The confidence gap survives being handed the score

Partly, and not in the way the phrase suggests. There is a measured gap in how women rate their own performance. It does not close when the confidence is corrected: in a study of more than 14,000 people it persisted after every participant was shown exactly how well they had done. What closes it is information about the bargaining range.

That single result reorganises the whole subject. If a gap survives being handed the score, then whatever produces it is not a belief about the score. It is something else, and the something else turns out to be tractable in a way that "confidence" never was.

The result that should have ended the confidence story

Exley and Kessler ran a large set of experiments published in the Quarterly Journal of Economics in 2022: over 4,000 online participants plus 10,637 school-aged youth, more than 14,637 people in total. Participants completed a task, then described their own performance to a prospective employer.

Gender gap in self-evaluation. Exley & Kessler, The Gender Gap in Self-Promotion, Quarterly Journal of Economics, 2022. Scale of 0–100.

Measure	Gap, in points
Self-evaluation of performance	12.68 lower
Willingness to apply	15.31 lower
Whether they would succeed	15.09 lower

Then the authors did the thing that makes this study different from the hundreds that preceded it. They gave participants perfect information about their own performance — absolute and relative, exactly how well they had done and exactly where they had come. The gap narrowed. It did not go away: still 7 to 11 points lower. The authors' own conclusion is that it therefore cannot be attributed to gender differences in confidence, understood as beliefs about performance.

Three further results are worth as much as the headline. The gap persisted when no employer was present and no strategic incentive existed, so it is not simply self-presentation under observation. No gap arose at all on a task associated with women rather than men — the effect is typed to the task, not to the person doing it. And it is visible as early as sixth grade.

7 to 11 points of self-evaluation gap remained on a 0–100 scale after participants were told precisely how well they had performed.

Self-esteem, measured properly, is a small effect

The background claim underneath most confidence writing is that women hold themselves in lower regard generally. That has been measured, at scale, for a long time. Kling, Hyde and colleagues published a meta-analysis in Psychological Bulletin in 1999 covering hundreds of studies since 1987 and roughly 150,000 respondents. The overall effect size was d = 0.21.

On the conventional scale, 0.20 is slight, 0.50 is moderate and 0.80 is large. So the finding is real, it is in the expected direction, and it is small — small enough that it cannot carry the weight of a 21% earnings difference or a 34% revenue difference on its own.

I should say how I know this. I have not read the 1999 paper; I have read the University of Wisconsin–Madison release describing it, and I would rather tell you that than let a citation do work it has not earned.

Women now negotiate more than men. The outcomes did not follow.

The most useful correction in this literature is also the least quoted. Kray, Kennedy and Lee, writing in Academy of Management Discoveries in August 2023, looked at MBA graduates from 2015 to 2019. Among them, 54% of women negotiated their job offers, against 44% of men.

That is not a historical constant. Men did negotiate more, once. The difference disappeared around 1994 and reversed from about 2007. The behaviour everyone is still being urged to adopt was adopted, roughly two decades ago.

The outcomes did not follow. More women attempted negotiation, and more women reported being rejected. Women MBA graduates earned 22% less overall: 88% of men's salaries immediately after graduation, falling to 63% ten years later. Run that arithmetic slowly, because it is the part that matters. Twenty-five percentage points of relative position lost over ten years, about 2.5 points a year, and the loss accelerates with distance from the moment of negotiation. Whatever is producing this is not confined to the offer conversation, which means an intervention aimed only at the offer conversation cannot fix it. The evidence base includes a 2019 alumni survey of 1,900 MBA graduates, experiments with 500 college-educated managers, and a meta-analysis of nine prior studies from 1982 to 2015.

Sometimes it does hurt to ask

Bowles, Babcock and Lai ran four experiments, published in Organizational Behavior and Human Decision Processes in 2007, on what happens to the person who initiates a negotiation. The negative effect of initiating was more than twice as great for women as for men in hiring evaluations, and 5.5 times greater for a woman candidate on measures of willingness to work with her.

The detail that makes the result difficult is who was doing the penalising. Male evaluators penalised only women for negotiating. Women evaluators penalised both sexes equally. The authors draw the obvious implication themselves: reluctance to negotiate may be an accurate reading of a real social cost rather than unwarranted timidity.

Per-experiment sample sizes are not stated on the source I have been able to access, so treat the multipliers as directional rather than precise.

Set that against the advice industry. A person is told she lacks the confidence to ask. The experimental evidence says that when she asks, in some rooms, it costs her — and that it costs her specifically because of who is evaluating. Calling that a confidence deficit is not just wrong. It relocates a cost imposed by one party onto the psychology of the other.

