Safety Net for Poorest Weakened After Welfare Law But Regained Strength in Great Recession, at Least Temporarily
A Decade After Welfare Overhaul, More Children in Deep Poverty
 Because the SPM is available only since 2009, it cannot be used for this analysis of trends since 1995, but a modestly modified version of the SPM can be used. As explained in footnote 5 and in Appendix C, the poverty measure used here differs from the federal SPM in several technical ways, including using a 2010 poverty line adjusted only for inflation rather than for changes in living standards (measured by what Americans spend on basic needs). (For an analysis that combines 2012 SPM data with corrections for underreported benefits similar to those used here, see Arloc Sherman and Danilo Trisi, “Safety Net More Effective Against Poverty Than Previously Thought,” Center on Budget and Policy Priorities, May 6, 2015, https://www.cbpp.org/research/poverty-and-inequality/safety-net-more-effective-against-poverty-than-previously-thought.)
 We correct the tendency of Census Bureau data to underreport income from three government assistance programs: Temporary Assistance for Needy Families, Supplemental Security Income, and SNAP (formerly food stamps). The corrections come from the Transfer Income Model (TRIM) policy micro-simulation model developed by the Urban Institute. TRIM starts with Census survey data but adjusts those data to more closely match actual numbers and characteristics of benefit recipients shown in program records.
 Differences between the poverty measure used in this analysis and the federal SPM largely reflect data limitations in the early years of our analysis. Among these differences: (1) we adjust the poverty threshold over time only for inflation, while the threshold in the SPM grows from year to year with expenditures for basic needs; (2) our income measure does not count WIC or subtract child support paid; (3) our poverty thresholds do not vary by homeownership status; (4) we must use approximations for the value of out-of-pocket medical and work expenses (which both our measure and the SPM subtract from income) as well as rent levels (used to adjust the poverty thresholds locally) and rent subsidies; and (5) although our measure (like the SPM) expands who is counted in the family unit when determining poverty status, we do not include everyone who is included in the SPM. For details, see Appendix C.
 Health insurance is not counted as income in the NAS-based measure we use, but having health insurance can reduce poverty and deep poverty rates under our poverty measure by reducing out-of-pocket medical expenditures.
 TANF’s low eligibility ceilings mean that many poor families aren’t eligible for TANF, but the ratio of families receiving TANF cash benefits to all poor families with children also fell dramatically during this period. In 1995, for every 100 poor families with children (using the official poverty statistics), 76 families received cash assistance from AFDC, TANF’s predecessor. By contrast, in 2005, for every 100 poor families with children, only 35 received cash assistance through TANF, and that figure is even lower today.
 The HHS estimates of declining participation rates in TANF do not include people affected by TANF time limits. Time limits — including the 60-month lifetime federal limit on receiving TANF assistance enacted in 1996 and, in many states, shorter state time limits — are an additional factor that contributed to the weakening of the cash welfare assistance component of the safety net.
 Stephen Freedman et al., National Evaluation of Welfare-to-Work Strategies Evaluating Alternative Welfare-to-Work Approaches: Two-Year Impacts for Eleven Programs, U.S. Department of Health and Human Services, June 2000, page ES-35, Exhibit ES-10, www.mdrc.org/publication/evaluating-alternative-welfare-work-approaches/file-full. Ten of the 11 programs showed some increase in the share of participants below half the poverty line (six of them statistically significant), while nine showed declines in the overall poverty rate (five of them significant). Taking a simple average across all 11 sites, deep poverty rose by 2.7 percentage points while regular poverty declined by 2.1 percentage points. The study measured poverty using participant earnings, welfare, and food stamp income from administrative sources.
 Freedman et al., Exhibit ES-9, page ES-34.
 Marianne Bitler, Jonah Gelbach, and Hilary Hoynes, “What Mean Impacts Miss: Distributional Effects of Welfare Reform Experiments,” American Economic Review, vol. 96, no. 4 (September 2006), www.ssc.wisc.edu/~scholz/Teaching_742/Bitler_Gelbach_Hoynes.pdf.
 Our poverty measure treats unmarried partners of the household head as part of the head’s family unit if they were consistently present, that is, living in the household both at the time of the survey and 12 months earlier. See Appendix C for details.
 Arloc Sherman, “Poverty and Financial Distress Would Have Been Substantially Worse in 2010 Without Government Action, New Census Data Show,” Center on Budget and Policy Priorities, November 7, 2011, https://www.cbpp.org/research/poverty-and-financial-distress-would-have-been-substantially-worse-in-2010-without.
