Let's talk about this chart, which you've never seen before:
But first, some context. In yesterday’s post, I noted that rather than address the drops in poverty among single mothers and among children after welfare reform, many detractors wield a chart that looks something like this one from Jared Bernstein:
This chart is intended to show that while the Temporary Assistance for Needy Families program (TANF—the main cash welfare program in the U.S.) did not expand during the Great Recession or after, the Supplemental Nutrition Assistance Program (SNAP—or food stamps) responded as antipoverty programs should. It rose with unemployment, then continued to rise even when unemployment fell. It did so because the slack labor market—as evidenced by the declining employment rate in the chart—failed to provide enough jobs.
In yesterday’s post, I demonstrated that the poverty-reducing impact of SNAP during the Great Recession was far less important than the effects of tax credits and especially unemployment insurance. Whatever conclusions are to be drawn from this chart about how SNAP performed during the recession and after, the fact of the matter is that, today, work, unemployment insurance and work supports in the tax code are the most important bulwarks against poverty.
But on top of this, welfare reform’s detractors are drawing the wrong conclusion from charts like Jared’s. Let’s start, first, with the fact that this chart is supposed to look at responsiveness to the recession, but it includes a single data point prior to the recession. It’s hard to tell, then, what represents a counter-cyclical response, as opposed to a continuation of earlier trends.
The chart also divides SNAP and TANF recipients by the total population of Americans. That clearly mutes any changes that occurred in TANF receipt—which looks like a flat line in Bernstein’s chart—because few Americans are eligible for benefits. Divide a small number by a big number and you get a small fraction. Dividing SNAP recipients by the total population isn’t right either, but many more people are eligible for SNAP, so the impact there isn’t as great. The effect is to make TANF look completely unimportant relative to SNAP. Even with SNAP, dividing by the total population hides an important trend, as we’ll see.
Finally, since SNAP benefits clearly continue rising as the unemployment rate falls, Bernstein includes the national employment-to-population ratio on a second axis. The idea is that, “The decline in the unemployment rate has overstated improvement in the job market,” as Bernstein says. He scales the second Y axis so that the trend in SNAP receipt looks like a mirror image of employment. When employment falls, SNAP rises—see, it’s just responsiveness.
What’s the problem? Apart from the creative scaling (and, as we’ll see, other unhelpful presentation choices), the drop in employment is affected by the retirement of the baby boomers. Retirees get SNAP benefits too, but clearly if the intent is to say the SNAP trend simply reflects the state of the economy, you don’t want to compare it to a trend that also reflects other factors.
We can do better. Here’s my version:
I start in 2000, which was when the big drop in poverty leveled off with the onset of the recession. That lets me look at the pre-Great-Recession trends. The bottom line shows the change in the unemployment rate. It rises between 2000 and 2003 and then falls most of the way back to its 2000 level by 2006. What happens to SNAP receipt?* The second line from the bottom shows that it rose and rose and rose.
Now, it’s true that employment did not recover to the extent that the unemployment rate did. The dark line toward the top of the chart shows that among 25- to 54-year-olds, the share of non-working people rose by about two points from 2000 to 2006. But it did fall from 2003 to 2006. SNAP rose.
Starting in 2008, with the onset of the Great Recession, SNAP receipt began rising again. But it kept rising for three years after unemployment peaked in 2010. In 2013 it was at a record high, and it came down only a little in 2014.
“But wait,” I can hear Bernstein say. “That unemployment drop hides how bad the labor market continues to be. Because more and more people have dropped out of the labor market—and so are excluded from the unemployment figures—we need to look at the employment-to-population ratio.”
To this I say, first, that some of the decline in labor force participation has been voluntary. Less than 40% of working-age men out of the labor force, for instance, say they want a job in the Current Population Survey. Some of these guys are on disability benefits, so the fact that they don’t want a job doesn’t necessarily mean they could get one if they tried. But some of them could. Others are out of the labor force because they are in school or caring for family.
