Three in Four Female Chess Players Reported Bias or Harassment: Auditing the FIDE Survey Number

A sentence arrived at my desk on Wednesday afternoon: three in four female chess players have said they faced gender bias or harassment. Above the sentence was the name of an institution — FIDE, the International Chess Federation. Below it there was nothing. No sample size, no fielding date, no geographic scope, no link to the primary document.

I work with numbers for a living, so I did what I always do: I opened my old tracking sheet. In 2026, when the ISL ran behind closed doors in Goa — 115 matches, zero fans — I logged 38 matches and roughly 2,900 passes on a twelve-column sheet. It carried PPDA, field tilt, line-break counts, and a column titled "entry list → withdrawal."

This time the column I needed was empty. I have never had entry-to-withdrawal data broken down by gender. Empty stands had made the broadcast audio sound like a spreadsheet with a pulse; now a number had arrived whose pulse I had no second column to check.

This report does not challenge the 75 percent. It audits it: what the figure measures, what it cannot measure, and which columns FIDE should have added. The diagram is not the truth; it is the question I ask the match. There is no match here — only a percentage.

What the material actually contained

The package that reached me had exactly one substantive chess claim: a FIDE survey found three in four female players reporting bias or harassment. The rest was a website's cookie and privacy notice, which has nothing to do with chess. That is what automated scraping of a news page usually produces — boilerplate, subscribe buttons, and a shadow of the paragraph rather than the paragraph. I record it because it bears directly on reliability: the headline statistic is likely not wrong, but the scaffolding around it is plainly incomplete.

FIDE is the sport's global regulator. It issues ratings, awards titles, runs the Olympiad cycle, and — the point that matters here — controls recognition, arbiters, and disciplinary process for over-the-board events. Its commissions on women's chess and on ethics handle such matters. That makes this survey something more than a shopkeeper's opinion poll: whatever it asks becomes the evidence base for policy.

The structural background has to sit alongside the number or the 75 percent will be misread. The gender rating gap at the top remains wide — the leading women sit roughly in the 2550–2600 band while the world number one runs above 2830, a gap of about 250–280 Elo (approximate and date-sensitive). The title record is starker: fewer than 50 women have ever held the full GM title, under two percent of the GM pool. Judit Polgar remains the only woman to have entered the open world top ten; Hou Yifan won the women's world title four times and peaked around 2680.

So the figure lands in a sport where women's presence at the top is already thin. Three explanations are usually offered: the participation pipeline, access to coaching and resources, and the quality of the competitive experience. The survey targets the third. That is why it should be read as a design, not a headline.

What 75 percent measures

The first problem is the question's construction. The headline fuses "bias or harassment." Those are different behavioural categories with different remedies. Bias in its broad form includes conduct that is not unlawful: doubting a woman's competence, casual comments, grading her performance against a lower bar. Harassment is conduct that may breach a disciplinary code and, in many jurisdictions, criminal law.

An aggregated percentage is rhetorically strong and analytically weak. It cannot tell an organizer whether the problem is culture or safety. The interventions differ: arbiter training, coach guidance and language codes for culture; reporting channels, suspension process and complainant protection for safety. One number will be used for both, and the likeliest outcome is that the harder problem — probably safety — gets funded second.

The second problem is precision. The public summary gives no sample size (n). If n is about 400, the 95 percent confidence interval around 75 percent is roughly ±4 percentage points; if n is about 100, it opens to ±8.5 points. The figure is directionally robust and precision-unknown — and policymakers work with intervals, because intervals decide which population a programme targets and what its success threshold is.

The third problem is the most serious: selection bias. Surveys distributed through federation and platform channels over-sample the most engaged, tournament-active, online-present players. That group is precisely the one most exposed. Prevalence estimates from self-selected online samples typically run above population estimates. This does not make the number false; it makes it the number of those who could speak. The fourteen-year-old who left after a district event is probably not in the sample — and she is the data point we most need.

The fourth gap is the missing comparison group. The standard way to isolate a gender-specific effect is to ask male respondents the same question. If men report 20 percent and women 75, the 55-point residue is the actual gender-specific burden. Reading 75 percent as a pure gender gap, with no baseline, is the error that will eventually cost the survey its credibility, because opponents will spend their energy filling that hole.

The fifth gap is stratification. The summary does not say whether the 75 percent is elite or amateur, over-the-board or online, which continent, which age band. For policy, missing strata matter more than sample size, because strata tell you where to intervene. If most harassment happens in platform chat and social media, arbiter training achieves nothing. If it concentrates between 13 and 17, the national championship's safeguarding protocol is the first priority, not the senior circuit's.

Jurisdiction: the board outside the board

One point gets too little attention. Much harassment in chess now happens online — platform chat, direct messages, stream comments, social media. FIDE's disciplinary reach runs mainly over events it recognizes and the officials attached to them. It has limited direct control over private platform chat. If a large share of the survey reflects online experience — and I think it does — FIDE is publishing data on conduct that largely occurs outside its own jurisdiction.

