Plate x in print

The Stereotype

statistical discrimination; the group taken as a proxy for the person, and the prophecy that fulfills itself.

Where the employer cannot read the worker, he reads the group the worker belongs to, and treats the average as the man. The judgment may be statistically sound and still fall unjustly on every member who departs from his group's mean; and worse, it can come true in the telling, for a people judged not worth the investment will receive none, and so grow into the very averages that first condemned them.

— the finding of this plate, after Edmund Phelps, 1972

The cures this volume has gathered all suppose that the buyer wishes to read the seller truly — the signal sent to be believed, the screen built to sort the honest from the rest. But the cheapest reading of a person is often not his own at all; it is the group he belongs to, and there the asymmetry takes its darkest turn. The study of discrimination began by laying it to taste: the employer who simply dislikes a group and will pay, in talent forgone, to keep his distance from it — a prejudice costly to him and unjust to them, but at least intelligible as malice. Edmund Phelps, in 1972, with Kenneth Arrow reaching the same ground by another route in the same years, showed that discrimination needs no malice whatever. It can fall out of cold rationality, from an employer who wishes only to guess each worker's worth as nearly as he can. The plate's question is the one this opens: when the individual cannot be seen and only his group can, who bears the cost of the average?

The mechanism is nothing but inference under a veil. An employer must set a wage on a worker whose true productivity he cannot observe; he has a noisy test, a thin record, and one thing more — the worker's visible membership in some group, about whose average he holds a belief, right or wrong. A man bent on accuracy will not discard it; the laws of probability forbid it. If he takes the group's average to be lower, or takes the test to mean less for its members, then the rational estimate of any one of them bends toward the group and away from the individual standing before him. He indulges no taste; he does arithmetic, blending the weak private signal with the group average just as a statistician would. And so two workers with identical records are offered unequal wages because they wear different labels — a discrimination produced not by feeling but by the honest pursuit of a good guess.1

1. Statistical discrimination is inference under a veil. The employer cannot observe true productivity; he has a noisy test plus the worker's visible group membership, about which he holds a belief concerning its average. Probability forbids ignoring it: if he takes the average to be lower, the rational estimate of any member bends toward the group and away from the individual. Not a taste but arithmetic — and two identical records draw unequal wages because they wear different labels.

The bending works through two channels, and the second is the subtler. The first is a gap in averages: believe one group less productive, and every member is marked down toward that lower mean, the able among them taxed for their company. The second needs no gap in averages at all, only a difference in how well the signal reads. If the test predicts worth reliably for one group and poorly for another — because it was built around the first, or fits the second less well — then for the second group the employer trusts the individual score less and leans on the group average more, so that its strongest performers, whose high scores he discounts as noise, are paid beneath their merit while its weakest are paid above theirs. Either way the individual dissolves into the average, the more so the less he can be known apart from it.2

2. Two channels. First, a gap in averages: believe a group less productive and every member is marked down toward that mean, the able taxed for their company. Second, subtler: even with equal averages, if the test reads reliably for one group and poorly for another, the employer trusts the individual score less and the group average more — the strong performers discounted as noise, paid beneath their merit. Means punish the group; unreliability blurs the deserving into the rest.
⁂

Here the doctrine reaches the discovery that gives the plate its name, and it is the cruelest part. The discrimination can make itself true. Suppose a group is judged, rightly or not, to be the poorer bet, and is therefore offered less — lower wages, fewer of the good positions, slighter odds that effort will be repaid. Then its members face a thinner return on the costly business of acquiring skill, and fewer of them, quite rationally, will pay that cost; and as fewer invest, the group's real average sinks, until it arrives at the very figure the employer assumed. The belief, acted upon, manufactures its own confirmation: a low expectation withholds the reward investment requires, the investment is withheld, and the prophecy is fulfilled — a baseless guess hardened into a fact, then cited as its own vindication. The group is locked, by nothing but the expectation of its poverty, into the poverty expected: a trap with no villain, sprung by arithmetic.3

3. The discrimination can make itself true. A group judged the poorer bet is offered less — lower wages, slighter odds effort pays; its members face a thinner return on acquiring skill, so fewer pay that cost, and the real average sinks to the figure assumed. The belief manufactures its confirmation: low expectation withholds the reward investment needs, investment is withheld, the prophecy fulfilled. A baseless guess hardens into a fact then cited as its own vindication — a trap with no villain, sprung by arithmetic.

This self-confirming character is what makes statistical discrimination so much harder to dislodge than the prejudice it is sometimes mistaken for. Mere distaste carries the seed of its own defeat: the employer who turns away good workers for their group leaves them to be hired, cheap and able, by a rival who does not share his scruple, who should in time undersell him — so taste-based discrimination is, in theory, competed away. The statistical kind enjoys no such cure. It is not a preference to be punished but a calculation to be repeated, and every profit-seeking employer has the same reason to make it; competition does not erase it but enforces it. Worse, where the prophecy has fulfilled itself, the averages really have diverged, so the discriminating inference is now statistically correct, and the market rewards rather than penalizes the firm that draws it. Rationality, profit, and even accuracy line up behind the practice, and none of the forces that grind down ordinary prejudice can find any purchase on it.4

4. Self-confirmation makes it harder to dislodge than prejudice. Distaste carries its own defeat: the bigot leaves able workers to be hired cheap by an unprejudiced rival, who should undersell him, so taste-based discrimination is competed away. The statistical kind is not a preference to punish but a calculation to repeat; competition enforces rather than erases it. And where the prophecy fulfilled itself, the averages truly diverged, so the inference is now correct — rationality, profit, and accuracy all behind it.

So the plate closes on a failure with no malice in it, only a want of knowledge. The same hidden quality that emptied the market for used cars works here upon people: when a worker cannot be seen for himself, his group stands in as the proxy, and he is priced by its average and not his merit — the able made to pay for their company, and whole groups held by the self-fulfilling turn in the low equilibrium that expectation digs for them. The asymmetry wears here its most human face: the cost of unequal information falling not on goods but on persons, judged by what their kind reveals because they cannot be read as themselves. And the answer to the plate's question follows from its diagnosis: the cure is not to scold the employer, whose inference is sound, but to lift the veil — to give the individual a signal strong and clear enough that he can be known apart from his group, which is precisely the office of the credential the volume met before. Where the worker can prove himself, the average loses its grip; where he cannot, the group decides, and the prophecy waits to be fulfilled.5

5. A failure with no malice, only a want of knowledge. The hidden quality that emptied the used-car market works on people: unseen for himself, the worker is priced by his group's average, the able paying for their company, whole groups held in the low equilibrium expectation digs. The cure is not to scold the employer, whose inference is sound, but to lift the veil — a signal strong enough that the individual is known apart from his group. Where he can prove himself, the average loses its grip.