Welfare Reform Myths and the Word That Carries Too Much

The most consequential of the welfare reform myths holds that a single statute ended welfare in the United States, and the claim spread because a real and dramatic contraction stood behind the slogan. A presidential pledge promised an ending, a caseload fell by roughly three quarters across the following two decades, and a program that had anchored public discussion of poverty for two generations shrank to a minor place in the federal budget. Each element was visible. Newspaper accounts could print the pledge, chart the falling caseload, and describe closed or quieter welfare offices, and a reader supplied with those three facts had little reason to ask whether the word welfare in the slogan named the same thing as the word welfare in the budget. The misunderstanding therefore had a structure rather than a single author. A legal term with a narrow statutory meaning traveled into general speech, gathered programs the statute had never governed, and carried a verdict about the whole safety net that the evidence supported for one component alone.

The statute at the center of every claim in this article has a precise identity. The governing law is the Personal Responsibility and Work Opportunity Reconciliation Act of 1996, Public Law 104-193, 110 Stat. 2105, signed August 22, 1996, principally codified in Title IV-A of the Social Security Act at 42 U.S.C. 601 and following. That act repealed Aid to Families with Dependent Children, created Temporary Assistance for Needy Families as a fixed block grant to states, imposed work participation standards and a federal time limit on federally funded assistance, and restricted benefit eligibility for noncitizens through separate provisions codified elsewhere in Title 8. Precision about that identity matters because several claims fail at the level of attribution. A rule about cash assistance gets credited with changes in food assistance. A restriction written into immigration law gets described as a rule of the block grant. A spending total for every means tested program gets quoted as the cost of monthly checks to families. Naming the statute, the title, and the codified section at the start removes a class of error before any figure enters the argument.

The method applied to each claim follows a fixed order. The claim receives the strongest form a holder of the claim would recognize, because a verdict against a weakened version settles nothing. The evidence then comes from a deciding source named in the text: the statutory text itself, an official dataset published by the agency that administers the program, or a named researcher or institution whose figures can be checked. A verdict follows at one of three strengths. A claim is false when the deciding source contradicts the claim as stated. A claim is contested when named researchers, working from identified data, reached opposed conclusions that the record through December 15, 2019 had not resolved. A claim is partly true when the documented record supports one half of the claim at full strength and contradicts the other half at full strength. No verdict in this article rests on an unnamed study, an unsourced percentage, or a figure quoted without the population and period that produced the figure.

One test organizes the verdicts, and the test has a name: check the program and the denominator. The program check asks which program a claim describes, because cash assistance, food assistance, health coverage, Supplemental Security Income, and refundable tax credits followed different rules and moved in different directions after 1996. The denominator check asks what quantity a figure has been divided by, or what population a count describes, because a caseload count, a share of families in poverty, a share of benefit dollars, and a share of total federal spending answer four different questions. A claim that passes both checks can survive. A claim that fails either check fails as stated, however accurate an isolated figure inside the claim may be, because the error in such claims lives in the movement between a correct number and an incorrect conclusion. Eight claims receive that treatment in the sections that follow, drawn from the assertions repeated most often about the 1996 act: an ending that did not happen, a refusal to work that the caseload composition does not support, a time limit described as absolute, a budget share that depends entirely on definition, a drug testing saving that state accounts never recorded, a migration motive that scholarship left divided, a fraud prevalence that agency measurement placed near one percent of benefits, and a block grant claim that the record supports in one half and contradicts in the other.

The End That Did Not Happen and the Share That Depends on a Definition

Two claims about the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 recur in accounts of the decades that followed enactment. The first holds that welfare reform ended welfare. The second holds that welfare accounts for a large share of federal spending. Each claim turns on the meaning assigned to a single word. In the debates of the 1990s and in the budget documents of the period that followed, welfare carried at least two distinct meanings. In the narrower meaning, welfare denoted cash assistance to families with children, delivered through Aid to Families with Dependent Children before 1996 and through Temporary Assistance for Needy Families after 1996. In the broader meaning, welfare denoted the full set of federal means tested programs, a category that includes food assistance, health coverage for people with low incomes, Supplemental Security Income, and refundable tax credits. A verdict that is accurate under one meaning can be false under the other. Each claim therefore receives its strongest form first, and each verdict rests on a statute, an official dataset, or a named institutional analysis.

The statutory baseline is not in dispute. On August 22, 1996, President William J. Clinton signed the Personal Responsibility and Work Opportunity Reconciliation Act, cited as Public Law 104-193, 110 Stat. 2105. Title I of the Act repealed Aid to Families with Dependent Children, a program authorized under Title IV of the Social Security Act of 1935, together with the related Job Opportunities and Basic Skills Training program and the Emergency Assistance program. In place of those programs, the Act created Temporary Assistance for Needy Families, financed through fixed block grants to states under 42 U.S.C. 603. President Clinton had campaigned in 1992 on a pledge to end welfare in the form in which it had operated since 1935. The question for the first claim is whether the record of the following twenty two years supports the proposition that welfare, in either meaning of the word, came to an end. The provisions behind that baseline receive full treatment in the guide to the welfare reform act, its repeals, and its replacement program.

The First Claim in Its Strongest Form

In its strongest form, the claim that welfare reform ended welfare rests on legal structure. Before August 1996, Aid to Families with Dependent Children operated as an individual entitlement. A family that met the eligibility rules of a state plan held a legally enforceable claim to benefits, and federal funding operated as an open ended match, rising automatically when state spending rose. The 1996 Act terminated that structure in three ways that holders of the claim regard as decisive. Federal law after 1996 imposed a sixty month limit on assistance funded with federal grants for most families with an adult recipient, under 42 U.S.C. 608(a)(7). Federal law imposed work participation standards on states under 42 U.S.C. 607, with financial penalties for states that failed to meet the standards. Federal funding arrived as a fixed annual grant whose amount did not respond to changes in poverty, unemployment, or the size of the caseload. On that legal account, the program that had existed from 1935 to 1996 did not merely contract. The entitlement at the center of the program was abolished, and the replacement was a capped funding stream administered under time limits and work rules.

The caseload record gives the strongest form of the claim its numerical force. The Congressional Research Service caseload series, built from Department of Health and Human Services data, recorded a peak of 5.033 million families and 14.161 million recipients in calendar year 1994. The same series recorded 1.196 million families and 3.150 million recipients in calendar year 2018. The decline in families between 1994 and 2018 was about 76 percent, and the decline in recipients was about 78 percent. The composition of spending inside the block grant moved in the same direction. The Center on Budget and Policy Priorities calculated that states devoted 71 percent of federal and state Temporary Assistance for Needy Families funds to basic assistance in fiscal 1997. The Administration for Children and Families reported that basic assistance accounted for 21.4 percent of combined federal and state funds in fiscal 2018. Holders of the claim also point to the behavior of the program during the recession of 2007 to 2009. The Congressional Budget Office recorded in June 2019 that outlays for family support and foster care programs, the category that contains Temporary Assistance for Needy Families, ranged between 30 billion dollars and 35 billion dollars in each year from 2009 to 2019, a flat path during a period in which food assistance outlays rose sharply. A funding stream that does not expand when need expands, holders conclude, no longer performs the function that defined the pre 1996 program. What that funding record produced for caseloads, employment, and poverty is measured separately in the outcomes record for the 1996 act.

Did Cash Assistance End or Contract?

Cash assistance shrank dramatically without ending. The Administration for Children and Families recorded 1.196 million families receiving assistance in calendar 2018, and federal law continued to appropriate a block grant for that purpose in every year from fiscal 1997 through fiscal 2019. A program reaching roughly one family in four of its 1994 scale had contracted, not closed.

The distinction between contraction and termination matters because the two descriptions imply different accounts of the intervening years. The Administration for Children and Families published annual caseload, expenditure, and recipient characteristic reports for every year of the period, describing a program that paid monthly cash benefits in every state in every year, under eligibility rules that states set within federal parameters. The contraction was nonetheless severe by every measure the official series contain. Child recipients of cash assistance equaled 61.7 percent of all children in poverty in calendar year 1994, according to the Congressional Research Service series. The corresponding figure for calendar year 2018 was 19.5 percent. The block grant itself lost purchasing power throughout the period because the nominal amount was fixed while prices rose. The Congressional Research Service, in tables updated July 1, 2019, calculated that the state family assistance grant lost 36.2 percent of its real value between fiscal 1997 and fiscal 2018. The Center on Budget and Policy Priorities, in an analysis dated April 2, 2018, projected that the loss would reach 39 percent by 2019 in the absence of further cuts. A smaller caseload divided a grant of declining real value, and states directed a falling share of the combined funds toward monthly cash payments. The kernel of the first claim, understood as a claim about cash aid alone, records a genuine transformation. The claim as stated, however, asserts an ending, and an ending is what the record does not contain.

The claim also fails under the broader meaning of welfare, because the programs outside cash assistance did not contract in parallel. Several expanded substantially during the same years in which the cash caseload fell by about three quarters. The safety net that existed in 2019 differed from the safety net of 1996 in composition, with a far smaller role for unrestricted monthly cash payments and a far larger role for food assistance, health coverage, and credits conditioned on earnings. Each major component of the pre 1996 safety net had a documented counterpart in operation in fiscal 2018, and the largest components had grown.

What Happened to the Rest of the Safety Net

Food assistance supplies the clearest contrast with cash assistance. The Food and Nutrition Service of the Department of Agriculture recorded an average of 25.5 million participants per month in the Food Stamp Program in fiscal 1996, at a total federal cost of 24.3 billion dollars. The same agency recorded an average of 40.8 million participants per month in the Supplemental Nutrition Assistance Program in fiscal 2018, at a total federal cost of 65.5 billion dollars. Participation reached a peak of 47.6 million in fiscal 2013, during the slow employment recovery that followed the recession of 2007 to 2009, and declined during the expansion that followed without returning to the levels of the 1990s. The 1996 Act restricted eligibility for some categories of participants, including many legal immigrants and unemployed adults without dependents, and Congress restored portions of that eligibility in legislation enacted in 1998 and 2002. The direction of the two decades after enactment was nonetheless toward a food assistance program serving more people at greater federal cost than the program of 1996, operating as a legal entitlement whose spending rose and fell with eligibility and economic conditions.

Health coverage followed a similar path. The Balanced Budget Act of 1997, Public Law 105-33, created the State Children’s Health Insurance Program one year after the 1996 Act, extending federally matched coverage to children in families with incomes above Medicaid eligibility limits. The Patient Protection and Affordable Care Act of 2010, Public Law 111-148, expanded Medicaid eligibility to most adults with incomes up to 138 percent of the federal poverty line in states that adopted the expansion, with coverage taking effect in January 2014. The Congressional Budget Office reported that federal Medicaid outlays reached 389 billion dollars in fiscal 2018 and that Medicaid spending rose by 32 percent between 2013 and 2015 as the expansion was phased in. The same agency recorded 46 billion dollars in federal subsidies for health insurance purchased through the marketplaces in fiscal 2018, a category of means tested spending that had no counterpart in 1996, together with 17 billion dollars for the Children’s Health Insurance Program and 28 billion dollars for low income subsidies within Medicare Part D. Means tested health programs accounted for 480 billion dollars of mandatory outlays in fiscal 2018 in the Congressional Budget Office series, a figure more than twenty nine times the size of the cash assistance block grant recorded for the same year.

The tax code carried a growing share of income support during the same period. The Internal Revenue Service Statistics of Income series, as compiled by the Congressional Research Service, recorded 19.464 million returns claiming the earned income tax credit for tax year 1996, with total credits of 28.825 billion dollars. For tax year 2018, the same series recorded 26.492 million returns and total credits of 64.924 billion dollars. The child tax credit did not exist when the 1996 Act was signed. The Taxpayer Relief Act of 1997, Public Law 105-34, created the credit at 400 dollars per child for 1998 and 500 dollars per child from 1999. Legislation enacted in 2001, 2003, and 2004 raised the credit to 1,000 dollars per child and broadened the refundable portion, and Public Law 115-97 raised the credit to 2,000 dollars per child for 2018. The Congressional Budget Office recorded 77 billion dollars in outlays for the refundable portions of the earned income and child tax credits combined in fiscal 2018. Supplemental Security Income, the means tested cash program for aged, blind, and disabled recipients administered by the Social Security Administration, operated in every year of the period and accounted for 55 billion dollars in mandatory outlays in fiscal 2018. Cash support through the tax system and through Supplemental Security Income therefore expanded or persisted during the years in which cash support through Temporary Assistance for Needy Families contracted.

The verdict on the first claim follows from that distribution of evidence. As a statement about welfare in general, the claim is false. Cash assistance continued without interruption from 1996 through 2019, food assistance and health coverage grew substantially, and the earned income credit and the child credit delivered tens of billions of dollars in each year of the later period. As a statement about the cash program alone, the claim captures a real contraction and mislabels the contraction as an ending. The accurate formulation is that the composition of the safety net changed. Unrestricted cash assistance to families with children, the dominant instrument of the pre 1996 system, became a minor component measured by spending and by caseload, while food assistance, health coverage, and earnings conditioned credits became the principal instruments.

The Spending Claim in Its Strongest Form

The fourth claim holds that welfare is a large share of federal spending. In its strongest form, the claim defines welfare as the whole of federal means tested spending and observes that the total rivals the largest familiar categories of the budget. The Congressional Budget Office, in the June 2019 edition of Federal Mandatory Spending for Means-Tested Programs, placed mandatory outlays for means tested programs at 747 billion dollars in fiscal 2018, adjusted to exclude the effects of payment timing shifts. That total excluded means tested spending delivered through discretionary appropriations, such as housing assistance, and excluded state spending, including the state share of Medicaid. Medicaid alone accounted for 389 billion dollars of the federal total. For comparison, the Congressional Budget Office recorded discretionary outlays for national defense at 622 billion dollars in fiscal 2018, so federal means tested mandatory spending exceeded federal defense discretionary spending in that year. The same Congressional Budget Office series placed means tested mandatory outlays at 500 billion dollars in fiscal 2009 and at 747 billion dollars in fiscal 2018, an increase of about one half across a decade that included a recession and a coverage expansion. A category that absorbs 747 billion dollars in a single fiscal year, holders conclude, qualifies as a large share of federal spending under any ordinary use of the phrase.

