Criticisms of the Seattle Minimum Wage Studies Miss the Mark

I recently did a write-up of two studies on the Seattle minimum wage increase. These studies threaten to overturn a lot of prior studies and prevailing attitudes about minimum wages. It's not likely that the "minimum wages have no ill effects" constituency would take this lying down. I wanted to see what I could find in terms of criticisms of this new research. I found this commentary by the Economic Policy Institute (EPI), written by Ben Zipperer and John Schmitt. Their initial summary:

The authors’ analysis, however, suffers from a number of data and methodological problems that bias the study in the direction of finding job loss, even where there may have been no job loss at all. One initial indicator of these problems is that the estimated employment losses in the Seattle study lie far outside even those generally suggested by mainstream critics of the minimum wage (see, for example, Neumark and Wascher [2008])—as the authors themselves acknowledge.

The authors of the Seattle study (Jardim et. al.) are very upfront about the limitations of their data set (details below). And there is a long discussion of why their results differ from previous results. In a nutshell, the new studies could accurately measure actual hours worked, while previous studies could only detect job losses. Also, the new studies could directly measure individuals’ wages, whereas prior studies had to settle for statistical aggregates (“teenage workers” or “restaurant workers”) that  have heterogeneous wages and thus dilute the impact of a minimum wage hike.

The first bullet point in the EPI piece:

The employment responses estimated by the authors are well outside the bounds of most published research, and indeed all of the research cited by the authors implies much smaller and even no employment changes in response to wage increases similar to those experienced so far in Seattle.

Once again, they explain why they get a different result. Intuitively, it makes sense that you’d see a bigger effect if you can measure hours worked instead of binary “employed/not employed.” It seems reasonable that it’s easier to cut hours than to actually lay off workers. Also, the study was able to actually measure wages. Like I said in my previous post, they were able to recover the standard effect sizes seen in the minimum wage literature when they looked at “restaurant workers” as a whole, instead of identifying low-wage workers according to their actual wages. This casts doubt on the prior minimum wage literature, not on the new study.

Second bullet point:

The study implausibly finds employment changes due to the minimum wage in parts of the labor market where there should have been none. The study’s own estimates inaccurately imply the minimum wage caused large gains in the number of jobs paying above $19.00 per hour and in the number of hours worked in those jobs—even though those jobs are well above the wage range where the $13.00 minimum wage should be having measurable effects.

This is a bizarre framing.  The study repeatedly cautions the reader that Seattle was undergoing an economic boom, which presumably was driving up high-wage employment. The authors of the study do not say or imply that increased employment above the $19/hour mark is due to the minimum wage. Only someone robotically interpreting their model to mean “everything after the minimum wage increase is caused by the minimum wage increase” would draw such a conclusion, and this would require ignoring the multiple warnings in the paper about Seattle’s booming economy. What’s stunning is that low-wage employment declined despite this boom in the labor market. See Figure 1: “Finding a Reasonable Threshold” and the discussion within the text of the paper in Minimum Wage Increases, Wages, and Low-Wage Employment: Evidence from Seattle. 

Here is some commentary from the paper:

While the preponderance of evidence suggests that a low-wage threshold slightly above the statutory minimum poses little risk of miscoding jobs as lost when they have actually been promoted to higher wage levels, in our preferred specifications we report findings based on a relatively conservative $19 threshold.

I'm reading this as saying that the disappearing jobs below the new minimum wage are mostly funneled into jobs slightly above the new minimum, insofar as they aren't completely lost. There's no way those disappearing jobs below the minimum wage were funneled into much higher wage categories.

Here is Figure 1 itself:

A few prominent features present themselves in this figure. "Hours Worked" at wages below the statutory minimum decline dramatically, as you can see at the left end of the graph. The dollar ranges just above the new statutory minimum increase substantially. (The EPI piece complains that they don't see a "spike" at the new statutory minimum. That seems nit-picky in light of this figure.) For the Synthetic Control (the two charts on the right), the right-most wages are variously positive and negative, significant and insignificant, implying no significant effect on total employment. The lines in the center of each bar are the standard errors; if the bar reaches beyond the standard errors, it's thought to be "statistically significant." The EPI piece insists that all categories above the $19/hour mark should be pooled for the sake of determining statistical significance. Fair enough. That seems like a subjective judgment. Interested readers can read the full paper and EPI piece to see if this is a valid criticism.  

Third bullet point from the EPI piece:

The study excludes an important group of workers, representing roughly 40 percent of the workforce: those working for employers with businesses in multiple locations. By omitting all multi-location businesses, such as chains, in Seattle, the authors bias their results toward showing job loss if there has been a shift in employment from small, single-location establishments toward larger firms with multiple locations.

The paper discusses this at length. Here’s some background. The authors tried to match employees to employers, and geocoded employers so they could decide if the employer was within Seattle city limits (affected by the minimum wage) or outside the city limits (not affected). Multi-site employers, with locations both inside and out, presented a challenge, because it wasn’t clear whether the employee was affected (worked within Seattle city limits) or not. Here’s a  quote from the Jardim et. al. paper:

Multi-location firms may respond differently to local minimum wage laws. On the one hand, firms with establishments inside and outside of the affected jurisdiction could  more easily absorb the added labor costs from their affected locations, and thus would have less incentive to respond by changing their labor demand. On the other hand, such firms would have an easier time relocating work to their existing sites outside of the affected jurisdiction, and thus might reduce labor demand more than single-location businesses. Survey evidence collected in Seattle at the time of the first minimum wage increase, and again one year later, increase suggests that multi-location firms were in fact more likely to plan and implement staff reductions. Our employment survey results may therefore be biased toward zero.

In other words, the authors of the study fully considered this, and they decided that it biased them toward finding a smaller employment effect than if they could accurately measure everyone’s exact location of work. Further down the page, the EPI piece acknowledges and discusses the author’s arguments. They don’t like the survey results and insist on some hard data demonstrating what multi-site employers actually do after a minimum wage hike. Fair enough, but it seems like Jardim et. al. went out of their way to address this limitation. Again, this seems like a mix of nit-picking and self-serving speculation on the part of the EPI. They also say:

A better approach would be to use the longitudinal nature of their data to see whether or not the Seattle minimum wage increase led low-wage workers to shift from single-site to multi-site employers in greater shares than was the case elsewhere in Washington, where the minimum wage remained constant.

Jardim et. al.’s second paper is a longitudinal analysis. It doesn’t directly address the question of shifting from multi-site to single-site employers, but it does suggest that workers in single-site locations benefit only modestly or not at all from the minimum wage increase. With the benefit of the second study, it's hard to argue that former single-site employees were massively upgrading their career prospects by transitioning to multi-site employers. The second study actually found a decline (of 8%) in job-turnover, making it harder to tell a "multi-site employers drove wages up" story. (The EPI piece was published before the second paper came out, so they didn't have the benefit of that analysis when it was written.) 

Back to the EPI piece:

We also note that another recent study of the Seattle labor market’s response so far to the minimum wage ordinance finds estimates of the employment effect in the restaurant sector that are in line with earlier research and concludes that to date the minimum wage increases have had no negative impact on employment.

Again, this misses a major theme of the Jardim et. al. paper: they were able to recreate these null-results in the restaurant sector using their data set. The paper is actually consistent with these previous findings if it imposes the limitations of other studies on its own data set.

The EPI piece looks to this observer like a mix of borderline-reasonable critiques and motivated reasoning. As far as I can tell, the results of the Jardim et. al. paper still stand.