Friday, February 22, 2019

Seattle Minimum Wage Studies

The recent decision by the government of Illinois to increase the minimum wage has renewed my interest in the topic. I've written on this a few times in the past. See here, here, here, here, here, and here. I've read Neumark and Wascher's book Minimum Wages, which was written in 2008. Enough interesting research has been published since 2008 that I thought I'd read a few papers on the topic.

Seattle recently introduced a minimum wage higher than the prevailing minimum wage in Washington state. Seattle's minimum wage changed from $9.47 to $11.00 on April 1, 2015 and from $11.00 to $13.00 on January 1, 2016, then to $15.00 on January 1, 2017. (Actually it varies by large vs. small employers, whether or not benefits are offered, whether or not employees receive tips, and it's indexed to inflation after $15.00.) Since Seattle had a higher minimum wage than the rest of Washington, this was a kind of natural experiment allowing us to test the effects of a minimum wage. Two excellent papers came out of this, both by the same group of researchers. (Two papers that I have read, anyway. If other papers have emerged from this episode, I haven't read them, so I can't attest to their excellence.) The first paper (first published, and the first I will discuss here) is called Minimum Wage Increases, Wages, and Low-Wage Employment: Evidence From Seattle. (Here is a link to the abstract; you can find a full pdf of the working paper with a Google search.) The second paper is called Minimum Wage Increases and Individual Employment Trajectories.

I came across these papers because I had remembered reading a summary by Jonathan Meer in 2017, which he posted to Facebook. See what he said, quoted here in this Marginal Revolution post by Alex Tabarrok. Pretty damning. The minimum wage increase had some pretty bad consequences for Seattle. But my Google search for "Seattle Minimum Wage Study" kept turning up headlines implying that the news out of Seattle wasn't so bad. What gives?

The First Paper

I'll lead with the conclusion from the first paper. The minimum wage hike to $13/hour caused hours worked in low-wage jobs to fall by 9.4% (3.5 million hours per quarter). The number of jobs fell by 6.8% (5,000 jobs). Wages for low-wage workers only increased by 3.1%. Coupled with the loss in hours worked, the average low-wage worker netted a loss of $125 per month. The labor demand elasticity came out to -3.0, which is in stunning contrast to most minimum wage studies that find values in the -0.1 to -0.3 range (or sometimes zero or even positive, implausibly implying a minimum wage increases demand for labor!).

[A later revision of their paper found that the loss was $74 per month. I don't know if I should regard this as a "correction" to the original paper, or if I should instead view it as one of two perfectly valid alternative point estimates for different model specifications. Maybe they caved to some criticisms on the "working paper" version of the paper, the one with the $125/month estimate, and got talked down to a lower estimate for the publication version. It doesn't matter too much. Both are substantial losses for a low-wage worker. I'm happy to have a discussion using either one as the "true" estimate.]