The strongest form succeeds, however, only by adopting the broad definition of welfare, and the popular force of the claim depends on the narrow one. In ordinary usage during the debates that produced the 1996 Act, welfare referred to the cash program, and the claim invites a picture of a program of checks to families. The budget identity of that program is specific and measurable. It is the state family assistance grant under 42 U.S.C. 603(a)(1), together with the state funds that federal law requires as a condition of the grant. Measured against the federal budget, that program occupies a position at the opposite end of the scale from the one the claim describes.

Is Cash Assistance a Large Share of Federal Spending?

For cash assistance, no. The state family assistance grants totaled 16.489 billion dollars in fiscal 2018, equal to about four tenths of one percent of the 4,109 billion dollars in total federal outlays recorded by the Office of Management and Budget for that fiscal year. Only the broader means tested total reached a large share.

The arithmetic behind that verdict has two parts, a fixed numerator and a growing denominator. The numerator was set in nominal dollars in the 1996 statute and was not adjusted for inflation, for population growth, or for changes in the number of families in poverty. The grant amount appropriated under 42 U.S.C. 603(a)(1) stood at the same nominal level in fiscal 2018 as the level enacted for the first years of the program, which is why the real value of the grant fell by 36.2 percent between fiscal 1997 and fiscal 2018 in the Congressional Research Service calculation. The denominator moved in the opposite direction. The Office of Management and Budget historical tables recorded total federal outlays of 1,601 billion dollars in fiscal 1997 and 4,109 billion dollars in fiscal 2018. A grant equal to about one percent of federal outlays in fiscal 1997 therefore equaled about four tenths of one percent of federal outlays in fiscal 2018, without any legislated reduction in the nominal grant. Even a generous construction of the numerator does not rescue the claim. Combined federal Temporary Assistance for Needy Families expenditures and state maintenance of effort expenditures totaled 31.3 billion dollars in fiscal 2018, according to the Administration for Children and Families, equal to about 0.76 percent of total federal outlays in that year, and the combined total includes state dollars that are not federal spending. The portion of the combined total devoted to basic assistance, 21.4 percent of 31.3 billion dollars, came to about 6.7 billion dollars in federal and state funds combined. The grant mechanics behind those totals are set out in the account of the TANF block grant structure. The federal cash core of the program, measured on the most inclusive accounting available, accounted for well under one percent of what the federal government spent in fiscal 2018.

Classification explains most of the distance between the claim and the record. The Congressional Budget Office category of family support and foster care, which contains Temporary Assistance for Needy Families together with child support enforcement, the child care entitlement, and related programs for children, totaled 32 billion dollars in fiscal 2018. Medicaid, inside the same means tested accounts, totaled 389 billion dollars, more than twelve times the entire family support category and more than twenty three times the state family assistance grant. A statement that treats the 747 billion dollar means tested total as if it described the cash program overstates the federal commitment to cash assistance by a factor of about forty five. A statement that treats the cash program as the measure of means tested spending understates the total by the same factor. Both errors appear in public discussion of the 1996 Act because the word welfare is made to carry both quantities at once.

The Same Dollars Against Different Denominators

A spending figure acquires meaning only against a stated denominator, and the block grant illustrates the point three times over. Against total federal outlays of 4,109 billion dollars in fiscal 2018, the 16.489 billion dollar grant equaled about 0.40 percent. Against discretionary outlays, the proportions rise without changing the underlying dollars. The Congressional Budget Office recorded discretionary outlays of 1,263 billion dollars in fiscal 2018, of which 642 billion dollars was nondefense discretionary spending. The grant equaled about 1.3 percent of all discretionary outlays and about 2.6 percent of nondefense discretionary outlays. Against the population the program exists to serve, the comparison changes character again. The Census Bureau counted 38.1 million people with incomes below the official poverty line in 2018, a poverty rate of 11.8 percent. Divided across that population, the block grant equaled about 433 dollars per person in poverty for the year, before any division among cash benefits, child care, work programs, and administration. The 747 billion dollar means tested mandatory total, divided across the same population, equaled about 19,600 dollars per person in poverty. The second quotient overstates the resources directed to people counted as poor, because eligibility for Medicaid, food assistance, and the tax credits extends to households with incomes above the official poverty line, and because health spending purchases medical services rather than income. A share quoted without a named numerator, a named denominator, and a named year carries no verifiable content.

The verdict on the fourth claim follows the same division as the verdict on the first. For cash assistance, the claim is false. A fixed grant of 16.489 billion dollars against total outlays of 4,109 billion dollars is a small share by any standard applied elsewhere in budget discussion, and the share had fallen by more than half since fiscal 1997 through a frozen numerator and a growing denominator. For the whole of means tested spending, a version of the claim is defensible with the definition stated openly. Mandatory means tested outlays of 747 billion dollars in fiscal 2018 equaled about 18 percent of total federal outlays and a little more than one quarter of all mandatory outlays, as the Congressional Budget Office recorded. That version describes a budget dominated within the means tested category by health programs, in which Medicaid alone exceeded the cash block grant by a factor of more than twenty three. The two claims fail in the same way. Each takes a part of the post 1996 system for the whole, or the whole for a part. Cash welfare contracted to a small fraction of its former scale without ending, and the safety net that replaced the pre 1996 composition grew around it, financed largely through programs that the word welfare, in its 1996 sense, did not name.

Claim Two: Recipients Generally Refuse to Work

The Claim in Its Strongest Form

The claim, stated in the form its holders would recognize, runs as follows. Cash assistance under Temporary Assistance for Needy Families, and under Aid to Families with Dependent Children before it, went in the main to adults who could have supported themselves through paid work and who chose assistance instead. On this account, nonemployment among recipients was a preference rather than a circumstance. Work was available, the adult was able to perform it, and a monthly grant removed the necessity that would otherwise have produced employment, so the adult declined jobs or declined to search with genuine intent. Low measured employment therefore revealed motivation, and the remedy followed: time limits, sanctions, and participation mandates were instruments for changing a decision the recipient was free to make differently. At its strongest, the claim rested not on individual cases, which can be found in any large program, but on a generalization about the caseload as a whole: that generally, as a description of the typical recipient, refusal explained nonwork. That generalization is testable against official data on composition, on conditions adult recipients faced, on employment during receipt, and on the construction of the federal measure.

Who the Caseload Contained

The Administration for Children and Families, through its Office of Family Assistance, published Characteristics and Financial Circumstances of TANF Recipients for fiscal 2018 on September 9, 2019. In an average month in fiscal 2018, about 492,400 adults and about 1.8 million children received TANF cash assistance. Children thus accounted for roughly four fifths of the approximately 2.3 million individual recipients in that average month. A claim that silently pictures an adult deciding whether to accept a job misdescribes most of the caseload at the level of arithmetic, and a count dominated by children cannot support a conclusion about the work decisions of adults.

The Office of Family Assistance also reported approximately 548,000 child only families in an average month in fiscal 2018, defined as families in which no adult received TANF cash assistance, accounting for 54.5 percent of the total TANF caseload. The denominator was the 1,004,923 families receiving TANF cash assistance in the average month, in Table 9 of the fiscal 2018 tables. In a child only family, assistance was paid on behalf of the children, and any adult present received no assistance in an individual capacity.

A second series, on a wider denominator, confirms the same direction. The Congressional Research Service, in The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions dated July 1, 2019, tabulated the fiscal 2017 national data files for a universe including families in separate state programs funded with expenditures countable toward the maintenance of effort requirement. On that universe of 1,402,871 families, families in which all adult recipients were not working numbered 434,602, at 31.0 percent, and families with an employed adult numbered 435,259, also at 31.0 percent. Child only families accounted for 38.0 percent: 14.1 percent of all families were in care by a nonparent caretaker relative, 9.5 percent in a noncitizen parent category, 9.0 percent with a parent receiving Supplemental Security Income, 0.3 percent in another ineligible parent category, and 5.1 percent in an unknown category. The two series differ because their denominators differ, and neither percentage should be transferred to the universe of the other. On either universe, a family headed by an unemployed adult recipient was a minority component.

The child only categories record situations grounded in status and caregiving. A parent receiving Supplemental Security Income had met a disability based standard under that separate program. A nonparent caretaker relative, most often a grandparent, aunt, or uncle on the descriptions in the Congressional Research Service report, received assistance on behalf of a child in relative care without receiving assistance as a parent recipient. A parent ineligible on citizenship grounds was not a recipient whose benefit created the disincentive the claim describes.

Health, Caregiving, and the Limits on Available Labor

The characteristics tables did not themselves measure diagnosis, so the strongest verified evidence on health came from survey based work by named agencies. The United States Government Accountability Office, in report GAO-02-37 of October 31, 2001, and in briefing report GAO-02-884 of 2002, analysed the Census Bureau Survey of Income and Program Participation for 1999 and found that 44 percent of TANF recipients aged 18 to 64 reported a physical or mental impairment, with 38 percent reporting a severe impairment and 29 percent reporting a mental impairment. The finding does not establish that every recipient reporting an impairment was unable to work: the Government Accountability Office recorded that about 20 percent of recipients reporting impairments also reported working in 1999.

A report issued by the Administration for Children and Families in February 2008, Assisting TANF Recipients Living with Disabilities to Obtain and Maintain Employment, summarized a common survey fielded in six states: the share reporting a mental health condition ranged from 21 percent to 41 percent across those states, the share reporting a learning disability ranged from 8 percent to 18 percent, and the share reporting a physical health condition ranged from 16 percent to 26 percent. The Women Employment Study conducted by University of Michigan researchers including Sheldon Danziger and colleagues in one urban Michigan county found that low education, limited job skills, mental health problems, care for a child with a health problem, and transportation problems were each present for at least 20 percent of the single parent caseload in that study population. These were bounded populations, not national estimates, but barriers that constrain hours, retention, and search were prevalent enough in each that an explanation built on preference alone leaves most measured variation unexplained.

In Table 9 of the fiscal 2018 tables, the age of the youngest child was under one year for 12.1 percent of the 1,004,923 TANF families, one year to two years for 18.5 percent, and three years to five years for 19.8 percent, so the youngest child was under six years of age in just over half of all families, at 50.4 percent on those component shares. Under 42 U.S.C. 607, a single custodial parent caring for a child under twelve months of age could be disregarded from the participation calculation for a limited period, and states administered exemptions and good cause rules for illness, incapacity, and care for a family member with a health condition. An exemption does not prove that every parent of a young child was unable to accept employment, and parents of young children in the wider labor force worked in large numbers in 2018 and 2019. The typical case involved care for a child under six, so employment depended on child care availability, cost, hours, and stability in a way employment for a childless adult did not.

The Office of Family Assistance reported that 35.9 percent of adult TANF recipients in an average month in fiscal 2018 had completed less than a high school education, 55.3 percent had completed high school or an equivalent credential and no further education, and 8.8 percent had education beyond high school, with adult recipients in that average month as the denominator. Low attainment does not equate to refusal. It bears on the kind of employment on offer, on wages, and on stability of hours, and it describes the labor supplied rather than willingness to supply it.

Employment During Receipt

On employment itself, families with an employed adult accounted for 31.0 percent of all assistance families on the wider fiscal 2017 universe, equal in share to families in which all adult recipients were not working. On the fiscal 2018 release, 27.3 percent of adult TANF recipients were employed in an average month, with adult recipients as the denominator. A monthly measure understates employment over a longer window by construction, because a recipient who worked in February and not in March appears as not employed in March. Earnings records in the leaver literature showed higher employment on an ever employed over four quarters measure than on any single quarter measure for the same populations, the expected arithmetic of short duration jobs.

Table 51 of the fiscal 2017 characteristics tables recorded 858,841 closed case adult recipients in fiscal 2017, of whom 26.5 percent were recorded as employed at closure. Employment at closure is not employment after closure, and a case can close for earnings, sanction, time limit, household change, or administrative reasons, so the figure cannot be assigned a single cause.

Cycling between work and assistance reconciled the monthly and longitudinal figures. A recipient could work in unsubsidized employment, lose hours or lose the job, remain on or return to assistance during the gap, and work again, with each monthly snapshot classifying the case by that month alone. The Congressional Research Service, in the July 1, 2019 report, described states meeting work standards in part by assisting families who were already working, through regular assistance for families whose earnings remained low enough to qualify and through earnings supplement arrangements. Subsidized employment, in which a state used TANF or maintenance of effort funds to support a wage in a placement, formed a smaller component and was recorded as a work activity rather than as unsubsidized employment. Unsubsidized employment during receipt is direct evidence against a general refusal. Subsidized employment should not be inflated into independent labor market employment, nor dismissed as nonwork, since the hours were real even though the wage was publicly supported.

Does a Low Participation Rate Mean a Family Did No Work?

No. The federal rate counted a family only when a work eligible adult completed the required hours in a countable activity with verified documentation in that month. Employment at fewer hours, employment in a noncountable activity, and work in another month fell outside the numerator, even though employment had occurred in fact.

The work participation rate under 42 U.S.C. 607 was a performance standard applied to a state, not a survey of individual behavior and not a legal duty that ran directly against a recipient in federal law. The statute set a standard of 50 percent for all families and 90 percent for two parent families, measured as the share of families with a work eligible individual in which the relevant adults met federal hours requirements in listed activities, over the month. States set individual obligations under state law, decided exemptions within the federal framework, and imposed sanctions under state rules. A state could meet or miss its standard for reasons that included documentation, the mix of activities counted, and the credit described in the next section, while individual employment followed a separate distribution the rate captured only in part.

Hours thresholds and countable activities drove a further wedge. A parent employed for fewer hours than the federal threshold, or in a month in which verification was incomplete, or in education beyond the period for which such activity could count, could be employed and uncounted at the same time. The converse also held: a parent counted toward the rate might be in job search or work experience, activities that were not paid employment. The rate was therefore neither a pure employment rate nor a pure activity rate, and quoting it as a measure of refusal misstates the instrument in both directions.

Families without a work eligible individual were also excluded from the denominator, including child only families with a parent receiving Supplemental Security Income, families in which children lived with a nonrecipient nonparent relative, and families with an ineligible noncitizen parent, on the exclusions in the Congressional Research Service report of July 1, 2019.

