User:Opabinia regalis/RfA data

Notes on various admin-related data and graphs.

Admin activity

 * In both cases, ~30-35% of admins take 0 logged actions.
 * 2014-15 has many more lightly active admins (just a handful of actions) and many fewer highly active ones compared to 2006-7. More work in fewer hands.
 * Hard to tell from the graphs, but the raw numbers of logged actions are much higher in the older data. (Note, excluding flagged bots does make a difference here - a lot of people used to run bots on their main accounts.)

RfA discretionary zone, 2008 to mid-2015

 * The de facto discretionary zone is 70-75%. Nobody failed over 75% since 2008. About a third in the 70-75% range passed. Two <70% outliers.

RfA outcomes

 * Distribution of outcomes has been surprisingly stable over time, despite massive decrease in raw numbers.
 * People don't withdraw more often now, and the fraction of NOTNOW/SNOW closes hasn't changed, despite popular belief to the contrary.

Talk page chatter

 * Number of RfAs has gone down, but the amount of discussion (as measured by posts, not bytes) per RfA at WT:RFA has gone up.
 * Not clear if that's discussion about the actual RfAs, or discussion about the drought.

RfA and rollback

 * This is not new - see User:WereSpielChequers/RFA by month for another presentation of the same data - but it's interesting to see it plotted out. The long-term declining trend in RfA candidacies began in early 2008, around the same time the rollback user right was introduced.
 * Separately plotting successful and unsuccessful candidacies suggests that there was a lag of a few months between the beginning of the decline in successful candidacies and the beginning of the decline in unsuccessful candidacies. It's not clear whether these unsuccessful candidates had expected to succeed based on typical criteria from just a few months earlier, or if candidates less likely to succeed were also less aware of the apparent changes in environment.

Admin tenure
In several 2016 RfAs, there was discussion of standards for wiki-tenure qualifying a candidate for RfA, with some editors suggesting extremely long tenure periods as their personal expectations. This prompted me to wonder what the actual distributions look like.


 * Currently active admins registered a long time ago. There are only seven active admins who registered in 2013 or later, and three of those are adminbot accounts. Put another way, a bit over 75% of our currently active admins had already arrived on Wikipedia by the end of 2006, and 99% had arrived by 2013.
 * ACE2016 voters served as a convenient comparison set, representing a pool of users who are at least engaged enough with the project to have noticed the talk-page mass message about the election and decided to act on it. (This group naturally has some overlap with the active admins.) Unsurprisingly, admin registration dates skew much earlier. About 40% of ACE2016 voters registered in 2011 or later, compared to about 4% of the admins.

Pipeline to adminship

 * It's inevitable the distribution of tenure lengths will get broader over time - the ceiling for tenure length grows at a rate of one year per year, after all ;)
 * The lower end of the tenure length range for successful RfAs is the interesting part and can be seen at either the real minimum or the bottom of the distribution (ie the bottom of the box plots, representing the lower quartile). What is apparently considered "minimum tenure" has been growing dramatically.
 * In 2006, the median tenure length was about one year. The most recent successful RfAs from editors with one year's tenure occurred in 2010. In the period from 2011 to 2016, six administrators were promoted with under 15 months' tenure, out of 170 total promotions. A single 2015 registrant was promoted in 2016. What was once the median tenure length is now vanishingly rare.
 * Minimum expectations are wildly out of sync with what was expected of the overwhelming majority of our current active admin pool.


 * The successful RfA lists were scraped from these pages, so any errors there were inherited by this analysis.
 * Registration dates were obtained by API queries on the usernames listed on the above pages. Some users have changed their names since their RfA. This results in either a null result, or a post-RfA registration date (if the name was registered again later). These were excluded.
 * I realized after the fact that I misformatted queries for usernames with special characters. These were also excluded.
 * Some usernames returned null for unknown reasons, and were also excluded.
 * Some early user logs are not accurate, and some early RfAs were documented in the pages above but conducted previously on the mailing list. Log oddities resulted in a handful of apparent very quick adminships in early years.