 Chuck Marr, Bryann DaSilva, and Arloc Sherman, “Letting Key Provisions of Working-Family Tax Credits Expire Would Push 16 Million People Into or Deeper Into Poverty,” Center on Budget and Policy Priorities, Updated February 20, 2015, https://www.cbpp.org/research/letting-key-provisions-of-working-family-tax-credits-expire-would-push-16-million-people.
 Arloc Sherman, “Why Isn’t Poverty Falling? Weakening of Unemployment Insurance Is a Pivotal Factor,” Center on Budget and Policy Priorities, October 7, 2013, https://www.cbpp.org/research/why-isnt-poverty-falling-weakening-of-unemployment-insurance-is-a-pivotal-factor.
 A more comprehensive examination of the missing benefits yields similar results. When one includes two additional benefits, SNAP and SSI, in addition to TANF, and adjusts for the effects of the growing national population by expressing the underreported amounts in dollars per capita, the inflation-adjusted value of benefits missed by the CPS fell from $123 per person in 1995 to $105 per person in 2005.
 This alternative method, which serves to check our TRIM findings, differs from our main method in three ways: it does not use TRIM adjustments; it uses official poverty guidelines rather than an NAS-based threshold; and, unlike the NAS method, it uses family disposable income (after taxes and noncash benefits) without netting out work expenses or medical out-of-pocket expenses or altering the family unit for which poverty status is determined. To adjust for underreported income, the method starts with the share of children below half the poverty line not including SNAP, AFDC/TANF, or SSI (11.0 percent in 1995 and 7.5 percent in 2005). It finds the share of children lifted out of deep poverty by each of the three programs in the uncorrected Census data for those years (cumulatively, 6.8 percent of children in 1995 and 3.2 percent in 2005). For each program, it adjusts those uncorrected shares by dividing them by the ratio of Census-reported benefit payments to true aggregate benefit payments (which we find to be 65 percent in 1995 and 55 percent in 2005 for SNAP; 70 and 48 percent for AFDC/TANF; and 71 and 78 percent for SSI). Finally, it applies those adjusted anti-poverty effects to the pre-transfer deep poverty rate to yield an adjusted deep poverty rate (1.0 percent in 1995 and 1.9 percent in 2005). Although these levels are lower than in our TRIM/NAS-based calculations (in part because this method does not subtract medical and work expenses and uses the lower, official poverty line), the increases in both methods are of the same magnitude.
 Christopher Wimer et al., “Trends in Poverty with an Anchored Supplemental Poverty Measure,” Columbia Population Research Center, December 2013; Yonatan Ben-Shalom, Robert A. Moffitt, and John Karl Scholz, “An Assessment of the Effectiveness of Anti-Poverty Programs in the United States,” prepared for the 2012 Oxford Handbook of the Economics of Poverty, chapter 22; H. Luke Shaefer and Kathryn Edin, “Rising Extreme Poverty in the United States and the Response of Federal Means-Tested Transfer Programs,” National Poverty Center Working Paper 13-06, May 2013.
 Bruce D. Meyer, Wallace K. C. Mok, and James X. Sullivan, “The Under-Reporting of Transfers in Household Surveys: Its Nature And Consequences,” National Bureau of Economic Research, NBER Working Paper 15181, July 2009, http://www.nber.org/papers/w15181.
 Arloc Sherman, “Safety Net Effective at Fighting Poverty But Has Weakened for the Very Poorest,” Center on Budget and Policy Priorities, July 6, 2009, https://www.cbpp.org/research/safety-net-effective-at-fighting-poverty-but-has-weakened-for-the-very-poorest.
 National Research Council, Measuring Poverty: A New Approach(National Academy Press: 1995).
 Housing assistance includes federal, state, and local housing vouchers and public housing. We calculate the value of rental assistance using a method that the Census Bureau developed for use with its NAS-based poverty measures. Specifically, for each assisted household, the annual value of housing assistance is 12 times the local monthly fair market rent reduced by the household’s required contribution, approximated as 30 percent of its annual cash income. (We approximate fair market rents using a weighted average of HUD local Fair Market Rent levels for each family's state, broken down further by whether the family lives in a metropolitan or non-metropolitan area, from data available at www.huduser.org/portal/datasets/fmr.html. We estimate the number of bedrooms in each housing unit using HUD occupancy rules.) When more than one family or unrelated individual shares the same apartment, we assign each its per capita share of the household’s housing subsidy. Since housing assistance cannot be used to meet other needs such as food or clothing, we, like the Census Bureau, place a cap on the value of housing assistance equal to housing's share of a poverty-level budget (that is, 44 percent of the poverty line for a family of that size and composition).
