World-Record Half-Lives: how long records stand

A world record has a lifespan. It is born the day it is set and dies the day someone beats it, and the records standing today simply haven't died yet. That is exactly the shape of a survival analysis: durations, some of them still running. So I turned a century of athletics into survival data, ~250 record "reigns" across nine events, and let the model tell us how long records live, why that's changing, and which current records look most exposed.

~1.5 yrmedian record half-life. Most records fall fast

≈ 2×longer that modern (post-1980) records survive

9events, each verified complete against the real world record

Records fall fast, but far less than they used to

The median record survives only about a year and a half: someone is usually right behind. But that number hides a heavy tail, and the tail is growing. Split the survival curve by era and a record set since 1980 clearly outlives an older one. The modern game is harder to move. A proportional-hazards model agrees: being a modern record roughly halves the chance of being broken, and records set by a big margin last longer still.

Kaplan-Meier survival curves for records set before vs after 1980 — Kaplan-Meier survival of world records. Post-1980 records (wine) outlast pre-1980 ones (grey).

Every progression bends toward a wall

Plot each event's record over time and the same shape appears everywhere: each new record shaves a little less than the one before, the curve flattening toward some human limit. For the women's 800 m and high jump that plateau has plainly arrived; for the distance events the curve is still dropping.

Record progressions for four events with fitted deceleration curves — Record progressions with fitted deceleration. Each jump is smaller than the last.

So which record falls next?

Because a record's risk is highest early, the cleanest forecast comes straight from the survival curve: given that a record has already survived so many years, what's the chance it falls in the next five? Fresh records score high. The 2023 marathon (~60%) in the super-shoe era, the 2024 women's high jump (~59%). Old records have earned their safety by surviving the dangerous years, and Kratochvílová's 1983 women's 800 m comes out near zero, the statistical picture of a record no one expects to fall.

Model probability each current record falls within five years — Five-year fall probability for each current record, from the survival model.

A note on the data

The progressions come from Wikipedia, but those pages bury the real record history among top-lists and side tables, scrape carelessly and you get truncated or fictional records. Every event here had to pass a hard test: its reconstructed best must equal the current ratified world record. Nine events passed; the rest were left out rather than shown half-true. The code and full method are on GitHub.