The Discrimination

Set down the central fact first, plainly, because everything that follows is an attempt to explain it. In the United States women earn somewhere between eighty-two and eighty-five cents for every dollar earned by men — the exact figure turning on whether one counts hourly or weekly or annual pay, and full-time workers or all — and Black workers earn substantially less than white workers of the same education.¹ Some of these gaps reflect real differences in the work: in hours, in occupation, in experience. But not all of them do, and the part that does not is the subject of this plate.

The plate on the margin held that the wage tends to the value of the marginal product; the plate on skill explained why that value differs across workers. Between them they account for an enormous amount of the variation in pay. But they share a single assumption, and discrimination is the violation of it: that two workers with the same marginal product are paid the same. Discrimination is the wage responding to who the worker is rather than to what their work is worth — and the question is how a market is supposed to permit such a thing, and why it so persistently does.

The founding economic theory of discrimination is Gary Becker's, and its first move was to take prejudice seriously as an economic preference. Some employers, Becker supposed, simply do not wish to employ workers of a certain group — they have, in his deliberately clinical phrase, a “taste for discrimination.” Such an employer behaves as if hiring a disfavored worker carried a cost beyond the wage: as if a worker who must be paid a wage w in fact cost w plus some further, psychic penalty — the measure of his distaste.² He will therefore take on the disfavored worker only at a discount, only if their wage falls below the favored worker's by enough to offset the penalty he feels. The wage gap, on this account, is the price the disfavored must pay to overcome the prejudice; it is, quite literally, the cash value of the distaste.

The crucial feature of Becker's taste — the one that makes it more than a description of bigotry — is that it costs the bigot. An employer who turns away productive workers, or pays a premium for favored ones he might have hired more cheaply from the disfavored pool, is leaving money on the table. Prejudice, in Becker's hands, is not free; it is a luxury, and the employer who indulges it pays for it in forgone profit. From this single observation flows the theory's most famous, and most contested, prediction.

If discrimination costs the discriminator, then in a competitive market it should not survive. Imagine, as the diagram does, a pool of disfavored workers facing a range of employers — some deeply prejudiced, some mildly so, some not at all. The unprejudiced employers feel no penalty; they will gladly hire the disfavored at their full productivity, for to them these are simply cheap and productive labor. So long as there are enough such employers to absorb the whole disfavored workforce, competition among them bids the disfavored wage up to its full value, and the gap vanishes. The prejudiced firms, refusing the cheaper workers, carry higher costs and earn lower profits; in a competitive market they contract, while the colorblind firms, enjoying their cheaper labor, expand. Competition, on Becker's logic, is the enemy of discrimination — it rewards the firm that ignores race and sex and penalizes the one that does not, and over time should grind the wage gap away.³

This is an arresting result, and it has been pressed — not always in good faith — into the claim that markets cure discrimination unaided, and that intervention is therefore needless. But the argument cuts the other way with equal force, and this is the edge the volume takes from it. For if competition should erode taste-based discrimination, then the survival of large wage gaps, decade upon decade, is itself a finding: evidence that one of Becker's premises fails. Perhaps there are too few unprejudiced employers to absorb the disfavored workforce. Perhaps competition is simply too weak to punish the prejudiced — which is to say the labor market carries the monopsony power of the eighth plate, and an employer sheltered by that power can afford his tastes as a monopsonist can afford anything. Or perhaps the prejudice does not reside in the employer at all, but in customers who will not buy from the disfavored, or co-workers who will not stand beside them — lodged outside the firm, where the discipline of competition cannot reach.⁴ The market does not reliably cure discrimination; that it has not is one more measure of how far real markets stand from the competitive ideal, and one more pointer to power.

There is a second theory of discrimination, and it is in some ways more disturbing than the first, because it requires no prejudice at all. Edmund Phelps and Kenneth Arrow set it out in the early 1970s. Suppose an employer cannot directly observe an applicant's productivity, but must guess at it from a résumé, an interview, a few imperfect signals. And suppose the employer believes — rightly or wrongly — that the average productivity of one group is higher than another's, or merely that it is harder to gauge for one group than another. A purely rational employer, with no animus whatever, will then use the applicant's group as a proxy for what cannot be seen, judging the individual by the statistics of their group. This is statistical discrimination, and it yields a wage gap from cold inference alone.⁵

The statistical kind is more insidious than the taste-based for three reasons. It needs no hatred, so it cannot be waved away by appeal to the good intentions of employers. Competition does not erase it, because using cheap group information is no costly indulgence but a rational economy — the colorblind firm has no advantage to exploit. And it can be self-fulfilling: if a group is paid and promoted less because employers expect less of it, its members have less reason to invest in skills whose reward they will not fully reap; they invest less; and the lower investment confirms the original belief, locking the group into a discriminatory equilibrium that no one need intend and that can persist indefinitely.⁶ Worst of all, it wrongs the individual even when the group average is accurate: the able worker from the disfavored group, judged by a statistic they exceed, is paid for what their group is rather than for what they are.

How much of the observed gap is discrimination of either kind? The question is far harder than it sounds, and the difficulty is itself instructive. The standard tool, devised in 1973, splits the raw gap between two groups into two parts: an “explained” portion, attributable to measurable differences between them — in education, experience, hours, occupation — and an “unexplained” residual, the gap that survives once those differences are accounted for. The residual is then often read as a measure of discrimination: the part of the gap that equal characteristics cannot explain.