The Credit That Set the Effective Standard

The caseload reduction credit at 42 U.S.C. 607(b)(3) supplied the largest single reason why measured rates and effective standards diverged from the headline numbers of 50 percent and 90 percent. As enacted in the Personal Responsibility and Work Opportunity Reconciliation Act of 1996, the credit reduced the standard a state was required to meet by the percentage by which its caseload had declined from the base year of fiscal 1995, subject to an adjustment for declines attributable to eligibility changes. The mechanism operated in percentage points: a fall of 30 percent produced an effective all families standard of 20 percent, and a fall of 50 percent or more produced an effective standard of zero percent. The credit measured decline from any cause that was not an eligibility change, including economic expansion, diversion, sanction, and time limit, and assigned the same credit to each cause. A state earned a lower required rate by having a smaller caseload, whatever the route by which families had left, and the required rate carried no information about the behavior of the families who remained.

The Deficit Reduction Act of 2005, Public Law 109-171, enacted on February 8, 2006, recalibrated the credit by substituting fiscal 2005 for fiscal 1995 as the base year, effective for fiscal 2007. Caseloads in fiscal 2005 stood far below caseloads in fiscal 1995, so the accumulated decline of the first decade after 1996 ceased to generate credit. The same statute broadened the population in the calculation to families in separate state programs and required federal definitions of countable activities and verification, changes that raised measured difficulty independently of recipient behavior. For fiscal 2017, the Congressional Research Service report of July 1, 2019, recorded an average all families participation rate of 53.0 percent and an average two parent rate of 69.5 percent, with only two jurisdictions, Nevada and Guam, failing the all families standard as adjusted, and nine jurisdictions failing the two parent standard as adjusted. Those averages describe performance against standards already reduced by each state credit.

In Congressional Research Service tabulations of Department of Health and Human Services data reproduced in TANF Block Grant reports available before December 2019, numerous states recorded a credit of 50.0 percentage points against the all families standard, producing an effective standard of 0.0 percent, while other states faced effective standards in single digits. A state with an effective standard of zero percent met its federal standard at any participation rate. An outcome of met or not met therefore certified only that a ratio over a restricted denominator had cleared a threshold the credit had lowered by an amount determined by caseload history since fiscal 2005.

Before the fiscal 2007 recalibration, the long decline from the fiscal 1995 base had driven effective standards toward zero in state after state, so compliance coexisted with modest measured participation, and a reading of either as a behavioral finding mistook an artifact of the formula for evidence. After recalibration, measured failure became more common in fiscal 2012, when sixteen states failed the all families standard on the Congressional Research Service account, and compliance recovered by fiscal 2017 as caseloads fell further from the fiscal 2005 base. The sequence tracks the formula and its base year, and no dataset cited in this section records a synchronized change in the adults themselves.

Work After Exit and the Evidence of the Leaver Studies

Evidence after exit supplies an independent test, because a caseload characterised by general refusal would be expected to show low employment once assistance ended. The leaver studies funded by the Assistant Secretary for Planning and Evaluation, synthesised in the Final Synthesis Report of Findings from ASPE Leaver Grants prepared by Gregory Acs and Pamela Loprest of the Urban Institute in 2001, covering fifteen studies of families that left TANF in the late 1990s, found the opposite pattern on incidence alongside a consistent pattern on earnings level. Across those studies, about three quarters of leavers worked at some point in the year after exit, about three out of five were working at any given point in time, and a little more than one third worked in all four quarters. Median earnings were about 2,600 dollars per quarter, and earnings rose between the first and fourth quarters after exit in most studies, by at least 200 dollars per quarter. Hourly wages averaged between seven dollars and eight dollars per hour, above the federal minimum wage of 5.15 dollars per hour in the study period and low in absolute terms.

Two qualifications apply. Leavers were a selected population, defined by exit through earnings, sanction, time limit, household change, or administrative closure, so leaver rates do not estimate employment for families that remained on assistance. The studies also relied on unemployment insurance earnings records, which missed informal work, self employment, and work across a state line. Subject to those qualifications, the leaver population showed substantial attachment to paid work.

The Counterweight: Low Earnings and Unstable Work

The same datasets defeat any suggestion that employment characteristically delivered self sufficiency. In the Acs and Loprest synthesis, only a little more than one third of leavers worked in all four quarters after exit, so most who worked did so intermittently, and median quarterly earnings of about 2,600 dollars annualise to an amount well below the poverty guideline for a family of three in the study period. Government Accountability Office report GAO-02-884 of 2002 found that leavers with impairments were one third as likely to be employed as leavers without impairments after adjustment for demographic characteristics and receipt of Supplemental Security Income. The fiscal 2018 figures add that the 27.3 percent of adult recipients employed in an average month were, by continued eligibility, families whose earnings had not carried them beyond the income limits of the program. Employment at low wages, with variable hours and gaps between jobs, was the modal form work took for recipients and leavers alike.

A monthly employment share of 27.3 percent in fiscal 2018 can be read as low only against an expectation of stable full time work. Against the quarters worked distributions in the leaver synthesis, it is consistent with a larger share working over a year in shorter spells. Neither reading licenses a conclusion about willingness, which administrative data do not observe. Hours offered, shift scheduling, child care for a youngest child under six in half of families, health conditions reported by 44 percent of adult recipients in the 1999 survey, and attainment below high school completion for 35.9 percent in fiscal 2018 each bear on employment through channels that do not pass through refusal.

Assessment of the Claim

On this evidence, the claim does not survive in its general form. Most individual recipients in fiscal 2018 were children. A majority of families, 54.5 percent, were child only families containing no adult recipient. On the wider fiscal 2017 universe, families with an employed adult equalled families with no adult working, at 31.0 percent each. Health conditions, care for a youngest child under six in half of families, and low attainment characterised the adult caseload in named surveys and tables. The participation rates quoted for the claim were state performance measures over a restricted denominator, judged against standards reduced, in many states to zero, by the credit at 42 U.S.C. 607(b)(3). Leavers worked at high incidence and low earnings. What the record supports is a narrower finding, stated without characterisation of any group: work among recipients and former recipients was common, intermittent, and poorly paid, nonwork was associated in the data with health, caregiving, and human capital measures, and the federal statistics reflected the design of the standard at least as much as the conduct of the families measured. Individual cases in which an adult declined available work cannot be excluded or quantified by aggregate data.

Limits That Bend and a Motive That Resists Proof

The Claim That Sixty Months Means Sixty Months

The third claim, in the strongest form its holders would recognize, runs as follows. The Personal Responsibility and Work Opportunity Reconciliation Act of 1996, Public Law 104-193, signed on August 22, 1996, replaced an open ended entitlement with a block grant built around reciprocity and duration. Work requirements, participation rates and a lifetime clock formed a single discipline, and the clock was the point. A family received assistance while it prepared for employment, the months accumulated whether consecutive or not, and at sixty months assistance ended. Exceptions were administrative detail. The force of reform depended on the limit being applied to every family in the same way, without a route around it through another funding label or a discretionary extension.

The verdict is false, and the falsification sits in the statute. Section 408(a)(7) of the Social Security Act, codified at 42 U.S.C. 608(a)(7), does not directly prohibit a family from receiving assistance for more than sixty months. The Congressional Research Service, in its primer on the Temporary Assistance for Needy Families block grant, describes the provision as a prohibition on the use of federal funds. A state may not use federal TANF funds for a family that includes an adult, or a minor head of household, who has received federally funded assistance for a cumulative sixty months. The distinction between a funding prohibition and a personal disqualification matters, because a rule about which dollars may be spent leaves open which families and which months enter the count, and which dollars a state may substitute when the federal count is exhausted.

Does the federal limit end all assistance after sixty months?

No. The federal rule barred federal funds after sixty cumulative months for a family containing an adult recipient, subject to a hardship exemption for up to twenty percent of the caseload, exclusions for child only cases and specified months, and state authority to continue aid with state funds alone.

The hardship exemption is the most direct qualification, and its size defeats any description of the limit as absolute. Under 42 U.S.C. 608(a)(7)(C), a state may extend federally funded assistance beyond sixty months for reasons of hardship, as the state defines hardship, or where a family includes an individual who has been battered or subjected to extreme cruelty. The Administration for Children and Families states the ceiling in operational terms. Families receiving assistance beyond sixty months under the exemption must not exceed twenty percent of the average monthly caseload. In a state with a caseload of fifty thousand families, the ceiling permits ten thousand extensions. The Department of Health and Human Services study titled Welfare Time Limits, prepared for the Office of Planning, Research and Evaluation, adds that because child only cases form part of the denominator, the share of adult headed cases extended can exceed twenty percent of adult headed cases alone.

The second qualification concerns who is counted at all. The federal clock applies to a family that includes an adult recipient or a minor head of household. It does not apply in the same way to a child only case, in which assistance is paid on behalf of a child and no adult receives assistance for that adult. The Congressional Research Service primer gives the standard illustration of children cared for by a nonparent relative, such as a grandparent or an aunt, where the relative receives no assistance for the relative. No adult recipient exists whose months accumulate, and assistance for the child may continue beyond sixty months without breaching the funding prohibition. The same logic reaches households in which a parent is excluded from the unit under other rules, including ineligibility on immigration grounds.

The third qualification concerns which months enter the count. Under 42 U.S.C. 608(a)(7)(D), a state disregards a month in which an adult lived in Indian country or an Alaskan Native village where at least fifty percent of the adults were not employed. Administration for Children and Families guidance for Tribal TANF restates the rule and pairs it with the child only disregard. Months within the disregard never accumulate, so two adults with identical calendar periods on assistance may stand at different positions on the federal clock. Certain one time or short term benefits and support services also fall outside the regulatory definition of assistance that starts the clock under 45 C.F.R. 264.1 to 264.3, so receipt of help from a TANF funded agency and accumulation of a countable month are not identical events. The statute measures something narrower than calendar time.

The fourth qualification concerns money. Because the prohibition governs federal funds, a state may continue assistance beyond sixty months with state funds kept apart from federal funds. Federal law expressly permits use of state maintenance of effort funds beyond the limit, and the Health and Human Services study of state policies describes two routes. A state may segregate state funds within its program, paying a family from state dollars so that the federal clock does not govern the payment, or operate a separate state program funded solely with state dollars for the families placed in it. Article 231 of this series documented both routes. The Congressional Research Service primer offers New York as an example of aid beyond five years using state funds, and California as an example of assistance for the adult ending at a state limit while assistance for the children continues without a lifetime end point. Neither pattern fits a description of sixty months as the end of assistance.

State variation completes the refutation, because states moved in both directions from the federal figure. The Welfare Rules Databook, compiled by the Urban Institute and funded by the Department of Health and Human Services, recorded state lifetime limits as of July 2013 that diverged sharply from sixty months. Arkansas and Arizona each set twenty four months for the entire unit. Connecticut set twenty one months. Idaho set twenty four months, and Indiana applied twenty four months to the adult portion. Florida and Georgia applied forty eight months, and California applied forty eight months to the adult portion while assistance for children followed the child only logic described in this section. At the other end, the Health and Human Services study of state policies in the first decade after enactment found that Michigan and Vermont had adopted no state time limit at the time of that analysis, a position Michigan later changed with a forty eight month limit effective October 1, 2007.

A narrower version of the claim survives. Sixty months was a genuine constraint on federal spending, backed by a penalty of up to five percent of the adjusted state family assistance grant for use of federal funds beyond the permitted margin, as the Congressional Research Service primer records. For many families, sixty months or a shorter state figure did mark the end of cash assistance. What the record does not support is the adjective absolute, because the statute that created the limit also created the exemption, the disregards, the child only exclusion and the state funded continuation.

The Claim That Benefits Explain Migration

The sixth claim concerns motive. In the strongest form its holders would recognize, public benefits are income available without employment, and a rational person comparing destinations compares total income, including benefits, alongside wages, housing costs and family or community networks. A jurisdiction offering more generous benefits therefore offers a higher total return to a person with low earnings or no earnings. Benefit generosity need not be the sole reason for a move, or the reason a migrant would name first, to be a real reason. Migration flows should, over time, tilt toward generosity, and restricting benefits for newcomers should reduce the incentive and the cost. The strongest form draws support from a genuine pre 1996 observation. Noncitizens formed a rising share of caseloads in several major programs during the first half of the 1990s, a pattern federal budget and audit agencies documented for Supplemental Security Income in particular, and Congress wrote a corresponding policy statement into section 400 of the 1996 act, codified at 8 U.S.C. 1601, favoring removal of the incentive for immigration provided by the availability of public benefits. A legislature that enacted a five year bar plainly treated the incentive as worth legislating against.

The verdict is twofold. As a practical claim about access, the statute heavily constrained the route described, for the population and period governed. As a claim about motive, the literature remained genuinely divided through December 2019, and the honest label is contested. A bar on eligibility proves nothing by itself about why any individual migrated, and a study of location choice inside the United States proves nothing by itself about the decision to enter the United States.

The statutory constraint is the firmer ground, and precision about its layers matters because public discussion routinely merges them. Section 401, codified at 8 U.S.C. 1611, provides that a person who is not a qualified alien is not eligible for a federal public benefit, subject to limited exceptions including emergency medical assistance. Section 402, codified at 8 U.S.C. 1612, sets limited eligibility for qualified aliens in specified and designated programs, including Supplemental Security Income and the Food Stamp Program, later renamed the Supplemental Nutrition Assistance Program, in subsection (a), and Medicaid, TANF and the Social Services Block Grant in subsection (b). Section 403, codified at 8 U.S.C. 1613, imposes the separate rule known as the five year bar. A qualified alien who entered on or after August 22, 1996 is not eligible for a federal means tested public benefit during the five years after entry in qualified status. The five year bar is 8 U.S.C. 1613, not 1612. Section 1612 governs whether a category is eligible for a particular program at all, while section 1613 governs the waiting period across federal means tested benefits.

Within that structure, the exceptions define the practical reach of the bar. Refugees admitted under section 207 of the Immigration and Nationality Act and asylees granted asylum under section 208 were eligible for Supplemental Security Income and food assistance for seven years after arrival under 8 U.S.C. 1612(a)(2)(A), and for TANF for five years, as the Congressional Research Service overview of noncitizen eligibility summarized the rules. Veterans, members of the armed forces on active duty, and their spouses and unmarried dependent children were excepted. A lawful permanent resident credited with forty qualifying quarters, generally ten years of covered employment, including quarters credited from a spouse or parent in defined circumstances, stood in a different position from a resident without that history. Cuban and Haitian entrants and Amerasian immigrants appeared in the exception lists, and emergency Medicaid remained available without regard to the bar. The result was a schedule in which category, date of entry, work history, military connection and program each changed the answer.

