On the Seattle Minimum Wage Study (part 2)

Previously: On the Seattle Minimum Wage Study (part 1)

Epistemic Status: If My Calculation Are Correct…


Some links:

Link to Paper: Original Working Paper

National Review: Critics Have Study Wrong

Forbes: Both Sides Might Be Right

Washington Post: $15 Minimum Wage Might Not Help Workers

Also Washington Post: Maine Tipped Workers Don’t Want Minimum Wage Increase

Bryan Caplan: No One’s Mind Will Change

Arnold Kling: Response to Caplan to Not Give Up

Charts from Part 1:


Quarters Under $13 Under $19 All  % Under $19
0 39807 92959 292640 31.8%
1 40706 94913 300892 31.5%
2 35421 89598 303089 29.6%
3 35085 90813 305229 29.8%
 4/1 ($11 min) 35087 92668 311886 29.7%
 5/2 33959 93382 320807 29.1%
 6/3 30002 87067 320195 27.2%
 7/4 24662 87112 321360 27.1%
 8/5 ($13 min) 24420 88431 331927 26.6%
 9/6 23232 86842 336517 25.8%

Hours worked:

Quarters Hours U13 Hours U19 Hours All U19 H% Avg U19 Avg All
0 14117 37408 130007 28.8% 402 444
1 14527 38565 132604 29.1% 406 441
2 11999 35589 136012 26.2% 397 449
3 11335 34269 132275 25.9% 377 433
 4/1 12174 37270 139197 26.8% 402 446
 5/2 11589 37472 142638 26.3% 401 445
 6/3 9924 34943 146960 23.8% 401 459
 7/4 7645 33031 140429 23.5% 379 437
 8/5 8315 35681 149514 23.9% 403 450
 9/6 8046 35867 153603 23.4% 413 456

Average hourly wages:

Quarters Avg U13 Avg U19 Avg 13-19 Avg 19+ All
0 11.14 14.14 15.96 46.14 36.93
1 11.15 14.15 15.96 47.44 37.76
2 11.27 14.37 15.95 48.79 39.78
3 11.28 14.41 15.96 49.77 40.61
 4/1 11.47 14.48 15.94 47.31 38.52
 5/2 11.54 14.58 15.94 48.83 39.83
 6/3 11.64 14.74 15.97 50.15 41.73
 7/4 11.82 14.97 15.92 52.80 43.9
 8/5 11.87 15.01 15.96 53.14 44.04
 9/6 11.87 15.03 15.94 52.30 43.6

With a day to contemplate the profound weirdness on display on that last chart, what conclusions can we draw?

The basic study design and conclusion tell a story. The minimum wage goes up. In response, some jobs are eliminated, and other jobs have their wages increased to compensate, in order to keep relative wages in line. This cascading effect levels off at around $19/hour: At that point, the number of workers getting a raise to $19, and the number getting a raise from $19 to something higher (combined with the number that are fired, although that seems like it should be quite small), are equally sized, so the number of workers earning about $19 an hour should be the same as it would have been without the minimum wage increase, unless that increase has a big enough impact to throw off the entire local economy.

If the average wage in the 13-19 range is not going up (if anything it is going slightly down) then that story doesn’t work. The distribution in the $13-$19 range remains about equal. If the upper end of the range isn’t going up, then neither is the lower end of the range.

One of these claims is wrong, but which? Let’s do another calculation. I neglected to have a column for the % of workers earning $13-$19 an hour, but here it is:

Quarters 13-19%
0 18.2%
1 18.0%
2 17.9%
3 18.3%
 4/1 18.5%
 5/2 18.5%
 6/3 17.8%
 7/4 19.4%
 8/5 19.3%
 9/6 18.9%

So here we see the number going up a little, as you would expect – the lower paid workers are being pushed up the wage scale, which should be bigger in this still-legal range than any negative effects.

But now we have the fact that the lower end of the range is getting bigger, and the higher end is claimed to not be getting much bigger (at least, it should be getting bigger slower), but the average value inside the range isn’t going down. That’s not possible. Since we have verified that the share of all workers in this range is increasing, and that wages overall went up a lot (for those above $19/hour, they went up from $46/hour to $52/hour in two years, much more than the national average or inflation! Boom town USA!) I only see one conclusion.

The increase in the minimum wage increased wages across the board. At a minimum, it did so far above $19/hour. The paper notes that this is unlikely, and I agree that this seems like it should be unlikely from the basic cascading raise mechanism. So this strongly suggests there is something else at work. Hold that thought.