 Using TAXSIM means that we give up some features of the Census model, such as imputation of itemized deductions, but in return we are assured of a consistent model over time. For details on the NBER TAXSIM version9 model, see www.nber.org/~taxsim. Using TAXSIM with the CPS requires determining which individuals file taxes together. In general, we treat each nuclear family as a tax filing unit; however, we assume that foster children are dependents of the head of the housing unit, as are certain clusters of relatives (called related subfamilies, typically a child of the head of household raising children) if they are living at home but have no earnings.
 TRIM is developed and maintained by the Urban Institute under contract with the Office of the Assistant Secretary for Planning and Evaluation at the Department of Health and Human Services. Documentation of the TRIM model is at http://trim3.urban.org/T3Technical.php. While the model was developed chiefly to allow users to compare current policies with proposed policies, we use data only for TRIM’s “baseline” (or current-policy) scenario.
 In producing the CPS files, the Census Bureau, like TRIM, also assigns benefits for some people with missing data. Unlike TRIM, however, Census does not use this process to try to match the actual number of recipients shown in program records.
 The CPS recently added questions on MOOP and child care expenses, but for the early years of our analysis, these data are not available. To ensure consistency, we approximate them for each year using formulas adapted from the Census Bureau and BLS. For out-of-pocket medical spending, we use the medical portion of the poverty line in a version of the NAS thresholds (described in the next bullet in the text) that includes out-of-pocket medical spending as a basic need. We start with the medical share of the poverty line in 2010 — that is, 7.23 percent of the NAS-based poverty line for a family of two adults and two children (www.bls.gov/pir/spm/spm_threshold_200910.xls), or $1,989 — and adjust that amount up or down using a set of ratios provided by the Census Bureau that depend on the family's size, health insurance status, and age, thus assigning to each family a level of medical out-of-pocket (MOOP) expenditures in 2010 typical for its family type. We deflate these values for other years using the overall CPI-U, as Census and BLS do when calculating NAS-based measures that use a fixed CPI-based threshold and include MOOP in the threshold.
Our estimates of work expenses also rely on formulas provided by the Census Bureau. The formulas are based on data on median weekly out-of-pocket child care expenses and other work-related expenditures from the Survey of Income and Program Participation (SIPP). Values vary depending on year, weeks worked, number and ages of children (in the case of the child care formula), and other family characteristics. (We use weekly values provided by the Census Bureau for 1996 through 2006; for other years, we extrapolate values using the CPI-U.) Whether the family used paid child care in a given year is estimated with a probabilistic model. The estimated value of work expenses is capped at the value of the worker’s earnings. A couple’s child care expenses are further capped based on the earnings of the lower-earning spouse.
 Short-term partners may contribute less to family finances than long-term partners, either because they feel less invested in the family or simply because couples in which the partner for whatever reason contributes less tend to break up sooner. We found evidence for this when we examined data on family hardships. In an unpublished analysis, we examined 402 single parents living below the poverty line (according to the official annual cash poverty definition) with an unmarried partner, using SIPP hardship questionnaires from 1998, 2003, 2005, and 2011 and income data for the preceding 12 months. Among these parents, having a partner with enough income to lift them and their children above the poverty line for the year was associated with significantly fewer material hardships, relative to having a poorer partner — but only ifthe partner was stable, that is, living with the family consistently for the previous 12 months. For other, non-stable partners, bringing in enough income (while cohabiting) to lift the family’s annual income above the official poverty line for the year did not reduce the number of reported hardships. This difference between stable and non-stable partners was statistically significant and suggests that the case for including long-term partners in the family unit for the purposes of poverty determination is more compelling than the case for including short-term partners. (The analysis examined four hardships: food insecurity, falling behind on the rent or mortgage, having phone service cut off for failure to pay, and falling behind on gas or electric bills. Results were similar after controlling for differences in race, ethnicity, gender, number of children, parent education, age, disability, and whether the partner was a biological parent of the children.)