But no matter. You can see from the chart that through 2010, the trend in non-work is basically the same as for the unemployment rate, which isn’t surprising because it is mostly driven by the unemployment rate. After 2010, non-work falls less than the unemployment rate does. But it did fall from 2011 to 2014. SNAP receipt? Kept on rising until it fell in 2014. The chart clearly shows the trend for SNAP receipt rising more rapidly from 2000 to 2014 than the trend for non-work. That’s also true from 2007 to 2014. Bernstein’s chart obscures that because he is showing the trend in employment (100 minus the share not working) and using a creatively scaled Y-axis (which runs from 0.50 to 0.64).
So SNAP rose quite a bit more than indicators of labor market strength would suggest it should have. It didn’t so much respond to the recession as it resumed the upward march that had begun in 2002.
But let’s not stop there. Eligibility for SNAP is based on a number of factors, but recipients generally have to have income below 130% of the poverty line. If SNAP receipt rose because poverty rose, then dividing SNAP beneficiaries by the number of low-income Americans should show little to no rise. If poverty doubles and SNAP receipt doubles, SNAP beneficiaries per low-income person shouldn’t change.
Oh, but it does. The dashed line in the chart divides SNAP beneficiaries by the number of people under 125% of the poverty line (which is readily available from the Census Bureau site). This rate rises more than when dividing by the entire American population. (Note that this new trend is measured on a second Y-axis, on the right side of the chart. That’s only so it fits in the chart with the other trends without obscuring them.) The gap between this line and the dark line showing the non-worker trend nearly disappears by 2013. The conclusion is that the number of SNAP recipients grew by quite a bit more than the number of low-income Americans.
There’s one other thing—non-work may be what the economists call “endogenous.” That is, the rise in SNAP receipt may have caused non-work to increase. Bernstein is impressed by the strong correlation between the employment-to-population ratio and SNAP-beneficiaries-to-population ratio, but the causality he’s inferring may run the other way. AEI’sRobert Doar summarizes research suggesting that non-work in the U.S. may have increased more than it did in the U.K. because the U.K. has had a more restrictive safety net policy since the recession.
Actually, though, Bernstein’s correlation isn’t nearly as high as he says. If you correlate two rates that have the same denominator within each year, you will exaggerate the extent to which the things in the numerator are correlated. Even two randomly generated sets of numbers will have a strong correlation if you divide each pair by the same number. This problem doesn’t manifest itself in Bernstein’s correlation of SNAP receipt to unemployment, because the denominator in the former (total population) is different from that in the latter (total labor force).
In my own estimates, I found that the correlation between SNAP beneficiary rates and employment rates from 2007 to 2014 was -0.94—even higher than Jared’s -0.87 for 2007 to 2015. But when I correlated the numerators, it dropped nearly in half to -0.52. That’s still a strong correlation—as we would expect—but the direction of causation remains ambiguous.
One last point related to TANF. In my chart, the downward-sloping line that starts near the top shows the trend in TANF receipt, looking at TANF families as a share of single-mother families. As I discussed in my last post, TANF receipt clearly fell over this period. That was the hope of welfare reform—that more families would become independent through work, thereby reducing the rolls. As I showed in that earlier post, things went as intended, and poverty stayed below its pre-reform levels. But my chart reveals that TANF did respond to the recession. TANF receipt rose in 2009 and 2010.
Think that bump looks lame? I’ll repeat the chart here that leads off this post. It starts with 2007, as Bernstein did:
.png)
Which program looks like it responded most rationally to the economic downturn—TANF or SNAP? And hold off before you say that, since TANF receipt didn’t rise until 2009, the program only responded with a lag. If you look at the monthly data, TANF families actually rose beginning in August 2008, before Lehman Brothers’ bankruptcy set off the (not-so) “Great” part of the Great Recession. Between December 2007 (the official start of the recession, according to the National Bureau of Economic Research) and December 2008, TANF families rose by 3.6%. SNAP beneficiaries? The number rose by 0.9%. The bump in TANF receipt in my chart begins in 2009 because TANF receipt had dropped enough prior to August that the 2008 average was lower than in 2007.
I ask again: Where is the evidence that welfare reform increased, rather than lowered poverty?
* SNAP figures from https://www.fns.usda.gov/pd/supplemental-nutrition-assistance-program-snap. Population figures from the Census Bureau’s Current Population Survey.
This piece originally appeared in Forbes
This piece originally appeared in Forbes