Meanwhile, as noted above, Seattle was booming. Seattle employment increased 13%, hours increased 15%, pay was up across the board, and the city had basically full employment the whole time.

Employment under $19/hour declined from 28.8% to 23.4% of hours worked, but this was at a time of rising wages. Pay for those not at all constrained by the minimum wage increased from $46 to $52, and since the later group includes some ‘promotions’ from the lesser group, if anything this is low-balling the increase at 13.3% if workforce composition stayed similar. If we simply assume that wages went up that much, and we already know we have a level distribution of wages between $13 and $19, then fully 36% of those workers should now be over $19/hour because the barrier is $16.81 previous wage! If we include the previously under $13 workers, we still get 22.6% of these jobs graduating to Over $19/hour due to rising overall wages!

It’s 6.5% of the hours worked. Previously, 28.8% of hours worked were under $13/hour. Now, 23.4% are, and if we add 6.5% to that we get… 29.9%! So now those old jobs have become a greater percentage of the workforce, and they’re earning more money per hour, and the overall workforce is expanding.

This would reverse the results of the study completely. We now have a picture of Seattle as a boom town where everyone is doing well and the little guy is doing best of all.

The only problems are that this is very, very difficult to reconcile with the data on what individual employers were actually doing, and is almost certainly being warped by a change in the composition of the population of Seattle, plus the paper does statistical analysis that needs to be explained.

Also, priors.

Though it would be premature to make causal inferences on the basis of this single-differenced data, both headcount and hours data suggest that reduced low-wage employment can be apportioned primarily, but less than entirely, to wage

Over this same period, overall employment in Seattle expanded dramatically, by over 13% in headcount and 15% in hours. Table 3 makes clear that the entirety of this employment growth occurred in jobs paying over $19 per hour. The impression of skewed growth – driven in part by rapid growth in the technology sector – extends to wage data. Average hourly wages at jobs paying less than $19 rose from $14.14 to $15.01 (a 6.1% increase), while average hourly wages at all jobs surged from $36.93 to $44.04 (a 19.2% increase).

As we’ve already seen, I looked at the same numbers and agree that the difference is due to wage increases but that it can be entirely explained by general wage increases across the board rather than by a shift in the minimum wage.

It seems simply incorrect to compare jobs “Under $19” in two periods where wage levels are clearly going up across the board (e.g. average job paying Over $19 going from $46/hour to $52/hour) and then claim that a reduction in jobs under $19/hour is a real reduction.

I do find their arguments around the restaurant industry convincing, but only interesting insofar as they explain results from previous studies, so not worrying about that right now.  They then discard methods that don’t hold up to a placebo test, but select some methods they say hold up well. I am going to take them at their word that they executed these tests correctly, and that they did pass the placebo test.

Table 5 presents our first estimates of the causal impact of the Ordinance for workers earning less than $19 per hour. Looking at results using both the synthetic control and interactive fixed effects methods, we associate the first minimum wage increase, to $11, with wage effects of 1.4% to 1.9% (averaging 1.7%). The second increase, to $13, associates with a larger 2.8% to 3.6% wage effect (averaging 3.1%). A 3.1% increase in the wage of these workers corresponds to a $0.44 per hour relative to the base average wage of $14.14.34. We do not find strong evidence that wages rose in anticipation of enforcement during the three quarters following passage of the law. The small coefficients range from 0.3% to 0.7% and most are statistically insignificant.

During a period in which wages in Seattle overall rose 13.3%, they estimate a combined 4.8% increase for lower-paid workers due to the minimum wage increase. That magnitude of effect seems very easy to confuse with the overall rise in wages due to Seattle’s booming economy, which may or may not have otherwise evenly distributed gains. Given that the controls are what we can safely call Not Seattle, no matter how well we choose those controls from the Washington State area, it seems like there’s a lot of noise here.

They note that their result here is smaller than other findings. My guess is that this can be at least somewhat explained by wages naturally rising quickly anyway, which should logically reduce the amount above that the law can boost low-wage wages.

All their other findings similarly seem to fall apart if you assume that a large percentage of the Under $19 wages naturally rose into the Over $19 category and thus ‘low-wage employment’ in an important sense did not decrease in Seattle even as a share of overall employment. If you buy that, then unless there were similarly booming other areas to use to construct the control, the control should fail. The fact that the control worked when tested on the past data, when Seattle was not booming in this way, doesn’t change that.

