The Gender Gap in Salary Negotiation: What the Research Actually Measures

Two accounts coexist here, and each closes the conversation.

The first fits in one line: women ask less, so they should ask more. It places the entire burden on the person concerned, and ignores what happens when she does ask.

The second says the opposite: individual negotiation can do nothing, it is all decided elsewhere. That is partly true, and it leaves without an answer the person who has a meeting next Thursday.

What the research describes is more precise than either, and more useful. It separates two mechanisms that combine, and that separation has direct practical consequences.

In the article that follows, I explore how anticipating a social judgement reshapes a request before it is ever made. If you recognise yourself in this theme, the tests on the platform help you take stock.

A caution: what this text will not do

Almost everywhere, this subject is handled with striking figures whose origin is rarely traceable — percentages copied from one article to the next, studies cited without dates, results from one context transposed to another.

So I will quote no percentage. This is not rhetorical caution: orders of magnitude vary sharply by country, sector, period and method, and a figure detached from its protocol tells you nothing usable.

What follows describes mechanisms documented convergently by several research teams. Mechanisms are more robust than numbers, and they are what you use to prepare a meeting.

One further point: this work describes statistical tendencies across groups. A group tendency predicts nothing about an individual — not about you, and not about the person across the table.

First mechanism: how often the request is made

This is the better-known one, popularised by the work of Linda Babcock and Sara Laschever. The starting observation is simple: in comparable situations, the probability of opening a pay negotiation is not the same across sexes, to women's disadvantage, and the gap appears from entry into the labour market.

Two clarifications change how that result should be read.

It is not a matter of negotiating skill. The work does not show women negotiating badly. It shows a difference in initiating the discussion. Once the negotiation is open, the gap narrows sharply. The gap varies a great deal with the setting. Where the possibility of negotiating is explicitly signalled — a line in the advert, a question at interview, a formalised procedure — the difference in frequency drops markedly. That result matters, because it indicates the mechanism turns largely on the ambiguity of the situation: when nothing indicates whether asking is expected or out of place, the anticipated social risk weighs more.

Put differently: the hard part is not the request, it is knowing whether it is permitted. And that doubt costs more for those to whom the context has less often signalled that it was.

Second mechanism: the social cost of the same request

This is the decisive point, and it is almost always missing from advice articles.

Hannah Riley Bowles, Linda Babcock and Lei Lai ran a series of experiments in which the content of the request was held constant — same words, same amount, same context — and only the sex of the person asking varied. The converging result: the request does not carry the same relational consequences depending on who makes it. The measured tendency is a lower willingness to work with the person, and a harsher judgement of their manner, when the request comes from a woman.

That result has three consequences worth stating plainly.

It moves the responsibility. What these experiments measure is not a shortfall of confidence in the person asking: it is a difference in reception. "Be more assertive" does not address that mechanism; it works around it. It makes the hesitation rational. If a request carries a higher relational cost, hesitating is not timidity: it is a correct reading of a real risk. That changes how it should be discussed — you do not "fix" a well-founded hesitation, you try to reduce the risk that grounds it. It does not excuse inaction. The measured cost is an average, not an individual fate, and the same work shows it depends heavily on the form the request takes — which is where the room for action lies.

⚠️ Two methodological reservations, present in the literature itself: much of this comes from experimental settings, and transfer to the field is never total; and effect sizes vary across studies, cultures and periods. The mechanism is solid, its magnitude is not.

What this changes in preparation

Here is the operational part, and it follows directly: if the risk lies in how the request is perceived, then what protects you is not assertiveness — it is what the request rests on.

A request resting on a person ("I think I deserve more", "I'm worth more than this") exposes the person entirely. A request resting on a dated file ("here are the duties added, on these dates, and here is what was announced in January") moves the discussion onto an external object.

That displacement is exactly what Fisher and Ury's Harvard method calls separating the people from the problem. The principle is usually presented as a rule of courtesy; here it has a sharper function: it removes the material for relational judgement.

That is the point of ScanMyJob. It does not supply lines to say and does not advise you to be firmer: it measures the room observable in your written exchanges and produces a dated record of facts — what was announced, when, by whom, what followed. The request stops being an opinion about you. The examples gallery shows four real cases with their records, including a pay-review commitment made orally then restated in writing, which was only recoverable from the messages.

Three practical effects of that shift:

The third is underrated. Managers move; a dated file stays, and stays identical.

What does not work

Imitating a style. The same body of work that describes differential treatment does not indicate that a harder register improves outcomes — rather the reverse. Adapting the form helps when it is legitimising (linking the request explicitly to the service's interest, explaining where your reference point comes from), not when it means hardening the tone. Attributing every refusal to the mechanism. A closed budget is still a closed budget, and a refusal may have nothing to do with this subject. Reading it that way systematically makes preparation less effective, not more. Dropping the underlying question. None of the above belongs to individual negotiation when the gap is structural: employers' obligations on equal pay, available remedies and protected characteristics are legal questions, handled with a lawyer, a union or the competent authority in your country — never with an article, this one included.

What this article does not do

It does not tell you what happened in your case. An individual gap can have many causes, and research on group tendencies supports no conclusion about a particular situation. It supplies no figure to quote. Deliberately. If you need orders of magnitude for your sector and country, national statistical sources are the place to start — and building a range is a separate exercise. It does not claim preparation settles the question. A record of facts does not offset a collective mechanism. It does one precise, limited thing: it removes the material for judgement about the person and replaces it with dates. That is little against the scale of the problem, and a great deal within a single meeting — particularly at the hiring stage, where the first figure holds for a long time.

Finally, if apprehension about asking occupies a space that goes beyond work, that is not a shortfall of method and it can be worked on: the assistant is there to talk it through, and so is a professional.