That is not an accusation; it is a structural fact. The sport's infrastructure moved from board to screen faster than its regulatory frame moved. The most useful reporting channel — platform moderation and ban data — does not sit with FIDE. Until a shared reporting arrangement exists, this survey will measure pain and leave the treatment at someone else's door.

The mechanism: bias does not show up in ACPL, it shows up in entry lists

Here is the core of it. I trust the pattern, then I go looking for the player who breaks it. The pattern here is clear: the effect of bias or harassment will not appear in move accuracy. It will appear in entry lists.

A chess career is built out of matches, matches require tournaments, and choosing a tournament is a decision full of non-chess variables: who the organizer is, where you sleep, who travels with you, what last time felt like. A player who has once felt unsafe at an event will skip it — and that skip is recorded nowhere. It is an invisible rating loss.

Which is why the next real research on this subject is a participation ledger, not a game analysis. Four metrics matter: withdrawal rates by gender (entered but did not play); participation continuity across a cycle; drop-out rates by age band; and error rates under time pressure compared by gender. None of these appear in the public summary. Until they do, 75 percent remains a sentiment metric, not a performance metric.

That distinction carries political weight. A sentiment metric lets an institution say the problem is large and run an awareness campaign. A performance metric forces it to say how many players it lost and at what level. The first is comfortable; the second demands accounting.

There is also a directional error hidden in reporting rates. Chess is a small, dense network. Coaches, arbiters, organizers and training camps all circulate inside the same circle, so a complaint has a price: the next camp recommendation, the next invitation, the next team selection. The most severe conduct therefore has the lowest chance of being recorded, while the mildest discomfort is easiest to report. Together they place the 75 percent in a strange position: a floor for severe conduct, a ceiling for mild discomfort.

India's funnel

I can see this part with my own eyes. When I built ChessBase India from 2026, the clearest signal was never a result — it was the count of entries in age-group sections. Under-9 and under-11 girls' sections look full. Parents, schools and state award schemes — Tamil Nadu's school pipeline, institutions like Velammal, cash prizes — have together created a genuinely measurable production line at that level. India now has more than 85 grandmasters, the deepest teenage cohort anywhere, and girls are a rising share of it. Koneru Humpy, Harika Dronavalli, Tania Sachdev and R. Vaishali's generation proves the top is reachable.

But past under-17 the same sections thin, and the number of women over 25 in open sections can be counted on fingers. That is the funnel. The question is not why Indian girls do not start chess — they do, in large numbers. It is what happens between 14 and 20. Three answers are usually offered: academic pressure, safe travel and accommodation, and environment. The first two are clearly measurable. The third is what FIDE is trying to measure, and the least measured of all.

One audience-side data point, because analysis that ignores what people watch stops describing its own market: ChessBase India's YouTube channel reached roughly 2.9 million subscribers, and women's chess content there draws numbers comparable to open-tournament coverage. Demand exists. What is missing is continuity — the girl visible in coverage at 15 is absent from the open board at 22. The content funnel and the player funnel narrow at the same point.

Contrarian: this is not a scandal, it is a budget document

My instinct is to treat a methodology-free number as a defect. Here I think the omission is deliberate and should be read as such.

Publishing a prevalence figure without strata is an administrative act. Institutions release such numbers when they need a mandate for the next budget cycle. The moment they break the figure into strata, they must also show improvement by stratum — that is, evaluate their own work. An aggregated 75 percent stays useful forever; a 75 percent that becomes 58 in four years is dangerous to its author.

We also cannot say the situation is worsening. A trend needs two points. Reported incidents may have fallen while reporting rose — good news that would make the 75 percent look larger. In social phenomena, reporting rate and incidence rate do not move together, and any analyst who plots them on one line is reading his own hope, not the data.

And the uncomfortable part: the number is most useful to those least able to act on it. A national federation can build policy on it. The fifteen-year-old at district level whose experience is inside that 75 percent cannot assemble the process, the language or the confidence to file a FIDE complaint. The number travels upward; the complaint stays stuck below.

So I set my threshold in advance. I will revise my view if any one of these becomes public: a stratified over-the-board, under-18, women-only figure below 30 percent on a sample above 1000; a three-year participation ledger showing a gender gap of under two percentage points between entry and actual play; or platform-level harassment data from a major online site confirming or contradicting the survey's direction. Absent all three, 75 percent stays where it is — an honest sentiment metric whose accounting route has not yet been built.

Takeaway

The best tactical note is the one that survives contact with the next match. The next match is the next national championship, and what to watch there is not chess. Three registers: the gap between entries and actual participation, the reason column on mid-event withdrawals, and the number of complaints logged with the chief arbiter together with how many days each took to resolve. If none of the three shows a distinct pattern for women, then 75 percent is measuring something else. If it does, FIDE's next report should print those three columns instead of a percentage.

And one question stays open at my desk: when the largest part of the sport has moved onto a screen, can the board's regulator keep the books on the screen's conduct — or will chess need a new shared register, where platforms, federations and organizers sit on the same complaint count?