So I’m not worried too much about reconciling my observations with their methodology; I think I see how we came to disagree, and why the exact cutoff of $19 doesn’t matter. Jobs will be pushed up over any threshold you choose, and the mistake is making the threshold a constant over time. Their other worries and caveats are things to worry about, but not if you’ve already explained the result in another way, plus their answers to them seem strong.

That leaves composition changes, survey results and priors.

The survey problem comes from this footnote, which I quoted in part one, was a large part of my motivation to look at the data, and also seems rather important:

The Seattle Minimum Wage Study surveyed over 500 Seattle business owners immediately before and a year after the Ordinance went into effect. In April 2015, multi-site employers were more likely to report intentions to reduce hours of their minimum wage employees (34% versus 24%) and more likely to report intentions to reduce employment (33% versus 26%). A one-year follow-up survey revealed that multi-location employers were more likely to report an actual reduction in full-time and part-time employees, with over half of multi-site respondents reporting a reduction in full-time employment (52%, against 45% for single-sit firms).

How do you respond to that? Even if the rest of the study can be explained away what the hell do you do with this? Half of all firms are cutting their full time minimum-wage workers, even more than were expecting to! If that’s true then how in the world could low-wage work not be devastated?

The result might be wrong, but let’s assume it isn’t. If we also knew that low-wage employment didn’t go down, what happened?

My answer is that some businesses cut low-wage jobs that no longer made sense, doing a combination of hiring higher-wage workers, extending hours for other workers, automation and simply doing less business. Meanwhile, other businesses are starting up and hiring those workers to do other things, because Seattle is booming and there are profitable things to do. I don’t even know that this is an unusual amount of churn. It seems like it is, but given that new firms by definition are always net hiring, and some old firms are also net hiring, it does not seem that crazy that you could see half of existing firms cutting low-wage jobs, and yet not lose net jobs when you count replacements in the $13-$19 range and other hiring.

Doing additional surveys, and knowing more about the breakdown of how employment changes work, would be crucial in seeing if that story makes any sense or not. The numbers working out does seem rather implausible, and I wouldn’t expect a normal booming economy to result in half of firms eliminating minimum wage jobs within a few years, but perhaps I am wrong about that. It would make sense that minimum wage jobs tend to be more transitory than higher paying jobs.

That would also tie into the changing composition of Seattle. A boom, and a general rise in wages and wealth, changes the demand for goods and services. Minimum wage jobs are likely to disproportionately be engaged in providing services to lower income customers, since higher income customers would prefer to pay more for higher quality goods and services, which means higher wages. So that effect might cause a lot of lower-end businesses to cut back or go out of business, reducing low-wage work, or to upgrade to higher-paid employees. Out with fast food, in with casual dining.

The demographic shift could also explain the results if higher-skilled workers are coming to Seattle to enjoy the higher wages and booming economy, which lower-skilled workers are being driven out of the city because the minimum wage means they cannot find jobs and presumably also issues with rising housing costs. Thus, when it looks like the whole wage scale is moving up by 13.3%, that might not be what is happening.

Instead, a lot of workers that get paid $50+ per hour are moving in (I have a friend who in fact took a great job in Seattle this past year, and falls into this category) and this drives averages up but doesn’t change the lived experience of any given person or the pay of any given job, since new high-end jobs are being created that don’t change what the currently average worker gets paid. That means that we can’t push a given person’s wage up 13.3% just because wages went up that much, which means we in turn do see a sharp decrease in low-wage work, which means that low-wage workers must be leaving since Seattle is remaining at full employment.

In this scenario, Seattle is doing something that is in fact bad for workers previously earning minimum wage, because a lot of them are losing their jobs and even being driven out of the city (although some would be driven out anyway by increasing costs; you can also tell the story that the minimum wage isn’t that binding since no one could have survived on the old minimum in the new Seattle either way). Here, the decrease in jobs and hours is real, and low-wage workers are becoming unemployed and being forcibly relocated – and under a more universal minimum wage increase, they might become permanently unemployable, which would be very bad.

Those stories then get reinforced by priors. The prior says that raising the minimum wage means you increase cost of low-wage labor, so demand goes down, and quantity declines. It’s hard to avoid that result, and simply saying ‘the data says that is not what happened’ is not really good enough. Until you can explain why that is not what happened, you need to remain deeply suspicious.

