A Pay Gap Is a Signal, Not a Diagnosis

This chapter examines what a gender pay-gap statistic can—and cannot—tell us about pay transparency. The practical goal is to read the number as evidence about a defined pay distribution, then ask which comparisons and additional data are needed before treating it as evidence of unequal pay or discrimination.

That discipline matters because the statistic is both powerful and incomplete. Across the OECD, the median full-time working woman earned about 90 cents for every dollar or euro earned by the median full-time working man in 2024. The average unadjusted gap had fallen by roughly nine percentage points since 1995, but the remaining difference was still large and progress was slow. The number makes an inequality visible. It does not tell us which decisions produced it, who had the opportunity to challenge those decisions, or what remedy would work.

What the headline number measures

A gender pay gap compares the pay distribution of women and men within a specified population. The population might be all employees in a company, full-time workers in a country, or workers in a defined occupation. The statistic might use the mean, which is sensitive to very high or low values, or the median, which identifies the midpoint of each distribution. It might describe hourly pay, weekly pay, annual earnings, basic salary or total compensation. Those choices are not cosmetic. They determine which features of inequality the number makes prominent.

An unadjusted gap is especially easy to misread. It describes the difference between the observed pay distributions without removing the effects of the jobs people hold, their hours, their tenure, their level of responsibility or the way work is organised. It can therefore reflect several patterns at once:

  • women may be concentrated in lower-paid occupations or levels of hierarchy;
  • women may work fewer paid hours or experience interruptions associated with unpaid care;
  • men and women may have different access to promotion, bonuses or high-paying assignments;
  • employers may pay differently for comparable work; or
  • discrimination, stereotypes and unequal bargaining power may shape recruitment, evaluation and progression.

These explanations are not mutually exclusive. A company can have a large overall gap because women are underrepresented in senior roles and also have a smaller but important gap among people doing comparable work. Another company can have a modest overall gap while still undervaluing a female-dominated occupation or excluding women from its most influential promotion route. The headline number tells us that the distribution differs. It does not select the explanation for us.

The causes also extend beyond the firm. The OECD links persistent gaps to the unequal distribution of unpaid domestic and care work, especially around the birth of a child. Reduced hours, part-time employment and career breaks can affect earnings and later progression. Horizontal segregation places women and men in different occupations and sectors; vertical segregation places them at different levels within organisations. Gender norms can shape both choices and judgments about appropriate work. Historical undervaluation can persist even when a job requires substantial skill, effort or responsibility.

This is why a pay gap has consequences beyond a single period of employment. Lower earnings reduce current economic independence and can compound into smaller savings and pension entitlements. But the same life-course importance makes interpretation more—not less—important. A statistic that combines occupation, hours, career stage and pay-setting may show where inequality is concentrated while leaving the intervention unclear.

The distinction between an aggregate gap and unequal pay for the same job is therefore fundamental. Equal pay for equal work asks whether people doing identical or nearly identical work receive different pay. That is an important legal and organisational question, but it is not the whole problem in a labour market where women and men often do different jobs. Looking only for identical job titles can leave a larger form of undervaluation untouched.

“Work of equal value” broadens the comparison. It asks whether different jobs make comparable demands on workers, even when their tasks are not identical. A truck driver and a nursing aide do not perform the same tasks, but both may face worker shortages, significant physical demands, risks and responsibility. If an evaluation system counts physical strain clearly while treating interpersonal, emotional or caring work as a vague personal quality, it may assign different value to comparable demands. The resulting pay difference can be reproduced by a classification system that looks neutral on paper.

This does not mean that every difference between occupations is unjustified or that a single national formula can price all work. It means that a credible diagnosis must make the comparison standard visible. Readers should ask whether the system compares jobs, individuals, or only broad occupational labels; whether it includes the full demands of the work; and whether the resulting categories actually connect to pay decisions.

flowchart LR A["Headline gender pay gap"] --> B{"What is being compared?"} B --> C["Observed mix of jobs, hours and careers"] B --> D["Workers in comparable jobs"] B --> E["Different jobs of equal value"] C --> F["Care, segregation and progression evidence"] D --> G["Within-job pay and reward evidence"] E --> H["Gender-neutral job evaluation"] F --> I["More complete diagnosis"] G --> I H --> I I --> J["Choice of correction and accountability"]

The diagram’s point is not that a headline gap can be decomposed mechanically into three neat boxes. It is that the same number can lead to different questions. If the concern is occupational sorting, the relevant evidence may involve representation by level, hiring and promotion. If the concern is pay for comparable work, the analysis needs job categories, pay components and a defensible comparison group. If the concern is work of equal value, the organisation must examine how it evaluates and classifies different jobs.

