After I responded to Professor Caplan’s post defending an orthodox interpretation of the minimum wage, we had a small conversation on Twitter:

Bryan: My argument is that we should consider all relevant evidence, and most minimum wage researchers don’t.

Ashok: But the evidence is only relevant if std theory applies, and unless one has strong priors that has been ‘refuted’, relevant evidence would be far more valuable w/o Krueger/Card, to the extent one trusts empirics

B: Empirical work “refutes” nothing. All it can do is raise or lower probabilities.

A: Absolutely agree, but then you can’t claim that your argument works for those of us w/o strong priors, as you do.

B: How do you figure? If you have 50/50 prior on employ effect of min wage, why doesn’t totality of evidence sway you?

A: Because of empirics from Krueger. Once my priors change, value of “relevant” empirics diminishes as it begs the question

B: Presenting new info never “begs the question.”

I’m using this blog post to explain more thoroughly the point I was trying to make earlier, and 140 characters won’t suffice. 

Here’s the argument for a standard approach to minimum wage and relevant effects on employment:

But suppose you disagree with me on both counts.  Suppose you have a weak prior about the disemployment effects of the minimum wage.  Suppose further that you think that the best empirical work in economics is very good indeed.  Doesn’t existing evidence then oblige you to admit that the minimum wage has roughly zero effect on employment?

Let’s break this down logically. Let the proposition P be that labor markets operate under orthodox theory and hence wage controls cause unemployment. Everyone has some prior on this proposition. As it happens, Caplan’s beliefs are strong, and mine aren’t. So I update my priors quite a bit more significantly considering Krueger and Card, but this isn’t really important to the debate at hand.

Caplan is arguing that I should change my beliefs based on price controls in general, labor market regulation, and immigration. Here’s why I think his argument is wrong:

• Caplan is arguing that P is true
• He is, further, arguing that a plethora of other evidence stand as fair empirics against the Krueger study which clearly shows no disemployment effects on minimum wage (or would at least adjust one’s priors to towards that conclusion).
• In citing the evidence that the labor demand curve for immigrants is downward sloping and hence suggesting that wage controls cause unemployment, one has to assume that labor markets operate under normal conditions, which is precisely the question at hand.

If wage controls are fundamentally different, as Krueger/Card shows, it really doesn’t matter what the rest of an economy works like. The idea that you can extrapolate how labor markets work from rent controls and agricultural subsidies is irrelevant, because we’ve defined our proposition specifically in the case of a labor market.

I was wrong in suggesting that this “begs the question”, considering the Bayesian nature of this debate. However, one would have to have really strong priors to ignore Krueger/Card and hence extrapolate empirics from other industries onto labor, but he can’t then claim that:

But suppose you disagree with me on both counts.  Suppose you have a weak prior about the disemployment effects of the minimum wage.  Suppose further that you think that the best empirical work in economics is very good indeed.  Doesn’t existing evidence then oblige you to admit that the minimum wage has roughly zero effect on employment?

Further, even in the empirics he cites, there’s a missing dimension in his reasoning: humans are not capital:

If you object, “Evidence on rent control is only relevant for housing markets, not labor markets,” I’ll retort, “In that case, evidence on the minimum wage in New Jersey and Pennsylvania in the 1990s is only relevant for those two states during that decade.”  My point: If you can’t generalize empirical results from one market to another, you can’t generalize empirical results from one state to another, or one era to another.  And if that’s what you think, empirical work is a waste of time.

Even if we establish that it’s okay to generalize empirical results from one market to the other – and, as was my point above, that’s not possible without assuming your conclusion – this seems like a pretty remarkable statement. There is nothing qualitatively different between New Jersey and Pennsylvania, or the 1990s and 1960s. There is something qualitatively different between a human being and a machine. For one, it’s illegal to enslave own one of them.

Because standard economic theory applies in one particular market certainly doesn’t suggest that every single commodity or good (if you can even call labor that) operates under these conditions. In the most comprehensive study conducted on wage controls, we have strong reason to believe that these very conditions are irrelevant.

I cede that if one has strong priors and a distrust of economic empirics, the dominant belief might still be an orthodox interpretation of the minimum wage. But if one has strong enough priors he can believe just about anything. When so much empirical evidence casts doubt on economic theory, I don’t understand the need to study “relevant” data, which is useful only to those with such high priors.