Updated April 9, 2017
Back in the field in 2010 to study the deeply poor, Kathryn Edin began to encounter something markedly different from anything she had seen in 20 years of canvassing poor communities: families with no visible means of cash income from any source. As Edin and I write in $2 a Day: “[W]hat was so strikingly different from a decade and a half earlier was that there was virtually no cash coming into these homes.”
Has there been a spike in the number of children going for periods with virtually no cash in the U.S.? We began to test this with the Survey of Income and Program Participation (the SIPP), where we saw a striking spike in the number of households with children reporting cash incomes of no more than $2 per person, per day over a month, a calendar quarter, and a year. To date, we have substantiated our core findings in two other major household surveys. Yet all of these surveys suffer from problems—some people may not want to reveal all their sources of income, they may forget, or they may misunderstand the questions. Perhaps our findings across three surveys were driven entirely by faulty data.
We tested this first by examining administrative records from the Supplemental Nutrition Assistance Program (SNAP, formerly food stamps)—which told much the same story as the SIPP. But another important source can help here—the federal government supports a micro-simulation model called TRIM, which is constructed by the non-partisan think-tank, The Urban Institute. TRIM corrects Current Population Survey (CPS) data for the “under-reporting” of public benefits—when people forget or choose not to say they receive benefits from a program like TANF. We can use it to examine data that is adjusted for these factors. Even with these corrections, survey data remain imperfect. And when extreme poverty is measured with an annual recall, you might expect it to be subject to a lot of error (people might not remember their income from January well when reporting after the end of a year; TRIM doesn’t correct for reporting in other forms of income besides public participation). But TRIM is a significant improvement over unadjusted survey data from the Current Population Survey, and we seek to determine if results using TRIM match our previous findings from other data sources.
Below we chart the number of children under 18 in households reporting annual cash incomes under the $2-a-day threshold, after correcting for underreporting in TANF and SSI (these estimates were generously provided by our colleague Danilo Trisi at the Center on Budget and Policy Priorities, an expert on TRIM). In the adjusted TRIM data, the number of children in annual extreme poverty rises from 415,000 in 1995 (0.6% of all children) to a peak of 1.3 million in 2011 (1.8%), and remained at 1.2 million (1.6%) in 2012. That’s roughly a tripling in the number of children in extreme poverty between 1995 and 2012.
How do these adjusted estimates compare to estimates of annual extreme poverty using unadjusted annual-recall survey data? Table 1 presents the adjusted and unadjusted counts for 1995 and 2012 using Trisi’s exact procedures (which are close but not identical to our own). As we would expect for reasons described in $2.00 a Day and our academic papers, the TRIM adjusted estimates of annual extreme poverty in any given year are lower than the unadjusted counts. We also find that the TRIM counts of annual extreme poverty are very much in line with our previous SIPP estimates of annual extreme poverty reported elsewhere. In those papers (and also in the SNAP administrative data), we also examine spells lasting for less than a year.
However, while the overall levels of extreme poverty are lower in any given year in the adjusted data, the magnitude of the change over the 18-year study period is much greater. If we were examining only the unadjusted data, we would conclude that annual extreme poverty nearly doubled between 1995 and 2012. But using the adjusted data, it appears that extreme poverty nearly tripled over this period. In short, correcting for underreporting doesn’t explain away the rise in $2-a-day cash poverty since 1996 that motivated $2.00 a Day. In fact, it makes the change over time look even more stark.
What can explain this? As shown in table 1, the answer appears to relate to the fact that the unadjusted CPS data overstated extreme poverty to a greater degree in 1995 (261% of the adjusted data) than it did in 2012 (174% above). But if problems with underreporting with our survey data are getting worse over time, how can that be true?
We argue that the answer to this puzzle is directly related to the decline of cash assistance, and phenomenon first identified by Arloc Sherman and Trisi at the Center. In essence, in any given year, some families responding to surveys fail to report that they got TANF, and this problem seems to be getting worse over time. But while the rate of underreporting is getting somewhat worse over time, the number of cash assistance dollars to underreport has also gotten a lot smaller.
Take the following simplified illustration. Let’s say that in 1995, 35% of AFDC recipients forgot to tell survey researchers they were on cash welfare, and in 2012, the same was true of 45% of beneficiaries. How many recipients are missed in these two years? There were roughly 12 million people on cash welfare in 1995, and 4 million in 2012. That would mean in 1995, the survey would miss 4.2 million cash welfare recipients (12 million * 35%). But in 2012, there were only about 4 million people on cash welfare. In this year the survey would miss 1.8 million recipients (4 million * 45%), less than half what was missed in 1995.
Thus, even though the rate of underreporting is worse in 2012, the number of missed cash welfare recipients is a great deal larger in 1995 (4.2 million missed compared to 1.8 million) because there are so many fewer cash assistance recipients overall in 2012. This phenomenon then mutes the magnitude of the change in extreme poverty in the unadjusted data. With their cash welfare dollars unaccounted for because of underreporting, more families wrongly appeared to be in $2-a-day poverty in 1995 than in 2012.
When looking at extreme poverty in data adjusted for survey problems, the magnitude of the rise in extreme poverty appears even larger than previously reported, and the explanation, we believe, is directly related to the decline of TANF. If we are right about this, then the adjusted data should have the biggest impact on counts of extreme poverty among children in single mother families—those most impacted by welfare reform and the decline of cash assistance. Stay tuned for our next blog post which will have a look at that.
- H. Luke Shaefer