 The underlying data set makes this study particularly credible. Washington's Employment Security Department collects data for administering unemployment insurance. Washington is one of four states that collects hours worked in addition to earnings, which allows them to calculate hourly wages. (Which are the other four? There is potential for future papers here.) In other words, someone with this data set knows everyone's wage, employment status, and hours worked at various points in time. They can identify who was actually affected by the minimum wage, because they know who earned less than or close to the new minimum. This is in contrast to other studies that use crude proxies for "affected by the minimum wage." Studies often look at employment in the restaurant industry as a whole, or teenagers as a whole, assuming that a large fraction of these workers are affected by a minimum wage increase. To the extent that these groups contain workers who earn well above the new minimum, measuring employment in these groups will dilute the true effect of the minimum wage on the workers actually impacted. In fact (and I think this is really cool) these researchers looked at "restaurant workers"in their data set, effectively imposing the limitations of other minimum wage studies on their own data set, and got results similar to those other studies. They are effectively saying: When I hobble my data set to be as restrictive as the data sets used in other minimum wage studies, I get very low effect sizes, too. But the data in its full detail and granularity tells a different story. This is really important. It casts doubt on a lot of prior work on the minimum wage and suggests that much more granular data gathering is required to study the matter. From the paper:
In summary, utilizing methods more consistent with prior literature allows us to almost perfectly replicate the conventional findings of no, or minor, employment effects. These methods reflect data limitations, however, that our analysis can circumvent. We conclude that the stark differences between our findings and most prior literature reflect in no small part the impact of data limitations on prior work.
There are a lot of other interesting technical details that I won't comment on too deeply here. Again, read the paper for the full details. They chose $19/hour as their threshold for "low-wage worker", and have a long exposition and many graphs justifying this threshold, plus some commentary on how their results are robust to choice of threshold. This was an interesting discussion. Who is "affected by the minimum wage?" You might be tempted to define this as "workers initially earning less than the new minimum wage." But those people are basically extinct after the new minimum wage goes into effect; if the new minimum wage is $13/hour, there are no longer any workers earning $12/hour. That doesn't mean they've all lost their jobs, nor does it mean they've all been bumped into the $13/hour bucket. Also, if you suddenly move everyone who is currently below $13/hour to the new statutory minimum, employers have to give at least a slight pay bump to people earning more than that to preserve merit/seniority differentials (this commonsense observation is consistent with the paper's findings). These researchers had to look at statistical aggregates to discover what was going on. If employment dropped in the "low-wage worker" category, that's a good sign that the minimum wage reduced employment for this group. And that's exactly what happened. (Their second paper actually follows individual low-wage workers over time, rather than bucketing the population into "low-wage" and other categories.)

There's also a long discussion about the appropriate control group. This is important. I think when most people read headline-news version of these studies, something like, "Minimum Wage Caused Seattle Wage-Earners to Lose $125 a Month", they think it's an objective fact being reported. As if someone's just listing off accounting entries from before and after the minimum wage hike, and discovering that total pay was lower afterwards. Economists need a control group to study these things. You have to compare "Seattle" to "King County excluding Seattle", or compare Seattle to several surrounding counties, or Seattle to some kind of synthetic-Seattle (other regions that have similar economic features and trends to Seattle but for the minimum wage hike). They actually test several of these options. They find that the two options using real geographic neighbors, "King County excluding Seattle" and three nearby counties (Snohomish, Kitsap, and Pierce Counties), don't work very well. They fail a "falsification test" (see paper for details). But the synthetic controls pass the falsification test. In my reading, this also casts doubt on some previous minimum wage literature, which compares a treatment region (say, a border county in a state that increases the minimum wage) to a contiguous control region (a county across the border, presumably not affected by the minimum wage increase). Reiterating: the "$125 a month" figure above isn't an objective fact. It's what actually happened in Seattle compared with what their model says should have happened in the absence of a minimum wage increase. From the paper:
Difference-in-difference specifications assume that the treated and control regions have the same trends in the absence of the policy (parallel trends assumption), and will generally fail to produce consistent treatment effect estimates if this assumption is not true. It is prudent to be especially cautious about the parallel trends assumption given that the greater Seattle region experienced rapid economic growth coming out of the Great Recession, and the pace of recovery could have varied  by sub-region. As we show below, our two difference-in-difference specifications fail a falsification test, which suggests diverging trends. 
This economic boom in Seattle is important. Both papers caution the reader multiple times that this economic boom might mean the results aren't generalizeable to other regions. Actually, some critics of the first study (invariably people who are pro-minimum wage) have implausibly claimed that the economic boom discredits the finding of large disemployment effects. If anything, and economic boom should make a region better poised to absorb a minimum wage increase. The critics say something to the effect that the economic boom raised all wages, which shifted everyone up the income distribution, so that's where all those low-wage workers and lost hours went. (See Ben Zipperer's comments in this NYT piece.) This doesn't make a lot of sense. I would expect an economic boom to generally raise everyone's position on the income distribution, but also to create more jobs at the low-end. And besides, the authors make clear that they tested several thresholds for defining low-wage work (landing on $19/hour), and that their results were pretty robust to these assumptions. It seems like this critique requires an implausibly large shift up the income distribution.

