This article originally appeared in Forbes.
To a greater extent than we often admit, policy debates between left and right often run aground simply because we lack evidence sufficiently strong to adjudicate between competing arguments
. So while one participant may convincingly beat back Claim A, the other retreats to Qualifier B or New Claim C. The empirical debate continues until Qualifier F is reached, at which point the honest response is often: “Maybe.”
I am involved in a small but important project with a group of bipartisan researchers, thinkers, and advocates who are trying to come to some consensus around the problem of poverty and to better understand and reconcile the values and evidence that each side brings to the table. Ultimately, the hope is that by building up camaraderie and learning from each other, we will be able to help forge a new, broadly popular (or, at least, acceptable) anti-poverty agenda that will make a difference in the lives of the poor. It is a bit of an experiment probing the extent to which political and ideological polarization can be surmounted.
The hard part for this group, I believe, will be what we do when we get to the point of “Maybe,” as we routinely will. At least that is my hope. The worst case scenario is that one or the other side—routinely—will be unable to agree that Claim A is just empirically wrong. In other words, we won’t even get to Qualifier B or New Claim C.
I thought of this project as I read Tom Edsall’s latest column on work-promoting anti-poverty policy, titled, “Cutting the Poor Out of Welfare.” I have deep respect for Tom, and his books on post-1980 political realignments have influenced me greatly. There is a debate to be had over the costs and benefits, the winners and losers of an American safety net policy that has become less generous toward non-workers and more generous to workers. Much of the research Tom cites in his column is unimpeachable, and he raises important issues about the ways we have redistributed aid in the last twenty-five years. But Tom’s central contention—that work-promoting welfare reform has been “devastating” for the poorest of the poor and ineffective for the rest—has almost no basis in fact.
Tom cites the work of the great poverty scholar Robert Moffitt. Moffitt, in research presented in his presidential address at the recent Population Association of America annual meeting, showed that anti-poverty spending at the federal level has become more generous over time. Between 1986 and 2007, on a per-capita basis it almost doubled. It even grew as a share of GDP, from 9 percent in 1985 to 12 percent in 2007. And that excludes Medicaid, Medicare, Social Security (including disability benefits), unemployment insurance, and worker’s compensation. Cash welfare to single mothers fell, but the Earned Income Tax Credit (a supplement for low-income workers), Supplemental Security Income (for poor children and adults with disabilities), the Child Tax Credit, and the Supplemental Nutritional Assistance Program (food stamps) all increased in generosity.
However, Moffitt also found that while the very poor still receive more means-tested benefits than other low-income groups, their share of such spending declined between 1983 and 2004. Anti-poverty policy has become more focused on the better-off part of the low-income population, largely because of the emphasis on helping working families. Tom cites estimates indicating a drop in the share of benefits received from 56 percent of benefits to 32 percent among those with pre-tax and -transfer incomes that put them less than halfway to the poverty line. (It’s not clear the group to which those figures refer, but it appears to be non-disabled, non-elderly families.) From the numbers in the presentation, you can compute that among this group of families, benefits dropped by 35 percent between 1983 and 2004 among those with pre-tax and -transfer income below half the poverty line.
Note, though, that this extremely poor group is defined with respect to their pre-tax and -transfer income. That is important because the poverty line is set after taking into account cash transfers. People whose income before transfers put them under 50 percent of the poverty line do not necessarily have after-transfer income less than 50 percent of the poverty line. Adding cash transfers to their incomes would put many of them above half the poverty line, and adding non-cash transfers (which the poverty line ignores) would put more above that level.
