Wikipedia talk:Wikipedia Signpost/2022-06-26/Essay
Discuss this story
A better way of looking at RFA trends is to look at some of the RFAs that have spectacularly changed direction. Usually these are negative trends - someone comes up with a reason to oppose the candidate that gets traction among the !voting community and the RFA changes direction. Usually these are very obvious when you read the subsequent vote rationales. Sometimes it happens becasue of a mistake the candidate or their nominator made during the RFA - one of the classics being when the nominator picked up the wrong laptop and made a comment while logged in as his nominee/girlfriend. Othertimes it happens when one of the few people who actually review the candidate's edits spots a problem and details it in their oppose. The minority of RFA participants who actually spend an hour or so checking the candidate's edits have a huge influence on RFAs. ϢereSpielChequers 15:24, 30 June 2022 (UTC)
- It seems to me that there is an error here somewhere in the argumentation re independence of samples. Assuming the value of each new sample in this process is partly dependent on the previous sample, then there is no requirement for the population mean to be affected to get biased results - we are not 'selecting samples of a certain value' from the pool, but 'modifying a drawn sample post selection'; 'p' doesn't seem to come into it. - Maybe a better approach to analysing this series would be treatment as an autocorrelated time-series. That would allow identification of inflection points, with some estimate of confidence. --Elmidae (talk · contribs) 15:27, 2 July 2022 (UTC)
- I expect that an autocorrelated time-series would still assume that the population of people who !vote in the first day, the second day and so on are roughly identical. But there is a confounding variable in RfA trend lines: the populations of strongest support !vote first. For instance, if I see the candidate's name every week, almost always in a context where they are creating good content or making comments that I agree with, then I'm going to vote immediately; if I see it in a more negative context then I might take more time to decide or take a while to build up my oppose rationale. If I've watchlisted the candidate's talk page then I will see the RfA quicker. The message to me is: trend line "haruspicy" is doomed to fail. We should just take the tallies as they are and crats should decide consensus on factors other than trend lines in cases where the tally isn't clear enough. (The exception is if there's a clear information gap e.g. undisclosed COI editing by the candidate was only unearthed on day 6.) And all of this trend line analysis adds to the excruciating pressure and overanalysis that discourages prospective RfA candidates. — Bilorv (talk) 13:53, 13 July 2022 (UTC)