How many Americans live on $2 a day? The biggest debate in poverty research, explained.

How many Americans live on $2 a day? The biggest debate in poverty research, explained.

No matter what — poverty is still way too high. By Dylan Matthews@dylanmattdylan@vox.com  Jun 5, 2019

How many people in America live on less than $2 a day?

That seems like it ought to be a simple question to answer. And for much of the last decade, Princeton sociologist Kathryn Edin and University of Michigan social policy researcher Luke Shaefer have been using survey data to argue that a significant and rising share of American children live in households earning less than $2 in cash income per person. This data and their ethnographic portraits of people living in extreme poverty in places like Cleveland and Mississippi formed the basis of their widely acclaimed 2015 book $2 a Day, which was widely written up by journalists like myself.

But since their research started circulating, some economists and sociologists have pushed back, arguing that the Edin and Shaefer’s research — which relied on surveys — underestimated the support very poor households get from welfare programs that provide benefits “in kind” rather than through cash.

That’s because many people frequently fail to tell surveyors about government programs they benefit from, meaning surveys can “underreport” assistance. The Supplemental Nutrition Assistance Program (formerly “food stamps”), in particular, is a crucial support for many of these families, and would place most above the $2/day line if respondents included them in their responses to surveys.

The most comprehensive response to date — by University of Chicago professor Bruce Meyer, his colleagues Derek Wu and Victoria Mooers, and the Census Bureau’s Carla Medalia — has just been publicly released, and concludes that true $2-a-day poverty, after adjusting the data properly, is extremely rare.

“Our best estimate of the extreme poverty rate,” they write, is 0.11 percent for individuals as of 2011. That implies that about 336,160 people are in extremely poor households, far lower than the couple million children estimated by Edin and Shaefer. The vast majority of those people, they argue, are childless adults, and the extreme poverty rate for parents is close to zero.

Because they used private IRS, Social Security, and other government data, Meyer and his colleagues have more precise estimates of what income people are earning and what benefits they’re collecting than surveys can provide. That gives their estimates a great deal of credibility.

There are two key reasons why these findings matter. The first relates to the long-simmering debate over the effects of the 1996 welfare reform law on low-end poverty. While the overall poverty rate fell in the law’s midst (which coincided with the best job market in decades), there have been several indications, including Edin and Shaefer’s work, that hardship for the very poorest increased. If Meyer and his coauthors are right, then extreme poverty is concentrated among childless people — not the single mothers most affected by welfare reform. That removes a key argument that welfare reform did damage.

The second relates to a more abstract question: how bad is poverty in America — and what do we have to do to tackle it? There has been a major push by poverty scholars in recent years to expand the availability of cash benefits, particularly a child allowance distributed to parents, in order to reduce hardship for families. The Edin/Shaefer research provided support for the idea that that change is necessary; the Meyer research removes one potential rationale for it.

Cards on the table: I have been enthusiastic about the Edin/Shaefer research for many years. It was a big piece of evidence in this review I did of the research on welfare reform, and I found Edin, Shaefer, and other poverty experts’ arguments for a child allowance incredibly compelling.

Meyer and company’s research has weakened my belief that very low-end poverty increased in the wake of welfare reform; we certainly need more papers on this, and I think there are plenty of other problems that made the 1996 law a mistake, but I’m no longer very confident there was an increase in extreme hardship as a direct result of the law.

But my belief that a child allowance would help considerably remains intact. The Meyer research still makes clear that there’s a struggling population in real material hardship that the current American safety net is not adequately addressing.

Two ways to estimate $2 a day poverty

The Edin and Shaefer estimates were based on something called the Survey of Income and Program Participation (SIPP), a longitudinal survey run by the Census Bureau, which seeks to track certain individuals and families over time and see how their incomes and participation in government transfer programs like Medicaid and food stamps evolve.

SIPP regularly picks groups of people and follows them for three years, enabling the creation of charts like this one, from Edin and Shaefer’s initial policy brief released in 2012:

What the data, analyzed this way, suggests is that a significant number of households with children were making under $2/person/day, and that the number was growing. The growth was less pronounced if you added in food stamp benefits, but it was still there.

But a well-known problem with SIPP and other surveys like it is that respondents tend to underreport the benefits they get. Due to a variety of reasons, from the stigma of being on government assistance to just not remembering, survey results suggest fewer people get benefits than administrative records collected by the government (which are generally thought to be more accurate). Edin and Shaefer were aware of this, but argued that the trends they found were robust to the problem.

Meyer is an expert on underreporting, and in the latest paper he brings some big administrative data to bear on the extreme poverty questions: IRS tax records. He and his coauthors link SIPP and the Current Population Survey, another frequently used household survey measuring poverty, to IRS and Social Security Administration data on earnings, tax credit receipt, and Social Security retirement and disability benefits; state food stamps records; and Department of Housing and Urban Development records on housing benefit receipt.

