Les drops de Q à l'épreuve des faits
Étude statistique des messages de « Q » (2017-2022) confrontés à l'actualité
Abstract
Between October 2017 and November 2022, an anonymous author signing "Q" posted 4,966 messages ("drops") on 4chan, 8chan and 8kun, the founding texts of the QAnon conspiracy theory. Most research studies QAnon's followers and platforms. We study the drops themselves. We check what they say against what actually happened.
Nothing suggests that Q anticipated the news. Trump's tweets do not follow the drops more often than chance once local activity is taken into account (4.2% vs 4.0%, p = 0.23). Of the checkable predictions, 28% came true (95% CI 21–36%), but only 16% once scheduled events are set aside. None of the announced arrests or indictments happened. In a blind "horoscope test", a language model must say which of two news lists goes with a drop. Three times out of four it finds no specific link. When it does, it favours the real period for the past and the future alike, and a second model sees a specific link in only 5 of 100 "future" comparisons. The timing of the drops is better described by a model of earthquake aftershocks than by a news feed. That model attributes most drops to earlier drops and almost none to Trump's tweets. Against the wider scene (six subreddits sampled every active day from 2016 to 2023, Wikipedia, Google Trends, 4chan's /pol/), Q draws on a small number of the scene's themes, and we detect no precedence either way at the scales we tested. Its own forums, though, react within hours. On /pol/, which we read in full (134.5 million posts), Q moved into threads already opened by Trump's "calm before the storm". It kept their conspiracies but not their explicit hate vocabulary, and its threads are calmer than the rest of the board.
The study combines descriptive statistics, text analysis, permutation tests, GDELT news data and a sourced check of 250 predictions. All data, code and verdicts are public.
The full text is in French (HTML on the French page, PDF above), and this English abstract summarises it.