Jeffrey Frankel explains the real reason that Republicans support sugar subsidies

by Andrew Gelman on September 20, 2012 · 8 comments

in Campaigns and elections

Jeffrey Frankel, a distinguished economist, public servant, and expert on international trade, made a common but, I hope, avoidable statistical error in a recent column. Frankel wrote:

Compared to “blue-staters,” those who live in red states exhibit less responsibility, on average, in their personal behavior: they are less physically fit, less careful in their sexual behavior, more prone to inflict harm on themselves and others through smoking and drinking, and more likely to receive federal subsidies . . . Blue-state residents, who tend to be more educated and have higher incomes than residents of red states . . .

This is all fine, but Frankel is making a common, and important, mistake, which one might call “personifying the states.” In fact, as we describe in detail in our book Red State Blue State, differences between state averages do not necessarily reflect individual differences. For example, blue state residents have higher incomes, on average, than red state residents; but Democratic (“blue”) voters are poorer, on average, than Republican (“red”) voters. Even Mitt Romney knows that!

Similarly, it is misleading to write, “Statistical analysis shows that states where more residents suffer from obesity, often because they get less physical exercise and eat more junk food, tend to vote Republican. To illustrate, a mere 1% decrease in a state’s obesity on average is estimated to raise the ratio of Democratic to Republican voters from 1.00 to 1.07”—an argument that conflates individuals and states and confuses correlation with causation.

Frankel’s mistake is an easy one to make; others who have confused state-level with aggregate patterns include respected commentators Nicholas Kristof, Michael Barone, and Tucker Carlson. Seeing this mistake made by a leading scholar and former member of the Council of Economic Advisors has motivated me to post on this topic once again.

P.S. See here for the sugar-subsidy story.


Paul Crowley September 20, 2012 at 11:31 am

You decided against the sarcastic version, in the end? It was fun, but that’s likely the right call – it’s my experience that sarcasm can hide flaws in the argument.

Mavis Beacon September 20, 2012 at 11:37 am

It seems like the point Frankel is trying to make, albeit poorly, is that outcomes like obesity and smoking rates that he labels hallmarks of personal responsibility have a lot to do with state-level policy. If you’re trying to make the point that blue state policy encourages good outcomes and red state policy does not, then state-level comparisons are appropriate.

Andrew Gelman September 20, 2012 at 11:40 am


Maybe I missed it, but I didn’t see anything in Frankel’s article about state-level policy.

Peter Hovde September 20, 2012 at 1:23 pm

And here I thought sugar subsidies were a function of the electoral college, and Florida’s swingy status and large number of electoral votes.

Kevin Hill September 20, 2012 at 6:20 pm

I liked the snarky version better.

BTW, I use this sort of stuff in my Graduate Methods class to illustrate the ecological inference problem.

My personal favorite is using the states as the unit of analysis, and then correlating the % vote received by Obama in 2008 with the state % black. The correlation is NEGATIVE (and anyone who has ever studied the South knows why).

Andrew Gelman September 20, 2012 at 8:29 pm

Now, if we could just get Jeffrey Frankel to take your class . . .

LFC September 20, 2012 at 10:46 pm

“To illustrate, a mere 1% decrease in a state’s obesity on average is estimated to raise the ratio of Democratic to Republican voters from 1.00 to 1.07”

This is misleading, as the post indicates, b/c it implies that obesity and voting patterns are somehow causally linked and b/c it could be accused of fuzzing the state-individual distinction (first clause referring to “a state’s obesity,” second clause to “voters”). None of Frankel’s other sentences quoted in the post, however, are misleading in this clear a way, it seems to me. He doesn’t say Democratic voters have higher incomes on average, he says residents of blue states do, which is true. If the reader infers that Frankel means Democratic voters when he doesn’t say it, that’s at least partly the reader’s fault.

As for “personifying the states”: Suppose state X has Y residents, 90 percent of whom smoke (unrealistic but just a hypothetical). State Z has Y residents, 10 percent of whom smoke. In this case there would be nothing wrong with referring to state X as “a smoking state.” In other words, while differences on average among states don’t necessarily reflect differences among individuals, the state-level differences may in particular instances reflect differences among individuals, and when that’s the case I can see nothing wrong with “personifying the states.” The problem is not with personifying the states per se but with doing so without first determining whether the practice is justified in a particular instance.

Jeffrey Frankel October 2, 2012 at 8:25 am

Thank you LFC. It didn’t occur to me that anyone would interpret the causal implication of that one sentence literally (the implication that raising the obese share of the population causes an increase in the Republican vote). I guess it should have occurred to me. I believe that the rest of my language in that op-ed is more careful — and also in related op-eds that I have written for the Christian Science Monitor, VoxEU, and my own weblog. (Commenters on those sites have not had the same interpretation problem.)

Since The Monkey Cage appears to have missed the point I was trying to make, I will state it more clearly: My point is to refute allegations from Mitt Romney (and a long line of other politicians from his political party) that Americans who vote Democratic tend to be more dependent on the federal government and to show less personal responsibility on average than other Americans. (Actually Romney didn’t say “on average.” or “tend to.” But I figure it is only fair to interpret him that way, since his comments were extemporaneous and private.) The statistics show, if anything, the reverse correlation. I was not offering any causal interpretation.

Believe it or not, I do understand the difference between correlation and causality. (For what it is worth, I studied econometrics at MIT and taught it at UC Berkeley.)

And, yes, of course statistics based on individual data are more reliable than statistics based on aggregate state data.


Comments on this entry are closed.

Previous post:

Next post: