When Is Discrimination Profit-Maximizing?
In the textbook world, many firms compete to fill identical jobs, many qualified workers stand ready to take them, information about productivity is perfect and free, and moving between jobs costs nothing. In that frictionless setting—Gary Becker’s benchmark—an employer that indulges a “taste for discrimination” pays for it. By excluding equally productive workers from group X, the firm either hires weaker workers or pays more to recruit from a smaller pool; non-prejudiced rivals draw on the full pool and win business. Competition, in this narrow model, grinds bias into dust.
The elegance of that result is also its weakness. Its assumptions are knife-edge. Once we relax them toward how labor and product markets actually work, discrimination can again be privately optimal even when firms face competition. The first crack appears when customers or co-workers carry the prejudice. If buyers prefer not to be served by members of a particular group, or if teams impose disutility from working with them, the firm’s profit function changes. Sorting workers into back-office versus front-office roles, segmenting stores or product lines, or differentially staffing locations can raise demand or reduce internal frictions without any animus inside the firm. That logic helps explain familiar patterns: mid-century “whites only” job ads and dual wage scales in the United States, racial steering and differential treatment in housing markets, and, in other societies, caste-based sorting in customer-facing services when clients insist on it. The point is not moral approval but equilibrium arithmetic: if buyers reward discriminatory sorting, a lone, virtuous rival cannot easily profit by defying revealed tastes.
A second crack opens when information is noisy. Edmund Phelps and Kenneth Arrow formalized statistical discrimination: when individual signals are imperfect or costly, Bayes’ rule tempts employers to use group averages as proxies. That move can be privately rational yet socially damaging. Worse, it can become self-fulfilling. As Coate and Loury showed, if a group anticipates biased screening or slower promotion, the expected return to investing in costly signals—schooling, certifications, even job search—falls. Lower investment then makes the biased prior look more accurate ex post. Customer-taste and statistical channels often braid together: biased customer responses depress measured performance in sales or service roles, which rationalizes biased evaluations and pay, which deepens the prior. What looks like “the market learning the truth” may be the market amplifying a belief it helped create.
A third crack is mobility. Costless switching is a fairy tale. Jobs arrive through networks; locations and schedules are sticky; non-wage amenities matter; contracts and norms limit flight. Standard search and bargaining models deliver residual wage-setting power—monopsony—whenever offers are lumpy and moving is costly. With that power, a firm can sustain group-specific wage policies or role assignments even in the shadow of competitors. A little frictions plus a little customer preference go a long way: small wedges can support durable gaps that no single entrant can profitably overturn.
The evidence aligns with these mechanisms rather than the frictionless ideal. Résumé audits find different callback rates for identical credentials once names cue race or gender. Field studies at car dealerships show persistent price gaps by buyer identity, consistent with customer-facing bargaining interacting with stereotypes. Online platforms that put photos and names up front see acceptance disparities that shrink when identity cues are delayed or standardized. Housing paired tests continue to document steering and unequal treatment decades after formal bans. Flip the design levers and behavior moves: when orchestras adopted blind auditions, women’s advancement rose—an empirical nudge that points to information design rather than markets “discovering” fairness unaided.
Policy and product design naturally follow from mechanism diagnosis. If the profitable channel is customer prejudice, laissez-faire does not solve the externality; civil rights law changes the objective function by increasing the cost of discriminatory strategies and protecting third parties who bear the harm. If the channel is informational, then blinding (anonymized résumés, standardized rubrics, delayed identity cues) and better signals reduce reliance on noisy priors at low efficiency cost. If search frictions and monopsony dominate, stronger outside options—transparent pay bands, limits on noncompetes, easier switching—restore competitive discipline. None of these instruments is a panacea; which one works best depends on where the distortion lives.
This leaves the tidy claim that “markets will fix it” on a much narrower footing. The benchmark needs three pillars to stand: only employers have tastes, information is perfect, and mobility is free. In the real world, customers and co-workers often carry the tastes, signals are noisy and sometimes strategically curated, and mobility is neither free nor universal. Under those conditions, discrimination can be profit-maximizing and persist in competitive equilibrium. That is not a counsel of despair. It is a call to precision: identify the operative channel, design a test that shifts the relevant margin, and choose the instrument—law, platform rules, or both—that removes the profit from bias at the lowest efficiency cost.
Two practical payoffs follow. First, each mechanism yields testable predictions: customer-taste discrimination should shrink when identity cues are muted or customer mix shifts exogenously; statistical discrimination should shrink when signal quality improves or is blinded; monopsony-supported gaps should shrink when switching costs fall. Second, the evaluation criterion can be welfare, not vibes: which interventions reduce unjustified disparities while preserving or improving allocative efficiency? Markets are capable but not magical. Where preferences, information, and frictions align to make bias pay, competition does not neutralize discrimination—it monetizes it. The job of policy and design is to realign the payoffs so that fairness and profit point in the same direction.
The Taste For Purity
If people prefer Brahmin cooks (for instance, see here, here, and here), the profit-maximizing strategy for restaurants is to look for Brahmin cooks (for instance, see here).