David R. Henderson  

Krugman's Graph on Food Stamps

Trading With Foreigners: What'... Abrams on Investing...

Check out Paul Krugman's graph of participation in the food stamp (SNAP) program and U6 unemployment. MAKE SURE YOU LOOK ONLY AT THE GRAPH. Don't read his words around the graph.

Then ask yourself what you think the relationship is between the two variables.

Then, and only then, read what Krugman writes about the relationship, specifically his last sentence. What do you think? Did he get it right?

My answer in the comments below.

Comments and Sharing

CATEGORIES: Labor Market

COMMENTS (30 to date)
Alex Godofsky writes:

Is this a trick question? It looks like unemployment is a leading indicator of SNAP participation, and (I would assume) that there is a direct causal relationship where more unemployed people => more people qualify/apply for food stamps.

And... that seems to be pretty much what Krugman is saying.

Was I supposed to come to a different conclusion?

Peter writes:

I came to the same conclusion as his last sentence, without looking at his interpretation first. Is that incorrect? Interested to hear other interpretations however.

Bacon Wrapped writes:

According to the graph U6 unemployment has recently declined, and meaning that after a lag SNAP participation should decline too.

Tim Worstall writes:

I wouldn't trust any correlation between those two series. Not when I recall that eligibility for food stamps has been massively expanded in recent years (twice I believe, once under Bush and once under Obama).

Roger writes:

Yeah, what gives, David? Seems okay to me too. In part, that's because we figure a priori that U6 would lead. If you switched the labels I might still suspect that U6 is driving SNAP. I guess SNAP could lead if people planned ahead before quitting their jobs or in anticipation of being fired, but even then they're moving together as you'd except. In any event, I don't see where eyeballing the graph gives us any particular reason to doubt Krug's remarks.

Daniel Kuehn writes:

What's wrong with his interpretation?

Daniel Kuehn writes:

Not sure if this is what you're getting at, but the response is stronger on the up-swing than the downswing. This isn't particularly surprising to me since you don't have to be unemployed to get food stamps and the last several recoveries have been weak - so people may be getting jobs again but still weak enough jobs that they qualify for food stamps.

But I don't see how that changes the general claim that it follows with a lag.

If a lot of the people are just leaving the labor market (and they are) we might see the same thing in the coming years - a weaker correlation until we have real economic growth again.

That's the only thing I can think of.

Eric writes:

I agree with Krugman, and the other commenters. I think SNAP followed U6 with a lag before looking at his words.

Paul Stagg writes:

You can't make a causal relationship claim by just looking at a graph, but assuming there is a causal relationship between the two, I don't see anything necessarily wrong with his claim.

I'd like to see more about the causal relationship, though. Seems like it would make sense, it also seems reasonable that SNAP participation would not drop as significantly as increase because once on SNAP you may be less likely to come off.

There are plenty of opportunities to show how stunningly wrong Krugman can be - I'm not sure this is one of them.

But I'm also not that smart.

Hazel Meade writes:

Makes sense to me.
People go to SNAP only after being unemployed for a while and exhausting savings, thus there is a lag between U6 and SNAP.

Andrew writes:

Ummm.... if food stamp participation is dependent on unemployment with a lag, why is the lag 3 years in one case? Unemployment changed from increase to decrease in 2003 yet SNAP didn't follow until 2006. You can see it once again in 2010. Unemployment decreases, yet 2 years later, still an increase in SNAP participation.

Hopefully some of you also noticed that in 2006-2007, SNAP participation EXCEEDED U6.

What I see from this graph is that increasingly more, fully employed people are continuing to recieve SNAP assistance.

RPLong writes:

Yes, I think Andrew has it right. Unemployment rises and falls, but SNAP participation just keeps increasing.

lupis42 writes:

It doesn't follow with a lag - it follows up with a lag, but the decreases in U6 in the 90's are cotemporal with the decreases in SNAP, and the decrease in U6 from 2003 - 2006 shows no matching decrease in SNAP, lagged or otherwise.

Eric writes:


"SNAP participation normally follows U6 with a lag, and is in that sense behaving normally."

This is what Krugman claims. He never said SNAP never exceeds U6 at a fixed time-point. The 2003 U6 peak is higher than the 2006 SNAP peak. The 2009 U6 peak exceeds the 2011 SNAP level. David Hendersons question was whether Krugman got the relationship between the variables correct and I think he did.


2006 is a peak and 2001 is a deep valley, so I see no real evidence that SNAP just keeps on increasing. I do think the trend-line on SNAP is steeper than the trend-line on U6 and variance of SNAP is lower than the variance of U6, but Krugman made no claims about trend-line or variance. He said SNAP follows U6 with a lag, and I suspect the largest lagged correlations are at 2-3 years in the past. Using correlations in first differences would yield an even greater correlation (I suspect from the graph without doing any calculations).

