Overestimates appear to be the product of unreliable survey data.
A recent dust-up between the U.N. and the Trump administration reinvigorated the debate over “extreme poverty” in the U.S. The U.N. claimed an astounding 18.5 million Americans were mired in this condition; the latter said the true number was a mere 250,000 — fewer than one in 1,000 Americans.
As I’ve noted previously, the U.N.’s number was absurd, relying on a definition of extreme poverty that no one else uses. But there is a legitimate debate over the true extent of extreme poverty — typically defined as a household income under $2 per person per day, which is something like one-tenth of the poverty line — with estimates ranging from a few hundred thousand to a few million. And in new research presented at the American Enterprise Institute Tuesday, the economist Bruce D. Meyer and two co-authors make a forceful argument that the lower numbers are the correct ones.
Their own estimate is that “at most one-quarter of one percent of households are living on less than $2/person/day” — about 326,000 individuals in total — and that these are overwhelmingly single adults, sometimes students. Indeed, the authors were unable to identify a single family with children that was extremely poor in their data.
Claims of rampant extreme poverty first rose to prominence with the 2015 publication of the book $2 a Day, the central claim of which was that 4 percent of American families with children fell below that cutoff — largely because the 1996 welfare reform made it harder to get cash assistance. The book itself made clear some serious limitations of this number, though. For one thing, it was limited to cash income, and the estimate fell by half when food stamps were included. And for another, it was based on survey data, meaning that individuals reported their own income. It’s well known that people tend to underreport their welfare benefits and off-the-books cash in surveys, and responses are particularly suspect when individuals claim to subsist on essentially no money.
In response, some critical researchers tried to correct the survey data to account for underreporting, or looked at households’ spending rather than their income, and reached radically lower estimates. But neither approach is entirely satisfactory. That’s where the new work of Meyer et al. comes in.
Meyer’s team has assembled an impressive data set that starts with the government survey used in $2 a Day (the Survey of Income and Program Participation, or SIPP), but also includes administrative data from various government agencies. As a result, instead of trying to statistically estimate whether a survey respondent is underreporting his income and benefits, Meyer et al. can simply look at federal and state records to see how much that individual received. This allows the authors to make more comprehensive and accurate adjustments to the survey data than anyone else has been able to. The biggest tradeoff is that they currently have administrative food-stamp data from just eleven states, so their fully adjusted national estimates need to be scaled up from those — but fortunately, these states are demographically similar to the country as a whole.
Their raw estimate, based only on cash income reported in the survey, is that 3 percent of all households (and nearly 10 percent of single-parent households) live in extreme poverty. Add in self-reported non-cash benefits and it’s down to 2.1 percent. Account for the fact that a small share of respondents claim to have little or no income despite working many hours at a paying job — clearly a mistake — and we’re at 1.3 percent. Reclassify low-income households that actually have substantial assets (such as $5,000 in cash or $25,000 in real-estate equity), and it’s 0.9 percent. And when you consult the administrative data to account for the underreporting of income and benefits, it falls more than two-thirds, reaching the final estimate of 0.24 percent. Incredibly, many of the individuals who move out of “extreme poverty” when these adjustments are made appear not to even be poor, much less extremely poor.
Meyer et al. demonstrate beyond a doubt that claims of extreme poverty are tremendously overblown.
The results are similar when they switch gears, repeating the analysis with the Current Population Survey’s Annual Social and Economic Supplement (CPS ASEC), another government survey they can match to their administrative data: Just 0.12 percent of households are in extreme poverty once the proper corrections are made. This is an estimate of full-year extreme poverty, whereas the SIPP estimates pertain to a four-month period, which likely explains why it’s so much lower.
This is not quite the final verdict on the extreme-poverty question. As the authors note, none of this accounts for off-the-books income that people might not report to survey-takers, and the authors’ administrative data are far from comprehensive; they aren’t able to correct for underreporting of unemployment insurance, veterans’ benefits, workers’ compensation, cash welfare, or the child tax credit. On the other hand, however, just as surveys undercount government benefits, they also neglect the homeless, which number in the hundreds of thousands on any given night. The authors suggest a “next step” to their research will be to mine additional sources of data to address both of these limitations.
At the event debuting the results, Laura Wheaton of the Urban Institute also raised some technical issues with the analysis, including the way the authors estimated earnings for people who reported working for pay but not earning much money (they multiplied the hours by the minimum wage, which may overestimate earnings for some self-employed and tipped workers).
Meyer et al. demonstrate beyond a doubt, though, that claims of “extreme poverty” are tremendously overblown. One can be concerned about the poor without claiming that such incredibly severe deprivation is common in this country — so perhaps it’s time to shift our focus accordingly.