Data Article · Attention & Memory
The Half-Life of Fame
When a famous person dies, the world looks them up — a hundred to three thousand times more than the day before. Then, with remarkable indifference, it looks away. What 719 celebrity deaths reveal about the physics of attention.
On January 26, 2020, the day Kobe Bryant's helicopter went down, his Wikipedia page was read 9.5 million times — about 1,300 times its normal traffic. Chadwick Boseman's death that August drew 9.9 million readers in a day, nearly 2,900× his baseline. These are the largest sudden redirections of human attention that Wikipedia records. And they are gone almost immediately: for the median famous person, the excess attention that follows death loses half its size in a single day.
This article measures posthumous attention for 996 people with major Wikipedia presence who died between late 2015 and the end of 2024 — everyone from Elizabeth II to Umberto Eco — using bot-filtered daily pageviews of English Wikipedia. Attention, it turns out, behaves like a radioactive element with a very short half-life and a very long tail of indifference. The interesting differences between kinds of fame are not in how fast the spike decays (that is universal) but in how big it gets, and how long a faint afterglow persists.
1 The universal shape
Normalize every person's daily pageviews by their own pre-death baseline and the curves become almost superimposable: a vertical wall on the day of death, a peak within 24–48 hours, and a collapse that gives up half the excess in a day and ~90% within a week. Pick anyone below — the shape barely changes; only the height does.
Anatomy of a posthumous attention spike
Daily pageviews as a multiple of the person's own pre-death baseline (log scale) · 120 largest spikes shown in gray
Data table (highlighted person)
The median half-life — the time for excess attention to fall by half from its peak — is one day in every single profession we can measure. Musicians, presidents, scientists: the decay clock is identical. Attention does not linger longer for the more consequential dead.
2 What differs: how loud, and how long the echo
If the decay rate is universal, what distinguishes kinds of fame? Two things. First, the size of the spike: the median athlete's death multiplies their traffic 449-fold; the median scientist's, 102-fold. Athletes and musicians die loudly; scientists die quietly, in part because their baseline readership is already steady and their deaths draw fewer casual searchers.
How loudly each profession's deaths are noticed
Spike ratio: peak daily views ÷ pre-death baseline (log scale) · one dot per person · tick = profession median
Data table
Second, the echo: how many days until traffic settles back to within twice its old baseline. Here the professions separate in a way that says something about how the dead are used. The median musician takes 36 days to fade back below 2× baseline — people keep returning while the albums play. Politicians and military figures are re-filed within ~2 weeks. Grief, apparently, has genres.
Loud versus long: spike size against echo duration
One dot per person · four professions highlighted, all others gray
Data table (highlighted professions)
3 The anniversary heartbeat
Individual fame decays; collective memory pulses. Pages for historical traumas beat once a year on their anniversaries with remarkable regularity — and the beat is strongly shaped by round numbers. The September 11 attacks page runs 26–85× its baseline every September 11; the 20th anniversary (2021) was the loudest pulse in the series, followed by a visible slump. D-Day surged for its 75th and 80th. Memory, institutionalized as commemoration, has its own arithmetic.
Six pages that beat once a year
Anniversary-day peak as a multiple of that year's ordinary traffic · note: y-scales differ per panel
Data table
4 Coda: attention is not consequence
It is tempting to read pageviews as a measure of importance. The clearest counterexample in our window: Roe v. Wade. The page spiked to 1.06 million daily views when the Dobbs draft leaked in May 2022, and 1.27 million the day after the decision that June. Within four days of each spike, more than 80% of the attention was gone. The legal and demographic consequences were, at that point, only beginning — and they compounded for years while the page traffic returned to a few thousand views a day.
Attention collapses; consequences compound
Daily pageviews, log scale · Aug 2021 – Aug 2023
Data table (weekly totals)
The half-life of fame is one day. The half-life of what fame attaches to — a body of songs, a court doctrine, a border drawn in a war — is measured in decades. The gap between those two clocks is, perhaps, the most consistent fact this data has to offer.
Data & methods
- Attention measure. Daily pageviews of English Wikipedia, all access methods, user agents only (bots and spiders excluded), from the Wikimedia REST API, 2015-07-01 through 2026-07-08. Views are per language edition, not per country; they measure worldwide English-reading attention.
- The panel. All humans in Wikidata with an English Wikipedia article, ≥ 40 interlanguage sitelinks (a stringent notability bar), who died between 2015-10-01 and 2024-12-31: 996 people. Of these, 719 are eligible for decay metrics — requiring ≥ 60 observed baseline days and a baseline of ≥ 100 views/day. The exclusions are mostly people whose article barely existed before their death, so spike ratios for marginal figures are understated, not overstated.
- Baseline & metrics. Baseline = median daily views over days −90 to −8 (the final week is excluded; the median resists illness-news bumps). Spike = max over days 0–7. Half-life = days from peak until excess views (above baseline) halve. Echo = days from peak until views ≤ 2× baseline, censored at +365.
- Professions. Wikidata occupations, majority-voted into 13 coarse buckets. People are messy; a fraction of assignments are debatable (opera-singing politicians exist). Medians over n ≥ 15 are robust to this.
- Anniversary panels. Peak = max within ±3 days of the fixed anniversary; annual baseline = that year's median excluding ±14 days around the peak.
- Honest limits. Title renames can split a page's history (spot-checked; negligible here). Wikipedia readership itself trends and mobile share grows over the window, so cross-year ratio comparisons carry some drift. 2025 deaths are excluded — they lack a full post-death observation year.