Scott Sumner  

Divine coincidence?

Michael Darda on Alan Blinder... Kevin Williamson on Social Sec...

Some of my progressive friends argue that markets are not efficient because information is a public good. Thus research in equity values will be under-provided. Others argue the financial sector is inefficient because greedy Wall Street types convince the gullible public to pour money into expensive managed funds, which underperform indexed funds. Thus investment analysis will be over-provided.

I sometimes say that both are wrong, because these two inefficiencies exactly cancel out. (I usually have a twinkle in my eye, when making this claim.)

Matt Rognlie is a grad student who occasionally leaves comments in various blogs. His typical comment is usually more interesting than the average published paper in an economics journal. I particularly liked one he left after a post by Nick Rowe discussing Larry Summers advocacy of minimum wages on "efficiency wage" grounds. (The basic argument is that a minimum wage may boost worker productivity, and thus not cost jobs. Although it's a bit more complicated than that.) Here's Rognlie's comment:

The Summers claim doesn't hold up very well empirically either: most studies show roughly 100% pass-through of higher minimum wages to prices. This means that the "second-order effect on total costs" argument doesn't hold when wages are increased across the board (presumably because of the fallacy you're pointing out) -- unless, for some reason, firms consistently pick the moment of a minimum wage increase to inflate their markups.

This reflects an interesting internal tension among minimum wage advocates in economics. If clean identification is the standard, they have the empirical high ground on the direct policy question -- "do minimum wages decrease unemployment?". But once we move from direct estimates to the underlying mechanisms, very few of the ideas floating around have any empirical support, and most of them can easily be rejected. Neither the Summers story nor the simplest variant of the monopsony story (which would imply that a binding minimum wage lowers marginal costs) seem consistent with full pass-through to prices. Or, to take another example, search-based monopsony requires that marginal recruitment costs increase dramatically with the scale of a firm's employee base -- which is hard to rationalize either anecdotally or using the formal evidence in the literature.

So it's an amusing contrast. By prevailing standards, the minimum wage itself has great empirical support -- but then pretty much every single explanation that's offered to rationalize these findings crashes and burns when you subject it to data and a little light theory.

An even more subtle but telling case of this disconnect is the following. The current empirical consensus among mainstream labor economists is that the employment effects of the minimum wage are close to zero. But most of the economic arguments offered to explain this finding do not specifically rationalize the zero -- instead, they suggest some force that pushes opposite the traditional supply-and-demand logic, and could in principle offset it by 50% or 150%, rather than just 100%.

In general, if X is some positive value and Y is some negative value -- with no a priori relationship between the magnitudes -- it's a remarkable coincidence for X+Y to always come out near zero. Yet, if we substitute "the traditional downward-sloping component of labor demand" for X and "the added effects of search-based monopsony" for Y, that's exactly what the minimum wage consensus seems to believe. After all, cleanly identified studies aren't giving sizable negative estimates in some settings and sizable positive estimates in others -- they're basically just giving zeros.

(Of course, one advantage of the Summers hypothesis is that -- when drawn out under some very particular assumptions -- it could explain a consistent zero. I'm curious whether this view is currently popular, despite its obvious other failings, because smart economists like Summers realize this advantage.

My guess is no, and that this is all a little too subtle. It will take a while for everyone to realize that proposing a negative Y to offset the neoclassical X isn't good enough -- you also need to explain why Y is so consistently of a magnitude that cancels out the neoclassical X. Or else you need to acknowledge that your point estimates might be driven by attenuation bias a la Sorkin, and that the Credibility Revolution isn't always and everywhere as credible as it purports to be.)

And as if that's not enough, he also left several other extremely interesting comments, later on in the comment thread. I'd encourage people to read all of them. (Warning, some of his comments are a bit difficult if you haven't studied economics.)

Comments and Sharing

COMMENTS (14 to date)
Daniel Kuehn writes:

The zero effects are of course not really zero - they're just not statistically different from it. They're also not "zero" everywhere. DLR has a nice graph showing the estimated distribution of county level effects IIRC, and of course state specific studies using the same methods show both positive and negative effects. This helps with the alleged perfect cancellation. But I do generally agree that we have a ton left to learn about the mechanisms in play.