Table 3b of their paper, breaking down what each adjustment they make does to the estimated extreme poverty rate for individuals, is incredibly dense but super important, so here it is:

Here’s what we’re looking at. Let’s start with single parents. In the initial survey data, 9.56 percent of single parents were counted as in extreme poverty. Including in-kind transfers like SNAP brings that share down dramatically, to 2.65 percent.

The researchers then apply a number of adjustments to the survey data meant to remove anomalous reports (like people who report many work hours but zero dollars in pay, and people with no income but lots of wealth) before using the administrative data. Adding in IRS data on earnings and Earned Income Tax Credit receipt, Social Security Administration data on retirement and disability benefits, and state data on SNAP and housing assistance brings the estimated share of single parents in extreme poverty down to literally zero.

Interestingly, the Meyer methodology suggests that the vast majority of individuals in extreme poverty aren’t parents, but childless adults. The extreme poverty rate they find for single childless adults is 1.12 percent, amounting to 252,336 people. A recent unpublished paper (that I really hope is put online soon so I can link to it) by sociologists David Brady and Zachary Parolin suggests that much of this population consists of unauthorized immigrants who are ineligible for programs like food stamps.

There’s an important caveat to both the Edin and Shaefer study and Meyer and company’s work on this: This survey data can’t reach Americans who are homeless. The Department of Housing and Urban Development estimates that about 550,000 people were homeless on a given night in 2018; some advocacy groups argue this is a dramatic undercount. While some people without homes have jobs working enough hours to arguably fall outside the category of “extremely” poor, there’s a huge overlap between extreme poverty and homelessness. Both the Edin/Shaefer and Meyer papers are, as they both state clearly, undercounting by not including this population.

The technical issues, and the philosophical issue

So, what accounts for such a huge discrepancy between the Edin/Shaefer count and the Meyer et al count? There are a number of technical, methodological assumptions that the two research groups make that set them apart a bit.

Edin and Shaefer are interested in households that fell below $2/person/day for some portion of the year, whereas Meyer only count as extremely poor households that fell below that line for the entire year.

Speaking of the in-person, ethnographic interviews with very poor households that he and Edin did in their research, Shaefer says, “I think virtually all the families in our book would have, under the way they do things in this paper, been pushed above the line … [Maybe] they were earning money, working a low-wage job for six months, and then lost their job and fell into extreme poverty.”

Shaefer also disputed the Meyer paper’s assumption that households with substantial assets were definitionally not extreme poor. A household with a property worth at least $25,000 where no one is getting cash earnings is still extremely hard up, and arguably deserving of the descriptor “extremely poor.”

But those don’t explain the bulk of the discrepancy. If you look at Meyer and company’s table, it shows that that the estimated share of single parents (the main population Edin and Shaefer are concerned with) in extreme poverty falls from 9.56 percent to 2.65 percent the minute you include in-kind benefits, particularly food stamps. That’s not the whole ball game, but it’s well over 70 percent of the ball game. That choice, to count food stamps as equivalent to cash, is by far the biggest point of divergence.

Meyer has long been a vocal advocate of measuring poverty by looking at consumption: how much actual housing, food, transportation, clothing, etc. are society’s poorest people getting? He and frequent coauthor Jim Sullivan made the case for it well in a 2009 paper:

Consumption reflects permanent income and thus captures the long-term prospects of a family better than current income. Consumption is more likely to capture the effects of saving and dissaving, the ownership of durable goods such as houses and cars, and access to credit. Consumption is also more likely to reflect private and government transfers. The consumption and income data available in the U.S. are both subject to error, but the consumption data provide more information than income data to impute noncash housing benefits and the service flow from vehicle and home ownership.

The advantage of consumption is that it’s comprehensive. It reflects all the various material resources that families utilize. And it offers a simple answer as to how to treat stuff like food stamps. While whether food stamps are a kind of “income” is arguable, the food purchased with them is definitely a kind of consumption.

Meyer and his coauthors actually aren’t the first to have attempted a consumption approach to estimating extreme poverty. In 2014, Cory Smith and Laurence Chandy at the Brookings Institution used the Consumer Expenditure Survey. It’s not an ideal survey for this purpose, but it’s probably the closest thing the US has to the kind of surveys that developing countries and the World Bank use in determining extreme poverty rates abroad.

Their best inference is that the extreme consumption poverty rate in the US in 2011 was around 0.07 percent, strikingly close to the Meyer estimate. Scott Winship, a sociologist who has challenged more pessimistic assessments of the state of poverty and inequality in America, looked at another income survey, the Current Population Survey, and concluded the same.

Edin and Shaefer have consistently pushed back on this view, emphasizing that there’s a difference between enjoying the flexibility that comes with cash, and being constrained by in-kind benefits like food stamps. Food stamps can’t pay your rent or buy clothing for your kids. “We spent a lot of time in the book on SNAP and the extent to which people sell their SNAP for 50 cents on the dollar because they need the accessibility of cash,” Shaefer says.

