Spikes in English Wikipedia views after deaths of notable people did not decay faster or slower from 2015 to 2025; the general slowdown of spikes comes mostly from fewer one-day spikes

Question: did the decay of collective attention on English Wikipedia change from 2015 to 2025 for a fixed kind of stimulus? Earlier SCI studies found slower, not faster, attention in daily top lists and linked this to one-day spikes, but they could not tell if these spikes are automated traffic or human bursts. Data: Wikimedia daily top-1000 lists (user agent) for all-access (4,127 days) and mobile-web (4,122 days), 2015-07-01 to 2026-10-21, and Wikidata dates of death. Method: we selected new spikes that reach the article top 50 and are at least 8 times the rank-1000 cutoff on all days from 7 days before to 30 days after the peak. Thus all values at or above 1/8 of the peak are exact. We measured the peak-day share of spike views, the next-day ratio, the days until views fall below 1/8 of the peak, and the fraction of one-day spikes. Death spikes have a Wikidata date of death 0 to 2 days before the peak. Intervals come from 2,000 bootstrap samples of calendar months. Results: in 1,496 all-access death spikes, the peak share changed from 0.582 (2015-2017) to 0.569 (2023-2025), a difference of -0.013 (95% CI -0.032 to +0.006). The days to 1/8 of the peak changed by -0.08 (CI -0.28 to +0.13). The result held for lag 0, lag 0-1 and two size groups. In mobile-web data, death spikes became slightly broader (peak share -0.039, CI -0.058 to -0.021). In all 15,900 all-access spikes, the peak share decreased from 0.719 to 0.563, and the fraction of one-day spikes decreased from 0.354 to 0.093. This change in the mix of spikes explains 72% of the decrease (66% in mobile-web data). In 2015-2016, 29-30% of one-day spikes got less than 10% of peak-day views from the mobile web, against 6% of other spikes. In 2026 this fraction increased again to 31%, mostly on obscure pages. Conclusion: we found no acceleration of attention decay, also for a stimulus class with a fixed cause. The large aggregate slowdown is mostly a loss of one-day, often desktop-only spikes. Limitations: English Wikipedia only, daily resolution, large spikes only, values below 1/8 of the peak not seen, and the user traffic class changed in 2020. We did not test causes.

collective attention · Wikipedia pageviews · social acceleration · attention decay · deaths · Wikidata · automated traffic · burst dynamics · top lists · computational social science