2014-02-04

Minimum Wage Analyses

Robert Murphy has a good article at EconLib discussing the competing academic analyses of minimum wage effects. Daniel Kuehn is convinced in some respects and unconvinced in others. Something that stood out to me in Kuehn's blog post was this:
What's nice is that Bob's piece brings up Neumark and Salas, which I haven't discussed here. They show that alternative time trend specifications reverse some of these results (although I don't know if this is with a contiguous counties sample - Bob, do you know?). I don't know the paper well - my one questions is whether there is an overfitting issue - basically the Meer and West critique could very conceivably apply to Neumark and Salas.
I don't want to put words in Kuehn's mouth - I know he is very sensitive to that sort of thing. So I will only speak to my impression of what he's saying. It seems slightly funny to me because I think many of the points he has been making recently on minimum wage analyses could likewise be criticized from an over-fitting angle. I raised this point in the comments section of Kuehn's blog, and he remained skeptical. Fair enough.

But in the comments sections of various blogs, Kuehn has pressed me to explain why I prefer the simpler analyses of minimum wage legislation to these newer studies, the ones that Murphy calls "revisionist."

To me, the question is context-specific. Identifying regional variation in the context of a discussion about the usefulness of raising the federal minimum wage makes very little sense to me. What matters regarding the federal minimum wage are the national employment effects. If there is evidence for raising certain regional living wages, then that is a much different question. So before we choose our model, we should know which question we are attempting to answer.

This morning, I asked Kuehn in a comment why he felt that contiguous counties were good comparators in this sort of analysis. Kuehn replied that Tarrant County (for example) might not be a perfect comparator for Dallas County, but it's better than Arlington (Virginia) County.

But this is only true if a raise in the living wage in Dallas County has no direct effect on Tarrant County's labor market, which is obviously false. In the short run, the labor market in Dallas can expand at the expense of the Tarrant County labor market by using the living wage to induce Tarrant County workers to work in Dallas. If the transition to a new equilibrium isn't fully accounted for by the time-series data, then we would expect to see the living wage increase look very good for Dallas County; but we'd be drawing the wrong conclusion.

In the long run, we would expect to see more low-skill jobs migrate out of Dallas County and into Tarrant County - but only those low-skill jobs that have the ability to migrate. Food service jobs will obviouly not migrate because people in Dallas and people in Fort Worth both enjoy eating at Whataburger, and Whataburger wants a presence in both markets. So, again, if our studies focus on Whataburger jobs (and the like), then we will expect to see the Dallas County workforce grow at the expense of the Tarrant County workforce.

The important thing is that it is happening at the expense of the contiguous county.

To account for this kind of detail, I would prefer to see a study of Dallas County and all its contiguous counties (amalgamated) as Group #1, compared to Arlington County, Virgina and all its contiguous counties (amalgamated) as Group #2. Then we'd be able to see how a given policy impacts the expanded labor market defined by Group #1, and compare it to a comparable labor market (Group #2) that did not raise its minimum wage.

[MINOR UPDATE] --- What I mean in the preceding paragraph is that we need to account for job substitution away from contiguous-but-affected markets toward the county that makes the living wage increase. Then, having thus accounted for substitution, we can compare that to a valid control group that did not raise its living wage and the surrounding contiguous labor markets that would have been affected, if it were not a "control" group. (It sounds confusing because I'm bad at describing it.) --- [END OF UPDATE]

Do the contiguous county studies behave this way? I don't know - I'm not an academic, and I don't have much time to read the work. But my follow-up questions to others have not answered my question. Perhaps one of my readers knows.