The Hagedorn, Manovskii, and Mitman working paper on the effect of unemployment insurance (UI) on employment has been getting a lot of press lately. In brief, they find that the end of the federal unemployment insurance extension accounts for about 1.8 million new jobs in 2014.
Mike Konczal does a useful deep dive on the paper here and is very skeptical of the result. In particular, he criticizes as implausible and empirically inaccurate labor market search models that imply employer monopsony power, which are essential to the plausibility of the result. These models are also essential to the revisionist literature on the minimum wage, holding that minimum wage increases do not reduce low-productivity workers’ employment. Curiously, Mike Konczal has defended search models in this aspect. He’s a smart guy and clearly thinks that applications of search theory to macroeconomic variables have problems that the application to the minimum wage doesn’t – but if search theory badly explains one phenomenon, it’s unlikely to do well explaining another. There’s a clear tension between claiming simultaneously that employer monopsony power explains why raising the minimum wage doesn’t reduce employment and that ending UI can’t have increased employment so much because employers don’t have that much monopsony power, even if the latter claim is limited to slack periods in the business cycle. (Why wouldn’t employer monopsony power be greater during slack periods in the business cycle? The Marxist concept of the “reserve army of the unemployed” comes to mind here.)
Another interesting parallel between the UI and minimum wage research is that the famous Dube et al. paper in Review of Economics and Statistics relied heavily on matched-border-county estimates, as does the Hagedorn et al. paper. Having looked at these data, I actually agree with Konczal that these models are inappropriate. The logic behind using matched border counties is that contiguous counties are alike in every relevant way other than the policy discontinuity associated with the state border (say, one county has a high minimum wage and the other does not). But border counties are actually usually quite dissimilar. Take Camden, N.J. and Philadelphia, Penn. These two counties are vastly different in size, so if Camden creates jobs at a higher rate than Philadelphia, Camden’s new jobs might still be a tiny percentage of Philadelphia’s. Yet the Dube/Hagedorn approach considers these counties to be equivalent, and takes the larger percentage increase in jobs for Camden as an indication of superior New Jersey employment policy. (See also David Neumark on empirical evidence that border counties are not appropriate control groups.)
In summary, if you are skeptical of the empirical strategy and theoretical justification of the literature saying the minimum wage has no negative employment effects, you should also be skeptical of the empirical strategy and theoretical justification of the new paper showing that unemployment insurance has big disemployment effects. If you like the Dube et al. minimum wage work, you should like the Hagedorn et al. UI paper. How many wonks are intellectually honest enough to adopt one of these two, ideologically inconvenient pairs of positions?
PILEUS 
Model, data, techniques. Generally you want two of them down solid. Dube and HMM have the technique down (let’s stipulate, I don’t have the expertise to comment on county-county stuff). Dube also has good data, and some room with the model. I don’t see HMM have those other two under control. Points:
– The data and the business cycle control is the real kicker here. LAUS can’t do the work they need it to, and it’s not clear they are distinguishing between UI receding and “recovery” more generally.
– Dube et al show that quits fall with a minimum wage increase, which gives empirical support to their search model (people keep their jobs longer, reducing unemployment levels). HMM don’t show that wages fell where employment grew the fastest, or given any other support for their search model. This model is doing double-work in this 2014 version, since they need it to explain how so many people without UI got jobs.
– There are a couple of stories that can get you a muted minimum wage effect on employment outside search models. For example, prices go up, and bosses get more productivity out of their workers. These stories make sense.
The basic model story isn’t very clear here: how does X getting UI extensions raise the (reservation) wages of uneffected Y? Also the central story in HMM hits a wall with the general empirical observation of the procyclicality of wages and job growth in recessions. The search models they rest on are very well known for not being able to explain business cycles. They should kick the tires more.
Thanks for the comment, Mike. The control variable/regression to the mean problem sounds important, I agree. I think Dube et al. face a similar problem, though, in the complicated polynomials of time they create. The more “kitchen-sink” your controls, the more you risk biasing down the coefficients on your variables of interest (overfitting). The quits result after minimum wage boosts could fit a standard competitive labor market model too, if workers see that it’s become more difficult for them to be hired elsewhere.
The last word has surely not been written on either topic; we’ll see where the research goes from here…
Also thanks for commenting – this is an important follow-through point to discuss.