Krugman versus the Audit Heuristic
By Bryan Caplan
My two favorite liberals have repeatedly told me, “Paul Krugman is right about virtually everything.” When my favorite liberals praise Krugman’s position about a topic I don’t thoroughly understand, I hold my tongue. Recently, though, Krugman staked out strong positions on a topic I know intimately: the ideological structure of public opinion. And he’s factually wrong.
Krugman notes that there are four logically possible combinations of social liberalism and economic liberalism. He calls their adherents liberals (high on both), conservatives (low on both), libertarians (high on social liberalism, low on economic liberalism), and hardhats (low on social liberalism, high on economic liberalism). Since social liberalism and economic liberalism are positively correlated, the latter two categories are relatively rare. So far, so good. But then he leaps to the claim that the latter two categories are absolutely rare – about as rare as they’d be if social and economic liberalism were perfectly correlated. Krugman:
You might be tempted
to say that this is a vast oversimplification, that there’s much more to
politics than just these two issues. But the reality is that even in
this stripped-down representation, half the boxes are basically empty…
“Basically empty”? Gallup, Pew, and the American National Election Studies all disagree. Cautious definitions put the libertarian share of the U.S. population at 9-14%. Broader definitions put the share much higher. Libertarians themselves have done most of the data analysis, but you don’t have to trust us. Nate Silver concurs. Or check out some social and economic questions in the General Social Survey for yourself, and see how they correlate. For example:
1. “Do you think the use of marijuana should be made legal or
not?” (GRASS, “Legal”=1, “Not legal”=2)
2. “Suppose this admitted homosexual wanted to make a speech in your
community. Should he be allowed to speak, or not?” (SPKHOMO, “Allowed”=1, “Not Allowed”=2)
3. “If such a person wanted to make a
speech in your community claiming that Blacks are inferior,
should he be allowed to speak, or not?” (SPKRAC, “Allowed”=1, “Not Allowed”=2)
1. “Please indicate whether you would like to see more or less
government spending in each area. Remember that if you say ‘much
more,’ it might require a tax increase to pay for it. b. Health.” (SPHLTH, “Spend much more”=1, “Spend more”=2, “Spend same”=3, “Spend less”=4, “Spend much less”=5)
2. “Please indicate whether you would like to see more or less
government spending in each area. Remember that if you say “much
more,” it might require a tax increase to pay for it. g.
Unemployment benefits.” (SPUNEMP, “Spend much more”=1, “Spend more”=2, “Spend game”=3, “Spend less”=4, “Spend much less”=5)
3. “Generally, how would you describe taxes in America
today… We mean all taxes together, including social security,
income tax, sales tax, and all the rest. a. For those with high
incomes, are taxes…” (TAXRICH,”Much too high”=1, “Too high”=2, “About right”=3, “Too low”=4, “Much too low”=5)
Here’s the correlation matrix.
Mind the coding. On the social questions, lower answers are always more liberal. On the economic questions, lower answers are more liberal for spending, but more conservative for taxes. Social liberalism therefore statistically comes as a package, and so does economic liberalism. The correlations between the social and economic questions, however, are much smaller, and some have the wrong sign.
No doubt the low and irregular correlations partly reflect measurement error. As I’ve emphasized myself, low correlations between individual issues don’t disprove the one-dimensional model. But even if you put responses on common scales and sum them to reduce measurement error, the correlation between social liberalism and economic liberalism is only .06. If you put in a hundred questions, measurement error would largely wash out, so you might get the correlation up to .2 or .3. But getting it up to .5 would be like pulling teeth. Plenty of data miners have tried! And with a .5 correlation, the libertarian and hardhat boxes remain very far from empty.
When I judge wide-ranging thinkers, I often use the following rule of thumb. Call it the Audit Heuristic.
1. Read what they say about a topic I know very well.
2. See how reliable they are on that topic.
3. Assume that they’re about as reliable on topics I don’t know well as they are on topics I do know well.
For the case at hand, Krugman fares poorly against the Audit Heuristic. I don’t expect my favorite liberals to change their mind about him because one of his posts is mistaken. But my favorite liberals should at least admit that Krugman has given me reason to doubt his reliability. And I’d trust my favorite liberals more if they were quicker to concede Krugman’s specific flaws rather than sing his general praises.