaining welfare reforms. Rebecca Blank described the statute, the credit, and the strongest labor market in a generation as reinforcing forces. The welfare bundle contributed, and the size of the contribution stayed contested.

Q: Was it welfare reform or the economy?

The evidence reviewed through 2019 supported a joint answer. The Council of Economic Advisers attributed about 44 percent of the caseload decline between 1993 and 1996 to economic growth and about 31 percent to state welfare waivers. Meyer and Rosenbaum assigned the dominant role in the employment increase to the earned income tax credit expansion of 1993, and estimated that falling unemployment raised the absolute level of employment among single mothers as well. Grogger assigned time limits about one eighth of the decline in welfare use after 1993. Blank concluded that the macroeconomy, welfare reform, and other policy changes, including the credit, reinforced one another, so a single cause account contradicts the decomposition record.

Q: Did welfare reform increase deep poverty?

On income based measures, deep poverty rose. Danilo Trisi and Arloc Sherman of the Center on Budget and Policy Priorities estimated that the share of children below half the poverty line rose from 2.1 percent in 1995 to 3.0 percent in 2005, using income that counted noncash benefits and tax credits and corrected for underreporting. Kathryn Edin and H. Luke Shaefer recorded a larger increase in extreme cash poverty. The measurement objection, developed by Bruce Meyer and James Sullivan, held that surveys missed a large share of benefit dollars, excluded major resources from cash measures, and diverged from consumption data that recorded improvement. Administrative linkage by Meyer and Nikolas Mittag supported part of that objection, and the finding stayed contested.

Q: How many poor families still get help after welfare reform?

The Center on Budget and Policy Priorities measured reach with a ratio of families receiving cash assistance to families with children in poverty. The ratio stood at 68 per 100 in 1996, with 4,380,430 families receiving assistance against 6,400,950 families with children in poverty. The ratio fell to 23 per 100 in 2015, with 1,510,211 families receiving assistance against 6,477,753 families in poverty, and reached 22 per 100 in 2018. A related Congressional Research Service measure recorded the share of eligible families receiving benefits falling from an estimated 82 percent in 1995 to an estimated 27 percent in 2015. Reach contracted far faster than poverty contracted.

Q: Did welfare reform work during the recession?

The recession of 2007 to 2009 tested responsiveness rather than employment incentives. LaDonna Pavetti, Danilo Trisi, and Liz Schott of the Center on Budget and Policy Priorities estimated that cash assistance caseloads rose by 13 percent between December 2007 and December 2009, while food assistance caseloads rose by 45 percent and the number of unemployed persons doubled. Participation in food assistance rose from 26,316,000 persons in fiscal 2007 to 40,302,000 in fiscal 2010 in the Department of Agriculture series. Ron Haskins, Vicky Albert, and Kimberly Howard judged the combined system against earlier recessions and reached a more favorable assessment. The underlying caseload figures were common ground, and the standard of comparison divided the assessments.

Q: What did the tax credit do compared to welfare reform?

The earned income tax credit paid only to a worker and rose with earnings, so the expansion enacted in 1993 raised the payoff to entering employment. Meyer and Rosenbaum estimated that the credit explained 62 percent of the increase in weekly employment of single mothers relative to single women without children between 1984 and 1996, and 35 percent of the increase in annual employment between 1992 and 1996, with waivers assigned about 14 percent to 20 percent across that shorter span. Grogger described the credit as a candidate for the single most important policy measure behind falling welfare use and rising work. Welfare reform operated through requirements, sanctions, and time limits, and the credit operated through the wage.

Q: Do researchers agree about welfare reform results?

Agreement covered direction on several findings and failed on shares and measurement. Named researchers disputed neither the caseload decline recorded by the Congressional Research Service nor the coverage decline recorded by the Center on Budget and Policy Priorities. Agreement also covered the direction of the 1990s employment rise, in which the earned income tax credit expansion of 1993 played a major part in the Meyer and Rosenbaum estimates. Disagreement covered the division of credit among the statute, the credit, waivers, and the labor market, and the deep poverty trend, where Edin and Shaefer and Meyer and Sullivan defended incompatible measures. Blank placed the consensus at joint causation with reinforcing forces.