I do not think, even if my calculations are correct, that we can tell whether Seattle’s increase in the minimum wage helped or hurt low-wage workers. The paper’s result seems to be explained by overall rising wages, but those rising wages might be due to changes in the composition of the workforce, and in general the changes going on in Seattle seem too big, and too different from the surrounding areas, for us to be confident in what changed here as the result of the minimum wage increase compared to a counterfactual Seattle, or what effect that had on the rest of the state or country. If the things I am seeing do not have the effects I think they do (and there are several steps where I could be wrong or missing something), and using a constant threshold for low-wage workers still makes sense, then it is quite likely that the paper’s original conclusion is correct. If that is true, then even a $13 minimum wage is backfiring on low-wage workers in Seattle, and we need to avoid minimum wages that come anywhere near that high relative to an area’s overall price level (but could likely tolerate something that is the local equivalent of the $11 level). Anything more would be devastating.

But what if the confounding effects I have considered are small, and my main line of exploration is right that you need to move the threshold up by 13.3%? In that case, Seattle low-wage workers benefited or at least didn’t suffer from the increased minimum wage (with the price level moving up so quickly, it’s not clear how binding the new minimum really is), and given how the city is otherwise doing, it does not seem to be doing any great harms to Seattle’s overall economy. A booming Seattle would then clearly be able to tolerate a $13 minimum wage, and might tolerate a higher one. That does not imply strongly what would happen in a poorer area, or one that was not booming, if such a minimum was introduced, but it does suggest that there is more slack in the system than one would, on priors, consider there to be.



This entry was posted in Uncategorized. Bookmark the permalink.

10 Responses to On the Seattle Minimum Wage Study (part 2)

  1. Pingback: Rational Feed – deluks917

  2. Peter Gerdes says:

    I very much appreciate your analysis here but I worry that this kind of detailed focus on the particular data from one study (without further qualification) risks muddying the water and falsely legitimizing the view that minimum wage increases benefit workers. Yes, studying social phenomena is hard and every study on the subject will be subject to a range of very plausible objections and potential confounding factors but its important not to give the impression that its all just a matter of evaluating data and that, absent a large preponderance of data favoring one side of the debate, both views are equally plausible/respectable.

    The truth is that in most areas of scientific inquiry it is theoretical frameworks and the resulting priors that do most of the work. Even relatively simply claims like the negative effects of low dose lead exposure levels on children’s cognitive development would be extremely difficult to establish if all we could do is examine sociological data and run regressions. For instance it would be almost impossible to eliminate potential confounders of socioeconomic status, political power, concern with children’s welfare etc.. (communities that take the care to ensure that children aren’t exposed to lead would arguably be different). However, we can be very confident in our conclusions here because we believe that the mechanism for lead toxicity isn’t something that only appears at high doses (e.g. it isn’t the result of metabolism to a harmful compound that occurs after the mechanism for benign metabolism is overwhelmed as in some toxins), that effects in animal studies tell us things about effects in humans, that neurons in the brain are responsible for cognitive function (so neural damage suggests cognitive harm) etc.. etc..

    As such, questions about the minimum wage need to be evaluated against our theoretical models and priors about the fact that when the price of a good goes up we expect its consumption to go down. Of course there are exceptions (Irish potato famine…higher potato prices increased consumption) but we feel that some specific explanation is needed in those cases. Yes, increasing the minimum wage increases the money minimum wage workers have to spend but, as some of their wages will go to goods produced by non-minimum wage workers or the costs of non-labor inputs (raw materials) our priors suggest it is implausible that savings from economies of scale will outweigh these losses so a minimum wage hike doesn’t increase the incentives to cut minimum wage hours.

    This doesn’t mean there aren’t respectable reasons to believe that minimum wage boosts are beneficial (indeed the priors I mentioned merely suggest minimum wage hikes will reduce hours not that they will do so enough to offset benefits). However, understood in this context we don’t just evaluate minimum wage changes and look at the data to see what happens we consider particular mechanisms which would invalidate or usual expectation that price increases decrease consumption and see if the data support these and IMO the consistent lack of support for such mechanisms at significant levels provides the strongest evidence for the fact that minimum wage hikes reduce employment (or at least hours worked) by the poorest individuals.