 Following the approach used in the SPM and TRIM data, we treat household members younger than 15 living with no relatives as family members of the head of household. This affects fewer than 200,000 poor children in our data in both 1995 and 2005.
 See “Poverty Thresholds for Two-Adult-Two-Child Family Following NAS Recommendations: 1999-2010,” downloaded from http://www.census.gov/hhes/povmeas/data/nas/tables/2010/web_tab5_povertythres2010.xls (column 3). Note that although the NAS poverty line was calculated by BLS and is found on the Census website, it may be too low. While it is designed to reflect a minimal spending level for food, clothing, shelter, utilities, and medical needs, it leaves out the portion of shelter expenses that homeowners pay for mortgage principal. NAS poverty lines have excluded any amount for mortgage principal payments since the original 1995 NAS report, which noted that these payments were being left out for reasons of “processing convenience.” Preferably, the report said, an amount for these payments would be included in the NAS poverty line. Recently, BLS staff have developed a method for doing so. This would raise the poverty line by about $2,000, which would raise the estimated number of people in poverty and deep poverty.
 See Thesia I. Garner and Kathleen S. Short, “Creating a Consistent Poverty Measure Over Time Using NAS Procedures: 1996-2005,” NAS Poverty Measurement Working paper, April 3, 2008, available at www.census.gov/hhes/povmeas/publications/povthres/experimental_measures_96_05v7.pdf.
 All else being equal, using the NAS-based poverty threshold instead of the official poverty line raises the poverty rate in 2010, as previously shown in Figure 2. It also raises the percentage of people who would be poor when government taxes and benefits are not counted (from 25.3 percent to 31.2 percent).
 For the sake of consistency with the original NAS recommendations, we subtract our estimate of MOOP, described earlier, from both sides of the poverty equation — thresholds and income. This has no effect on families’ poverty status but lowers the poverty threshold and, conceptually, treats medical expenditures as a deduction rather than a basic need.
 The NAS panel recommended updating poverty lines each year using a method that tends to rise slightly faster than the CPI. (Under that approach, the poverty line is adjusted each year by the percentage change in median family expenditures on basic needs for U.S. two-child two-adult families.) This alternative adjustment method is not comparable across recent years, however, due to improvements introduced in 2007 in the Consumer Expenditure Survey, the survey used to track spending on basic needs. Therefore, while our analysis uses the NAS threshold for 2010 as a starting place, we adjust it over time using the CPI. If we had used the NAS method of adjusting our thresholds over time (as we did in our original 2009 analysis of deep poverty), our poverty line would have risen somewhat faster, our estimated number of people in poverty would have fallen somewhat less, and the number in deep poverty would have risen somewhat more.
 Adjustments for family size use the “three-parameter” scale described by Garner and Short (see footnote 31 above) applied to the non-medical portion of the NAS poverty threshold. Geographic adjustments are based on the ratio between the local and national HUD fair market rent (described in footnote 23), applied to the housing portion of the threshold.
 Income below 75 percent of the poverty line, as measured by our approach, is a more stringent definition of deep poverty than it may appear. Traditionally, analysts have considered those below 50 percent of the poverty line to be in “deep” poverty. Typically, however, the 50 percent threshold has been used in conjunction with a measure of cash income. That is, a family with cash income below 50 percent of the official poverty line was considered to be in “deep poverty” even if the family received food stamps or housing assistance that lifted its total purchasing power above this threshold. The 50 percent threshold represents a much stricter definition of deep poverty when one adopts, as we do, a broader poverty measure that includes noncash and tax-based benefits as income and corrects for underreporting of benefits. The percentage of the population with cash below half the poverty line in 2010 using the official measure (6.7 percent) was closer to the percentage below 75 percent of the poverty line by our measure (7.4 percent) than to the percentage below half the poverty under our measure (3.3 percent). In that sense, a 75-percent-of poverty standard under an NAS-based TRIM-corrected measure may be more comparable in severity to the traditional 50-percent-of-poverty standard under the official cash measure.
 These calculations assume that a family’s other income would remain unchanged. Although people’s behavior might change in the absence of government assistance, a recent comprehensive review of the literature by some leading economists in the field concludes that this would have a “negligible” effect on overall poverty rates. Yonatan Ben-Shalom, Robert A. Moffitt, and John Karl Scholz, “An Assessment of the Effectiveness of Anti-Poverty Programs in the United States,” prepared for the 2012 Oxford Handbook of the Economics of Poverty, chapter 22.