Brian writes:

As Daniel says, the net effects are not really zero. As far as I've seen, the overall effects seen in the literature are that the minimum wage has a definite but modest disemployment effect, so there's no reason to look for exact cancellation. But more to the point is that, for any given case, the minimum wage changes/differences are so small that the effects are lost in the noise. This means there's no point in understanding the size of each effect more precisely and accurately because the noise kills our ability to test them anyway.

The smallness of the minimum-wage changes is purposeful, of course, since it allows the politicians who favor the changes to maintain plausible deniability on the disemployment effects.

Scott Sumner writes:

Daniel and Brian, Thanks for that clarification. In fairness to Matt I think the claim was that findings of zero occurred at different minimum wage levels, i.e. studies that looked at different periods of time. But your point is well taken.

Having said that, I think the more interesting point is that the cost of the increase is passed on roughly one for one to consumers. In that case whether it's exactly one or just close to one makes little difference (unlike the zero effects on employment finding, which is an approximation, as you say.)

Daniel Kuehn writes:

And I'm not even saying it's an approximation so much as I'm saying that it's an average effect and the local effects are much more dispersed, as we'd expect with lots of local labor markets with differing price levels.

Brian writes:

"I think the more interesting point is that the cost of the increase is passed on roughly one for one to consumers."


I agree. This observation is interesting, if true (I've not seen the literature on it). It is completely counter to the claims of minimum-wage advocates, who like to argue that the additional money for wages comes out of profits. (Hence it's OK, in their opinion.) In contrast, 100% pass-through says that the money comes out of the consumer's pocket, and hurts those who frequent minimum-wage businesses. Not coincidentally, these tend to be the folks who are at or near minimum wage in the first place.

Consider the academic studies of the minimum wage done by Liberal/Progressive labor economists (LPEs). Have any of these studies been reviewed and discredited by other LPEs? Is there any indication that LPEs have done an unbiased, scientific review of those results?

If not, then is it really possible that all studies of the minimum wage are correct when done by LPEs? Along the same lines, is it possible that all studies by Conservative/Classical labor economists (CCEs) are incorrect to the point of being useless, as determined by LPEs?

I ask, because some CCEs have debunked the revered Card-Kreuger study of the minimum wage, but it continues to be (to this layman's eyes) a central reference supporting the likelyhood that an increasing minimum wage increases overall employment.

The debunking points out that Card-Kreuger looked at eployment only in large franchised restaurant chains, and did not look at overall employment in the fast food sector (which contains a vast number of small, non-chain restaurants). Thus, it is possible that employment went up a bit in the chains because of pressure put on their small-fry competition.

It makes sense to me that LPEs want to politically support public policy to raise the minimum wage. It seems that they want it so much that no supporting study is invalid under review. That would be a remarkable event in the history of science.

Extraordinary results require extraordinary proof. Minimum wage studies are filled with probable statistical error, from sampling, timing, failure to measure a wide enough population, and no follow through (say for a year) to measure long term effects. Yet, every such supportive study is hailed as proof that long-established classical results relating price to demand just don't apply to labor and business.

Where are the studies computing that (say) $9/hr is too little, $15/hr is too much, and $10.10/hr is just right, based on an economic analysis of reality?


Excess Teen Unemployment
02/16/13 - AEI Ideas by Mark Perry
=== ===
[edited] The minimum wage rose 41% in three stages between 2007 and 2009. This had a disastrous effect on teenagers. The jobless rate for ages 16-19 increased from about 16% to more than 26% (10 percentage points). The overall US jobless rate also increased from about 5% to 10%.

The graph attempts to isolate the effect on teenagers by plotting “excess teen unemployment" the difference between the teenage and overall jobless rates.
=== ===

See the graph at the link. Excess teen unemployment closely traced increases in the minimum wage. Teens are a measurable segment as part of all people of low experience, knowledge, and productivity.

Daniel Kuehn writes:

[Comment removed for rudeness. Email the to request restoring your comment privileges. A valid email address is required to post comments on EconLog and EconTalk.--Econlib Ed.]