The Second Paper

As I hinted above, the second study, called Minimum Wage Increases and Individual Employment Trajectories, was longitudinal. It followed individuals over time, rather than following statistical categories (defined as above or below the $19/hour cut-off). It found that you could basically define two groups of low-wage workers: experienced and inexperienced. The paper actually explores ten deciles, ordered by increasing levels of experience, but the qualitative "experienced vs. inexperienced" categories work well to describe their findings. From the abstract:
On net, the minimum wage increase from $9.47 to as much as $13 per hour raised earnings by an average of $8-12 per week. The entirety of these gains accrued to workers with above-median experience at baseline; less-experienced workers saw no significant change to weekly pay. Approximately one-quarter of the earnings gains can be attributed to experienced workers making up for lost hours in Seattle with work outside the city limits. We associate the minimum wage with an 8% reduction in job turnover rates as well as a significant reduction in the rate of new entries into the workforce.
 So what's going on here? The first paper says that low-wage workers earned $125 (or $74) less per month due to lost hours, and the second paper says they earned $8-12 more per week. I was initially confused, because my Google searches were turning up headlines triumphantly declaring the Seattle experiment a victory for minimum wage supporters. But really, it's all there in the abstract. There is no contradiction here. There were significant reductions in new entrants. The less experienced half of low-wage workers didn't see significant gains or losses in their earnings. The more experienced half of low-wage workers saw a $19/week increase in their earnings. Some people seem to have "benefited" (in terms of sheer total earnings, if that's all we care about), for others it was pretty much a wash, and still others who weren't yet in the job market seem to have lost out. This is only a "success" if you don't count the people who would have entered the labor market but didn't (couldn't?).

Some people are claiming that the less experienced group in the second study are actually benefiting, because they are earning essentially the same income for fewer hours of work. That this is a "benefit" is not at all clear. We don't know how working conditions have changed in light of the new minimum. Maybe employers are instituting more draconian work protocols during working hours, then dismissing the workers when they're not needed. Whereas previously they might have paid a lower wage and kept the worker on the clock all day under a more relaxed schedule. Are workers having to clock out for non-peak hours, then clock back in later in the same day? Do they have to go somewhere in the meantime? Does this entail paying for gas and parking? Calling this a "win" for the low-experience group is incredibly naive. Doing so ignores the lessons of the study itself. You know those high-experience low-wage workers, the ones who benefited more from the wage increase? They were grouped according to total number of hours worked (specifically, hours worked in a reference quarter just prior to the minimum wage increase). Extra hours of work confer a benefit by pushing people into the "high-experience" category. If low-wage workers are seeing their hours cut, that probably means this pipeline has been slowed down. The best thing for people in this cohort is to earn lots of experience at whatever wage the market will offer. Conceivably, most of the benefits from that work are not the actual wages, but the basic job skills learned (things like showing up on time, following instructions, losing the bad attitude, etc.) that lead to higher wages in the following years. Those hours aren't just spent earning wages. They're spend building human capital.

Besides, let's concede that the "pro-" argument has changed significantly. Were minimum wage advocates saying to low-wage workers, "This new law won't affect your income, but it will give you back more free time."? No. That's not how these policies are sold on the political market. You often hear nonsense about how the new minimum will raise people's incomes and lift them out of poverty, offering them a "living wage." Minimum wage advocates who are chalking this one up as a "win" should acknowledge that the argument has changed rather dramatically, and that we got something very different from what was initially promised. By the way, while it's true that prior minimum wage studies often failed to find a negative impact on employment, it's just as true that those earlier minimum wage studies often failed to find a positive impact on earnings. Most minimum wage advocates only concern themselves with the first part. Well, if we're doing away with neoclassical economics (really, just the common-sense observation that when something costs more you buy less of it) and pretending we're slaves to theory-free naive empiricism, you have to take both parts of that research together. You have to admit the part about the benefits failing to manifest themselves. (See this table summarizing the literature from the book Minimum Wages, particularly the part about Income Distribution.)