Later legislation restored parts of what the 1996 act removed, and each restoration was partial, dated and program specific. The Balanced Budget Act of 1997 restored Supplemental Security Income and related Medicaid for defined groups lawfully residing on August 22, 1996, and the Agricultural Research, Extension and Education Reform Act of 1998 restored food stamp eligibility to defined pre enactment groups, including certain children, older persons and persons with disabilities. The Farm Security and Rural Investment Act of 2002, Public Law 107-171, enacted the broadest food assistance restoration. Under the sequence the Department of Agriculture described in its implementation assessment, eligibility for persons with disabilities took effect in October 2002 regardless of date of entry, eligibility based on five years of residence took effect on April 1, 2003, and eligibility for lawful permanent resident children took effect on October 1, 2003 regardless of date of entry. The Children Health Insurance Program Reauthorization Act of 2009, Public Law 111-3, addressed health coverage. Section 214 gave states an option to cover lawfully residing children and pregnant women in Medicaid and the Children Health Insurance Program without the five year wait, notwithstanding sections 1611, 1612 and 1613. No restoration repealed the bar as a general rule, and by December 2019 a newly arrived qualified alien without an exception faced unavailability of TANF cash assistance, Supplemental Security Income, food assistance and nonemergency Medicaid during the first five years, subject to those restorations.

Any demographic statement in this area requires a denominator. A share of a caseload is a ratio between a group receiving a benefit and all persons receiving that benefit. A poverty rate is a ratio between persons in poverty in a group and all persons in that group. A participation rate among persons in poverty is a third and different ratio. The United States Census Bureau reported an official poverty rate for the whole population of 11.8 percent in 2018, down from 12.3 percent in 2017, in Income and Poverty in the United States for 2018. A comparison of benefit use that omits the corresponding poverty rates compares groups with different eligibility pools, because means tested eligibility follows income, and the literature divides in large part over which comparison is relevant and over the unit at which use should be counted.

What Did the Named Studies Find?

The named studies divided along the method each used. Borjas found immigrant households clustered in high benefit states in a pattern consistent with a welfare magnet. Zavodny and Kaushal, conditioning on networks and labor markets, found no such pull. Camarota found higher household level use; Nowrasteh and Ku, changing the unit of comparison, found lower use.

George J. Borjas of Harvard University set out the welfare magnet hypothesis in Immigration and Welfare Magnets, published in the Journal of Labor Economics in 1999. Borjas argued that a person who has already decided to bear the fixed cost of an international move faces a small additional cost in choosing one state over another, so benefit differences among states should weigh more heavily for new immigrants than for the native born, for whom an interstate move carries the full fixed cost. Empirically, Borjas reported that immigrant households receiving welfare clustered in states with higher benefits, and read the clustering as evidence consistent with a magnet effect. The objection raised against that reading concerns identification. States with higher benefits also contained gateway communities, large labor markets and networks of relatives and earlier migrants that independently predict settlement, and clustering in California, New York or Texas is consistent with a benefit motive, a network motive and a labor market motive that cross sectional data do not separate.

Madeline Zavodny, in work published in 1997 on welfare and the locational choices of new immigrants, reported weak or inconsistent evidence that benefit levels explained settlement once employment conditions and the prior distribution of immigrants from the same origin were taken into account. Neeraj Kaushal of Columbia University published the sharpest statement of that position in New Immigrants Location Choices, subtitled Magnets without Welfare, in the Journal of Labor Economics in 2005. Kaushal found that new immigrants located where earlier immigrants from the same origin had located and where labor market conditions were favorable, while state benefit levels did not exert the predicted pull. The objection raised against Kaushal and Zavodny runs in the opposite direction from the objection to Borjas. Controls for the existing immigrant population may absorb the channel through which benefits operated historically, because earlier settlement may itself have followed earlier benefit patterns, and compressed interstate benefit differences in parts of the period limit the power of a test to detect a response.

A separate body of work asked how much assistance immigrant households used, and divided over the unit of analysis and over adjustment for income and education. Steven A. Camarota of the Center for Immigration Studies, in Welfare Use by Immigrant and Native Households, published in 2015 and based on the Census Bureau Survey of Income and Program Participation for 2012, reported that fifty one percent of households headed by an immigrant reported use of at least one welfare program, compared with thirty percent of households headed by a native born person. Camarota counted a household as using welfare where any member used a cash, food, Medicaid or housing program, and attributed the gap principally to education and earnings rather than to unwillingness to work. Alex Nowrasteh of the Cato Institute, in a detailed 2015 response, objected that the household head method counted benefits received by United States born members, including citizen children and spouses, as immigrant use, reported no dollar value, and ignored larger household size, which raised the probability that at least one member used a program. On an individual comparison weighted by value and adjusted for income, Nowrasteh found materially lower use by immigrants. Camarota replied that eligibility and consumption operate at household level, and that excluding citizen children removed the fiscal question of whether a household formed by immigration could support its members without taxpayer funded programs.

Work that adjusted for observable characteristics added a distinction between cash and near cash programs and health coverage. Kristin F. Butcher and Anne Morrison Piehl, and Marianne P. Bitler and Hilary W. Hoynes, in research on immigrant participation across the period after 1996, found immigrant participation in cash and near cash programs falling below native participation after the 1996 restrictions, while health participation followed a different path shaped by later coverage legislation for children. Leighton Ku and Brian Bruen, in research published through the Urban Institute and the Kaiser Commission in 2013, reported that low income noncitizen adults used benefits at lower rates and consumed lower average amounts than low income citizen adults, once the comparison was confined to adults of comparable low income. Camarota and other researchers at the Center for Immigration Studies objected that adjustment assumes away the policy question, because admission of large numbers of persons with low education produces low income as a foreseeable outcome. Ku and Nowrasteh replied that an unadjusted comparison mislabels a poverty effect as an immigration effect.

No named study in the literature through December 2019 measured motive directly. Surveys about reasons for migration, application records and econometric studies of location each capture a different object, and none captures the counterfactual at the center of the claim, namely whether the same person would have remained in the place of origin, or chosen a different destination, under a different benefit schedule. The Danish evidence often cited internationally illustrates both a possible response and the limits of transfer. Ole Agersnap, Amalie Jensen and Henrik Kleven studied a Danish scheme that reduced benefits for specified immigrants from outside the European Union by up to fifty percent beginning in 2002, with repeal in 2012 and reinstatement in 2015, and presented Denmark as a rare case in which benefit variation was plausibly independent of other determinants of location. Denmark operated a national schedule and a different admission structure from the United States, where admission categories, state variation, sponsor obligations and the five year bar interacted, so the Danish result does not settle the size of any response under the American schedule.

The balanced verdict follows. The practical half of the claim, that a person could enter after August 22, 1996 in an ordinary qualified category and obtain federal means tested benefits on arrival, is contradicted by sections 1611, 1612 and 1613 for the first five years, subject to the exceptions and restorations set out in this section. The motive half, that benefits cause migration or destination choice in a measurable degree, is contested. Borjas found clustering consistent with a magnet. Zavodny and Kaushal, conditioning on networks and labor markets, found no welfare magnet in American location data. Camarota found higher household level use, attributed principally to education and earnings. Nowrasteh and Ku, changing the unit, the valuation or the comparison group, found lower use than by comparable native born persons. The statute is settled. The scholarship, through December 2019, was not.

Waste, Testing, and the Difference Between Error and Fraud

Claim Five in Its Strongest Form

The fifth claim holds that drug testing applicants for and recipients of public assistance saves public money. In the strongest form presented by Governor Rick Scott of Florida, who signed the Florida testing statute in 2011, and by legislators in other states who enacted screening statutes, the argument ran as a chain of three propositions. Use of illegal drugs occurred at a materially high rate among applicants for cash assistance, high enough for testing to identify a substantial number in each application cohort. A test, or a screening instrument followed by a test, would identify those people accurately at a modest unit cost. The benefits withheld from people identified, together with any deterrent effect on later applications by people who expected to fail, would then exceed the cost of the tests, any reimbursement owed to applicants who passed, staff time, and litigation. Stated in that form, the claim is testable against published state figures for a defined cohort, a positive count, and a cost. The assessment uses Florida, Utah, and Tennessee, the states for which a state agency source supplied those elements.

What Did Florida’s Program Find?

Florida tested 4,086 TANF applicants from July to October 2011 and found 108 positives, a rate of 2.6 percent. Reimbursing applicants who passed cost $118,140, which exceeded the benefits withheld from those who failed by $45,780 on the figures released by the Department of Children and Families.

Section 414.0652 of the Florida Statutes took effect on July 1, 2011 and required each applicant for Temporary Cash Assistance, the Florida program financed through the federal Temporary Assistance for Needy Families block grant, to submit to a urine test at the applicant’s expense, with reimbursement for an applicant whose test was negative. A positive test made an applicant ineligible for one year, subject to a route back through treatment, and a parent who failed could designate another person to receive benefits on behalf of a child. The Department of Children and Families reported that 4,086 applicants were tested in the period from July through October 2011, that 108 tested positive, most often for marijuana, and that about 40 additional applicants cancelled a scheduled test. The Substance Abuse and Mental Health Services Administration, through the National Survey on Drug Use and Health for 2011, estimated that 8.7 percent of the population of the United States aged twelve or older had used an illicit drug in the month before the survey interview. A survey estimate of self reported past month use and a urine test with a shorter detection window are not the same measurement, and the populations differ, so the comparison cannot establish an exact difference in prevalence. The comparison did defeat the premise that the applicant cohort presented an exceptional prevalence, because the state offered no evidence in the litigation that followed that prevalence among applicants exceeded prevalence in the general population, and the only measured applicant rate was 2.6 percent.

The fiscal arithmetic was direct, because reimbursement made the cost of testing the passing majority a state cost. The state paid $118,140 to reimburse tests taken by applicants who passed, and the Department figures put the net loss at $45,780 for the four months of operation. Scale alone could not reverse that result while the positive rate held, because each cohort of one hundred applicants at a 2.6 percent positive rate would generate about ninety seven reimbursements against about three withholdings. A deterrent effect, under which a person who expected to fail refrained from applying, would have produced a saving outside the tested cohort, but the state record gave that possibility no support. An internal Department document concerning caseloads for July through September 2011 recorded that the testing requirement had not led to fewer cases, and reporting on the released state figures in 2012 recorded no reduction in applications attributable to the requirement.

Screening Models in Utah and Tennessee

Utah enacted a screening model in 2012. An applicant for the Family Employment Program, the Utah cash assistance program, completed a written instrument titled the Substance Abuse Subtle Screening Inventory, and an applicant whose result indicated a reasonable likelihood of a substance use disorder was required to take a drug test. The Utah Department of Workforce Services reported in August 2013 that the Department had screened 4,730 applicants from August 2012 through the summer of 2013, had required 466 to take a drug test, and had recorded 12 positive tests. The state spent nearly $6,000 on the screening instruments and more than $25,000 on the drug tests, a combined cost of more than $30,000 to identify 12 people. A positive test in Utah did not by itself terminate assistance. The statute directed the applicant into substance abuse treatment, and benefits were at risk for failure to participate in treatment rather than for the positive result alone, so no amount of benefits withheld as a direct consequence of a positive test was available in the Department release to offset the cost. A later Department report showed the same pattern in a different cohort. For August 1, 2014 to July 31, 2015, the Department screened 4,225 applicants, gave a urine test to 460, and recorded 18 positives, at a cost of $5,281.25 for screening and $23,281 for drug tests, a combined $28,562.25.

Tennessee enacted its statute in 2012, and the Tennessee Department of Human Services implemented a suspicion based model on July 1, 2014 for Families First, the Tennessee cash assistance program. Each adult applicant answered a three question screener about use of, or problems relating to, marijuana, cocaine, methamphetamine and amphetamines, and opiates during the preceding three months, and an applicant whose answers were deemed to indicate use of a controlled substance was required to take a drug test. The Department provided figures for the first eighteen months of operation in February 2016, as reported by the Associated Press and the Tennessean on the basis of Department data. Of 39,121 people who applied in that period, 65 tested positive, a share below 0.2 percent. The Department reported that 609 applicants had been asked to take a drug test, that 116 applicants had refused the initial screening questionnaire and were disqualified on that ground, and that the outsourced drug tests had cost about $23,000. That cost figure covered the outsourced tests and was not presented as a full program cost including staff time and case processing, so the Tennessee figure cannot be placed beside the Florida reimbursement total as a measure of the same bundle of costs. The Department release contained no figure for benefits withheld and no statement of a net saving.

Design did not rescue the fiscal claim. Universal testing in Florida maximized the number of tests purchased and produced a positive rate of 2.6 percent. Screening in Utah and Tennessee reduced laboratory tests by an order of magnitude and reduced direct testing cost, but screening selected the testing pool through an instrument or through self reported answers, and the absolute yield fell to 12 positives in the first Utah year and 65 positives across eighteen Tennessee months. A program that tests fewer people identifies fewer people, and benefits withheld are a function of the number identified. In none of the three state records did the state publish a net saving for the period reported, and in Florida, the one state whose published figures permitted a direct statement of net position, the position was a loss.

The Constitutional Limit on Suspicionless Testing

Florida also encountered a legal limit that screening models were designed to avoid. Luis Lebron, an applicant who refused a test, challenged the statute under the Fourth Amendment to the Constitution of the United States. On October 24, 2011, Judge Mary S. Scriven of the United States District Court for the Middle District of Florida issued a preliminary injunction against enforcement, and on December 31, 2013 that court declared the statute unconstitutional. On December 3, 2014, a three judge panel of the United States Court of Appeals for the Eleventh Circuit affirmed in Lebron v. Secretary, Florida Department of Children and Families, 772 F.3d 1352. The Eleventh Circuit held that the state had not demonstrated a substantial special need sufficient to justify a suspicionless search of every applicant, had not established a demonstrable drug use problem among applicants that differed from the problem in the population at large, and had not shown a program well designed to identify or deter drug use that would affect employability, endanger children, or drain public funds. The Congressional Research Service recorded that the state did not seek rehearing by the full Eleventh Circuit and did not appeal to the Supreme Court of the United States. The verdict on the fifth claim follows. On the published state record, drug testing as operated in Florida, Utah, and Tennessee produced no saving, and Florida lost $45,780 in four months.

