Tuesday, June 21, 2011

Why Would Wal-Mart Hire Any Men?

If you read the comments (yeah, I know) to this blog post about the Wal-Mart case, you will come across the old chestnut about sex discrimination in earnings. It goes like this:
If women were really as good workers as men but were paid less because of discrimination, why would any firm hire men at all?
The implication is an obvious one: Wal-Mart does NOT discriminate against women. The reason women earn less is that they deserve less, being less capable workers. If this were not the case, wouldn't Wal-Mart just hire all women at those lower wages?

Or put in different terms, a firm run by someone who doesn't discriminate against women would use this information to hire an all-female labor force. Because women would cost less to employ for the same levels of productivity, that firm would have higher profits than the discriminating firms. Over time firms of the former kind would take over the industry, and -- presto! -- discrimination would die a natural and painless death.

This argument was initially made by Gary Becker, an economist, a very long time ago. It is not an uncommon argument from conservatives (or from certain types of anti-feminist sites.) That does not mean that it shouldn't be discussed. So let's do that by looking at what is unrealistic about the specific conclusions.

To do that, the model to begin with is Becker's basic model. He covered the concept of discrimination (whether based on race, gender or other group characteristics) as a dislike (or hatred) by either the owner/managers of the firm, by the co-workers or by the customers. AND he covered discrimination in an artificial model where all participants know all relevant information. That information includes the true productivity of every single worker.

Most of the time Becker treated the three possible discriminating identities separately.
Thus, in the first model only owner/managers have a dislike towards workers from a particular group. That, my friends, is the model from which the above conclusion comes, though even then it would only work to eradicate discrimination from the whole industry if the industry was essentially a competitive one. If the industry is not sufficiently competitive, the bigoted owner/managers can hang on and practice discrimination.

Did that read as rather dry? I can't think of a juicier way of telling the story, but basically Becker argues that if the only problem we have consists of some bigoted owner/managers, while everyone else is just so sweet, sufficiently well-lubricated markets can get rid of those nasty bigots, always assuming that everybody knows everything relevant about everyone else.

There are no misconceptions in the model. Even the bigoted owner/managers of a pizza parlor, say, know that Joe and Jane are equally good pizza-bakers. They just hate Jane and are willing to hire her only if they can get her for less money.

This cannot last if we can find at least non-bigoted nice owner/manager.

That's the background of the old chestnut. Becker, having become immersed in the imaginary world of his simple model, concluded that competitive industries would never exhibit any long-run sex or race discrimination. Only oligopolies or monopolies could survive with at least some bigoted owner/managers.

But what happens if we let not only the owner/managers dislike poor Jane but also the customers Jane serves?

Suppose that some customers storm out when Jane serves them but none do when Joe serves them. In what sense are the two still equally productive? Now Jane won't bring as much revenue to the firm, despite being an equally good pizza-baker. In fact, from the firm's point of view she is a worse employee than Joe, because she will contribute less to the firm's profit. And this has nothing to do with her baking ability.

The situation has changed, has it not? Even a sweet non-discriminating pizza parlor owner might have to get rid of Jane now. Even if her baking skills exactly equaled Joe's baking skills.

The point of this sub-model may be the fact that perfect information about a worker's productivity, a completely unrealistic concept, still wouldn't be sufficient to keep the discrimination against Jane separate from her ultimate productivity. Jane would end up either fired or earning less than Joe.

Which sorta cracks the chestnut, at least in service occupations.

The various Becker models can be played with to produce differing predictions. But they all suffer from that perfect information assumption. Information about people's actual productivity is not freely available. If it was, no firm would ever hire someone unsuitable or subject new workers to trial periods and such. Indeed, if everybody knew everything else life would be very different from how it actually is.

More realistic models allow for missing information about the true productivity of workers, at least in the short-run. What takes its place? One possibility is statistical discrimination.

Suppose that Jane has applied for a job, gotten an interview, and now walks into the firm's human resources office, clad in a neat suit, carefully made-up and carrying her resume. How will the representative of the firm on the other side of the desk assess her likely productivity?

One possibility is that the interviewer will use both the information in Jane's resume, the information produced in the interview and the interviewer's own ideas about how people who "look like" Jane have fared in the past. Or how that person regards them in general. That Jane is female is part of that information.

This might mean that the way the interviewer sees women in general could affect the way Jane is viewed. If the company has never hired women before, Jane presents it with a risk. If the interviewer doesn't think much of women's abilities in the advertised job opening, Jane may be rated lower than an equally qualified applicant who just happens to be called Joe.

More realistically, bigoted views about a certain group of workers can seldom be fully separated from their assessed productivity. Someone who dislikes Jane because of her gender will find fault with her pizzas, too. There are few jobs where the measurement of productivity is easily disentangled from the overall impression a worker gives someone. And it's the latter which often determines pay raises and promotions.

Which is a long way to say that firms mostly cannot tell the true productivity of their workers. It's bound up in the same tight bundle with all sorts of views, including possibly discriminatory ones, as well as personal likes and dislikes. A Wal-Mart manager who expects women to quit to have children will promote fewer women to management, not because she or he hates women but because of the use of the class "women" in the evaluation of all individuals in it.

And yet the basic Becker model doesn't even allow for such motives!

Neither does it include social norms. Social norms matter. A lot. Yet they are absent in that old chestnut, as well as in Becker's general work.

To see why they matter, suppose that in a certain society the social norm states that women should not supervise men at work. Suppose, also, that there are quite a lot of trained women in that society who could do the job of supervision, defined in the technical sense. Borrowing from Becker's world, wouldn't an owner/manager who doesn't care about this social norm have a great opportunity there? Just hire all these women for management jobs! They will work for less, given the lack of opportunities the social norm causes, right?

To make things really simple, suppose, finally, that there are enough male workers who don't mind a female supervisor and enough nice customers to patronize this firm. I have set up the whole thing so that great profits are an obvious conclusion.

Until the firebombs and the broken windows and the death-threats, perhaps, to make things lively. If the rest of the society cares enough about the upholding of this particular social norm, the fact that all the direct players don't care about it is irrelevant. The firm which violates the social norm will be punished. Perhaps not with violence but economic boycotts would work, too.

The point of this example is that well-lubricated markets and non-bigoted direct participants may not be enough to end discrimination against women or minorities.

This post has gotten very long, for which I apologize. I could have written even more about the problems of taking simple economic models for reality and the mistakes that causes. And on empirical findings which contradict the conclusions of the old chestnut. And so on.