Two workers can bring the same skill to the same work and still be treated differently because of who the employer takes them to be. The preceding plate gave unequal pay a possible explanation in the conditions of the job; this one holds those conditions steady and asks what happens when the worker's identity enters the price. The question begins before the pay slip, in who is considered, interviewed, hired, and allowed to acquire the experience that later counts as worth. A wage gap observed between groups gathers all these stages together, along with differences in the work and the workers themselves. To understand the gap we need to distinguish a preference for excluding someone from a belief about what that person will do, and to distinguish both from the quantities our evidence actually measures. Gary Becker gave the first mechanism a precise economic form: prejudice can enter the chooser's preferences and change the bargain even when the worker's contribution is fully known.
Becker's employer acts as though hiring a disfavored worker carries a cost beyond the wage. An employer who feels a money-equivalent penalty of two dollars an hour for employing a member of a particular group compares that worker's ten-dollar wage with an effective cost of twelve. The penalty is borne in the employer's preference, while its consequence falls on the worker's opportunities: to be equally attractive, the disfavored worker must accept less, offer more, or find an employer who does not impose the penalty. This is the taste for discrimination. It makes the same productive service worth different amounts to different purchasers because their preferences include the identity of the person supplying it. Prejudice can therefore coexist with deliberate, consistent choice. The employer may understand perfectly well that the rejected worker is able, and may be willing to sacrifice profit to avoid the contact. The model takes that willingness seriously enough to put a price on it, and then asks how the cost is divided between the person who indulges the preference and the people against whom it operates.1
The resulting market gap depends on which employers the workers must reach to find employment. Imagine a range of firms, some willing to hire without the identity penalty and others imposing larger ones. A disfavored worker who can move to a welcoming firm has an alternative to accepting the prejudiced firm's discount. As the demand from welcoming firms expands, that alternative becomes stronger. But the relevant question is how many workers those firms can employ efficiently, with the required mix of skills and capital, and at what wage. The mere existence of an unprejudiced employer supplies no guarantee that every excluded worker can find a place there. The distribution of prejudice and the structure of production meet in the market: the employers whose hiring is needed to absorb the workforce can affect the terms available to it. The prejudice of the most hostile firm therefore need not set the group's wage, and a given amount of hostility can have very different effects in different economies. Access to another purchaser is part of the mechanism, just as it was in the first plate's bargain.2
Competition can make employer prejudice expensive, while leaving other forms of discrimination profitable to accommodate. A firm that forgoes equally productive, less expensive workers raises its costs relative to a rival willing to hire them. Entry and expansion can press on that disadvantage. Yet the customers may themselves refuse to buy from certain workers, or fellow employees may demand compensation for working alongside them. A manager seeking profit then faces the revenue or staffing consequences of other people's tastes. The discipline of competition does not automatically remove those preferences. Becker emphasized the conditions on the long-run result, including production possibilities and the persistence of employee and customer discrimination. A separate mechanism arises when the employer uses group membership to infer an individual's hidden productivity. The Asymmetry volume follows that mechanism in The Stereotype, including the investment choices through which expectations can become self-confirming. The two mechanisms can operate together. Seeing a discriminatory outcome tells us to investigate the preferences, beliefs, and constraints that produced it; the outcome by itself supplies no complete account of the employer's mind.3
An observed wage gap cannot be divided cleanly into worth and discrimination by controlling for a list of worker characteristics. The familiar accounting exercise separates a part associated with measured differences in education, experience, hours, or occupation from a remainder those variables leave unexplained. The remainder can include productive characteristics the researcher failed to observe. The explained part can also include the consequences of earlier discrimination: exclusion from training, promotion, or a better-paying occupation changes the very characteristics on which the comparison is conditioned. If a barrier channels otherwise similar workers into different jobs, controlling for the job can remove that channel from the measured gap. The calculation remains useful, but the question it answers depends on what has been held fixed and on how those circumstances arose. The decomposition divides the gap; it cannot tell us on which side of the division the discrimination fell. A smaller residual may reflect a more complete account of productive differences, or a comparison that has set aside more of the process we wanted to understand. Causal interpretation requires an argument about that process, beyond the arithmetic.4
A correspondence experiment can isolate unequal treatment at a particular gate by changing what the employer sees through random assignment. In their 2004 study, Marianne Bertrand and Sendhil Mullainathan sent fictitious applications to help-wanted advertisements in Boston and Chicago. They randomly assigned names associated with White or African-American applicants to resumes, including within the levels of credentials they constructed. The White-associated names received about fifty percent more callbacks. Random assignment gave the name signal a causal interpretation in this application setting: the difference in callbacks followed the manipulated name rather than a systematic difference in the submitted qualifications. The measured outcome was an invitation to continue, before any interview or work had taken place. It establishes unequal access to that next stage. It leaves open how much of the response arose from tastes or from what employers inferred through the names, and it gives no direct estimate of the wage that an eventual hire would receive. Those boundaries make the finding more intelligible. The experiment identifies a consequential act of differential treatment, at a named stage, without asking a wage residual to reveal an entire working life.5
Discrimination changes the terms on which a worker can turn ability into a livelihood, and it can change the opportunities through which that ability develops. The wage then carries the history of doors opened and closed, alongside the price of the service presently offered. Becker's contribution was to show how the exclusion could enter purposeful economic behavior, while the evidence teaches us to locate the act we can actually observe: a foregone interview, a different offer, a blocked route to a better job. The practical implication follows the mechanism. More alternatives can weaken an employer's ability to impose a personal preference; customer prejudice, information problems, and barriers to acquiring skills call for attention to their own channels. Equal treatment cannot be inferred from competition alone, any more than every raw pay difference can be read as its absence. The next plate introduces a different institution into the bargain: a minimum payment required by law. It establishes a floor beneath the wage of the covered worker; who reaches that job, and on what terms beyond the wage, remains part of the inquiry.