Wednesday, June 21, 2006

"The economy is a terrible experimental design" 

I call your attention to this Bogus Gold Editorial in which Doug Williams tries to tear a new one for the STrib in its rather ill-informed editorial in favor or a minimum wage increase. He is particularly incensed by this line from the STrib:
Does the minimum wage punish employers and reduce hiring? In theory yes, in reality no. After the last minimum-wage increase took effect, in 1997, the nation experienced the strongest jobs recovery in 30 years. During the last decade, states with high minimum wages have had better job creation than states with low minimum wages.
Of course the second two sentences are pure post hoc fallacious reasoning, and Doug gets off a good shot or two at this silliness. But I will come a little bit in the STrib's defense, insofar as the most cited work on the minimum wage in the last fifteen years does indeed say that we have relied too much on theory and not enough on the data. The book is almost always referred to when someone wants to make an argument for raising the minimum wage. "Just a little, it won't hurt," will be the refrain.

The problem is, the book is at least anomalous, if not just plain wrong. Russ Roberts reminds me of an old review of the Card and Krueger analysis -- the whole basis for the STribs claims about "reality" -- which is an attempt to generalize claims from an earlier paper by the same two economists that relied on a phone survey over eight months of New Jersey fast food workers. That's the "reality" on which the evidence is based. The review by John Kennan that Roberts links to (it's a subscriber link so most of you not on campuses won't get access to it) demonstrates that the "natural experiments" that Card and Krueger argue refutes theory have been done before and have shown opposite results. The other method more traditionally used -- using data over time from a full range of states, rather than looking for effects of one-off changes in minimum wages in a single state, still hold up the general theory that a 10% increase in the minimum wage reduce employment of teens by 1-3% (and more recent results put it at the lower end.) We economists are pretty sure the effect of increasing minimum wages isn't very negative, but we're pretty sure it isn't positive either. The problem is that you are trying to detect a relatively small signal of minimum wage effects in a whole lot of noise from teen wages (which fluctuate quite a bit.) One study finds a positive effect (maybe) for NJ teenage burger-turners and extrapolates to the whole of the country, while ignoring other studies using the same methodology that don't show the same results (Kennan demonstrates this for California retail salespeople.)

The point of all this is that, as the title of this post says (it's a quote of Nobel Laureate Robert Lucas that was in the Kennan article), natural experiments are seldom as clear cut as we make them to be. They are very seldom generalizable.

One more part of the STrib editorial that grates:
Granted, Congress needs to be careful when meddling with market forces. But it also needs to assure that the economy is delivering for all Americans, and today it plainly is not.
Sirs, please name me an economy that did, ever. All economies have diffusion of results. The boom years of the 1920s were a bad time for buggy-whip manufacturers; the 1990s saw losers to economic restructuring. Such pablum is an excuse for exactly the meddling you think should be done with care.