I did some number crunching on the moderator election results to see how closely they lined up with the rankings in the moderator nomination thread. I was going to post this as a comment on the SO blog, but I decided I'd share my findings here for those of you who enjoy statistics and graphs. If you are impatient, I'll cut to the chase and tell you that there was very little correlation between ranking in the moderator nomination thread and ranking in the actual election. Which wasn't very surprising to me, but I thought I'd share anyway.
Update: Just as I went to post this, the title auto-search showed me another post on election statistics. In the comments there, Jeff suggest a strong correlation between the user's rep and their election ranking. So I added that into my graph. Sure enough, there is a correlation: approximately half of the voting behavior is likely based on the user's reputation.0 If we take Jason Cohen out of the results, reputation explains about 77% of the voting behavior (see my second update below).
The table below shows the user's rank in the actual election1, as well as the number of upvotes and downvotes they received in the nomination thread on Meta.2
Rank User Upvotes Downvotes Net votes ------------------------------------------------------- 1 Gumbo 14 8 6 2 Jason Cohen 14 5 9 3 Jonathan Sampson 43 19 24 4 Will 15 7 8 5 mmyers 57 8 49 6 Jonathan Leffler 19 5 14 7 Kyle 46 14 32 8 Dave DeLong 11 5 6 9 John Topley 42 7 35 10 Welbog 27 24 3 11 George Stocker 42 21 21 12 John Rudy 15 13 2 13 jjnguy 18 13 5 14 Kip 13 6 7 15 Robert C. Cartaino 22 9 13
I have also created a chart, showing the user's rank in the election, as well as the rank if selections were made based solely on: number of upvotes in the nominations, net vote score in nominations, lowest number of downvotes in nominations, and reputation on Stack Overflow.3
http://img189.imageshack.us/img189/9676/electionrankings.pngAnd here is a summary of the data in the chart, with some statistical analysis. The "average difference" statistic is the averge of the absolute difference between a user's rank in the election and the user's rank by the given ranking method.
Ranking Correlation Average Method coefficient r-squared difference -------------------------------------------- Election 1.0000 1.0000 0.0 Net votes 0.2894 0.0837 4.3 Upvotes 0.0507 0.0026 5.0 Downvotes 0.2658 0.0706 4.7 Reputation 0.7036 0.4950 2.0
If you're wondering why downvotes have a positive correlation, it is because the "downvotes" ranking represent what would happen if we had chosen moderators based solely on who had the least downvotes on the nomination thread. If we chose based on most downvotes, the correlation would indeed be negative.
Update 2: I noticed that Jason Cohen was a major outlier in the ranking by rep. If you remove him from the stats, you get correlation coefficient = 0.8813, and r-squared = 0.7767, which is a very strong correlation.4
0 At least, I think that's what r-squared tells you. It's been a while since I've taken a statistics course.
1 Based on the inverse of the order in which they were eliminated by the Meek STV algorithm. For the top three users, relative ranking is based on transferred votes.
2 Data was retrieved on 2010-02-09, sometime roughly around 7:00 PM GMT.
3 Note: in the event of an n-way tie for ith place, all tied users are assigned the rank i, and the next-highest-ranked user is assigned i+n. For example, John Topley and George Stocker both received 42 upvotes, which ties them for 4th place when ranked by upvotes. The next highest user, with 27 upvotes, is Welbog, so he is assigned 6th place, and no user is assigned 5th place.
4 Of course this isn't entirely accurate, because we don't know how people who voted for Jason would have voted had he not been there.
