A response to Matt Yglesias’s response. I understand from email that his original post responding to me was intended to be read together with an earlier post, where he separates out questions of freedom and economic efficiency, and argues, more or less, that the best way to increase the bargaining power of labour is by pushing full employment. This means that he does not, after all, treat market outcomes as being in some way natural. So consider those specific objections withdrawn. But I still think that there is something fundamentally wrongheaded about the way that he is analyzing these questions. And not only that – but Matt Yglesias himself (2004 vintage) would seem to agree with me.
As Matt says in his updated post, specific arguments for regulations intended to help the plight of the poor ought to be considered on their merits, given the trade-offs – they can’t simply be deduced from some generic ethical position. You’ll get no objections from me on that. But where you will find me objecting, and quite vociferously too, is to the suggestion that we ought to employ simple econ 101 style analysis in order to figure out the tradeoffs, and the appropriate solutions. This style of analysis has an awful lot of presuppositions built into it, and these presuppositions are politically loaded.
Let’s take the case of labour relations. If, as Matt argues, you ought to start with a model of firms, each of which has a cost function such that the total compensation they can offer is fixed, and any increase in costly rights (such as not having your body searched to stop you smuggling widgets out between your arse-cheeks) is inevitably associated with a proportionate decline in wages, then, indeed, you will end up concluding that compulsory rules forbidding body searches will lower overall choice without really benefitting anyone. But you only get to that result because of what you are assuming in the first place. You are assuming that there isn’t any real distributional action happening within the firm – and in particular, that the firm’s owners don’t have any surplus (that they are able to extract because e.g. of workers’ weaker bargaining position) that could be reallocated through regulation. In other words, you are only ‘refuting’ the people you disagree with, because you assume away the problem that they are worried about. Moreover, as Brad DeLong (!) points out sarcastically, the logic applies more or less identically to any economic concession that employers might want to extract from their employees in exchange for higher wages. The model isn’t an argument against particular forms of regulation – it’s an argument against any extraction-preventing form of regulation at all.
The funny thing is that once upon a time, Matt Yglesias himself recognized that this mode of argument was extremely dodgy, and pushed back when Alex Tabarrok tried to claim that regulations requiring landlords to provide hot water to their tenants were not in those tenants’ interests. The contrast between Matt’s argument then:
Now of course the example doesn’t work as well if you can’t simply assume a perfect equilibrium price arrived at courtesy of the perfect information and zero transaction costs that characterize the rental market in Economoville but not, say, Washington, DC.
Simply pointing out that an “economics 101” model of frictionless labor markets and full employment is an abstraction does not eliminate its analytic utility in pursuing these questions. If anything, the existence of frictions makes me more skeptical of labor market regulations. In a frictionless marketplace regulations would reduce wages but with frictions it’s likely to increase long-term unemployment which is worse in social welfare terms.
is quite striking. It is even more striking, because the back-up claim in the second quote has no compelling empirical support. Comparisons across advanced industrial democracies find no evidence that increased labour market regulation (including a variety of regulations that one might expect to have much more direct consequences than e.g. body searches) lead to higher unemployment. Perhaps Matt might admit that the evidence is lacking, but argue, like James Heckman that we should still accord priority to the standard economics 101 story.
In the absence of better data, and better measurement frameworks, prior beliefs will continue to dominate how one interprets the evidence. This is not as much about dogmatism or conspiracy as it is about good science. In the absence of empirical evidence, logically consistent stories that accord with intuition have great appeal. At both an intuitive level and at the level of formal economic theory, incentives matter. If a person is paid not to work, the person will likely not work. If the costs of hiring a worker rise, fewer workers are likely to be hired.
However, this seems neither an attractive nor a compelling approach to me. A logically consistent story that accords with intuition plus four dollars or so will get you a horrendously overpriced venti latte at Starbucks.
The lesson here is straightforward. Simple economic models can be quite useful, but they should be employed with very considerable caution. In particular, one should always think carefully about whether the assumptions of your model blind you to factors that are important to the debate that you are applying them to. As a secondary matter, you should also look to the empirical support for your model – intuitively appealing models are frequently wrong.
As Deirdre McCloskey argued before she went loopy, one shouldn’t think about economic models as scientific proofs of anything. One should instead think of them as a particular form of rhetoric, where the parameters one picks at the start will have profound influence on the outcomes one predicts. If the math is right, the conclusions should be compatible with the starting assumptions, but that is about all one can say (internal consistency, while a good and useful thing, is obviously no guarantee whatsoever of external validity). More pungently, as dsquared put it many years ago
I’m pretty sure that it was JK Galbraith (with an outside chance that it was Bhagwati) who noted that there is one and only one successful tactic to use, should you happen to get into an argument with Milton Friedman about economics. That is, you listen out for the words “Let us assume” or “Let’s suppose” and immediately jump in and say “No, let’s not assume that”. The point being that if you give away the starting assumptions, Friedman’s reasoning will almost always carry you away to the conclusion he wants to reach with no further opportunities to object, but that if you examine the assumptions carefully, there’s usually one of them which provides the function of a great big rug under which all the points you might want to make have been pre-swept.