Q: Did welfare reform reduce poverty overall?

The official Census Bureau measure fell during the first years and rose afterward. The Center on Budget and Policy Priorities, summarizing the Census Bureau report for 2000, recorded child poverty falling from 22.7 percent in 1993 to 16.2 percent in 2000, the lowest level in that series since 1979. The Department of Health and Human Services summary of the same series recorded 22.0 percent in 2010. Analysts attributed the 1990s fall in part to employment growth, the earned income tax credit expansion of 1993, waivers, and the 1996 statute, and attributed the later rise in part to two recessions and reduced reach of cash assistance. The official measure counted pretax cash income and excluded the credit and noncash benefits.

Q: What happened to families who left welfare without jobs?

Leaver studies and disconnectedness research described a substantial group. Pamela Loprest and Austin Nichols of the Urban Institute defined disconnected single mothers as low income single mothers with no earnings, no cash benefit from Temporary Assistance for Needy Families or Supplemental Security Income, and no enrollment in school, and estimated that one in eight met that definition in 1996, against one in five during 2004 to 2008. Rebecca Blank and Brian Kovak estimated that such mothers comprised up to one fourth of low income single mothers. Blank and Kovak also recorded less education, younger children, and higher rates of health problems and domestic violence histories among disconnected mothers than among mothers who left for work.

Q: Did welfare reform raise earnings for single mothers?

Earnings rose for employed leavers during the 1990s expansion, within limits set by low wages and uneven retention. Robert Moffitt, summarizing a Department of Health and Human Services review of state leaver studies, reported wages of about 7 dollars to 8 dollars per hour and earnings of about 3,000 dollars per quarter for leavers who worked. Ron Haskins reported about 8 dollars per hour in 2000. Gregory Acs and Pamela Loprest reported median wages across state studies ranging from 6.50 dollars to 9.00 dollars per hour. Grogger found no significant effect of time limits on earnings or income, while the earned income tax credit raised total income for working families beyond wages alone.

Q: How did food assistance respond compared to cash assistance in the downturn?

Food assistance expanded with need and cash assistance expanded modestly. The Department of Agriculture series recorded average monthly participation rising from 26,316,000 persons in fiscal 2007 to 40,302,000 in fiscal 2010, an increase of about 53 percent, after Congress raised maximum benefits by 13.6 percent in the American Recovery and Reinvestment Act of 2009. The Center on Budget and Policy Priorities state collected series recorded food assistance caseloads rising by 45 percent between December 2007 and December 2009, against 13 percent for cash assistance. Marianne Bitler and Hilary Hoynes found food assistance and other noncash programs significantly more responsive to the economic cycle after reform, with no comparable increase in responsiveness for cash welfare.

Q: Did the block grant keep pace with inflation?

No. The Congressional Research Service recorded that the basic block grant totaled 16.6 billion dollars, remained at the same nominal amount from fiscal 1997 through fiscal 2016, and carried no adjustment for inflation, caseload size, or changes in the poverty population. The Research Service calculated a real decline of 33.1 percent across the span in which the nominal amount held unchanged. The constant dollar table in the state section of the record recorded the state family assistance grant falling from 16.567 billion dollars in fiscal 1997 to 10.564 billion dollars in constant 1997 dollars in fiscal 2018, a cumulative decline of 36.2 percent in purchasing power. Each state share rested on spending in the middle 1990s rather than on later poverty counts.

Q: How far did the reach of cash assistance fall?

Reach fell on both available measures. The Center on Budget and Policy Priorities ratio of families receiving cash assistance to families with children in poverty fell from 68 per 100 in 1996 to 31 per 100 in 2006, 27 per 100 in 2010, and 23 per 100 in 2015, before reaching 22 per 100 in 2018. The Congressional Research Service recorded the share of eligible families receiving benefits at an estimated 82 percent in 1995 and an estimated 27 percent in 2015, and stated that most of the caseload decline after 1994 resulted from a reduction in that share rather than from a reduction in the number of families meeting state definitions of need. Geography widened as reach contracted, with fourteen states at 10 per 100 or less in 2015.