Scott Sumner writes:

My prior is to expect the minimum wage to cost jobs. That's based both on theory (if McDonald's raises prices, how can that not cost jobs?), and on the limited empirical work that I did on the subject (dealing with labor policies in the 1930s). I also think the higher unemployment rates among the young in Southern Europe (even during boom periods) is suggestive, although there are other labor market interventions that might also explain that result. And I am also aware that teen unemployment rose more during this recession (relative to adult unemployment), than during the 1982 recession. That's also suggestive.

But I'm certainly open to the view that a fairly low minimum wage may have insignificant employment effects. There is of course research on both sides of the question.

ThomasH writes:

@ Brian,

Minimum wages are usually criticized here in terms of the harm to people who lose/do not gain jobs because of the higher wage. I have not seen much discussion of the harm to firm owners and their customers. This would be consistent if in their model demand for "low wage" intensive products is high and substitution between low wage labor and other factors of production is high.

If firms can pass on the higher wage costs (almost) 100% this implies that the substitution between goods and services produced with large inputs of low wage labor and other goods and services is quite low and the substitution of other factors of production for low wage labor is low.

In this case, a minimum wage becomes an almost pure transfer to low wage workers in those activities, financed by a tax on consumers of "low wage" goods and services. If this is the case minimum wages should be discusses mainly in terms of the fairness of that kind of tax and transfer versus other kinds of taxes instead of the costs to low wage workers who lose their jobs.

Daniel Kuehn writes:

Andrew -
I think it's wrong to characterize the studies as being by liberals or conservatives per se. Some of the people that are often referred to by other people as "conservatives" that find disemployment effects are not clearly conservative. Some of the people that are often referred to by other people as "liberals" also produce research suggesting a disemployment effect. And most of the people that find no disemployment effect aren't especially liberal at all.

Don Boudreaux writes:

If not directly on point, this February 22, 2013, EconLog post by David Henderson on minimum-wage legislation and monopsony power is more than tangentially related to this discussion.

Matt Rognlie writes:

I’m a little late to the party here, but let me clarify. I didn’t mean that all the point estimates were literally 0, or maybe 0.01. That’s obviously false; indeed, the standard errors on most studies are generally much larger than this, and there would have to be something calamitously wrong with these studies if they kept hitting almost the same exact point estimate within wide confidence intervals. Someone other than me would already be speaking up about that problem. : )

Instead, I meant that generally there were no statistically significant deviations from zero within the “cleanly identified” literature — and that furthermore, when you pool the estimates, this literature appears to be honing in on zero as the most credible signal amid all the noise. For instance, in the now-canonical Dube, Lester, and Reich (DLR), Table 2 presents several estimates for the labor demand elasticity. A bare-bones regression shows a negative elasticity, but using either the authors’ preferred approach (county-pair * period dummies) or alternative controls (MSA*period dummies; or even just Census division*period dummies, with or without state-level trends added in), nothing is statistically different from zero anymore. The story is similar with a different dataset in Table 5.

Daniel alluded to DLR’s Figure 5, which shows a kernel density plot of the distribution of elasticities estimated from various individual border segments. This doesn’t narrowly peak at zero; instead, it’s roughly a bell curve around zero. But this doesn’t have much to do with my point, since Figure 5 seems like roughly what we’d expect from statistical error alone, and DLR discuss it on that basis. Figure 5 does not demonstrate — and to my knowledge never purported to demonstrate — that the underlying effect of the minimum wage is actually positive in some identifiable areas of the US and negative in others.

And that’s really what I find unsettling about the literature. We’d expect all kinds of complicated forces to collectively produce the classical, supply-and-demand effect of the minimum wage; and we’d expect all kinds of intricate features of the labor market to matter for the effect of a hypothesis like monopsony. We certainly would not expect the exact balance between these two forces to be the same regardless of the industry or demographic or region or country we’re looking at. And yet that seems to be the theme of the cleanly identified literature.

