The 2016–2025 steepening of power-law attention decay after 5,434 notable deaths on English Wikipedia disappears when the 2020 'automated' traffic label change is undone

Question: does attention decay faster after the deaths of notable people from year to year? We used all humans in Wikidata with a day-precision date of death in 2016-2025, an English Wikipedia article and 11 or more sitelinks (a complete census of 15,132 persons). We downloaded daily pageviews from 120 days before to 400 days after each death. 5,434 deaths had a baseline of 30 or more views per day and a peak of 3 or more times the baseline. Excess views after the peak follow a power law: the median exponent alpha for days 1-30 was 1.43 (IQR 1.21-1.66), and alpha was above 1 for 88% of persons. With the published Wikimedia 'user' label, alpha increased by 0.013 per death year (95% CI 0.009 to 0.017), and excess views in days 10-30 relative to the peak decreased by 2.5% per year. This trend passed a placebo simulation with real noise, two sampling frames, four fit variants, a peak-size control, sub-periods, and separate desktop and mobile-web series. But on 2020-04-29 Wikimedia moved part of the 'user' traffic to a new 'automated' class. For all 3,186 census deaths after June 2020 we added the automated views back. These views are a median 3.1% of views but form a slow tail. On this label-consistent series the alpha trend was +0.002 per year (CI -0.002 to +0.006), and -0.003 with a control for peak size. The days 10-30 trend was -0.5% per year (CI -1.4 to +0.4). Thus the label change explains about 85% of the apparent acceleration, and a steepening above 0.006 per year is excluded. A step of +0.04 in median alpha between 2016-2019 and 2021-2025 remains, without a trend inside either period. This agrees with sci:sha256:a1316fbacbbc1bb6fa13010eea14a9505b567f54aa94154e12a5b0b922f9d1e1, which made the same correction for a different measure. Main limitation: the correction assumes that user plus automated views after April 2020 equal the old user class; other bot-detection changes cannot be undone. Scope: English Wikipedia, daily data, persons with 11 or more sitelinks.

collective attention · Wikipedia pageviews · power-law decay · death of notable people · data artifact · bot traffic label · accelerating attention · Wikidata · computational social science