Claim Seven in Its Strongest Form

The seventh claim holds that fraud in public assistance programs is widespread, meaning that a large share of benefits is diverted from eligible households through deliberate deception. In its strongest form, the claim rested on three kinds of support. Organized trafficking prosecutions, in which a retailer exchanged benefits for cash over a long period and a single case reached into the millions of dollars, were offered as a sample of a larger hidden volume. The size of the Supplemental Nutrition Assistance Program was invoked, because a program issuing tens of billions of dollars in benefits in a year loses a large absolute sum to even a modest percentage of fraud, and the absolute sum was then presented without the percentage. The payment error rate was quoted as though every dollar in that rate had been obtained by fraud. Each support contains a fact. Organized cases existed, the program was large, and the error rate was published. The question is whether the agency series support widespread as a description of the share of benefits lost to deliberate deception.

Does the Trafficking Measure Show Widespread Fraud?

The Food and Nutrition Service estimated that trafficking diverted 1.0 percent of SNAP benefits from 2006 to 2008 and 1.3 percent from 2009 to 2011. Those shares corresponded to an estimated $330 million annually in the first period and $858 million annually in the second period.

Trafficking has a precise meaning in that series, and the governing rules sit inside the administration law of the food assistance program. The Food and Nutrition Service, the agency within the United States Department of Agriculture that administers the Supplemental Nutrition Assistance Program, defined trafficking as the exchange of benefits for cash, ordinarily at a discount, between a recipient and a food retailer. The study titled The Extent of Trafficking in the Supplemental Nutrition Assistance Program: 2006 to 2008, by Richard Mantovani and Hoke Wilson and published by the Food and Nutrition Service in March 2011, estimated $330 million diverted annually, about one cent of each benefit dollar. The successor study for 2009 to 2011, by Richard Mantovani, Eric Sean Williams, and Jacqueline Pflieger, prepared by ICF International for the Food and Nutrition Service and published in August 2013, estimated $858 million annually and a rate of 1.3 percent. The 2013 study attributed the larger dollar total in substantial part to growth in the program, because total redemptions rose from $36 billion in 2008 to $55 billion in 2009 and to $73 billion in 2011. Quoting $858 million without the 1.3 percent rate and without the redemption totals presents the same estimate in a form that sounds far larger.

The method and the unit of measurement matter as well. The Food and Nutrition Service built the estimate from investigations of retailers, including covert investigations and analysis of electronic benefit transfer records, and adjusted results from investigated stores, which had been selected for suspicious patterns, to reflect the full population of authorized stores. The studies produced a benefit trafficking rate of 1.0 percent for 2006 to 2008 and 1.3 percent for 2009 to 2011, and a separate store violation rate, meaning the estimated share of authorized stores that engaged in trafficking, of 8.2 percent for 2006 to 2008 and 10.5 percent for 2009 to 2011. Both rates can be accurate, because benefits were concentrated in large stores and trafficking was concentrated in small stores. The 2013 summary reported that small grocery and convenience stores accounted for about 15 percent of redemptions but 85 percent of trafficking redemptions during 2009 to 2011. The store rate weights a small store and a large store equally. For a claim about benefit dollars diverted, the benefit rate is the relevant measure. The 2013 summary also placed the benefit rate in a longer frame, stating that trafficking had declined from nearly 4 percent of benefits in the 1990s to the rates reported for the two study periods.

Error Is a Different Measure From Fraud

The payment error rate measures something else. The quality control system required each state agency to draw a random sample of cases each month and reexamine eligibility and benefit amount, after which the Food and Nutrition Service reviewed a subsample to validate the findings. Under Food and Nutrition Service Handbook 310, an error was a determination in which the action taken by the state agency differed from the findings of the quality control review, because a household was ineligible, because an allotment was incorrect, or because a decision to deny, terminate, or suspend benefits was incorrect. Intent was not an element of that determination. A case could enter the rate because an agency worker miscalculated a household expense, because a household failed to report a change in hours or wages by a deadline without any finding that the failure was deliberate, because verification arrived late, or because data entry produced a wrong allotment. A case could also enter the rate in the opposite direction, when an eligible household received too little.

For fiscal 2013, the Food and Nutrition Service quality control series reported a national overpayment error rate of 2.61 percent, a national underpayment error rate of 0.60 percent, and a combined national payment error rate of 3.20 percent, the lowest combined rate in that series extending back to fiscal 1981. The composition of the figure defeats a reading of the figure as a fraud rate. The underpayment component represented eligible households that received too little, a loss to the household rather than a gain obtained by deception. The overpayment component included benefits to ineligible households and excess benefits to eligible households, and an analysis of the same Food and Nutrition Service data by Dorothy Rosenbaum at the Center on Budget and Policy Priorities, published on July 2, 2014, calculated that less than 1 percent of benefits in fiscal 2013 went to households that were ineligible at all. The Food and Nutrition Service, in its public explanation of the quality control system, stated that improper payments identified through that system were usually the product of unintentional mistakes, giving as examples an agency staff member who incorrectly calculated a household expense and a recipient who did not know that a small change in an hourly wage had to be reported. Fraud required a separate determination of an intentional program violation, through an administrative disqualification hearing, a court process, or a signed waiver or consent, based on an intentional false or misleading statement or an intentional act constituting a violation, including trafficking. A quality control count built without an element of intent cannot be converted into a fraud count by renaming it.

The trafficking and quality control series are also Supplemental Nutrition Assistance Program series. Neither measures fraud in Temporary Assistance for Needy Families cash assistance, in Supplemental Security Income, or in Medicaid, and a nutrition program figure cannot be transferred to those programs as though a common rate had been measured across all of them.

What Fraud Occurred and at What Scale

Fraud occurred, and a verdict against the widespread claim does not require a claim that fraud was absent. Cases investigated by the United States Department of Agriculture Office of Inspector General in the period covered were concentrated in retailer trafficking and in organized schemes built around a store. In cases announced by that office in January 2013, a Minnesota store owner was sentenced in federal court on November 29, 2012 to 41 months in prison in a case in which the investigation found trafficking of Supplemental Nutrition Assistance Program benefits and Special Supplemental Food Program for Women, Infants, and Children benefits in excess of $3 million, and an Indiana store manager was sentenced on December 17, 2012 to 41 months in prison in a trafficking ring in which agents seized more than $500,000 in cash. A store redeeming benefits for many households over a long period can accumulate a total in the millions while the share of national benefits trafficked across all redemptions remains at the level the Food and Nutrition Service reported.

The semiannual report of the Office of Inspector General for April 1, 2014 to September 30, 2014 recorded that almost 60 percent of the investigative resources of that office were dedicated to the Supplemental Nutrition Assistance Program, that investigations in that program during the six month period led to 271 convictions, and that recoveries and restitutions totaled $44.9 million for the period. The account of that report recorded that most cases in the period involved grocers or convenience store owners who had exchanged benefits for cash or for ineligible items. A conviction total for a six month enforcement period is not a prevalence estimate, because it counts cases that investigators selected and brought to conclusion in that period, often for conduct from earlier years, and it supplies no denominator of recipients. As a record of enforcement output, the figure establishes that retailer fraud was real and that the federal government devoted a majority of the investigative resources of that office to the program. As a measure of how common fraud was among the tens of millions of people who received benefits in a year, it establishes no such thing.

Recipient fraud, retailer fraud, and administrative error remain distinct categories with different actors and different remedies. Recipient fraud included selling benefits to a retailer, concealing income or household composition, and other intentional violations. Retailer fraud included paying cash for benefits at a discount, redeeming benefits for ineligible items, and concealing trafficking through redemption patterns. Administrative error included the agency and household mistakes counted by the quality control system, most of which carried no finding of intent. Merging those categories into a single total produces a figure that no agency published, for a population that no study sampled, over a period that no series defined. The verdict on the seventh claim is false as stated. Trafficking was measured at 1.0 percent of benefits for 2006 to 2008 and 1.3 percent for 2009 to 2011. Payment error was measured at a combined 3.20 percent for fiscal 2013, composed of 2.61 percent overpayment and 0.60 percent underpayment, and Food and Nutrition Service documentation described that error as predominantly agency and household reporting error rather than fraud. Organized retailer schemes were prosecuted and produced individual totals in the millions and an enforcement record of 271 convictions in six months of 2014. Those facts describe fraud present at a measured rate of about one percent of benefits, concentrated in a minority of stores, and pursued through a large enforcement effort. They do not describe a large share of benefits diverted by deliberate deception.

The Partly True Claim About Block Grants and Better Service

The eighth claim examined in this article held that block grants let states serve people better. Stated in the strongest form that holders of the claim would recognize, the argument ran as follows. A state government stood closer than a federal agency to the labor markets, child care markets, transportation patterns, and family circumstances within its own borders. A fixed federal grant, spent under broad federal purposes rather than under a detailed federal formula, allowed that state to move money toward the obstacle that actually blocked employment in its communities. In one state the binding obstacle might have been the price or supply of child care. In another state the obstacle might have been transportation, training capacity, or case management for parents moving from assistance into paid work. Flexibility allowed a state to shift funds out of a single category of monthly cash payments and into child care assistance, work activities, and work supports that reached families in employment as well as families receiving cash aid. Flexibility also allowed a state to act without waiting for a federal formula to catch up with a local change, because the state controlled the allocation among permitted uses. On that account, better service did not mean a larger cash grant alone. Better service meant a package fitted to local conditions, assembled by the level of government closest to the conditions, and directed toward work and family stability rather than toward maintenance of a caseload.

The statutory design gave that argument a real foundation. Section 401 of the Social Security Act, codified at 42 U.S.C. 601, declared that the purpose of the Temporary Assistance for Needy Families program was to increase the flexibility of states in operating a program designed to provide assistance to needy families so that children could be cared for in their own homes or in the homes of relatives, to end dependence of needy parents on government benefits by promoting job preparation, work, and marriage, to prevent and reduce out of wedlock pregnancies, and to encourage the formation and maintenance of two parent families. The Personal Responsibility and Work Opportunity Reconciliation Act of 1996 replaced Aid to Families with Dependent Children, an open ended matching program, with a fixed block grant, a substitution examined directly in a comparison of the AFDC and TANF welfare models. Under the block grant enacted in 1996, the federal payment arrived in a set amount, states contributed maintenance of effort funds, and states chose the mix among basic assistance, child care, work activities, and other permitted uses that served the statutory purposes. The claim therefore described a genuine change in governing structure rather than a mere slogan. Whether that change served people better was a separate question, and the spending and reach records between 1996 and 2018 supplied evidence on both sides.

Did Flexibility Let States Serve People Better?

The record through fiscal 2018 gave a divided answer. States used flexibility to expand child care and work activities substantially during the early years, and spending in those categories rose sharply. States also used the same flexibility to reduce the share of funds devoted to basic cash assistance sharply.

That divided answer set the standard for the verdict. A verdict of wholly true would have required the expansion of services to stand alongside a reach and a cash assistance effort that held up across the period from 1996 to 2018. A verdict of wholly false would have required the service expansion itself to be illusory, marginal, or unrelated to the block grant structure. Neither verdict fitted the documented record. The Center on Budget and Policy Priorities documented a more than fivefold increase in child care spending during the first years of the program and a substantial rise in spending on work related activities over the same span. The Administration for Children and Families documented, for fiscal 2018, a distribution in which basic assistance accounted for about one fifth of combined funds, far below the share recorded for fiscal 1997. The Congressional Research Service documented a fixed nominal grant whose inflation adjusted value fell by more than one third between fiscal 1997 and fiscal 2018. Taken together, the findings supported a verdict of partly true and partly contradicted, with each half stated at full strength.

The Supported Half, in Which Flexibility Expanded Child Care and Work Support

The clearest support for the claim came from child care. The Center on Budget and Policy Priorities, in an analysis titled How States Have Spent Federal and State Funds Under the TANF Block Grant, reported that spending on child care from federal Temporary Assistance for Needy Families funds and state maintenance of effort funds rose from 1.1 billion dollars in fiscal 1997 to 5.9 billion dollars in fiscal 2000. The increase was more than fivefold in three years, and fiscal 1997 was a phase in year, so the comparison captured the moment at which states first exercised the new allocation authority. The share of combined funds devoted to child care moved in the same direction. The Center on Budget and Policy Priorities reported that child care accounted for 5.4 percent of combined funds in fiscal 1997 and 21 percent in fiscal 2000. A shift of that size represented a deliberate reallocation rather than a bookkeeping drift. States directed a larger portion of a fixed pool toward a service that allowed a parent to accept a job, keep a job, attend training, or comply with a work activity assignment. For a parent whose employment depended on care for a child during working hours, child care assistance addressed a concrete barrier that a cash payment did not address in the same direct way.

Transfers between block grants formed a second channel for the same expansion, and the Congressional Research Service documented that channel in a report titled Child Care: Funding and Spending Under Federal Block Grants. Between fiscal 1997 and fiscal 2000, states transferred a total of 6 billion dollars from the Temporary Assistance for Needy Families block grant to the Child Care and Development Fund, more than 9 percent of the allotments awarded over that four year period. In fiscal 2000 alone, states transferred 2.4 billion dollars, equal to 14 percent of the fiscal 2000 allotment. The same report recorded that, in fiscal 2000, states also spent 2.2 billion dollars of federal funds and related state funds directly on child care, in addition to the transfers. The two channels mattered because each reflected a distinct use of flexibility. A transfer placed money inside the child care system and under the rules of that system. Direct spending kept the money inside Temporary Assistance for Needy Families accounts while applying the money to the same practical need. A federal formula confined to cash assistance would have permitted neither move at that scale.

