Gender probability by name

probability is a ratio, not a mood

The probability field is an integer from 50 to 100, and it reports the observed dominant-gender share in the underlying registration data. When it says 84 for Jordan, 84% of the 62,240 recorded people named Jordan were male. It is not a classifier expressing how sure it feels.

That distinction decides how you should use it. A ratio inherits all the weakness of the sample it was computed on, which is why the sample size travels with it in a separate field rather than being folded into a single score.

{
  "name": "Jordan",
  "gender": "male",
  "probability": 84,
  "total_names": 62240,
  "confidence": "high",
  "source": "db"
}

The confidence tiers, and the exact thresholds

confidence is derived mechanically from total_names. There is no judgement between the count and the label, so you can reproduce the tier yourself from the same response.

Why a correct answer can still say unverified

Counted registration data comes from seven national statistics offices — the US SSA, the UK ONS, France INSEE, Statistics Canada, Spain INE, Ireland CSO and Statistics Norway. Everywhere else, coverage comes from WGND 2.0, which spans 195 countries and territories and carries no counts at all.

So a Turkish, Italian or Indian name usually returns the right gender with total_names: 0 and confidence: unverified, because those countries do not publish name registration volumes. The label is a statement about auditability, not about correctness. Our measured accuracy in those markets is published on each country page.

Probability from a countless source is capped at 95, so a 100 always means a counted, unanimous sample.

Where to put your threshold

There is no universal cutoff, only a cutoff per use. These are the ones we would defend:

The reporting row is the one people get wrong most often. When you are counting rather than addressing, filling the unresolved rows in with the majority class does not remove the bias — it moves it somewhere you can no longer see it. Names that fail to resolve are not a random sample: they cluster by script and by country, so imputing them systematically inflates whichever group your data already covers best.

Whatever threshold you pick, record it next to the column. A gender field with no note about how it was produced becomes an unquestioned fact within about two years.

Questions

Related pages