Discrimination in Labor Markets: How Economists Measure It

What this means

In the basic model of a labor market, an employer pays a worker according to what that worker's output is worth. Two workers with identical productivity should command identical wages, because any employer offering less would lose them to a competitor. Labor market discriminationTreating equally productive workers differently on the basis of a characteristic unrelated to their productivity. is the case where that does not happen: workers who are equally productive are treated differently because of race, gender, age, or another characteristic that does not affect what they produce.

Economists model this through two distinct mechanisms, and keeping them separate is essential because they predict different things. Taste-based discriminationDiscrimination arising when an employer, coworker, or customer has a preference against interacting with members of a group. treats the preference as a cost the discriminating party is willing to bear. An employer who declines to hire a productive worker gives up that worker's output, which means discrimination is expensive for the firm doing it. This yields a testable prediction: in fiercely competitive markets, firms bearing that cost should lose ground to firms that do not, so taste-based discrimination should erode over time. Whether it erodes fast, or at all, depends on how competitive the market actually is.

Statistical discriminationDiscrimination arising when an employer uses group averages to predict an individual's productivity because individual information is costly to obtain. works differently. Here the employer is not acting on dislike but on incomplete information. Screening applicants is expensive, so an employer may lean on average characteristics of a group as a proxy for an individual's likely productivity. The individual applicant is then judged by a group average that may not describe them at all. Note the uncomfortable implication of this model: because it can be profitable rather than costly for the firm, market competition alone does not necessarily eliminate it.

The measurement problem is where the real intellectual work sits. A raw earnings gap between two groups is not evidence of discrimination, because it bundles together occupation, hours, experience, education, industry, and geography along with any discrimination present. Economists therefore use residual analysisA method that statistically controls for measurable productivity-related factors and examines the earnings difference that remains., comparing workers matched on observable characteristics and examining what gap survives. The residual is not a clean measure of discrimination either, since it also contains anything relevant that was not measured, and researchers say so explicitly. To get closer, economists run correspondence studiesField experiments sending otherwise identical applications that differ only in a signal of group membership, then comparing callback rates., in which matched fake resumes differing only in a name or a graduation year are sent to real job postings. Because everything else is held identical by construction, differences in callback rates isolate the effect of the signal itself.

The standard's second sentence connects this to the previous two lessons. If a group faces reduced hiring, lower wages, or slower advancement, the effect compounds across a career and across generations, which shows up as persistence in the distribution and as reduced mobility.

Why it matters

This is a topic where people argue past each other constantly, and almost always because they are citing different quantities. One person cites a raw gap, another cites an adjusted residual, and both are accurate about their own number while describing different things. Knowing which quantity is on the table, and what it does and does not control for, is what lets you evaluate the claim instead of picking a side.

It matters practically too. Age discrimination in hiring is a live issue for workers in their fifties and sixties, and the mechanisms above are what employment law attempts to address. Understanding the difference between a preference-driven refusal and a screening shortcut explains why different policy tools are aimed at each: anti-discrimination enforcement targets the first, while credential transparency, skills-based hiring, and blind screening target the second.

Try it

1. Start with the definition, because the whole analysis depends on it. Write the economic definition of labor market discrimination in your own words, and then explain why "workers in group A earn less than workers in group B" does not by itself satisfy that definition.
2. Build the causes list for a real raw earnings gap. Find a published raw gap between two groups from the Bureau of Labor Statistics or the Census Bureau, record the source and year, and then list every factor that could contribute to it: hours worked, occupation, industry, years of experience, educational attainment, field of study, geography, and firm size, along with discrimination. This list is the reason the raw number cannot be interpreted directly.
3. Now find an adjusted or residual estimate for the same comparison. Record what the researchers controlled for. Compare the adjusted figure to the raw figure and describe what accounts for the difference between them.
4. Interrogate the residual honestly and in both directions. Explain why the unexplained portion may overstate discrimination, since unmeasured productivity factors land in it. Then explain why it may understate discrimination, since some of the controls, such as occupation, may themselves be shaped by discriminatory access. This second point is the one students miss, and it is why economists debate which controls belong in the model.
5. Distinguish the two mechanisms with concrete cases. Write one scenario illustrating taste-based discrimination and one illustrating statistical discrimination. Then state, for each, whether it costs or benefits the firm, and what that implies about whether competition would erode it.
6. Design your own correspondence study. Specify the occupation, the exact matched pair of applications, the single characteristic you would vary, your sample size, and your outcome measure. Then write down what your design controls for by construction and what it still cannot rule out, including that a callback is not a job offer.
7. Trace effects across all three stages the standard names. For hiring, wages, and career advancement, describe how discrimination would operate at each stage and what evidence would detect it. Note that advancement is the hardest to measure and explain why, since promotion decisions involve subjective evaluation and small sample sizes within any single firm.
8. Connect to the adjacent lessons explicitly. Explain the mechanism by which reduced hiring, lower pay, or slower promotion at one stage of a career compounds into a lower lifetime earnings path, and how that shows up in distribution data and in mobility data as separate effects.
9. Evaluate remedies analytically rather than normatively. For each of anti-discrimination law with enforcement, pay transparency requirements, blind or skills-based screening, and increased labor market competition, identify which mechanism it targets, state the predicted effect, and name a cost or limitation. Note honestly where empirical evidence on effectiveness is mixed.
10. Write a short methods critique. Take one published claim about discrimination from any source and evaluate it purely on its evidence: what quantity is it reporting, what design produced it, what does that design establish, and what would you need to see to be more confident.

Check yourself