Adjusted and Unadjusted Pay Gaps Are Different Claims

Why the adjusted figure is always smaller

Start with the mechanism rather than the numbers, because the arithmetic is the easy part. The unadjusted gap takes everybody and compares them. The adjusted gap first matches people on a list of characteristics — occupation, industry, experience, hours, education, location — and then compares within those matched groups, reporting whatever difference survives.

Since occupation and hours explain a substantial share of the raw difference, holding them constant necessarily removes most of it. That is not a trick or a correction of an error; it is what the method is designed to do. An adjusted figure in the low single digits and an unadjusted figure near seventeen points are entirely consistent with each other and usually describe the same underlying data.

Which means a disagreement between two published figures is very often not a disagreement at all. It is two studies answering different questions, both reported as though they had answered the same one. Whichever outlet found the more useful number for its argument tends to be the one quoting it, and neither study has done anything wrong.

What each is actually good for

The unadjusted gap is the right instrument for a question about a labor market or a society. Does the way careers develop, occupations sort and hours get allocated produce different earnings for men and women in aggregate? That is a real and important question, and the unadjusted figure answers it directly.

The adjusted gap is the right instrument for a question about an employer or a hiring practice. Among people doing comparable work with comparable experience in the same place, is there a difference that the observable factors do not explain? That is closer to what equal pay law asks, and it is the figure an employer conducting an internal pay review will produce.

Neither of them is the more honest number, and treating one as the truth and the other as spin is the standard mistake. They are different tools built for different jobs. Using the wrong one produces a confidently wrong answer rather than an approximately right one, which is the worse of the two failure modes.

The trap in adjusting

Here is the part that gets least attention and matters most. Every variable you control for is a variable you have decided is a legitimate explanation rather than part of what needs explaining. Controlling for occupation assumes that occupational sorting is neutral — but if women are steered away from higher-paying fields, or pushed out of them, then controlling for occupation removes the discrimination from the measurement along with everything else.

The same applies to seniority. If promotion decisions are where the unequal treatment happens, then matching people on seniority level compares women to men who were promoted at a different rate and calls the remaining difference small. The control has absorbed the mechanism.

This is not an argument against adjusting. It is an argument for reading which variables were included, because that list is a set of assumptions about what counts as a legitimate reason for a pay difference. A study controlling for job title inside one company is making a very different assumption from one controlling only for broad occupation and years of experience.

Both can be true at once, and usually are

The most useful way to hold these together is that they describe two stages of the same process. The adjusted gap describes what happens at the point of pay-setting for comparable people. The unadjusted gap describes what happens across a whole career of hiring, sorting, promotion, hours and interruption.

An economy could have almost no adjusted gap and a large unadjusted one. That would mean employers pay comparable people comparably while the distribution of who ends up in which job is very uneven indeed. It is roughly what the data suggests, and it locates the interesting question in careers and sorting rather than in payroll decisions — which is a different problem with different remedies.

It could equally have a small unadjusted gap and a meaningful adjusted one. That would point at pay-setting itself rather than at sorting, and would call for pay audits rather than for changes to hiring and promotion. Knowing which of the two situations you are looking at is the entire practical value of understanding the distinction at all.

A worked reading

Suppose a company reports an unadjusted gap of 22 percent alongside an adjusted gap of 1.5 percent, and presents the second as the meaningful figure. Both numbers are probably accurate, and the framing is doing an enormous amount of work. Nothing has been falsified; a question has been quietly swapped for a narrower one.

What the pair actually tells you is that people in comparable roles are paid comparably, and that men and women are distributed very unevenly across the roles. The 22 percent is not explained away by the 1.5 percent — it is explained by the sorting, which is a fact about the company’s hiring and promotion rather than about its payroll. A reader who accepts the adjusted figure as the answer has been told something true and has stopped one question short of the interesting one.

What an employer’s own pay audit is doing

Most large employers now run an internal pay analysis at least annually, and it is worth understanding what that exercise actually is, because you may be shown its results. A pay equity audit is an adjusted analysis by construction: it groups employees into comparable roles, controls for the factors the employer considers legitimate, and reports the unexplained residual.

The interesting choice in any such audit is the grouping. Comparing people within narrow job codes produces small residuals and finds little, because the grouping has already absorbed the level differences. Comparing across broader families finds more and is harder to act on. Neither approach is wrong, and an audit that reports its results without describing its grouping has not told you the main thing.

If an employer publishes an adjusted figure, the reasonable follow-up is what it was adjusted for and how roles were grouped. Employers doing this seriously answer readily, because the methodology is the part they are proud of. An unwillingness to describe the grouping usually means the grouping is doing the work.

What to ask when you meet a figure

Three questions settle almost any pay gap statistic. Is it adjusted or unadjusted? If adjusted, which variables were controlled for? And what population does it cover — one employer, one industry, or a whole economy?

A source that answers all three is worth reading whatever the number turns out to be, because you can tell what it means. A source that answers none is quoting a figure it has not interrogated. The number could be almost anything, and so could the conclusion drawn from it.

This is general information about how these measures work rather than legal advice. If your question is about your own pay against a specific colleague, that is a legal question with its own tests rather than a statistical one, and it is covered separately in this section.