Work related spending supplied the third element of the supported half. The Center on Budget and Policy Priorities reported that investment in work related activities rose from 724 million dollars in fiscal 1997 to 1.9 billion dollars in fiscal 1999, and then reached 3.3 billion dollars in each of fiscal 2001 and fiscal 2002. The category included education and training, subsidized employment, and other work activities, together with supports connected to employment, and transportation assistance formed one example of a support that a state could fund when distance blocked a job placement. The Administration for Children and Families financial data for fiscal 2018 showed that work, education, and training activities accounted for 10.7 percent of combined federal and state funds in that year, while child care accounted for 17.0 percent of a combined total of 31.3 billion dollars. That distribution demonstrated that the categories states had expanded in the early years occupied a substantial place in program spending in fiscal 2018, even though the balance among categories had changed. Flexibility had produced a program in which services connected to employment stood beside cash assistance as a major use of funds, rather than as a minor supplement.

A fourth element concerned the range of families that service spending could reach. The Center on Budget and Policy Priorities explained that states did not have to limit child care assistance financed from Temporary Assistance for Needy Families funds and maintenance of effort funds to families that received cash assistance. States could use the funds for families that had left assistance for work and for other families with low incomes and paid employment. That breadth answered a specific weakness attributed to the predecessor structure, under which child care help tied closely to receipt of cash aid could fall away at the point when employment made care most necessary. Under the block grant, a state could continue child care help across the transition into a job and could extend help to a working family that had never entered the cash caseload. The Administration for Children and Families reported that, in fiscal 2018, child care was the largest single spending category in sixteen states, basic assistance was the largest category in twelve states and the District of Columbia, and child welfare services formed the largest category in six states. That variation offered direct evidence of tailoring in the limited sense that states made visibly different allocation choices under a common federal grant. The supported half rested on that combination of documented expansion, transfers, work spending, and variation among states.

The Contradicted Half, in Which Reach Collapsed and the Grant Shrank

The contradicted half began with reach. The Center on Budget and Policy Priorities measured reach through the TANF to poverty ratio, which compared the number of families receiving cash assistance with the number of families with children living in poverty. In 1996, the final year of Aid to Families with Dependent Children, 68 families received assistance for every 100 families in poverty. In 2015, 23 families received assistance for every 100 families in poverty. In 2018, 22 families received assistance for every 100 families in poverty. The decline from 68 to 22 meant that the system of 2018 reached about one third of the share of poor families that the system of 1996 had reached. The same series recorded a ratio of 41 in 2001, 31 in 2006, and 27 in 2010, before the figures of 23 in 2015 and 22 in 2018, and each step rested on poverty data from the Current Population Survey and on caseload data from the Department of Health and Human Services and from state agencies. A program that served people better in the aggregate would have had to reconcile expanded services with a cash assistance reach that fell across that long interval, through periods of economic expansion and recession alike.

The spending distribution told a parallel story. The Center on Budget and Policy Priorities reported that states spent 14 billion dollars, or 71 percent of combined federal and state funds, on basic assistance in fiscal 1997. The Administration for Children and Families, in financial data released on August 7, 2019, reported that basic assistance accounted for 21.4 percent of combined federal Temporary Assistance for Needy Families funds and state maintenance of effort funds in fiscal 2018. The fall from 71 percent to 21.4 percent amounted to a decline of nearly fifty percentage points in the share of the combined pool devoted to monthly cash aid. The Administration for Children and Families also reported that the amount used for basic assistance fell by 337 million dollars between fiscal 2017 and fiscal 2018, while the amount used for child care rose by 309 million dollars and the amount used for work, education, and training activities rose by 46 million dollars. Those movements illustrated the mechanism at the center of the contradicted half. Within a fixed pool, an increase in one category and a decrease in basic assistance could occur at the same time, and the interval from fiscal 1997 to fiscal 2018 showed a sustained reallocation away from cash aid. Flexibility explained how the reallocation happened, without establishing that the resulting distribution served poor families better than the distribution of fiscal 1997.

The fixed character of the grant compounded the effect of the reallocation. The Congressional Research Service, in tables for the report titled The Temporary Assistance for Needy Families Block Grant: Responses to Frequently Asked Questions, updated on July 1, 2019, recorded the block grant in nominal terms across the span from fiscal 1997 to fiscal 2018. Expressed in constant fiscal 1997 dollars adjusted with the Consumer Price Index for All Urban Consumers, the fiscal 2018 grant equaled 10.564 billion dollars. The cumulative change came to a decline of 36.2 percent between fiscal 1997 and fiscal 2018. The statute had set the grant without an adjustment for inflation, for the size of the caseload, or for changes in the population in poverty, and the Congressional Research Service account recorded that the amount remained unchanged in nominal terms from fiscal 1997 through fiscal 2016. A state in fiscal 2018 therefore exercised flexibility over resources worth more than one third less, in real terms, than the resources available in fiscal 1997. Tailoring a smaller pool could improve the fit between spending and local conditions at the margin, and the child care record showed that states used the early years of the grant to do exactly that. Tailoring could not restore purchasing power that inflation had removed.

State incentives pointed in the same direction, and one statutory mechanism illustrated the direction with particular clarity. Section 407(b)(3) of the Social Security Act, codified at 42 U.S.C. 607(b)(3), provided the caseload reduction credit. Under that provision, the minimum work participation rate that a state had to meet fell by the percentage points by which the state caseload had fallen, subject to rules that excluded reductions required by federal law and reductions that resulted from changes in state eligibility criteria. The Congressional Research Service summarized the effect as a reduction of the participation standard by one percentage point for each percentage point decline in a state caseload. A state could therefore move closer to compliance with the federal work standard through a smaller caseload, independently of the share of the remaining caseload engaged in work activities. The credit did not order any state to reduce a caseload. The credit did attach a compliance benefit to caseload decline as such, within a structure in which funds freed by a smaller caseload remained available for other permitted uses. Read beside the TANF to poverty ratio and the basic assistance shares, the credit helped explain why flexibility and contraction of cash assistance advanced together in the record. The contradicted half rested on that combination of a collapsed ratio, a reduced basic assistance share, a grant that lost 36.2 percent of real value, and a mechanism that rewarded caseload reduction.

A Counter Reading That Concedes What the Record Shows

A distinct counter reading applied to the whole field, and the counter reading cut against any tidy debunking of claims about the 1996 law, including a tidy debunking of the block grant claim itself. The first concession concerned scale. Cash assistance shrank dramatically between 1996 and 2018, and no interpretation of the service expansion altered that fact. The TANF to poverty ratio recorded by the Center on Budget and Policy Priorities moved from 68 families receiving assistance for every 100 families in poverty in 1996 to 22 in 2018. Basic assistance moved from 71 percent of combined federal and state funds in fiscal 1997 to 21.4 percent in fiscal 2018 in the data released by the Administration for Children and Families. Those two measures, drawn from different sources, described the same contraction from different angles, one through families reached against families in poverty and the other through dollars assigned to cash aid against dollars assigned to every permitted use. An assessment that treated the contraction as a minor side effect, or as a change fully offset by services of equivalent value to every family that lost cash aid, would depart from both measures at once. The contraction belonged at the center of any honest account of the period from 1996 to 2018.

The second concession concerned deep poverty research. Kathryn Edin and Luke Shaefer reported a finding of households with children living on cash incomes at the level of two dollars per person per day, and the finding formed part of the published research record on poverty after the 1996 law. The magnitude of that finding was contested in scholarly and policy debate during the years after publication, and the full examination of the dispute belongs to a later section of this article. For purposes of the counter reading, two statements held at the same time. The research existed, rested on named investigators, survey data, and fieldwork, and entered the record as a serious finding rather than as a rhetorical invention. The size of the population described by the research remained a matter of dispute among researchers who examined the underlying data and methods. A debunking that waved the finding away entirely would misrepresent the record, and a debunking that treated the contested magnitude as settled would also misrepresent the record.

The third concession concerned state spending as a matter of documented record. State spending shifted away from cash aid between fiscal 1997 and fiscal 2018, and the shift appeared in the official financial data that states reported to the Administration for Children and Families and in the analyses that the Center on Budget and Policy Priorities built from federal data. The shift was not an inference drawn from anecdotes about individual state programs. The Administration for Children and Families data for fiscal 2018 placed basic assistance at 21.4 percent of combined funds, within a combined total of 31.3 billion dollars, and the Center on Budget and Policy Priorities placed basic assistance at 71 percent of combined funds in fiscal 1997. Between those two dated observations lay a reallocation that touched every state, though the degree differed sharply among states. Acknowledging the reallocation did not settle whether the new distribution helped or harmed any particular family. Acknowledging the reallocation did settle that the distribution changed fundamentally, and that any verdict on the block grant claim had to reckon with the change as a fact rather than as an interpretation open to dismissal.

Those concessions carried a consequence for the verdict on the eighth claim. An assessment that conceded the dramatic shrinkage of cash assistance, the existence and the contested magnitude of the deep poverty finding reported by Edin and Shaefer, and the documented shift of state spending away from cash aid kept faith with readers who knew those points from the public record of the period ending on December 15, 2019. Such an assessment could defend the supported half without strain, because the child care increase from 1.1 billion dollars in fiscal 1997 to 5.9 billion dollars in fiscal 2000 and the rise in work related spending from 724 million dollars in fiscal 1997 to 3.3 billion dollars in fiscal 2001 and fiscal 2002 rested on named sources and dated figures of the same standing as the figures for contraction. The claim was partly true, because flexibility funded real expansions of child care and work support that reached families beyond the cash caseload. The claim was partly contradicted, because the same flexibility operated inside a fixed and shrinking grant, alongside a caseload reduction credit that rewarded smaller caseloads, while the share of poor families receiving any cash assistance fell from 68 for every 100 families in poverty in 1996 to 22 in 2018. Both halves belonged in the verdict, at equal strength, because both halves rested on the documented record.

Checking the Program and the Denominator

The analytical framework of this article can be stated in one sentence. Check the program and the denominator. Nearly every false or misleading claim examined in the preceding sections failed one of two tests, and most failed both. The first test asks which program a speaker is describing. Cash assistance under Temporary Assistance for Needy Families was one program among several, and after 1996 it was a shrinking one. A claim about cash assistance could be accurate for that program and false for the safety net as a whole, because the Supplemental Nutrition Assistance Program, Medicaid, the earned income tax credit, and the child tax credit moved on different schedules, under different rules, and in several periods in the opposite direction. The second test asks what quantity a figure has been divided by, or what population a count describes. A caseload count, a share of families in poverty, a share of benefit dollars, and a share of total federal spending are four different quantities, each answering a different question. A claim that moved between them without naming the move converted a correct figure into an incorrect conclusion, and the conversion was where the error lived.

The Program Test

The clearest program failure involved the size of cash assistance itself. The Congressional Research Service, in funding tables updated on July 1, 2019, recorded 16.489 billion dollars in state family assistance grants for the Temporary Assistance for Needy Families block grant. Quoted alone, that figure invited two opposite misreadings. Read as the cost of the safety net, it understated federal support by an order of magnitude, because it excluded nutrition assistance, health coverage, and refundable tax credits entirely. Read as a large sum in isolation, it overstated the weight of cash assistance in the federal budget. The Office of Management and Budget historical tables reported total federal outlays of 4,109 billion dollars in fiscal year 2018. Divided by that denominator, the state family assistance grants equaled about four tenths of one percent of federal spending. The claim became true or false at the moment a speaker specified, or failed to specify, which quantity the 16.489 billion dollars was being compared against.

A second program failure treated a fall in one caseload as a fall in assistance generally. The Congressional Research Service, drawing on Department of Health and Human Services data, recorded 14.161 million recipients of Aid to Families with Dependent Children and its successor program in 1994, the peak year of that series, and 6.143 million recipients in 2000, a decline of about 56.6 percent in six years. During a later and overlapping period, the Food and Nutrition Service recorded average monthly Supplemental Nutrition Assistance Program participation of 17.194 million persons in fiscal year 2000 and 44.709 million persons in fiscal year 2011, an increase of about 160 percent, driven in part by the recession that began in December 2007. A speaker who cited the first series as evidence that assistance had contracted, without mentioning the second, had described one program and implied a conclusion about all programs. A speaker who cited the second series as evidence that the 1996 law had changed nothing had made the mirror image error, because growth in nutrition assistance said nothing by itself about the reach of cash assistance, the component the 1996 law had restructured.

The earned income tax credit and the child tax credit produced a third version of the same failure, because both were delivered through the tax system rather than through a welfare office, and both were therefore easy to omit from a count described as welfare. A measure of support that counted cash assistance and nutrition assistance but excluded refundable credits understated the resources reaching households with earnings, and a measure of work incentives that counted the credits but excluded benefit reductions inside the other programs misstated the combined schedule a household faced. The program test required only that a claim name its programs, so that a reader could tell whether a change reflected household circumstances or a change in the list being counted.

The Denominator Test

The caseload series illustrated the denominator failure in its purest form. The fall from 14.161 million recipients in 1994 to 6.143 million recipients in 2000 was a statement about a numerator, the number of people receiving cash assistance, with no denominator attached. It could not by itself distinguish a fall caused by rising employment and falling poverty from a fall caused by narrower eligibility or stricter administration, because each cause produced the same numerator. The Center on Budget and Policy Priorities supplied the missing denominator by constructing a ratio from Current Population Survey poverty data and Department of Health and Human Services caseload data. On that measure, 68 families received cash assistance for every 100 families with children in poverty in 1996, and 22 families received cash assistance for every 100 families with children in poverty in 2018. The ratio showed that the count fell faster than the population in poverty across the longer period from 1996 to 2018, so that the reach of the program per family in poverty contracted to about one third of the 1996 level.

Benefit trafficking in the Supplemental Nutrition Assistance Program supplied the sharpest denominator contrast in the record. The Food and Nutrition Service estimated that trafficking, defined as the exchange of benefits for cash, accounted for 1.3 percent of benefits during the period from 2009 to 2011, down from about 4 percent in the 1990s and close to the approximately 1 percent figure the Department of Agriculture cited for the longer decline. The same department reported a national payment error rate of 6.8 percent for fiscal year 2018, up from 6.3 percent in fiscal year 2017. The trafficking rate divided an estimated dollar value of benefits exchanged for cash by total benefits, and it described intentional conduct investigated through transaction analysis and undercover work. The payment error rate divided overpayments and underpayments found in quality control case reviews by total payments, and it described the accuracy with which state agencies determined eligibility and benefit amounts, including payments that were too small as well as payments that were too large. A claim that fraud ran at 6.8 percent had attached the second figure to the first concept. A claim that only 1.3 percent of payments were wrong in any respect had made the reverse substitution. Each percentage was accurate only for the numerator and denominator its source had defined.

