Do musicians really
die at 27?

The claim: musicians die disproportionately at 27 — Hendrix, Joplin, Morrison, Cobain, Winehouse, Amy Winehouse’s own father invoking it at her funeral. It is one of the few statistical folk-claims that names its own sample size (five, sometimes six) and calls it a pattern. This run pulls every musician on Wikidata with a recorded birth and death date, born 1900 or later — 26,215 people, not five — and asks whether age 27 is a real bump in that record, or just the age the story picked.

First, the shape of the data. Musicians who died in their 20s and 30s are not a fixed population at risk each year; the count of notable-enough-to-have-a-Wikidata-page musicians who die at a given age rises steadily through this range, because more people have entered public life by 35 than by 20. A plain quadratic fit to log(count) across ages 20–40 tracks that rise closely — R²=0.929, linear term +0.305 (95% CI [0.160, 0.450], p=3.3e-04). That smooth curve is the baseline age 27 has to beat.

It doesn’t beat it. Fitting the quadratic on the other 20 ages and predicting age 27 from that held-out curve: 62 musicians observed died at 27, against 52 predicted — a standardized residual of z=+0.96, p=0.35. That is not a small effect that fails to reach the conventional bar; it is unremarkable on its face, 15% above the mean of its four immediate neighbors (54), well inside the noise a curve this bumpy throws off at every age.

Run the identical test on all 21 ages in the window and rank them: age 27 comes in 6th of 21 for |z|. The single most anomalous age in the whole 20–40 range is 20 (an undercount, z=-2.37, raw p=0.030) — and even that fails once corrected for testing 21 ages instead of picking one after the fact (Bonferroni p=0.63). Ages 31 and 35 also swing harder than 27. Nothing in this window survives multiple-comparisons correction, including the age with the most dramatic raw deviation. A quadratic already explains the data; there is no local anomaly left for any age to own.

The caveat that cuts the other way, stated plainly: this dataset’s selection mechanism, if it biases anything, biases toward the myth, not against it — a musician who dies at 27 is more likely to get a Wikipedia/Wikidata biography because the 27 Club framing exists, not less. A null result under a thumb on the scale in the claim’s own favor is a stronger null than one measured on neutral ground. Separately: "musician" on Wikidata spans classical, jazz, and session players alongside the rock icons the folk claim actually means, which this run does not separate out — a narrower rock/pop-only slice is a different, unrun question, not a re-fit of this one.