What additional information changes the diagnosis

Reporting can become more useful when it describes the path by which the gap appears. In the OECD stocktake, 15 of the 21 countries with private-sector pay-gap reporting required some disaggregation by job category. At least 24 countries also required gender-disaggregated non-pay information. These measures can include the gender composition of the workforce, seniority, contract type, hiring, promotion, termination, training, hours and parental-leave uptake. Each field answers a different diagnostic question.

Suppose women and men receive similar hourly pay within a job category, but women are rarely promoted into the next level. The overall gap may be generated by progression rather than by the starting wage in the current role. Suppose women are more likely to work part time or on temporary contracts. The pay gap may be tied to the organisation of hours and access to secure work. Suppose men and women are present in similar proportions at a level, but men receive more bonuses. The relevant investigation shifts toward eligibility rules, performance assessments and discretionary rewards. A number that is accompanied by these patterns can guide action more effectively than an average alone.

The UK reporting requirement illustrates both the value and the limit of this approach. Firms above the threshold had to publish mean and median hourly-pay gaps, bonus gaps, the share of men and women receiving bonuses, and the proportion of women in each pay quartile. Those indicators reveal distribution and access to variable pay. They can show whether women are clustered near the bottom of the firm’s pay structure or underrepresented at the top. They do not, on their own, identify which jobs are of equal value or why a particular bonus or promotion decision was made.

The choice of categories is consequently a form of institutional judgment. A category that is too broad may place very different jobs together and conceal their internal distribution. A category that follows an employer’s inherited titles may reproduce an old hierarchy rather than compare the demands of work. A category that is too narrow may contain so few workers that a reliable comparison is impossible or privacy would be compromised. The OECD notes that some countries exclude categories with too few workers to calculate robust statistics. That is a legitimate constraint, but it also means that “not reported” should not be read as “no gap.”

Privacy creates another boundary. Detailed individual salary disclosure may help workers compare themselves with peers, but it can also expose identifiable information in small groups. A system may therefore publish an aggregate result while reserving more detailed data for workers’ representatives, regulators or an audit. This is not merely a technical choice between more and less transparency. It determines who can see the problem, what they can infer, and whether they have a safe route to challenge it.

The UK paper makes a related point from the perspective of research. Its payroll source records pay, hours and tenure and can identify the size of the employer, but it observes only a small sample of employees at each firm. That makes it possible to study changes in individual workers’ outcomes while preventing the researchers from constructing a complete firm-level pay gap from that data alone. The public indicators and the administrative payroll data answer different questions. One shows what firms disclosed; the other helps estimate how pay changed for workers around the reporting threshold.

A careful diagnosis therefore uses several layers of evidence:

  1. The distributional signal. Mean or median pay gaps show where women and men sit in the observed pay distribution.
  2. The composition. Representation by occupation, level, contract, hours and tenure shows how workers are distributed across opportunities and constraints.
  3. The reward path. Bonuses, raises and promotions reveal how pay changes after entry.
  4. The comparison standard. Job evaluation and classification determine whether different jobs can be compared as equal in value.
  5. The institutional response. Worker access, representation, management review, public scrutiny and enforcement show whether the information can produce a correction.

These layers should not be confused with a universal checklist. A small employer may not be able to publish the same detail as a large public company. A country with strong collective bargaining may route information through representatives rather than individual employees. A regulator may prioritise employer-level comparability, while an audit needs context-specific detail. The relevant question is whether the chosen design is sufficient for the decision it is meant to support.

More information can also create a false sense of precision. A company may publish a sophisticated index whose categories do not correspond to equal-value work. A high score may reflect compressed wages at the bottom while women remain absent from senior roles. A calculable number may still be a poor measure if the underlying job descriptions are incomplete or the classification system assigns comparable jobs to different grades. Conversely, an incalculable result is not proof that the organisation has no inequality; it may indicate that the population is too small or the system is not designed to produce a robust comparison.

This is the central measurement lesson for pay transparency: detail is useful only when it preserves the question being asked. If the aim is to reveal whether women are paid less within comparable jobs, the categories must support that comparison. If the aim is to expose unequal access to advancement, the report must include progression and representation. If the aim is to address the undervaluation of women-dominated work, a headline firm average is not enough; the evaluation of work itself must be open to scrutiny.

The lesson also changes how a reported gap should be used by different actors. A worker may reasonably treat a large gap as a reason to ask how pay is set, but cannot infer from the percentage alone that their own pay is unlawful or that every colleague is comparable. A manager can use the statistic to locate a pattern, but needs job, reward and progression evidence before choosing a remedy. A regulator or representative can use it to target review, while still requiring the employer to explain its categories and decisions. Public audiences can create pressure, but public pressure is most constructive when the underlying comparison is intelligible.

The next empirical question is what happens when a government turns this signal into a repeated public obligation. The UK case is valuable not because its firm-level indicators diagnose every cause, but because its reporting threshold creates a setting in which the consequences of publication can be estimated. With the difference between signal and diagnosis kept in view, the result can be read for what it identifies—and not for more than it identifies.

References