The statistic everybody quotes and nobody can find

You have read that women apply for jobs only when they meet 100% of the qualifications, while men apply at 60%. It appears in keynotes, HR training and roughly every article on this subject.

There is no study. The claim traces to Sheryl Sandberg's Lean In in 2013, and journalists who went looking established that it rested on a speculative comment by a senior Hewlett-Packard executive with no quantitative data behind it. No locatable report, no sample, no methodology, no year.

Separately, the Behavioural Insights Team analysed recruitment data across four large UK organisations and found that women were more likely than men to be hired when applying for the same positions in three of the four. That is not a clean refutation and I will not present it as one: BIT published no sample sizes, and noted that the organisations held no data on candidate qualifications.

I am not raising this to score a point against a book. I am raising it because of where the claim gets used. It is deployed precisely at the moments when a structural explanation was available and inconvenient, and it displaces one. That is what makes an unsourced number expensive rather than merely untidy.

What actually moves the outcome

Mazei and colleagues published a meta-analysis in Psychological Bulletin in 2015 covering 51 studies and 10,888 participants — 4,656 women and 6,232 men, across several countries. Its value is not the average effect. It is the moderators: the conditions under which the gap in negotiation outcomes shrinks, disappears or reverses.

There are three. When women have negotiation experience. When they have information about the bargaining range. And when they are negotiating on behalf of another person — in which case women outperformed men.

Gender roles no longer give men a bargaining advantage if women are trained in negotiation, have information about the bargaining range and are negotiating for other individuals.

One methodological note, because the temptation to quote a headline effect size here is strong. I am not stating an overall d for this meta-analysis. The sample and the moderators are verified through the American Psychological Association's own release; the headline effect size is not, and I will not repeat a number I have not seen at source.

The field has already moved on from fixing the person

A recent Annual Review of Economics synthesis of negotiation interventions describes the shift plainly: initial efforts to encourage women to negotiate more like men have given way to altering the conditions of negotiation itself. Structural interventions — salary-history bans, wage transparency, removing the negotiation requirement altogether — have proved more successful in securing pay equity than individual behavioural ones, though they carry unintended consequences that warrant caution at implementation.

I should attribute that carefully. I have the synthesis and its argument; I have not been able to confirm the authors or the year, so I am citing it as an Annual Review of Economics synthesis and nothing more precise than that.

What this means in practice

The commercially useful part of all this is a single phrase from the meta-analysis: information about the bargaining range.

That is not a state of mind. It is three specific pieces of analysis, and each of them can be done before anyone walks into a room. What is this work worth to this buyer — measured in the revenue it protects, the cost it removes or the decision it makes possible for them, not in the hours it takes you. What has comparable work actually cleared at, in this market, at this scale, with these buyers — realised prices, not published rate cards. And where does your own walk-away sit, expressed as a number and a reason, so that the floor exists before the pressure does.

Notice what that does to the third moderator, the one where women outperformed men: negotiating on behalf of someone else. The obvious reading is to hire an agent. The more useful reading is to ask what an agent has that a principal usually does not. An agent has a mandate — a defined range, a defined floor, and a rationale for both that was settled in advance and is not about them personally. That mandate is a document. It can be written before the meeting by the person who will be sitting in it.

The pattern across the whole body of evidence is consistent. Where the gap has been closed experimentally, it was closed by changing what the negotiator knew or what the situation required — never by changing how the negotiator felt. The same logic runs through the gap between salaried and self-employed earnings, where the number is set by whoever holds the pen, and through the revenue gap between firms with identical survival rates.

What this cannot tell you from the outside

The limits here are real. Exley and Kessler is experimental, with online participants and school students describing performance on a set task, not owners of companies pricing complex work. Kray and colleagues studied MBA graduates taking salaried employment, which is a different transaction from selling a service to a client. Bowles and colleagues ran laboratory hiring evaluations and the per-experiment sample sizes are not available to me. Mazei is a meta-analysis of experiments, verified through a press release rather than the paper. The Annual Review synthesis is cited without confirmed authorship. Kling is 1999 and I have it second-hand.

More to the point: not one of these studies knows what your last ten proposals were priced at, which of them closed, what was discounted and by how much, or what the buyer's alternative would have cost them. Those are the facts that determine whether your next negotiation goes well, and they are all sitting in your own files.

Which produces the odd situation this piece has been circling. A woman running a firm can find, without much difficulty, a coach who will work on how she feels about asking. She will have considerably more trouble finding anyone who will tell her what the work is worth to the buyer and what comparable work has cleared at — because her accountant reconciles what already happened, her lawyer makes the contract sound, and her bank reads the outcome. Nobody in that group has been engaged to value the thing before it is sold.

The evidence says the second job is the one that closes the gap. It is also the only one nobody is doing.