How Sourcing Errors Travel

Errors of these kinds replicated because each republication stripped away a qualifier. A figure that began life inside a table, with a year, a population, and a definition attached, lost those qualifiers as republication moved it from speech to news report to later account. Four patterns accounted for most of the propagation in the record assembled for this article, and each pattern had a named instance a reader could check against its source.

The first pattern paired figures from different years as if they described a single moment. The record itself showed the hazard. The most widely cited trafficking estimate covered the period from 2009 to 2011, while the payment error rate most often placed beside it covered fiscal year 2018, and the caseload peak belonged to 1994 while the trough comparisons belonged to 2000. A speaker who assembled those figures into one portrait had combined three different economies, three different administrative regimes, and, in the caseload case, two different statutes, since the 1994 figure described Aid to Families with Dependent Children and the later figures described its successor.

The second pattern reported an error rate as a fraud rate. The Department of Agriculture anticipated the substitution in its release of the fiscal year 2018 payment error rate, in which department officials stated expressly that the rate was not a measure of fraud but a representation of how accurately states determined eligibility and benefit amounts. The Congressional Research Service made the same separation in its September 28, 2018 report on errors and fraud in the Supplemental Nutrition Assistance Program, which treated recipient error, agency error, and intentional program violations as distinct categories with distinct remedies. The substitution persisted because the error rate was published annually, by state, with a precision that fraud estimates never carried, and because a single percentage was easier to repeat than a definition.

The third pattern quoted a nominal dollar figure across two decades without the price level. The Temporary Assistance for Needy Families block grant was the standing instance. The Congressional Research Service recorded that the basic block grant amount established in 1996 remained unchanged in nominal terms from fiscal year 1997 through fiscal year 2016, and that its inflation adjusted value fell by 33.1 percent over that period. The Congressional Research Service further calculated that the grant in fiscal year 2018 stood 36.2 percent below its fiscal year 1997 value in inflation adjusted terms. An account that reported the grant as unchanged between 1996 and 2018 was therefore accurate about the statute and inaccurate about the resource, because the same nominal sum purchased about one third less at the end of the period.

The fourth pattern reported a screening stage as if it were a final result, and the state drug testing programs of the 2010s supplied a precise instance. Tennessee began administering a screening questionnaire to applicants for its Families First program on July 1, 2014. The figures in this instance describe the same Tennessee program examined in the drug testing section of this article, at an earlier reporting point, and the two sets of figures cover different reporting windows rather than conflicting counts of a single window. In the first year of operation, according to figures the Tennessee Department of Human Services provided and the Tennessean and ThinkProgress reported in October 2015, 28,559 people applied, 468 applicants answered a screening question in a way that required a urine test, and 55 applicants tested positive. The positive rate was 11.7 percent when the denominator was the 468 people tested and 0.19 percent when the denominator was the 28,559 people who applied. Reports that carried the first percentage as the share of applicants who used drugs, or that carried the number screened as the number who failed, had collapsed a three stage process into one stage. Florida supplied the contrast case, because Florida tested applicants directly rather than after a questionnaire. Between July and October 2011, according to Florida Department of Children and Families figures, 4,086 applicants were tested and 108 tested positive, a rate of 2.6 percent of those tested, in a program where those tested and those applying were nearly the same group.

Did Deep Poverty Rise After Welfare Reform, and by How Much?

The question is contested, and the record through 2019 did not settle it. Edin and Shaefer reported a large rise in cash income poverty at the two dollar threshold. Meyer, Sullivan, and Winship, using consumption data, administrative corrections, and broader resource definitions, reported far smaller levels and different trends.

Kathryn Edin and H. Luke Shaefer built their case in two stages. In a paper published in Social Service Review in 2013, Shaefer and Edin used the Survey of Income and Program Participation, the Census Bureau survey designed to track income and program receipt month by month, and measured households with children whose reported cash income fell at or below two dollars per person per day, a threshold adapted from the World Bank measure of extreme poverty in developing economies. They calculated that the share of nonelderly households with children in that condition rose from 0.8 percent in 1996 to 2.7 percent in 2011 on a quarterly income basis, and that the number of such households grew by 159 percent between 1996 and 2011. In their 2015 book, $2.00 a Day: Living on Almost Nothing in America, Edin and Shaefer reported that in early 2011 about 1.5 million households, including about 3 million children, lived on cash incomes at or below that threshold in a given month. The book paired the survey estimates with fieldwork in Chicago, Cleveland, and rural communities in Appalachia and the Mississippi Delta, documenting spells without cash income from earnings or cash assistance and expenses that required cash, including rent, utilities, and transport.

Edin and Shaefer stated their case with its strongest supporting evidence included rather than omitted. Their calculations counted cash from every household member and from private sources as well as public programs. In calculations that counted Supplemental Nutrition Assistance Program benefits as income, the measured level of extreme poverty fell and the measured rise became smaller, and Edin and Shaefer reported that version as well. Winship, in a 2016 report discussed below, summarized their benefit inclusive results as a rise in extreme poverty among children from 1.4 percent in 1996 to 1.6 percent in 2012. In a 2016 response to critics, Shaefer and Edin also pointed outside the survey record to administrative data from the nutrition program itself, in which the number of households receiving benefits and reporting no cash income had quadrupled since 1996. Their interpretation connected the trend to the structure enacted in 1996. Cash assistance had become harder to obtain and easier to lose, the block grant had not grown with prices or with the population in poverty, and the programs that expanded after 1996 delivered food assistance and tax credits rather than cash, leaving a household without earnings exposed in the specific dimension, cash income, that their measure tracked.

The objection, stated at full strength, began with the survey instrument rather than with the threshold. Bruce D. Meyer of the University of Chicago and James X. Sullivan of the University of Notre Dame argued across a series of papers, including work published in 2011 and 2012, that consumption was a more reliable indicator of material circumstances at the bottom of the distribution than reported income, because households drew on savings, credit, durable goods, and assistance that income questions captured poorly. On their consumption based measures, material circumstances at the bottom improved after the mid 1990s and deep consumption poverty fell. Meyer, Wallace K. C. Mok, and Sullivan supplied the measurement foundation for that objection in Household Surveys in Crisis, published in the Journal of Economic Perspectives in 2015. Comparing survey reports with administrative totals for nine transfer programs, they found that in the years they examined, more than half of welfare dollars and nearly half of food stamp dollars were missed in several major surveys, that the missed amounts took the form of recipients recorded as nonrecipients, and that the underreporting had grown over time. If surveys missed a rising share of transfer income, then a rising share of households would appear to live on almost no cash income even when benefit receipt had not changed.

Scott Winship of the Manhattan Institute extended the objection in Poverty After Welfare Reform, published in 2016, by rebuilding the income measure itself. Winship added noncash benefits and tax credits to reported income in the Current Population Survey, corrected for underreporting of transfer income using the Transfer Income Model developed at the Urban Institute, and adjusted prices with the personal consumption expenditures deflator used by the Congressional Budget Office and the Federal Reserve rather than the consumer price index used in the official measure. On his adjusted measure, child poverty fell from 13.1 percent in 1996 to 7.8 percent in 2014. For extreme poverty specifically, Winship found levels far below the Edin and Shaefer estimates and argued that genuine two dollar a day poverty among children was so rare in his adjusted data that the available surveys could not reliably establish whether its prevalence had risen or fallen. The dispute then continued on the choice of yardstick itself. In a 2018 working paper for Poverty Solutions at the University of Michigan, H. Luke Shaefer and Joshua Rivera compared the income based and consumption based poverty series against material hardship indicators, including food insecurity rates and hardship measures from the Survey of Income and Program Participation, and argued that the income based trends tracked those hardship indicators more closely than the Meyer and Sullivan consumption series, whose level of poverty they argued was implausibly low against the hardship record. Meyer and Sullivan answered Shaefer and Rivera in a 2018 comment, arguing that the correlations at the center of the comparison suffered from a spurious regression problem, that correcting the problem reversed the central findings, and that the broader body of evidence supported consumption as the more accurate reflection of circumstances at the bottom.

The persistence of the dispute had a technical explanation that both sides acknowledged. The competing measures did not measure the same quantity. Cash income, income augmented by noncash benefits and tax credits, and consumption were three different concepts, and a household could register near zero on the first while registering materially higher on the second and third, without any measurement error at all, because food assistance could not pay a rent bill and because consumption could be financed from resources an income question did not reach. Layered over that conceptual difference sat a measurement problem concentrated exactly where the dispute lived. Income at the very bottom of the distribution was the quantity household surveys measured worst, because the amounts were small, irregular, and drawn heavily from the transfer programs whose underreporting Meyer, Mok, and Sullivan documented, and because a reporting error of a few hundred dollars a month moved a household across the two dollar threshold while leaving a household in the middle of the distribution in the same statistical position. This article therefore records the deep poverty question as contested and unresolved on the evidence available through December 15, 2019. Edin and Shaefer established that reported cash income at the bottom of the Survey of Income and Program Participation fell to levels their threshold captured, and that the count on that basis rose sharply after 1996. Meyer, Sullivan, and Winship established that broader resource measures and consumption measures produced far lower levels and materially different trends, and that survey underreporting was large enough to account for a substantial part of the gap. Which of those findings described the material circumstances of the households in question was, on the record through 2019, a question the available instruments could pose more precisely than they could answer.

The Claim Ledger

Claim	Verdict	Deciding source	Full treatment in this series
Welfare reform ended welfare	False	Public Law 104-193 text and the CRS and HHS caseload series	welfare-reform-act-1996-guide
Recipients generally refuse to work	False	ACF characteristics data for fiscal 2018 and 42 U.S.C. 607(b)(3)	welfare-reform-1996-outcomes
The five year limit is absolute	False	42 U.S.C. 608(a)(7)	welfare-reform-act-1996-guide
Welfare is a large share of federal spending	False	OMB Historical Tables for fiscal 2018 and 42 U.S.C. 603(a)(1)	tanf-block-grant-structure
Drug testing recipients saves money	False	Florida Department of Children and Families results for 2011 and Lebron v. Secretary, 772 F.3d 1352	welfare-reform-act-1996-guide
People immigrate to obtain benefits	Contested	8 U.S.C. 1613 and the named research literature	welfare-reform-act-1996-guide
Program fraud is widespread	False	USDA Food and Nutrition Service trafficking studies for 2006 to 2008 and 2009 to 2011	snap-administration-law
Block grants let states serve people better	Partly true	ACF financial data for fiscal 2018 and the CBPP coverage series	afdc-vs-tanf-welfare-models

What a Careful Reader Should Believe

A careful reader should believe the eight verdicts at the strength the evidence earned, and no stronger. Six claims failed as stated. Welfare did not end, because cash assistance operated without interruption and the programs around cash assistance grew. Recipients did not generally refuse to work, because most recipients were children, most families contained no adult recipient, and leavers worked at high incidence at low wages. The five year limit was not absolute, because the statute that created the limit also created a hardship exemption, disregarded months, and a state funded route past the limit. Cash assistance was not a large share of federal spending, because 16.489 billion dollars in state family assistance grants stood against 4,109 billion dollars in total federal outlays for fiscal 2018. Drug testing did not save money, because Florida recorded a net loss and no state account in the record reported a saving. Program fraud was not widespread, because trafficking diverted about one percent of food assistance benefits in the measured periods and the larger error rate counted mistakes without any element of intent. One claim, the migration motive claim, stayed contested, because Borjas, Zavodny, Kaushal, Camarota, Nowrasteh, and Ku worked from identified data and reached opposed conclusions. One claim, the block grant claim, split in half, because flexibility funded real expansions of child care and work support while reach collapsed inside a shrinking grant.

The two-check test explains why the verdicts landed where the verdicts landed, and the test travels beyond the eight claims in the ledger. Check the program: a statement about Temporary Assistance for Needy Families describes one program, and a statement about the safety net describes several programs that moved on different schedules under different rules. Check the denominator: a count describes a numerator, and a rate describes a numerator divided by a named population, and confusing the two manufactured most of the errors examined in this article. A trafficking dollar total sounded large until the total was divided by all benefits redeemed. A payment error rate sounded like fraud until the definition separated intent from arithmetic mistakes. A block grant described as unchanged sounded stable until prices entered the calculation and a third of the value disappeared. Each correction required no new ideology and no new dataset. Each correction required the source, the population, and the period to stay attached to the figure during republication.

Three concessions cut against tidy debunking, and a careful reader should hold the concessions beside the verdicts rather than beneath the verdicts. Cash assistance shrank dramatically, by roughly three quarters on both the family measure and the recipient measure between 1994 and 2018, and the shrinkage was a transformation of the safety net rather than a bookkeeping change. State spending shifted away from cash aid, from 71 percent of combined funds devoted to basic assistance in fiscal 1997 to 21.4 percent in fiscal 2018, and that reallocation touched families that services did not reach in the same way cash reached families. The deep poverty dispute stayed unresolved, because Edin and Shaefer measured a large rise in households living on cash incomes at the two dollar threshold while Meyer, Sullivan, and Winship, using consumption data and broader resource definitions, measured far smaller levels and different trends, and the instruments available through 2019 could pose that question more precisely than the instruments could answer the question. Debunking that erased those three facts would commit the same offense as the myths, by selecting the half of the record that fitted a verdict.

The route onward runs through the sources rather than around the sources. The statutory questions resolve in Title IV-A of the Social Security Act and in Title 8 for noncitizen eligibility. The spending questions resolve in the financial data published by the Administration for Children and Families and in the historical tables published by the Office of Management and Budget. The research questions resolve in the named studies, read with the population and method of each study kept in view. A reader assembling that material for study can keep your statute notes, citations, and case chronologies together free on VaultBook. The claims will keep circulating, because each myth carries a figure that is accurate somewhere inside a table. The defense against circulation is the discipline applied throughout this article: name the program, name the denominator, name the source, and let the verdict follow the record rather than the slogan.

Frequently Asked Questions

Q: Is it true welfare reform ended welfare?