    In short, I’m again not disagreeing with you anywhere, just pointing out the danger of letting it seem like its just a matter of looking at correlations between minimum wage hikes and worker hours. Its the same danger we see in global warming debates when skeptics of anthropogenic global warming try to suggest that their hypothesis that warming trends are the result of changes in solar flux should have equal standing and the strength of claims of anthropogenic global warming. No, this all happens against a backdrop where we have extremely strong theoretical reasons to think that CO2 should result in increased temperature so rejecting that hypothesis requires both showing that some other cause could give rise to the observed temperature trend and offering a convincing account of why the expected effect of CO2 isn’t seen.

    • Peter Gerdes says:

      TL;DR Great analysis of the data but we shouldn’t let focus on the data cause us to ignore the fact that considerations of priors means that those who support minimum wage hikes EITHER need to:
      1) Overcome a presumption that we should expect increasing the cost of labor to decrease its consumption with either overwhelming empirical evidence or strong theoretical considerations supported by data.

      2) Admit that minimum wage hikes will reduce employment among minimum wage workers but explicitly claim the tradeoff is worth it, e.g. by arguing that the utility gain by the workers who are better off is more than the utility loss for workers who no longer have a job…or by suggesting (again contrary to what our priors/models tell us about fixed costs of training employees or risk of theft) that this decrease in consumption won’t result in less total hires only less hours worked by each hire.

    • TheZvi says:

      Peter, appreciate your response, both in terms of getting one (when one works on things like this it can feel like talking into the void, worrying no one cares) and its thoughtful content.

      I agree with you 100% that there are tons of confounding factors, that there is a gigantic prior that one needs to overcome, and that there is huge risk that I’m simply muddying the waters and doing harm. I struggled with that. Eliezer is busy ranting in giant blocks about how people don’t get it and we’re going to destroy people’s lives and how maybe we can just find the right Magic Words to make people understand, and I can totally see the argument for that approach.

      We do have the info we have, though, and if one starts looking at it, one must report back what one finds, even if it isn’t what was expected. I totally expected to draw the conclusion that what Seattle did was really bad, I can’t back out and not publish (as it were) when I find the data implying the opposite, just because I don’t like the conclusion.

      To that end, I made it clear what my priors are, and I tried to make it clear at the end that we can’t draw any firm conclusions from this data and there’s still a huge bar to overcome. I struggled with this, and whether I should have done more and been more explicit at screaming “defy the data if you have to” or if I should have done less of that so people would think harder about the data and what it implies. My guess is still that there are interesting things going on underneath the surface here that we don’t properly understand, and if we look hard at the clues and details we can figure out what questions would uncover them. If instead people just argue back and forth whether this was ‘good’ or ‘bad’ we’ll never know what they are!

      Or, for TL;DR, I decided I’d rather Do Science than Do Politics, be the change I want to see in the world and hope it all works out.

      • Peter Gerdes says:

        Two points. First, I read all your posts even if I don’t always have something to say so keep up the good work.

        Secondly, I think you are overreaching on the do science not politics bit. All science builds in priors and indeed that was exactly my point. One doesn’t figure out how where a rocket will end up by doing a bunch of regressions on previous launches…one uses priors about what theory is true. Anyway that’s just a little nitpick here I still very much appreciate the analysis.

  3. IvanFyodorovich says:

    Really great analysis man. This void-dweller is pleased. Perhaps in the next iteration of analysis we’ll have a sense of, well, what happened. Like, were there massive losses in low wage restaurant sector counteracted by gains in the high wage restaurant sector during this period? Were a lot of warehousing jobs lost in Seattle proper, and were a similar number of jobs gained in neighboring counties or did the jobs just vanish? The big tables only tell you so much, and the people who wrote the working paper have access to data that could give them some sense of this. The problem of “there is no control Seattle” remains but we could get a better sense of the facts on the ground so to speak.

    • TheZvi says:

      We will learn more as more time goes by, the effects have longer to kick in and the minimum wage gets pushed higher. I think your instincts are right and we need to get into more of the details of the data, to learn exactly what is happening – hopefully the data being gathered has the details, but given they were able to split out restaurants, I’m guessing it’s out there.

  4. Pingback: Seattle Minimum Wage Study (Part 3): Tell Me Why I’m Wrong, Please! | Don't Worry About the Vase

  5. Pingback: Best of Don’t Worry About the Vase | Don't Worry About the Vase

  6. Pingback: The Case Against Education: Foundations | Don't Worry About the Vase

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s