FIDE survey: Three in four female chess players report bias or harassment — sample size still undisclosed

In a FIDE survey, three in four female chess players said they had faced gender bias or harassment. That sentence is clearer than the evidence behind it. The survey question folds two different things into one line: bias, which usually arrives as a look, a remark, or an assumption of incompetence; and harassment, which in many jurisdictions is a criminal category. Merging them produces a sharper headline and a blunter instrument. An organiser reading the number still cannot tell whether the problem in their hall is cultural or a safety problem. The remedies diverge. Culture is fixed through arbiter accountability, tournament regulations and training. Safety is fixed through a time-bound complaints process and independent investigation.

The real story here is the sample size and the fielding window, not the 75%. The secondary report that carried the figure listed a website cookie and privacy notice among its "key information points." That is not chess information. It is the fingerprint of automated extraction from a page. The central claim does not become false because of it, but confidence that the report is complete drops sharply. I left the vendor to learn which assumptions the model card could not hide. Here there is no model card at all.

What is known fits in one sentence. The International Chess Federation commissioned a survey and found that roughly three of every four female respondents reported bias or harassment. What is unknown fills a much longer list: total respondents, fielding period, countries covered, recruitment method, number of questions, whether a male-respondent baseline question was asked, and a link to FIDE's primary document. None of these appear in the available material.

Start with the denominator arithmetic. If 400 women answered, the 95% confidence interval around a 75% estimate is roughly plus or minus 4 percentage points. If 100 answered, it widens to about 8.5 points. The figure is directionally robust and precision-blind. For a reader, that means "three in three" and "three in five" could both travel under this headline, and nobody downstream would notice.

Selection bias is the biggest structural risk. Surveys distributed through federation and platform channels reach first the players who are already most active, most tournament-going and most online. That is also the group most exposed to both bias and harassment. Prevalence estimates drawn from self-selected samples typically run high against population estimates. The opposite force is just as real: in a small, tightly networked professional community where coaches, arbiters and organisers overlap, the most severe incidents are the least reported, because a complainant puts a career at stake. Read the 75% as a floor, not a ceiling.

Missing strata matter more than missing sample size. The aggregate cannot separate elite from amateur, over-the-board from online, or one region from another. A policymaker needs those splits because each requires a different intervention. Online chat harassment is a platform moderation problem. In-hall bias is an arbiter accountability problem. A national federation complaint mechanism is a third problem entirely. A single unstratified percentage pours all three into one bucket.

The survey also needed a male baseline. A parallel question to male respondents — have you experienced harassment in chess? — would have told us whether this is a chess problem or a gender-specific problem. That distinction matters in a sport whose competition runs through mixed open sections alongside separate women's events. Without a baseline, the figure does not establish that chess is worse than other cognitive-competition domains; it establishes how many women in this sample reported such experiences.

Jurisdiction is the least discussed and most decisive layer. Much harassment now happens in platform chat, direct messages and social media, which sit outside FIDE's direct jurisdiction. If the survey pooled online and over-the-board experiences, the federation is drafting policy over conduct it only partly controls. That is not a weakness to hide; it is the picture.

If the survey was meant to connect to performance, the indicators that show up at the board were left unmeasured: withdrawal rates by gender, participation continuity across a cycle, drop-out by age band, and error rates under time pressure. Without them, 75% stays a prevalence metric, not a performance metric. The mechanism linking harassment to results exists in other cognitive sports, but this material contains no chess-specific measurement. The strongest channel is the least visible one: event selection. Players decline invitations, avoid certain venues and step out of mixed opens. That never shows in a rating. It shows in an entry list.

Structural background points the same way. Fewer than roughly forty women have ever held the full grandmaster title, under two percent of all GM titleholders. The top women's ratings have clustered around 2550 to 2600 while the world number one has sat above 2830, a gap of 230 to 280 Elo points. Treat those as approximate and pending verification, but the direction is not in dispute. Elo measures results, not development. Calling that gap a talent gap would be an error; it is a pipeline, access and environment gap.

The institutional read matters too. No player is named, no incident is dated, no national federation is quoted. This is not the architecture of a scandal response. It is the architecture of a policy-framed report, the kind an institution publishes when it wants data to justify an incoming rules package. My guess is that the primary document contained a follow-up section on planned measures, cut in condensed secondary coverage. The actionable material usually sits in exactly that cut.

For India, the number is useful only if measured the way player development is measured. Tamil Nadu's pipeline, schools such as Velammal, the Tal Chess Club lineage running back to Manuel Aaron, and the country's most generous state cash awards are all supply-side facts. The demand-side question is how many girls enter open sections and how many stay. I interviewed Viswanathan Anand in December 2026, minutes after his second GM norm, and the same pattern has repeated on open boards since: girls' numbers look healthy in age-group events and thin out after round seven. That thinning is not one incident. It is the sum of ordinary days. Federations that publish press releases about grassroots should be asked to account for it.

Here is where I have to be careful. This figure does not prove chess has a harassment crisis. It proves chess has a measurement crisis: an institution has published a prevalence statistic with every sampling detail stripped out. A prior is only a story until the sample makes it pay rent. This sample has not paid rent, so treating 75% as either truth or falsehood is an error. The part that survives should be stated plainly. One woman in four in that survey did not report such an experience. That quarter matters twice over. It shows the experience is not universal, which means intervention can work. It also shows the question was not framed so tightly that every respondent was pushed toward yes. Both facts make the number usable rather than weaker.

I trust the backtest that survives the transfer window, not the one that survives the tweet. The real test of this survey comes in the next 12 to 18 months, when FIDE publishes its rules package. Until then, three numbers deserve tracking. First, women's share of open-section entries, state by state. Second, withdrawal rates by gender, especially after round five. Third, complaints filed through official channels and how many were resolved. The market moves first; the model monk waits for the closing line to speak. The question is whether chess moves on a 75% headline or waits for the three numbers that would actually measure change.