Shaefer and Meyer have a related dispute, defined in an exchange of papers last year, about whether income or consumption measures do a better job of correlating to concrete metrics of material hardship that the federal government collects data about: if people report having difficulty paying for enough food or keeping up with their mortgage, if they report putting off doctors’ visits for economic reasons, etc. Shaefer and his coauthor Joshua Rivera argue that these metrics correlate better with income-based poverty measures than consumption-based ones. Meyer and Sullivan argue that the opposite is true.

So … what should I think now?

This can be a confusing dispute to follow, not least because of the political stakes. Edin, Shaefer, and Meyer are all highly respected researchers, but Edin and Shaefer make no secret of being left of center and supporting a more generous safety net, while Meyer is affiliated with the American Enterprise Institute and receives funding from the Charles Koch Institute. I think Meyer does extremely good, careful work, but there’s obviously a reason that the Charles Koch Institute wants to fund this line of research, and by the same token there’s a reason supporters of a bigger safety net would be drawn to research suggesting that extreme poverty is widespread. There’s a broader political debate outside the narrower methodological debate.

The two major debates on which this evidence weighs are, in brief: (a) did welfare reform increase extreme poverty and (b) how bad is deep poverty in America?

The Meyer paper shows pretty definitively that welfare reform did not increase extreme poverty as defined as “the share of households with children consuming less than $2 per person per day.” That phenomenon appears to be extraordinarily rare in the United States, thankfully.

And the methods used in the paper can be used to adjust other papers that have used survey data to suggest that low-end poverty increased in the wake of the 1996 law:

• The Urban Institute’s Pamela Loprest and Austin Nichols and University of Wisconsin Madison’s Rebecca Blank and Carnegie Mellon’s Brian Kovak used SIPP and CPS to estimate that the number of “disconnected mothers” — single parents moms who get neither cash benefits nor earnings from work — increased in the wake of welfare reform.
• A 2011 paper by Hopkins’s Robert Moffitt, Mathematica’s Yonatan Ben-Shalom, and University of Wisconsin Madison’s John Karl Scholz estimated that the share of families living under half the poverty line increased by nearly 50 percent from 1993 to 2004, covering welfare reform’s implementation.
• A 2016 analysis by Harvard’s Christopher Jencks and Columbia’s Liana Fox and Chris Wimer used the CPS to find that economic resources for Americans at the 2nd percentile of the income distribution fell dramatically after welfare reform.

Those are all good papers by careful researchers, but I’d love to see them rerun in a way that takes account of the underreporting problems Meyer has identified. Given how much that changes the story for the Edin/Shaefer data, it might change the story for the rest of the literature as well.

What other evidence do we have to suggest that well-being fell for folks at the very bottom in the wake of welfare reform? Shaefer and Edin have pointed to administrative data suggesting that the number of SNAP households with kids reporting zero dollars in earnings rose from 316,000 in 1996 to 599,000 in 2005, and then to 1.3 million in 2013. They note that this data matches the trendline of their original $2/day survey data quite well, suggesting that something concerning is happening at the very bottom of the income distribution. Randomized controlled trials conducted around welfare reform have also suggested heterogeneous effects: no positive impacts on people close to the poverty line, but damage done to people at the very bottom.

It’s not that there’s no evidence welfare reform substantially hurt people at the very bottom. But whereas in 2016 I would have probably described the evidence base as pretty uniform and strong, it now looks considerably more mixed. There are plenty of other reasons to be skeptical of welfare reform beyond its impact on the very bottom of the distribution, but my confidence in that particular argument has eroded.

What about the “how bad is poverty in America” question? It seems clear to me that if there is a population in the US as materially deprived as the median Haitian or Congolese person, it’s small. The Smith and Chandy estimates are, while limited, the best apples-to-apples comparison we have, and they suggest that something like 300,000 people fall below that consumption threshold at any given time. The real number is likely higher than that if you include homeless people that the surveys missed.

That’s tiny as a share of the overall US population, and thus hard to target. It’s real, and if there’s more we can do to reduce that number and get it as close to zero as possible, that’s worth pursuing in my view. But if we’re using “$2 a day poverty” to mean the same thing it means in foreign countries, it’s not as widespread as a naive look at the Edin and Shaefer numbers would imply.

But both the Meyer and Edin/Shaefer numbers suggest that there’s a slightly “richer,” but still incredibly deprived population in the US: people who struggle to pay basic utility and mortgage bills, who report food insecurity, and so on. This is especially true for the original population identified by Edin and Shaefer: people who fall below $2 per day if you don’t include food snaps, but rise above that line when they’re included.

“We find that the households removed from extreme poverty by in-kind transfers continue to be significantly worse off than poor households,” in terms of hardship metrics, Meyer and his coauthors write — likelier to report struggling with bills, getting food, etc.

Meyer and his coauthors conclude that this suggests food stamps are doing a good job of targeting the most deprived Americans. That’s definitely true — food stamps are extremely good. But the fact that this population struggles with affording necessities could also be used to argue for the kind of increased cash assistance that Edin and Shaefer have backed, particularly a child allowance that would offer cash support on top of the restricted food budget that food stamps provide people.