Q: Did welfare reform affect children?

Direct national evidence on the 1996 statute alone was limited, and most child evidence came from earlier or parallel programs. Pamela Morris, Aletha Huston, Greg Duncan, Danielle Crosby, and Johannes Bos, in a 2001 MDRC synthesis of random assignment programs, found that programs raising both parental employment and family income through earnings supplements improved school achievement among elementary school age children. Programs mandating employment activities without raising income produced few effects for that age group in the same synthesis. Lisa Gennetian and coauthors, in an MDRC synthesis covering adolescents, found worse reported school performance and higher grade repetition among adolescents in program families. No synthesis in the record claimed a direct estimate for a national cohort followed into adulthood by 2019.

Q: Why do income and consumption measures disagree about extreme poverty?

Three differences drive the disagreement. Income surveys miss benefit dollars, and Meyer, Wallace Mok, and Sullivan reported that more than half of welfare dollars and nearly half of food stamp dollars went unrecorded in several major surveys in the final years covered by that study. Cash income measures also exclude food assistance, refundable tax credits, and housing subsidies, which Edin and Shaefer showed cut the measured level of extreme poverty sharply when counted. Consumption, recorded in the Consumer Expenditure Survey, smooths across short income gaps through savings, borrowing, and help from relatives. Meyer and Sullivan treated consumption as the more accurate record for the most disadvantaged families, while Shaefer and Joshua Rivera argued that income poverty tracked hardship measures more closely.

Q: Did caseloads fall because families left or because fewer entered?

Both flows contributed. Grogger, Haider, and Klerman, in the American Economic Review in 2003, found that the 1990s reduction resulted from both faster exits from state assistance programs and fewer entries into state assistance programs, in the summary given by the Congressional Research Service. Exits included earnings closures, sanction closures, time limit terminations, and closures for missed redeterminations, and each recorded event subtracted one case without recording income after closure. Entries fell through formal diversion, including lump sum payments and applicant job search requirements, and through informal diversion during application. A leaver study alone cannot observe families deterred before entry, so flow evidence and leaver evidence answer different parts of the same decline.

Q: What did the state waiver experiments show?

Waiver experiments showed that defined packages changed behavior under demonstration conditions. Florida Family Transition Program combined a time limit of 24 months in a 60 month period for most recipients with financial incentives, mandates, and rich services, and MDRC found employment and earnings gains during the first two years, before participants could reach the limit. Minnesota Family Investment Program paired a generous earnings disregard with mandatory services and child care payment, and MDRC found increased employment, earnings, and total income and reduced poverty among long term recipients, alongside increased welfare receipt. The Congressional Research Service recorded 83 approved waiver applications from 43 states and the District of Columbia between January 1993 and August 1996. Demonstrations in selected counties during a strong labor market could not predict statewide performance during a recession.

Q: How did employment among single mothers change after the boom ended?

Employment retreated from the 2000 peak without returning to the level of the middle 1990s in most series before the recession trough. The Legal Momentum analysis titled Single Mothers Since 2000, drawing on Bureau of Labor Statistics data, recorded single mothers employed in an average month falling from 76 percent in 2000 to 73 percent in 2007 and 68 percent in 2009, with full time year round employment falling from 51 percent in 2002 to 45 percent in 2009. Haskins stated in Brookings Institution testimony that employment among females heading families had declined almost to the prereform level during the deep downturn. During the same span the earned income tax credit continued to pay only to workers, so families without earnings drew no support from the largest work based program.

Q: Can researchers separate the statute from the labor market?

Separation is partial and rests on assumptions rather than on a national experiment. Researchers used differences in waiver timing among states, differences in earned income tax credit amounts by family size, random assignment inside demonstrations such as the Florida Family Transition Program, and comparisons between single mothers and single women without children. Meyer and Rosenbaum built the central decomposition on that comparison, and Grogger used the age of the youngest child to isolate time limit effects in national survey data. Blank cautioned that a waiver indicator could stand for a bundle of administrative and political changes inside a state. No study in the record reviewed through 2019 observed the 1996 statute under a slack labor market with an unexpanded credit, so the counterfactual rested on model structure.