Matched controls show no median year-2 rise in English Wikipedia views after 1,026 sampled deaths (2017–2024), a fall of about 10% per decade of age at death, and inflated one-year plateau fits

Question: after a death, does attention to the English Wikipedia article stay above the level expected if the person were alive? We sampled at random 2,000 of 60,678 Wikidata humans with an English article, at least 3 sitelinks and a day-precision death date from July 2016 to June 2024, and 1,500 of 653,384 living people born 1925-1999. We downloaded daily user pageviews from July 2015 to September 2026. For 1,026 deaths with at least 1 view per day in the year before death, we matched 5 controls in the same calendar windows on occupation group, gender, birth year and the level and trend of pre-death views. We used three control designs: living people, and people who died more than 365 or 90 days after the outcome window. The outcome is the change in log10 mean views from days -365 to -31 to a later window, minus the same change for the controls. Results: the median premium was about 16 to 18 times in days 0-29 and +17% to +19% in days 30-364. In days 380-730 it was +1.5% (95% CI -0.7% to +4.0%), -0.6% (-4.6% to +2.7%) and -1.0% (-4.2% to +2.2%) in the three designs, and only 49% to 52% of people had a positive premium. Pre-death popularity did not change the premium. Age at death did: -0.043 to -0.047 log10 units per 10 years (about -10% per decade; all CIs exclude zero), from +30% to +44% before age 60 to -4% to -7% at 80 or older. Placebo runs with pseudo-dates gave no age gradient and small biases (-2.3% to -0.6%). A power-law-plus-offset fit to weeks 1-52 gave a lasting offset of +11% to +14%, because the excess still decreased after one year. Thus plateau fits to the first year overstate the long-term premium. This is an independent replication of an earlier SCI study with a larger sample. It agrees on the null mean year-2 effect and on the age gradient. Limits: only current titles are counted, living status comes from missing Wikidata death dates, the bootstrap ignores control reuse, and the samples are moderate because of API rate limits.

Wikipedia pageviews · collective attention · collective memory · death · matched controls · not-yet-treated controls · placebo test · age at death · power-law decay · replication · Wikidata