No. The Personal Responsibility and Work Opportunity Reconciliation Act of 1996, Public Law 104-193, ended Aid to Families with Dependent Children as an individual entitlement and replaced that program with the Temporary Assistance for Needy Families block grant, but cash assistance operated without interruption afterward. The Congressional Research Service series based on Department of Health and Human Services data recorded 1.196 million families and 3.150 million recipients receiving assistance in calendar 2018. Food assistance, Medicaid, Supplemental Security Income, and the earned income tax credit also operated throughout the period, and several of those programs grew. A contraction of cash assistance to roughly one quarter of the 1994 caseload is a transformation, not an ending.

Q: Do most welfare reform recipients refuse to work?

No. The Administration for Children and Families characteristics data for fiscal 2018 recorded about 1.8 million children among roughly 2.3 million individual recipients, and child only families, in which no adult received assistance, accounted for 54.5 percent of families. Among adult recipients, 27.3 percent were employed in an average month of fiscal 2018. The Government Accountability Office analysis of the 1999 Survey of Income and Program Participation found that 44 percent of adult recipients reported a physical or mental impairment. The leaver synthesis prepared by Acs and Loprest in 2001 found that about three quarters of families that left assistance worked at some point in the year after exit. Nonwork in the data tracked health, caregiving, and education measures rather than a measured refusal.

Q: Is the five year welfare reform limit absolute?

No. Section 408(a)(7) of the Social Security Act, codified at 42 U.S.C. 608(a)(7), prohibits the use of federal funds for a family containing an adult recipient after sixty cumulative months, and the same provision creates the routes around an absolute reading. A hardship exemption at 42 U.S.C. 608(a)(7)(C) permits federally funded extension for up to twenty percent of the average monthly caseload. Child only cases, in which no adult receives assistance, fall outside the clock. Months lived in Indian country or an Alaskan Native village meeting the statutory employment test are disregarded under 42 U.S.C. 608(a)(7)(D). A state may also continue assistance past sixty months with state funds kept apart from federal funds.

Q: Is welfare reform spending a big share of the budget?

For cash assistance, no. The state family assistance grants under 42 U.S.C. 603(a)(1) totaled 16.489 billion dollars in fiscal 2018. The Office of Management and Budget historical tables recorded total federal outlays of 4,109 billion dollars for fiscal 2018, so the grant equaled about four tenths of one percent of federal spending. Combined federal Temporary Assistance for Needy Families expenditures and state maintenance of effort expenditures totaled 31.3 billion dollars in fiscal 2018 in Administration for Children and Families data, equal to about 0.76 percent of total federal outlays, and that combined figure includes state dollars. A large share appears only when welfare is redefined as all means tested spending, a category dominated by Medicaid at 389 billion dollars in fiscal 2018 in the Congressional Budget Office series.

Q: Does drug testing save money under welfare reform?

No state record in the published evidence reported a saving. Florida tested 4,086 applicants between July and October 2011 under the statute later invalidated in Lebron v. Secretary, Florida Department of Children and Families, 772 F.3d 1352, and 108 applicants tested positive, a rate of 2.6 percent. The Florida Department of Children and Families figures put reimbursements to applicants who passed at 118,140 dollars and the net loss at 45,780 dollars for the four months of operation. Utah screened 4,730 applicants in the first year of the screening model reported in August 2013, required 466 tests, recorded 12 positives, and spent more than 30,000 dollars on screening and tests combined. Tennessee recorded 65 positives among 39,121 applicants in the first eighteen months reported in February 2016. Low positive rates left too few withholdings to cover testing costs.

Q: Did welfare reform stop immigrants from coming?

The verdict is contested. The statute settled access rather than motive. Under 8 U.S.C. 1613, a qualified alien who entered on or after August 22, 1996 was ineligible for a federal means tested public benefit during the first five years after entry in qualified status, subject to exceptions for refugees, asylees, veterans, and residents credited with forty qualifying quarters, and subject to partial restorations enacted in 1997, 1998, 2002, and 2009. The motive question divided named researchers. George Borjas, in the Journal of Labor Economics in 1999, reported clustering of immigrant households in higher benefit states consistent with a welfare magnet. Madeline Zavodny in 1997 and Neeraj Kaushal, in the Journal of Labor Economics in 2005, found that networks and labor market conditions explained location choices while benefit levels exerted no predicted pull.

Q: Do block grants let states serve people better under welfare reform?

Partly. Flexibility under the block grant enacted at 42 U.S.C. 601 and following funded real service expansions. The Center on Budget and Policy Priorities reported child care spending from federal and state funds rising from 1.1 billion dollars in fiscal 1997 to 5.9 billion dollars in fiscal 2000, and work related spending rising from 724 million dollars in fiscal 1997 to 3.3 billion dollars in fiscal 2001 and fiscal 2002. The contradicted half concerns reach and resources. Administration for Children and Families financial data for fiscal 2018 placed basic assistance at 21.4 percent of combined funds, against 71 percent in fiscal 1997. The Center on Budget and Policy Priorities TANF to poverty ratio fell from 68 families receiving assistance for every 100 families in poverty in 1996 to 22 in 2018.

Q: Is food assistance fraud common after welfare reform?

No, on the agency measure designed to answer that question. The Food and Nutrition Service defined trafficking as the exchange of Supplemental Nutrition Assistance Program benefits for cash and estimated trafficking at 1.0 percent of benefits for 2006 to 2008 and 1.3 percent of benefits for 2009 to 2011. The dollar estimates, 330 million dollars annually in the first period and 858 million dollars annually in the second, grew mainly because total redemptions grew. Fraud in that series meant deliberate deception measured through retailer investigations. The separate payment error rate, 3.20 percent combined for fiscal 2013, counted agency and household mistakes without any element of intent and included underpayments to eligible households. Neither series supports a description of fraud as widespread.

Q: Did falling caseloads mean child poverty fell just as fast?

No. The caseload and the poverty rate measured different populations and moved at different speeds. The Congressional Research Service series recorded families receiving cash assistance falling by roughly three quarters between 1994 and 2018. The Census Bureau official child poverty rate fell through 2000 and rose during the following decade, a path that a monotonic caseload decline cannot mirror. The Center on Budget and Policy Priorities supplied the connecting ratio: 68 families received assistance for every 100 families with children in poverty in 1996, against 22 in 2018. Receipt therefore fell far faster than poverty, because eligibility, application, diversion, sanction, and time limit decisions intervened between a family living in poverty and a family recorded on a caseload.

Q: Do children lose assistance when the five year limit is reached?

Not necessarily. The federal limit at 42 U.S.C. 608(a)(7) governs the use of federal funds for a family containing an adult recipient, and child only assistance falls outside that clock because no adult receives assistance in the adult capacity. State practice at the limit varied in the record assembled by the Congressional Research Service and the Department of Health and Human Services study of state time limit policies. California removed the adult from the assistance unit at the state limit while assistance for the children continued. New York continued aid beyond five years using state funds. Other states, including Arkansas and Arizona at twenty four months, set shorter limits for the whole unit. The outcome for children therefore depended on the state rule and on whether state funds replaced federal funds.

Q: Did the act require every recipient to work immediately?

No. The work participation standard at 42 U.S.C. 607 ran against states as a performance measure, expressed as 50 percent of all families and 90 percent of two parent families containing a work eligible individual who met federal hours rules in countable activities. Federal law did not impose a direct duty on each recipient, and states set individual obligations, exemptions, and sanctions under state law. A single custodial parent caring for a child under twelve months of age could be disregarded from the participation calculation for a limited period. Child only families contained no work eligible adult at all. The caseload reduction credit at 42 U.S.C. 607(b)(3) also lowered the standard a state had to meet, in many states to zero before the fiscal 2007 recalibration enacted in the Deficit Reduction Act of 2005.

Q: Is the work participation rate the same as the share of recipients who work?

No. The participation rate under 42 U.S.C. 607 counted a family only when a work eligible adult completed the required hours in a countable activity with verified documentation in that month, and families without a work eligible individual were excluded from the denominator entirely. Employment at fewer hours, employment in a noncountable activity, and employment in a different month all fell outside the numerator even though work occurred. The Administration for Children and Families characteristics data for fiscal 2018 measured employment directly and recorded 27.3 percent of adult recipients employed in an average month. The Congressional Research Service recorded an average all families participation rate of 53.0 percent for fiscal 2017, judged against standards already reduced by the caseload reduction credit. The two figures describe different instruments and cannot substitute for each other.

Q: Did the act cut off all benefits to legal immigrants permanently?

No. The restrictions enacted in 1996 were severe and categorical rather than permanent and universal. Section 401, codified at 8 U.S.C. 1611, barred persons who were not qualified aliens from federal public benefits subject to limited exceptions. Section 402, codified at 8 U.S.C. 1612, set program by program eligibility for qualified aliens. Section 403, codified at 8 U.S.C. 1613, imposed the five year bar on federal means tested benefits after entry in qualified status. Refugees, asylees, veterans, and residents credited with forty qualifying quarters stood inside exception categories from the start. Congress then restored defined portions of eligibility in the Balanced Budget Act of 1997, the Agricultural Research, Extension and Education Reform Act of 1998, the Farm Security and Rural Investment Act of 2002, and the Children Health Insurance Program Reauthorization Act of 2009.

Q: Do most families stay on assistance until the time limit?

No. The leaver record and the caseload arithmetic point the other way. A caseload is a stock produced by entries and exits, and Jeffrey Grogger, Steven Haider, and Jacob Klerman, in the American Economic Review in 2003, found that the 1990s decline resulted from both faster exits and fewer entries. The fifteen leaver studies synthesized by Acs and Loprest in 2001 covered families that left through earnings, sanction, household change, and administrative closure, most well before any federal limit could bind, because a limit enacted in 1996 could not produce exhaustion terminations until several years after state programs began in 1997. Table 51 of the fiscal 2017 characteristics tables recorded 858,841 closed case adult recipients in a single year. Long stays existed, and time limit terminations occurred, but exhaustion of sixty months described a minority path through the program.

Q: Did states keep caseload savings inside assistance?

The spending distribution answers no for the combined record. When caseloads fell, the fixed block grant did not fall with the caseload, so funds freed by fewer cases stayed available for every permitted use under 42 U.S.C. 601, including child care, work activities, child welfare services, administration, and transfers to other block grants. The Center on Budget and Policy Priorities reported transfers of 6 billion dollars from the Temporary Assistance for Needy Families block grant to the Child Care and Development Fund between fiscal 1997 and fiscal 2000 alone. The Administration for Children and Families financial data for fiscal 2018 placed basic assistance at 21.4 percent of combined federal and state funds totaling 31.3 billion dollars, against 71 percent devoted to basic assistance in fiscal 1997. Freed funds moved across the permitted categories rather than remaining inside monthly cash assistance.

Q: Is a benefit error the same thing as fraud?

No. The Food and Nutrition Service quality control system defined an error as a case in which the action taken by the state agency differed from the findings of the review, because a household was ineligible, an allotment was incorrect, or a decision to deny, terminate, or suspend benefits was incorrect. Intent formed no part of that determination. For fiscal 2013, the combined payment error rate of 3.20 percent consisted of a 2.61 percent overpayment rate and a 0.60 percent underpayment rate, and the underpayment component represented eligible households that received too little. Fraud required a separate determination of an intentional program violation through an administrative disqualification hearing, a court process, or a signed waiver, based on an intentional false or misleading statement or act.

Q: Did the act end Aid to Families with Dependent Children?

Yes, as a program and as an entitlement. Title I of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996, Public Law 104-193, repealed Aid to Families with Dependent Children together with the related Job Opportunities and Basic Skills Training program and the Emergency Assistance program, each authorized under Title IV of the Social Security Act. Temporary Assistance for Needy Families replaced the repealed programs from 1997 under Title IV-A, financed through fixed block grants under 42 U.S.C. 603 rather than through open ended federal matching. The distinction explains several neighboring myths. The predecessor program ended completely. Cash assistance to families with children did not end, because a successor program paid monthly benefits in every state in every year from 1997 through 2019 under different financing, work, and time limit rules.

Q: Were benefits for unmarried teenage parents ended?

No. The 1996 act conditioned rather than ended assistance for minor parents. Federal law required an unmarried minor parent receiving assistance to live in the home of a parent, legal guardian, or other adult relative, or in an adult supervised supportive living arrangement, subject to exceptions where no such placement was available or appropriate. Federal law also required a minor parent receiving assistance to participate in educational activities or an approved alternative, with defined exceptions. Those conditions sat inside the eligibility structure of Title IV-A of the Social Security Act as enacted by Public Law 104-193. A categorical claim that benefits for unmarried teenage parents ended confuses a living arrangement condition and a schooling condition with a termination of eligibility, and the caseload composition data published by the Administration for Children and Families continued to record minor parent cases under those conditions.

Q: Did the caseload fall mainly because families were diverted or pushed off rather than leaving for work?

The question is contested, and the flow evidence supports parts of both readings. Grogger, Haider, and Klerman, in the American Economic Review in 2003, decomposed the 1990s decline into faster exits and fewer entries, so diversion and deterrence before entry formed part of the mechanism by construction. The Brookings Institution review of the period recorded that states received caseload reduction as the central early message and responded to that message, and observed that sanctions accounted for a major part of the decline in some states. The Congressional Research Service reported that most of the decline after 1994 resulted from a reduced share of eligible families receiving benefits rather than from fewer families meeting state definitions of need. Leaver studies, including the Acs and Loprest synthesis of 2001, documented substantial employment after exit, but leaver samples cannot observe families diverted before a case opened.

Q: Did the earned income credit do more than welfare reform to raise employment?

The attribution is contested, and the strongest named estimates favor the credit while assigning the statute a real share. Bruce Meyer and Dan Rosenbaum, in the Quarterly Journal of Economics in 2001, estimated that tax changes centered on the earned income tax credit accounted for about 60 percent of the increase in employment rates of single mothers between 1984 and 1996 in one summary of their results, with welfare reform accounting for about 15 percent across the same span. Jeffrey Grogger, in the Review of Economics and Statistics in 2003, estimated that time limits accounted for about 7 percent of the rise in employment after 1993 and described the credit as a candidate for the single most important policy measure behind the increase in work. Rebecca Blank, in the Journal of Economic Literature in 2002, surveyed the literature and treated the forces as reinforcing rather than separable.