Gender pay gap: measures and mechanisms

The first question is what was measured.

A gender pay gap compares earnings for groups. It does not automatically compare people doing identical work, and it does not become meaningless because the groups differ. A broad earnings measure describes a social outcome; a narrower comparison helps investigate particular mechanisms. They answer related but different questions.

The OECD indicator uses the difference between men's and women's median earnings relative to men's median earnings, generally for full-time wage and salary workers.^[1] Other measures use hourly pay, annual earnings, means or different populations. Comparing percentages without comparing definitions can create a false disagreement.

Decomposition is a model, not a verdict.

The BLS explicitly cautions that its broad comparisons do not control for factors such as experience, skills and job responsibilities.^[2] Analysts can compare more similar workers using statistical models, but the resulting “adjusted gap” depends on available data, measurement quality and choices about what to hold constant.

Occupation illustrates the difficulty. Holding occupation constant can help ask whether earnings differ within a field. It can also remove a pathway through which unequal access to training, recruitment or promotion affects earnings. A variable can be useful for one question and conceal part of the mechanism relevant to another.

Likewise, an unexplained residual is not a direct measurement of discrimination. It can include discrimination, omitted factors, measurement error and model limitations. Nor does a smaller residual prove that the whole labor market is fair: some included factors may themselves reflect constrained opportunities.

A hiring slogan cannot settle the issue.

The claim that firms would simply hire only women if discrimination existed assumes away frictions that need investigation. Employers may have biased beliefs, workers may lack bargaining power, information may be incomplete, and hiring or promotion networks may reproduce unequal access. Competition can create pressure against some costly biases without proving that every form of unequal treatment disappears.

At the same time, a group-level statistic cannot establish a particular employer's legal liability. In the United States, the EEOC describes distinct legal protections against compensation discrimination.^[3] A concrete case requires evidence about the relevant work, decisions and applicable law. This article explains measurement, not how an individual claim would be resolved.

Workplace analysis combines pay data with starting salaries, raises, promotion, allocation of high-value work and flexibility. Comparisons require comparable responsibilities and examination of access to those responsibilities.

The aim is neither to make one number explain everything nor to explain the number away. Clear definitions make disagreement more productive: a weekly earnings gap, an occupational pipeline problem and unequal pay for comparable work can all matter without being the same phenomenon.