There is a lot of narrative discussion in this paper on "the two models" of low-wage workers and the effects of minimum wages. One model is that low-wage work is an entry-level job, from which point workers gain skills and gradually move up to a higher-paid job. A minimum wage knocks out the lower rungs of this ladder, making it harder for job market entrants to get their start. The other model is that low-wage jobs are dead-end jobs, with workers permanently stuck at low levels of income. A minimum wage makes these permanent jobs pay a little more and allows these workers to earn a "living wage." The authors suggest that both models are true of some workers in some jobs, and that the study somewhat supports this dichotomy. The low-experience group, the one that didn't benefit, corresponds to the first model, while the high-experience group corresponds to the second model, in their telling. But some of the information in the paper undercuts the second model. From the paper ("This model" refers to a neoclassical theory of low-wage jobs and skill-building):
This model implies individual employed in low-wage jobs at any given point in time will likely mature out of that market within a short time period. Prior studies have supported this conclusion, finding that 50 to 70 percent of workers earning exactly the minimum wage transition to a higher hourly rate within a year.
So, of the 30 to 50 percent of workers who remain at the minimum wage for a given year, how many of those are a permanent underclass? Do 70% have a transition probability of 1, with the other 30% a transition probability of zero? Or is there just some random-but-reliable churn across the spectrum? Given this transition probability, plainly most of the people who have ever earned minimum wage eventually move on to higher paying jobs. So the "permanent underclass" theory of minimum wage workers doesn't make much sense. I'm sure that some fraction of these people remain minimum wage earners year after year, but it's a much smaller fraction than 30% or 50%.

The authors explain the superficially contradictory conclusions of their two papers (emphasis mine):
The results of this longitudinal analysis contrast strikingly with repeat cross-sectional analysis suggesting sharper reductions in hours worked in Seattle's low wage labor market in 2016 (Jardim et al, 2018a.).These findings could be compatible for multiple reasons. First, longitudinal analysis by necessity excludes workers who enter Seattle's low-wage workforce after the baseline period, or who never enter at all. Second, individual workers may be making up for lost work in Seattle by adding employment outside the city limits. Our decision to include all Washington state employment and earnings might mask steeper employment and hours declines inside the city.
Job seekers who lack any labor market experience in Washington are invisible to us, as they do not appear in administrative records. We can infer their trajectories by studying aggregate statistics on the number of new entrants into the Seattle low-wage labor market. 
Here is a figure from the paper showing how new entrants in the low-wage labor market declined in Seattle compared to the rest of Washington state:

To be clear, the second study, which is longitudinal, is not picking up any effect on new entrants in the low-wage labor market. The first study, which is measuring the effect on statistical aggregates (employees with wages under $19/hour), does pick up these effects. I think this explains the superficial "contradiction" between the two studies. I believe this figure and the discussion about new entrants reflects the influence of a recent paper by Jonathan Meer and Jeremy West, titled Effects of the Minimum Wage on Employment Dynamics. This paper concluded that disemployment effects are small (perhaps even zero) for workers who are already employed, but minimum wage increases slow the rate of growth of new jobs. There was much academic debate over this study, but the new studies from Seattle seem to back the Meer/West story.

What These New Studies Mean

Assuming these new studies are credible, I think there are some important lessons for minimum wage policy. What follows is my own commentary, drawing from sources beyond the paper. If I get something wrong here, it's not the fault of the authors of these two excellent papers. I will jump from topic to topic in the following paragraphs, so apologies if you experience some whip-lash reading them. First off...

The demand curve for low-wage labor slopes downward! Both studies find that the minimum wage caused a significant reduction in hours. If these results generalize, it settles a debate in the minimum wage literature. Some studies have found that the demand is completely inelastic (meaning increasing the minimum wage has no effect on demand for labor), or even that the elasticity is positive (meaning that increasing the minimum wage increases the demand for labor). This has led to some speculation that employers of low-wage labor hold monopsony power. "Monopsony" means single buyer, similar to a "monopoly" which is a single seller. Just as a monopolist can unilaterally raise prices, a monopsonist can unilaterally lower them. Economic theory tells us that a government imposed price floor (like the minimum wage) can cause a monopsonist to buy more than they would if they could use their unconstrained market power to set the price. (See this post by David Henderson for a brief explanation of this argument. In addition, any good introductory microeconomics text will explain this argument.) Some commentary on the minimum wage literature has suggested that the near-zero or positive elasticity of demand implies that monopsony power is rampant in the market for low-wage labor. I think common sense militates against the monopsony story. Even in small towns, the main strip is usually populated by several chain restaurants, all competing for low-wage workers. Even setting common sense aside, the Seattle studies discredit the monopsony story. The higher minimum wage reduced hours worked. The monopsony story would have predicted an increase in hours worked, assuming monopsonists were previously successful in keeping wages down. Apparently stories of low-wage employers keeping wages low with "market power" are vastly overblown.

