mininum wage econometrics is hard

This is nominally about the minimum wage, but I encourage reading it in a more broadly epistemological cast of mind — this is work with which I was previously familiar, but it's well-described here. The upshot is that, with noisy data and confounding factors, blunt attempts to control for the confounds may well "control for" most of the signal as well, leaving mostly noise — only noise if you try to average out the noise in the wrong way.

More narrowly, the recent evidence suggests a way out of an interesting paradox in some of Card's work on the response of labor markets to presumably exogenous shocks:

• A paper looking at a large influx of immigrants to Miami in 1980 found no effect on local wages; in traditional terms, this suggests that labor demand is extremely elastic.
• In a famous paper using data from New Jersey and eastern Pennsylvania before and after an increase in the state minimum wage, he suggested that the increase in state minimum wage had in fact increased employment in low labor jobs.
• A later paper by Card, with more data from different time points, suggested that the data series for New Jersey and eastern Pennsylvania are both noisy enough and weakly enough correlated over shortish periods of time that any statistical significance in the previous study was illusory; to the extent that we call this "there is no net effect on employment from raising the minimum wage", again forcing it into a comparative statics supply-and-demand framework, this suggests that labor demand is extremely inelastic.

All of this — the elastic (short-term) response to an increase in supply, the inelastic (short-term) response to a minimum price, and the weak (short-term) coupling between series that might be expected to be similar — makes a fair amount of sense in the modern search-theoretic models of labor that capture the fact that (most) people don't get up each morning, immediately congizant of which employer is willing to pay the most money for their labor that day, and go to work for that employer; instead there's a lot of what you might elsewhere call "repeat business".  It appears that the main effect of higher minimum wages is, in fact, to reduce growth in employment, and that it takes a while for differences in levels to show up; moreover, if you "correct for" the difference in job growth between different jurisdictions, you can swamp much of the effect, even if you take some care not to.

If you believe in a strong income effect among minimum wage employees, perhaps the right thing to do, then, is to raise the minimum wage when it is expected that job market conditions will, for other reasons, be improving over the next year or two; you provide a boost in income now, and create slack in the job market in the future.  (This suggestion is meant to be a bit cheeky; even to the extent that I think these partial-equilibrium Keynesian arguments are correct, and that welfare losses from the business cycle are sizable compared to microeconomic deadweight losses, I'm skeptical that policy can be timed well; cf. the famous projections of the unemployment rate with and without the 2009 stimulus.)