For instance, a successor paper to DLR looks at teen employment rather than restaurant employment and finds broadly the same answer of zero effect. Other entries in the literature look at different countries (most notably the UK), and as far as I know there’s no pronounced dispersion along that dimension either — if you took a poll of mainstream labor economists in the UK, they would tell you that the effect was about zero there too. Indeed, to my knowledge, there is no strong sentiment among labor economists of the form “well, if you look at country/region/industry/demographic X, it seems to have a positive effect from the minimum wage, while if you look at country/region/industry/demographic Y, it seems to have a negative effect”.

Matt Rognlie writes:

It’s hard to imagine this is all just a coincidence. And remember that it’s layered atop another suspicious coincidence, which is that these offsetting forces happen to produce zero effect, rather than 0.1 or -0.1 or some other number. In fact, I’d go further: even if all you’re willing to stipulate is that experts agree that the effect is near-zero in the US as a whole, that one observation tells us quite a bit. Imagine that we’re Bayesians trying to sift through three hypotheses:

(A) the minimum wage on average leads to not-very-severe disemployment due to one of the commonly cited mechanisms (e.g. monopsony),

(B) the minimum wage on average leads to not-very-severe disemployment due to a labor market mechanism that we’d specifically expect to deliver near-zero elasticities, or

(C) the minimum wage only appears to produce not-very-severe disemployment empirically, because of a bias (e.g. Sorkin) that we’d specifically expect to deliver near-zero elasticities from these kinds of studies.

From (B) and (C), we’d expect the aggregate effect in the US to be near-zero; (A) is consistent with a much wider range of values, because there’s no reason to expect two very different forces to combine to produce almost exactly zero. If we do observe an aggregate near-zero, we should be revising our priors massively toward (B) and (C) and away from (A).

The key question, then, is just what the relative probabilities of (B) and (C) are. Right now my guess is that (C) is more likely, because I’ve seen spectacular attenuation bias in other settings with many of the same characteristics. But it’s certainly possible that (B) is true too, and to move the debate forward we need to think more carefully about exactly why the minimum wage should produce zero effect.

(By the way, if I’m wrong about the consensus that large-scale, cleanly-identified studies produce effects near zero, then I certainly have company: in a widely circulated review with the revealing title “Why Does the Minimum Wage Have No Discernible Effect on Employment?”, John Schmitt of CEPR includes as Figure 1 a very popular funnel graph that shows an overwhelming tendency toward near-zero point estimates whenever the standard error is low; the graph is also discontinuously asymmetric around zero, partly backing up the widespread claims that there is publication bias driving some of the negative estimates.)

Plucky writes:

Not to the guy that naively mentions the Obvious Explanation, but doesn't long run market equilibrium in a typical micro 101 scenario predict more or less exactly this?

In a long run competitive EQ, each individual firm operates at optimal scale, with a firm-level rate of return that is the minimum rate to have a stable capital base. Changes in demand are met on the supply side by change in number of firms, without any effect on price. Consumers capture all surplus above long-run marginal cost (where this is construed to include the minimum acceptable return on capital)

In micro 101, when the minimum wage goes up, the long-run eq in an industry employing minimum wage labor and minimal firm differentiation is that long run marginal cost goes up, price goes up by the same amount, firm profitability remains unchanged, and the number of firms shrinks as determined by demand elasticity. Large-scale firm closures (and employment losses) only occur if the cost increase is large enough to wipe out the consumer surplus, or if there is high elasticity due to substitution effects to a product unaffected by the cost increase.

Empirical results showing 1-for-1 price increase and minimal or a small decrease in overall employment could simply be interpreted as showing that 1) low wage industries generally conform to long-run competitive equilibria with low demand elasticity and 2) the the time lag from short to long-run eq in low-wage industries is pretty quick. For industries like say, fast food, this ought to be a "Duh" result.

The verbal anxiety and political resistence from the members of these industries also conforms to this interpretation- a fast food manager/owner is keenly aware of having minimal or no market power (consistent with a competitive eq), and so sees a cost increase that ceteris paribus would push him below the stay-in-business profitability threshhold (also consistent with a competitive eq) as an existential threat, and fights with accordingly strong fervor. But when all the competitors face the same cost increase, the new eq just happens at a higher price, with the handful of exits indicating low demand elasticity, and its mainly the customers who lose out

What's wrong with the micro 101 explanation?

Comments for this entry have been closed
Return to top