Methodology

Every number on this site is computed from the public files listed on the data sources page, using the formulas below. Nothing is estimated by feel, and nothing is written by a model. If a formula here does not produce a number you see on a page, that is a bug — tell me.

Where the data comes from

US counts are the Social Security Administration's national and state baby-name files, covering births from 1880 to 2025. These are counts of Social Security card applications, which is close to but not identical to counts of births. Three consequences matter:

• Births before 1937 are undercounted, because Social Security did not exist yet and cards were issued to those cohorts later, if at all. Early years should be read as a sample, not a census.
• Any name with fewer than five births in a given year and sex is omitted entirely. A name can therefore look like it vanished when it simply dropped below five.
• The files count births, not living people. Turning births into a living population is the calculation described under living estimate.

England & Wales counts come from the Office for National Statistics, covering 2019–2025. ONS suppresses counts below three. Scotland and Northern Ireland publish separately and are not included, which is why the site says "England and Wales" and never "UK".

Definitions

Rank

Position within a single birth year and sex, ordered by count, ties sharing a rank. Rank 1 in 2025 means no other name of that sex had more recorded births that year. Ranks are not comparable across sexes.

Living estimate

For each birth year y and sex s, the recorded count is multiplied by the chance a person born in that year is still alive, then summed:

living = Σ_y,s count(name, y, s) × l_x,s(2025 − y) ÷ l_x,s(0)

The survival curve is the SSA actuarial period life table for 2023, which runs to age 112; anyone who would be older is counted as deceased. This is a period table used as a cohort approximation: it applies today's mortality rates to people born a century ago, rather than the rates those cohorts actually lived through. That understates deaths in older cohorts, so estimates for very old names run somewhat high. It is a simplification, and this paragraph is where it is disclosed rather than hidden.

Two further limits: the estimate counts people who were born in the US and issued a card, so it misses immigrants who arrived with the name already, and it inherits the five-birth floor, so it misses everyone whose birth year fell below it. Results are rounded to the nearest 10 below 1,000 and the nearest 100 above, because the underlying precision does not justify a exact-looking figure.

Typical age and modal decade

The same survival weights produce a distribution of living bearers by birth year. The median age is the age at which half that weighted mass falls on either side; the mean is the weighted average. The modal birth decade is the decade holding the largest share of living bearers.

Location quotient

LQ = (state births with the name ÷ all state births) ÷ (national births with the name ÷ all national births)

Computed over the ten years 2016–2025, and reported only where the name has at least 20 births in that state and an LQ of 1.3 or higher. An LQ of 3.0 means the name is three times as common in that state as it is nationally — it says nothing about the raw number of children, which is why the tables show both.

Trend and momentum

The five- and ten-year trends are natural logs of the ratio between the latest count and the count n years earlier, shown on the page as a percentage change. If the earlier year has no recorded births, the trend is undefined and no sentence appears. Momentum is the change in rank against the previous year within the name's dominant sex; a positive number means the name climbed.

Gender share and unisex index

unisex index = 1 − |2 × (girls ÷ all births) − 1|

An index of 1.0 is a perfectly even split and 0.0 is entirely one sex. The site calls a name unisex when the index over the last ten years is 0.6 or higher, and mentions the split at all when it is 0.15 or higher. A name is flagged as having changed sides when its majority sex differs between its first and last decade of use by more than 30 percentage points.

Era label

A rule, not a judgement. The base label comes from the peak year: before 1930 is early 20th-century, 1930–1959 mid-century, 1960–1989 late 20th-century, 1990–2009 millennial-era, 2010 onward current. Two modifiers can attach: revival when a name that first appeared before 1950 has more than doubled over ten years, and perennial when it has spent 50 or more years in the top 100.

Similar names

Names sharing the dominant sex and either the same metaphone key or a Levenshtein distance of two or less with the same first letter, ordered by total recorded births and capped at twelve. This is a similarity of spelling and sound computed inside this dataset. It is not a claim that the names share an origin.

Origin

Shown only when a source says so. The sources, strongest first: an English Wiktionary entry filed under a given-name-from-language category; a Wikidata item carrying a language-of-name statement; or a native-script Wiktionary given-name entry whose own transliteration matches the spelling. The third is weaker than the first two and is reviewed by hand before it is used to publish a page.

Where no source exists, the section does not appear. This site does not generate meanings, etymologies or "personality traits" for names, and the build fails if an origin section renders without a source row behind it.

Where to see this applied

The definitions read more easily against real numbers. Ryan renders almost every section on one page, and California is where the location quotient does most of the work. The data stories are each built from one of these metrics alone.

What is not published

A name gets a page when there is demand for it and enough data to answer more than one question about it: at least six of the sections above have to render. Roughly 8,979 names currently clear that bar. Everything else stays in the database and is reachable through the data files, but does not become a URL. Pages that would be combinatorial — every name crossed with every state, every pair of names compared — are not built at all, because they would be near-duplicates of each other.

Rebuilding these numbers yourself

Every figure can be recomputed from the source files with the formulas above. The data sources page lists the exact downloads, and each name page links to a JSON endpoint with its own numbers. If you recompute something and get a different answer, I want to know.