These studies fundamentally change the academic debate. The debate on the minimum wage used to be about whether the effect on employment was statistically significant (though small), insignificant, or slightly positive. The effect on "hours worked" is substantial in both of the Seattle studies. The debate is no longer whether increasing the minimum wage reduces employment. Once again, both studies saw reduction in hours worked for all demographics. The new debate is about whether the reduction in hours worked are offset by the increased wages. For low-wage workers as a whole, the answer is "No."

These studies provide conservative estimates for what a minimum wage increase is likely to do. Seattle was particularly well situated to absorb a minimum wage increase. Seattle has high median incomes (median household income of $100,630 in 2017) compared to the rest of the United States ($59,039 in 2017) or Washington state ($70,979 in 2017). Those rich tech sector workers can afford the price increases that come with higher labor costs. That won't necessarily be true of the rest of the country. The same minimum wage increase in, say, Mississippi (with a median wage of $14.22), will almost certainly cause more disruption in the labor market. (Does anyone think all of these counties would be able to absorb a $15/hour minimum wage?) Also, as both papers caution repeatedly, Seattle was in the middle of an economic boom over the period when the new minimum wage went into effect. Two excerpts from the second paper:
Settle transitioned from a period of growth to a period of stasis or modest decline [in the low-wage labor market] once the minimum wage increase took affect, despite its booming economy. Washington state outside King County continues to see growth in the number of new entrants, stabilizing to a higher level in mid-2015.
As a final caveat, we emphasize that during the period under study Seattle was undergoing an exceptional economic boom driven by rapid expansion of its high-skilled workforce. This may have driven the wages of less-paid employees up relative to the remainder of Washington State where our matched controls reside.  
Some critics of these studies have claimed that Seattle's economic boom simply means the results aren't generalizeable elsewhere. The economic boom simply "contaminates" the study, so we have to throw it out. Nope. Seattle's boom means that these studies are a very conservative estimate of what an increase in the minimum wage would do elsewhere. The boom isn't just a contaminant, adding random noise to the study. It's a countervailing force pushing employment and quarterly pay up, while the treatment under study here is pushing those things down. We should take these results as a kind of lower-bound for what is likely to happen elsewhere. Say, in a state or city with a higher median wage, perhaps not undergoing an economic boom or indeed undergoing a decline. These studies should give pause to any local governments considering a similar move.

Some people might be surprised by how few hours these low-wage workers actually worked.
Approximately 93% of the control sample logged less than full-time full-quarter work (i.e. 520 hours) in the baseline period.
Less experienced workers worked 108.6 (142.2) hours during the baseline quarter for cohort 1 (2), while more experienced workers worked 367.1 (432.2) hours during he baseline quarter for cohort 1 (2).
These people mostly aren't working full-time jobs. The less-experienced group is only working 1/5 to 1/4 of a full-time job. The fabled full-time minimum wage worker isn't represented in this group. (Supposing one in ten of these "less experienced" workers works full-time (520 hours/quarter), that means the other nine-in-ten average out to only 63 hours a quarter. The "full-time minimum wage worker" archetype might be better represented in the more experienced group of workers, who work more hours on average.) What these people need are more hours of work, not higher wages. There seems to be this myth that everyone's trying equally hard, but "the rich" are richer simply by virtue of their higher salaries. Actually, if you look across income quantiles (splitting into five quintiles each containing 20% of households is common), households in the higher quantiles log more hours worked, more full-time workers, and so on. Hours matter more than wage rates. It's barking up the wrong tree to focus on getting slightly higher wages for these jobs (higher wages which are apparently offset by reductions in working hours).

