My hypothesis was usernames of newbies for which the last character is a digit are more likely to replace out than usernames for which this is not the case.
I looked at the player lists for all the newbie games going back to 1900.
If a slot was in the bottom 3 slots, it was ignored (that's where the SEs go). If the fourth to last or fifth to last slot were also marked SE, same treatment, entire row is ignored. If someone replaced into a newbie slot, I treated them as a newbie unless the mod marked their entrance SE, even in cases where I knew the player was SE. Science is about removing bias more than removing inaccuracy.
It would have been more tedious to count unique identities of players and I think if I did the correlation may have been even more dramatic, the same few ends-in-number players constituted a lot of the numbered completions, but that happens a lot for unnumbered players too, it might come out the same.
The results can be made a little more
Ends-in-alpha game completions: 320
Ends-in-alpha game replace-outs: 146
Ends-in-number game completions: 38
Ends-in-number game replace-outs: 36
Percentage of alpha players who end their involvement in a game by replacing out: 31.3%
Percentage of number players who end their involvement in a game by replacing out. 51.4%
Number players are responsible for 10.6% of game completions, but 19.8% of the replace-outs.
I don't have much reason for posting about this besides amusement and I have little to no idea why the trend trends this way. I guess it can be useful to predict a replace-out as scum but it's way less useful than the greeting tell someone did with data diving.
There are other name characteristics that might be correlated with replace-out rate or possibly other elements, like strength of play, and I would actually default to assuming other users besides myself with multi-word, not-camel-case names are a bit more likely to replace out and maybe a bit weaker than average, even though I don't replace out, and I'm massively weaker than average
.