In that regard, it is worth pointing out that Moffitt’s figures on federal benefits do not include Medicaid, which has become more generous since 1983. The drop in spending for Moffitt’s poorest group would look less severe (smaller than 35 percent) if Medicaid benefits were included, though there is disagreement about how to value those benefits. The decline in the share of benefits received by the bottom would also look smaller. Moffitt says that for consistency, if Medicaid benefits are to be added in, so should employer-provided health insurance, which would leave the distribution of benefits little changed. But that seems clearly wrong. Including the full value of employer-provided health insurance in pre-tax and -transfer income would move many families above twice the poverty line, which is the cut-off for inclusion in Moffitt’s study, so the share of benefits going to people above the poverty line but below 200% of the poverty line would presumably fall. For that matter, only the part of employer-provided benefits that corresponds to a federal tax break should be counted as a government benefit, not the whole amount.
Not only is this poorest-of-the-poor group better off—and better off over time—than Moffitt’s figures suggest, it is a smaller group over time. The number of families with no pre-tax and -transfer income has surely diminished as our anti-poverty policy has come to promote work heavily.
More to the point, the central question is whether there are more or fewer families living in poverty or extreme poverty than in the past. Here, Tom’s figures obscure what has happened. He cites excellent research from a Columbia University team showing that while the poverty rate dropped from 27 percent to 16 percent between 1967 and 2012, it would have risen to 29 percent absent federal transfers and tax policy changes. But these are figures for all adults, and the impact of transfers here is overwhelmingly the impact of Social Security benefits for the elderly. That is irrelevant to the debate over work promotion and how it has affected poverty. Figure 8 of the paper Tom cites shows that even before taking taxes and transfers into account, child poverty is no higher today than in the early 1980s or early 1990s. Incorporating them makes the poverty rate among children fall from about 30 percent in 1983 to less than 20 percent today. Figure 9 tells the same story for the share of children living under 50 percent of the poverty line.
This research is based on a less-widely used (but preferable, in my view) poverty measure than the one most poverty researchers prefer. The latter—the “supplemental poverty measure”—incorporates the spending of non-poor families into the poverty line and thus is affected by changes in inequality too. But the version of this measure created by the Columbia researchers in a second paper also shows child poverty no higher today than in the early 1980s or early 1990s even before taking taxes and transfers into account (see Figure 5a). Deep poverty among children is lower today than in the early 1990s before taxes and transfers (Figure 7).
To further argue that hardship at the very bottom has increased, Tom leans heavily on a paper by Luke Shaefer and Kathy Edin that finds “extreme poverty” rising between 1996 and 2011. I am a fan of Edin, whose ethnographic research provides invaluable insight into the lives of poor men and women. I don’t know Shaefer, who is the quantitatively-oriented half of the duo. But this paper simply isn’t credible.
Their measure of extreme poverty is living on $2 per day or less for an entire month or for three months. On its face, it is difficult to believe that anyone in the United States gets by on $2 per day for an entire month using any reasonable definition of “income” unless they have significant savings. At the extreme, even the homeless are generally able to scrounge $2 a day, whether that comes from in-kind assistance from shelters and soup kitchens or from the change given to them by strangers. To be clear, I am not asserting the homeless somehow live well or deserve their poverty, only pointing out that reported monthly incomes of $60 in household surveys (which often exclude shelters and other “group quarters”) are to be viewed very skeptically. In fact, Shaefer and Edin indicate in a footnote that two-thirds of families supposedly getting by on this “income” receive federally subsidized health insurance, which they do not include as income. That alone is surely worth $60 a month to those families.
Apart from the income definition, underreporting of incomes, particularly among the poorest households, is a very real problem in household surveys such as that used by Shaefer and Edin. While the survey they use is probably better than the alternatives, the research of Bruce Meyer and James Sullivan demonstrates that the underreporting of incomes among poor families in household surveys dwarfs whatever differences there are between surveys. (Indeed, Edin’s early research on welfare recipients, conducted in part with her and my dissertation advisor, Christopher Jencks, is some of the most compelling evidence on that point.)
More relevant for Tom’s argument, the under-reporting problem appears to be growing over time, which makes income trends look worse. In one influential paper, Meyer and Sullivan found that the reported income of the poorest tenth of single-mother households in the early 2000s was 25 percent lower than the income of the poorest tenth in the mid-1990s. But consumption in the poorest tenth (whether ranked by reported income or reported consumption) was unchanged.