But the decomposition cannot cleanly deliver what we ask of it, and it fails in two opposite directions at once. The residual overstates discrimination insofar as it absorbs productivity-relevant things we did not measure — effort, specific skills, job demands invisible in the data. And it understates discrimination insofar as the “explained” portion is itself a product of discrimination. This second problem is the deeper one. If women are channeled into lower-paying occupations, or Black workers into lower-paying firms, by discrimination, then “controlling for occupation” obediently subtracts the very effect we set out to find — it explains the gap by the discrimination and pronounces the discrimination explained. The same holds for the pre-market gaps in skill the plate on human capital described: a group that, anticipating a lower return, invests less in schooling will show a skill deficit that is counted as a difference in worth when it is partly a residue of expected discrimination.⁷ The decomposition divides the gap; it cannot tell us on which side of the division the discrimination fell.

For that, economists turned to a cleaner instrument: the audit, or correspondence, study. Send out pairs of fictitious job applications identical in every particular — same qualifications, same experience, same wording — differing only in a name that signals the applicant's race or sex, and count the callbacks. Productivity is now held exactly constant by construction, so any difference in the employer's response is discrimination and nothing else. The results are unambiguous and have proven stubbornly stable. In the best-known American study, résumés with white-sounding names drew about fifty percent more callbacks than identical résumés with Black-sounding names; a later synthesis of decades of such studies found that hiring discrimination against Black applicants had scarcely declined since the late 1980s.⁸ Whatever the decomposition leaves unresolved, the audit studies settle the threshold question. Discrimination in the labor market is not a hypothesis. It is measured, and it is real.

The gender wage gap is worth dwelling on, both because it is the most studied of all wage gaps and because its modern anatomy overturns the simplest story about it. The raw gap — the eighty-two to eighty-five cents already cited — sounds like a tale of women paid less for the same work. Mostly, it is not. Compare men and women in the same job at the same firm and the gap nearly closes: by one large estimate women earn about ninety-nine cents on the dollar for the same role.⁹ And among the young, before careers diverge, the gap is small — women aged twenty-five to thirty-four earn around ninety-five cents on the dollar. The raw gap is built far less from unequal pay for identical work than from men and women holding different jobs, different hours, and different rungs on the ladder. Which only moves the question back a step: why do they?

The answer that has reshaped the field is Claudia Goldin's, and it turns on time. The labor market, she observed, does not pay in proportion to hours; it pays a disproportionate premium for long, continuous, and unpredictable ones — for the lawyer always available, the manager who never leaves. These “greedy jobs” reward a kind of temporal commitment very hard to combine with the running of a home and the care of children.¹⁰ When a couple has a child, someone must take the flexible, interruptible role and someone may take the greedy, lucrative one; and overwhelmingly, still, it is the woman who takes the first and the man the second. The gap, on this account, is less a paymaster discriminating against women than the structure of work penalizing flexibility — with the need for flexibility falling, by custom and default, on mothers.

The data show it with uncommon clarity. Track the earnings of men and women around the birth of their first child, and the two, which moved together before, split abruptly at the cradle: the mother's earnings drop and never fully recover, while the father's continue as though nothing had happened. This child penalty is now the single largest component of the gender earnings gap in the rich countries, and it has barely narrowed even as the gaps in education and experience have closed or reversed.¹¹ It is a subtler thing than a prejudiced paymaster, and in a way more stubborn, because it is built into the joint architecture of the labor market and the family — into which jobs pay most, and who is expected to step back when a child arrives. This is a different mechanism, too, from the one the eighth plate found, where women's weaker propensity to quit handed employers monopsony power and a deeper markdown; here it is the structure of hours and the unequal burden of care. The two compound, and neither is the wage rewarding worth.

Return the plate to the bargaining range, and discrimination finds its place in the volume's argument. The wage, the opening plates held, settles within a band: the ceiling set by the value of the work, the floor by the worker's outside options, the position decided by power. Discrimination works on all three at once, and against worth at every point. It lowers the floor — the disfavored worker's outside options are worse, because the prejudice that depresses their wage at one firm depresses it at the next, leaving them less able to walk away and so more exposed to whatever this employer offers. It can lower the position within the range, as a prejudiced or monopsonistic employer presses the disfavored wage down toward that lowered floor. And by shaping who invests in skill and who is steered into which work, it reaches even to the ceiling, depressing the measured worth of those it has discouraged. Discrimination is the clearest case in the whole volume of a wage set by something other than the value of the work — a verdict passed on the worker's group, not on their product.

What, in the end, can be said with confidence, and what cannot? That discrimination exists, the audit studies establish beyond reasonable doubt: identical workers are treated unequally by trait. That it is in part a taste and in part a statistic, the theory makes plain, with the proportions varying by group, by market, and by era. That competition has failed to erase it tells us how far the labor market sits from the frictionless ideal in which it would — that there is power in it, and prejudice lodged where competition cannot reach. What cannot be said with confidence is the precise share of any given gap that current labor-market discrimination explains, as against unmeasured differences in the work, or the long shadow of past discrimination working through skill and expectation and the unequal division of care.¹² The honest course is to report the gap, the audit evidence, and the bounds — and to resist both the denial that calls it all choice and the overstatement that calls it all bias. The truth lives, as usual, in the contested middle, and shifts with the case. The remedies differ accordingly, for a taste and a statistic do not yield to the same cure.¹³

But the volume's argument does not need the precise share. It needs only what is not in doubt: that some of the wage, for some workers, is set by who they are rather than by what their work is worth. That is enough to finish what the plate on skill began — to show that the wage is not, and has never been, a clean measure of worth. The ceiling may be set by the value of the work. The floor, for the disfavored, is set lower than worth alone would set it; and where, in the narrowed band above that lowered floor, the wage finally comes to rest is decided — here as everywhere in this volume — by power.