Scott writes:

[Comment removed for supplying false email address. Email the webmaster@econlib.org to request restoring your comment. We'd be happy to publish your comment. A valid email address is nevertheless required to post comments on EconLog and EconTalk.--Econlib Ed.]

Glen Smith writes:

Looks to me that the most likely conclusion is that U6/SNAP relationship is changing in a way favorable for Krugman's arguments assuming his priors are true.

David R. Henderson writes:

My answer: Krugman is right for most of the time series. But look at the last 2 or 3 years. The relationship disappears. As U6 unemployment falls, food stamp participation rises. Krugman could argue that the length of the lag has changed dramatically, but he didn't argue that.

Krugman has, essentially, made the same mistake that Mitt Romney and Paul Ryan made during the 2012 presidential campaign. All 3 see unemployment driving SNAP, which was the traditional relationship. All 3 failed to take account of the changes in SNAP eligibility that were put permanently into law over George W. Bush's veto late in his second term. Casey Mulligan has written about this extensively.

DJ102010 writes:

David, how is this different from the earlier section?

In 2003 U6 peaks and SNAP doesn't stop increasing until 2006. In 2010 U6 peaks, but SNAP is still rising.

I'm not sure I follow your "The relationship disappears" claim given the justification that U6 is falling but SNAP is rising.

Ken P writes:

There was also a big push to increase participation rates of those eligible which bottomed at 54% in 2002 and were up to 75% by 2010.

There's an alternative interpretation for the lack of a drop in snap following the u6 drop. Many of those going back to work may have not been eligible for SNAP due to net worth.

Eric writes:


I agree with DJ102010. I see no qualitative difference between 2003-2006 and 2009-2012.

Jason writes:

Agreed with DJ102010.

Alex Godofsky writes:

Chiming in to also agree with DJ102010.

Mike Rulle writes:

He does not mention if graph is lag adjusted.

I am surprised no one referenced commenter "tanstaalf" who noted that U6 rose in 2013 while unemployment dropped in 2013----thus charging PK with cherry picking data. If true, DH's argument is stronger. I don't know who leads or lags between 1994-2000, but slope of U6 is more extreme downward. I assume this is related to "changing welfare in our time" during the Clinton Gingrich détente years. This could mean U6 "led" unemployment during those years.

Also, during 2003-2007, U6 is inversely correlated with unemployment as well as in 2010-2012/2013(if tanstaalf is correct). We should not be surprised there is correlation. The more irritating thing about PK is his insistence on being ideological and not try to explain why the change in "Munching Moochers" is more sticky than unemployment. Further, these two factors are "entangled" and have some mutual cause and effect depending on how U6 is administered.

Ricardo writes:

I agree with DJ202010 also.

Chad writes:

Inclined to give Krugman the benefit of the doubt on this one. If SNAP fails to respond to declining U6 in the next few years or so, there might be something to this point, but it seems unfair to force an assumed estimation of the lag on him in a one paragraph blog post.

ThomasH writes:

My conclusion is that that the availability of SNAP is not driving U6 as House Republicans seem to think. That is the reason that they were removed from the Farm Bill, right, to fight unemployment?

MikeM writes:

Snap benefits look essentially flat to me from 2005-2008, will be interesting if declining unemployment for this recession continues that trend or if overall snap participation to unemployment will be higher once we get lower unemployment.

JC writes:

Lag @2000 is 1yr. Lag @2003 is 3yrs. Lag @ 2006-07 is 0-1yrs. Current trend following U6 inflection @2010 is consistent with a 2-3yr lag. You may be wrong, David.

Tages Haruspex writes:

To my jaundiced eye, the lag going down is not equivalent to the lag going up. The inflection points on the down swing indicates that unemployed are on food stamps within a year but inflection points on the upswing indicate anywhere from 3 to 4 years to get off food stamps. This relationship would appear consistent. I must be missing something, it cannot be so simple.

Tages Haruspex writes:

I would love to see this graph over a longer timeline. It appears to me that some sort of disconnect occurred as the curves approach 2000. Until then the two curves appeared pretty synchronous. It would appear that the nature of U6 changes... as employment goes down, the need for SNAP not only persists, but increases; indicating the "gainfully" employed are in fact less than "gainful". One simple suggestion for the ensuing divergence is the surge in outsourcing that occurred in the late 1990s. Meanwhile, there are roughly 5 million fewer American manufacturing jobs now than at the start of 2001, and those that remain are paying lower wages according to any definition of purchasing parity. The ripple effect through the economy would indicate a dramatic increase in the ranks of America’s “working poor”. In other words, DH is correct; the relationship is progressively breaking down and SNAP participation in future will probably assume an autonomous upwards trajectory far less dependent of U6.

Comments for this entry have been closed
Return to top