But put all that aside and assume that there is no underreporting to speak of and no problem with the income definition. Even then the Shaefer/Edin estimates would be too imprecise to rely on. Remarkably, though the paper was published in a social work journal, there are no references to “standard errors,” “confidence intervals,” or “statistical significance” in the paper. That is to say, the authors apparently did not check (and were not asked by reviewers to show) that their results may be safely assumed to be real rather than the result of the chance makeup of the households actually interviewed. When sample sizes are small, there is always a risk that the particular sample of households interviewed does not actually represent well the broader population of interest. In that case, any result obtained could just reflect the weirdness of the sample rather than a real-world finding.
Shaefer and Edin report a couple of sample sizes in a footnote, and they may be used to roughly estimate whether their finding of a rise in extreme poverty might have simply been a statistical fluke. When I look at their results that include food stamps, tax credits, and housing subsidies (Medicaid is omitted), the increase from 1.1 percent of households in extreme poverty to 1.6 percent of households is way too small to count on given how few households there are who report such low incomes. I’m not even sure the 1.1 and 1.6 percent figures are reliably different from zero, let alone whether the 0.5 point change is.
In short, the Shaefer/Edin study should not raise doubts about the careful work of the Columbia University researchers. There is no reason to think that extreme poverty has grown.
In addition to arguing that work promotion has worsened poverty, Tom argues that it has been unsuccessful on its own terms—a smaller share of the poor work today than in the past. This is like arguing that an online dating service is ineffective at matching 30-year-olds up by pointing to all the single 45-year-olds using it who haven’t met their match. The most successful work promotion policy possible would result in 100 percent of the poverty population consisting of non-working families, because it would move everyone who can work into employment and raise them above the poverty line. The poverty population would be much smaller, and that would be a great achievement.
It simply makes no sense to look at employment rates among the poor to assess how work-promoting changes in anti-poverty policy have affected disadvantaged families. The way to approach this question is to look at employment rates of less-educated people, under the (conservative) assumption that welfare reforms haven’t increased educational attainment. A Department of Education analysis of Bureau of Labor Statistics data shows the basic trends:
The charts (Figure 4 in the report) show the percentage of each group that is employed, for men and women between the ages of 25 and 64. Work promotion as a policy has primarily been aimed at single mothers receiving cash assistance; there are few work requirements for food stamps, none for housing assistance in all but a few places in the country, and none for other means-tested programs. There has been little decline in employment, if any, among working-age women without a high school diploma (and certainly not until the recession). Even among men, employment among high school dropouts was as high before the recession as it was in 1990. The data just do not tell the story that Tom wants to tell.
To repeat, despite the redistribution of federal spending from the poorest of the poor to the better-off poor, there has not been an increase in the number of the poorest poor and there has been a decline in poverty that is only partly due to more generous transfers. As I have argued in the recent Room to Grow volume published by the YG Network, the federal safety net apparently did not do much to reduce the child poverty rate until the 1990s—when work-promoting safety-net reforms began in earnest. Neither changes in work nor changes in poverty provide reason to think that the poorest of the poor are worse off today than they were before welfare reform, and the poverty figures suggest that many are better off than they would have been.
Now, Tom or someone else may have a Qualifier B or might concede that Claim A was wrong but New Claim C is something we need to take seriously. Fine, we can then turn to B or C and maybe get to Qualifier F where we can admit that we have taken the empirical question as far as it can go. That is the only way we will bridge left-right disagreements. If evidence is important for policy formulation and evaluation, we must.
Scott Winship is the Walter B. Wriston Fellow at the Manhattan Institute for Policy Research. You can follow him on Twitter here.
Interested in real economic insights? Want to stay ahead of the competition? Each weekday morning, e21 delivers a short email that includes e21 exclusive commentaries and the latest market news and updates from Washington. Sign up for the e21 Morning eBrief.