To be clear – I don’t for a moment think that Matt is Milton Friedman. But I do think that he has internalized some Friedmanesque ideas over the years. Starting assumptions have consequences. And when one starts with a family of models that presupposes no real hierarchy in the workplace, no distributional struggles over how bosses treat workers etc, one isn’t likely to get results that are helpful or relevant to aforementioned problems of hierarchy, bosses’ treatment of workers and so on. That’s a real problem.
Update Alex Tabarrok responds vigorously and inadvertently provides me with a very nice illustration of the problem that concerns me. After huffing indignantly about how “Henry seems to think that economists have never thought about their assumptions or tested their models” (I actually never say, nor imply this), he discusses a couple of Gruber papers which indeed appear to show a substantial wage-benefit substitution effect, and then gives us this.
And, dare I mention it, here is Joni Hersch in a recent paper in the AER on sexual harrasment:
Workplace sexual harassment is illegal, but many workers report that they have been sexually harassed. Exposure to the risk of sexual harassment may decrease productivity, which would reduce wages. Alternatively, workers may receive a compensating differential for exposure to sexual harassment, which would increase wages. Data on claims of sexual harassment filed with the Equal Employment Opportunity Commission are used to calculate the first measures of sexual harassment risks by industry, age group, and sex. Female workers face far higher sexual harassment risks. On balance, workers receive a compensating wage differential for exposure to the risk of sexual harassment.
Henry goes wrong because he doesn’t want to conclude that sexual harassment is ok but he thinks that the only way to deny that conclusion is to deny that wages are higher so he rejects the model, he rejects the assumptions that he thinks (incorrectly!) are driving the model and he assumes without looking for any evidence that wages are in fact not higher. (Talk about being blinded by assumptions!).
On the general question, see Peter Dorman. On the specific question of sexual harassment, an ungated version of the relevant article is available here for those who don’t have subscriptions to the AER. It’s indeed a remarkable piece. One of Alex’s commenters suggests that the author doesn’t do anything to convince anyone that the causal story is a good one; this is (very, very slightly) unfair. The author does mutter a bit about not having an instrumental variable, and waves vaguely in the direction of a strategy to identify causation. But what she does not do, at any point whatsoever, is to think about whether or not complaints about sexual harassment serve as a good proxy for actual sexual harassment in the workplace. As Alex’s commenter notes, there is a strong case that the causal story is precisely the reverse of what the author claims to have found- less well paid employees may be less likely to complain about sexual harassment. Not only that, but this prima facie provides a much better explanation of the findings than the author’s own argument. This bit from the paper is genuinely priceless.
The log wage difference between a job with zero sexual harassment risk and a job with the mean sexual harassment risk is 0.0155, or about 25 cents per hour for women, and 0.0252, or about 50 cents per hour for men. The large compensation for sexual harassment risk for men is surprising. One possible explanation is that since men infrequently file sexual harassment claims, those claims that are filed are particularly egregious, and exposure to such risk warrants a larger compensating differential than received by women.
So, on the one hand, we have a Tabarrok-world story in which power relations don’t exist, everyone is contracting fairly with everyone else, people are compensated for the higher risk of sexual harassment in particular industries with higher wages, and men demand and receive particularly high compensation for the risk of being sexually harassed in harassment risky industries. I don’t think that this claim passes the laugh test, myself. Clearly Alex disagrees, and I look forward to his fuller explication of this quite striking empirical finding. On the other, we have a story in which different levels of industry compensation are associated with different levels of bargaining power and reinforcing social norms, so that the likelihood of reporting sexual harassment varies substantially with compensation. This provides a far more plausible explanation of the awkward finding (it’s prima facie believable that working class guys are less likely to report sexual harassment, especially by male supervisors, than are men in better paid professions), but one that is invisible to a particular kind of economist, with the accompanying set of intellectual biases, for whom, as Alex so amply demonstrates, this is further evidence of how markets compensate for sundry forms of unpleasantness. “Talk about being blinded by assumptions!” how are ya. It perhaps says something more general about the economics profession that this paper got published in a high impact journal, despite its obvious problems (although in fairness, while Alex doesn’t mention it, the paper appeared in the AER’s annual papers and proceedings round-up rather than in the more strictly refereed part of the journal).