All that said, I'm left with a feeling of "Meh, maybe it's not so bad." The fraction of workers who are affected by the minimum wage is very small. We're talking 2-3 percent of the population earning at or below the federal minimum wage, and the "or below..." includes restaurant workers whose tips generally bring them above the minimum wage. In Seattle, in 2014 quarter 2, 13.6% of workers were earning below $13/hour, and 31% were earning below $19/hour (the cutoff used to define "low-wage workers" in the study; I'm using numbers pulled from Table 3 of the first paper). So most workers are escaping the effects of existing minimum wages. Most would escape even substantial minimum wage increases. This is kind of hopeful. It means that the social engineers probably won't screw things up too badly. It's still a very bad policy, and the effect on inexperienced low-wage workers (particularly new entrants) is perverse. But it's not going to disrupt the economy as a whole. To preempt another argument about the minimum wage, some people talk about "inflation" caused by higher minimum wages, as in "The cost of labor goes up, so prices in general go up." If low-wage workers are a small fraction of the labor market, then prices probably won't rise very much. If a minor input of production (in this case low-wage labor) increases slightly in cost, the price increases passed on to consumers won't be very large. Even anti-minimum wage economists (like Don Boudreaux and Neumark and Wascher) will go out of the way to point out that the "inflation" argument against minimum wages is a fallacy. (It must be a common one, though. The book Scratch Beginnings makes this argument against raising the minimum wage, and I've heard it several times in casual conversation.)

The optimistic take here is that employers can adjust along many margins to absorb the effects of a minimum wage hike. They can demand more work out of fewer hours with stricter workplace protocols, insist on shorter breaks, lower their tolerance for slacking, make capital investments, and so on. They can trim fringe benefits. They can be more selective about whom to hire. (Jonathan Meer has a long discussion of the various margins along which employers can adjust to a minimum wage increase in this excellent podcast.) My "the glass is half full" take is that large numbers of people don't get suddenly cast into permanent unemployment. But I still think this is a bad policy. Workers and employers have come to a set of mutually agreeable terms about work hours, scheduling flexibility, work intensity, and fringe benefits like training, free parking, free meals, medical/dental insurance, and so on. The minimum wage goes up, and these employees are forced to take a different deal from the one that everyone previously agreed to. Minimum wage advocates are basically telling them, "I know you accepted lower monetary compensation in exchange for better working conditions and fringe benefits. Nope! I'm insisting you take more of your compensation in money in exchange for the fewer fringe benefits." They are effectively deciding they have the right to veto these agreements and give the workers the deal the social engineers think they should want. These other margins of adjustment can be hard to measure. If the two Seattle studies are credible, then even employment and compensation are hard to measure, even though we have government statistics on them. Read the paper and make a list of all the caveats they make about their data set, which is still superior to almost any other study's. To end on a sour note, I think the glass is more than half empty. Minimum wage increases are not casting millions of people into permanent unemployment, but they are making work less pleasant and less rewarding in a thousand little ways that are difficult or impossible to measure. So minimum wage advocates will always be able to claim the empirical high-ground, insisting that there's "no evidence" of ill effects. But we know that these painful adjustments are happening, because the kinds of businesses that employ low-wage workers subsist on slim profit margins. (Go look up profit margins by industry to see what I mean, or simply note the very high turnover rate in the restaurant industry.) They don't have giant piles of money sitting around such that they can simply absorb a cost increase. Call it speculative, but it's safe to infer that these businesses are adjusting along these other hard-to-measure dimensions.

My final thought is that maybe this really is "just one more study." (Even though it's two papers, it's effectively one big study.) It's a study of a single law passed in a single city over a few short years, so I shouldn't get too excited about it. Then again, it does seem to be the most detailed study with the richest data set. And recall that it was able to replicate the small-to-zero employment effect in the restaurant industry. My feeling is that this is an important pair of papers. It presents a road map for future research: start tracking hours worked in addition to earnings, and do so anywhere that a minimum wage increase is being considered.

1 comment:

  1. Do these or other studies show evidence of worker demographic substitution? That is, has anyone documented evidence that min wage encourages businesses to hire, say, affluent housewives that get drawn into the market by higher wages over, say, less skilled/poorer/under-credentialed people?