Pay gaps between demographic groups are often discussed as though two people in the same role were paid differently. A substantial part of the measured gap comes instead from the distribution of people across occupations.
Two different gaps get the same name
Researchers usually distinguish the unadjusted gap, which compares all workers in a group, from the adjusted gap, which compares workers with similar jobs, hours and experience.
The unadjusted figure is almost always larger. The difference between the two is mostly a description of who works where, not of pay decisions inside one employer.
Because both numbers are called the pay gap in ordinary conversation, arguments frequently run past each other. One side is describing sorting, the other is describing pay setting.
How sorting into occupations happens
Occupations differ sharply in typical pay, and groups are not evenly spread across them. Some fields draw heavily from one demographic while others draw from another.
The sorting has many contributing routes: which training programs people enter, where jobs are physically located, scheduling demands, informal referral networks, and expectations formed long before anyone applies.
These routes accumulate. A field that becomes strongly associated with one group tends to recruit from that group again through referrals, which makes the composition self-reinforcing over time.
Why field pay levels are themselves contested
Occupations do not have fixed values. Pay reflects bargaining power, credential requirements, public or private funding, and how easily an employer can replace a worker.
Some analysts argue that fields dominated by one group are undervalued relative to the skill they require. Others hold that pay tracks market demand and that composition follows pay rather than causing it.
Evidence is used on both sides, and the question of which direction the causation runs remains genuinely unsettled. Careful writing on the subject usually reports the disagreement rather than resolving it.
What within-job comparisons can and cannot show
Comparing people in the same job title at the same firm isolates one decision: how that employer sets individual pay. It is the narrowest and most controlled comparison available.
That narrowness cuts both ways. Controlling for occupation removes exactly the sorting that many people consider part of the inequality being measured.
An analyst who controls for everything correlated with the outcome can shrink almost any gap toward zero. Choosing which controls belong in the model is an argument about meaning, not about statistics.
Why the distinction changes what a remedy touches
Measures aimed at pay setting, such as structured salary bands or audits of individual pay decisions, act on the within-job gap and leave the distribution of jobs unchanged.
Measures aimed at sorting, such as apprenticeship access, scheduling changes or training pipelines, act slowly and show up in aggregate figures only after years.
Anyone reading a claim that a gap did or did not close should first ask which of the two gaps was measured, because the same underlying situation supports very different headline numbers.