Some of the recent discussion on Stef and SCSN's great news thread left me wondering about what Dominion's player base looks like.
I ran some elementary analyses on the full unfiltered Isotropish leaderboard for Dominion 2.0 (as of 13th of April 2016), this means that I took into account:
- 2p games only (which may mask/underrate players who play mostly 3p+)
- Only botless games.
- Isotropish level (mu-3sigma) rather than mu, which will severely underrate players with a low volume of plays. I also compared functions of number of plays with Isotropish level (also a function of play number), so some analyses might be skewed. Also, Isotropish level is a the result of an algorythm I don't know much about, but I suspect some of the results are artifacts due to how TrueSkill works.
- Only games played since Goko 2.0 went online (around June 2015, I believe) were taken into account.
I'm no statistician, and this was the first time I used array formula in excel, so if anybody spots errrors or has suggestions, please speak up!
Anyway, to the numbers!
Since June 2015, 36296 players have played 1787686 2-player games of Dominion Online. Of these players, only 12,174 played at least 20 games, which drops to 5,892 if you set the inactivity treshold to 100 games.
Minimum level is -75, maximum is 65.
As you might expect, median and average level are very low in this population since it contains a lot of extremely-low volume players (in addition to being inexperienced ones). The mean is -22, median is -26. (incidentally, these are very similar to starting values of mu-3sigma=0-25, is that significant? Who knows.)
Trying to get more useful data out of the leaderboard, I tried to calculate the mean after filtering the values by minimum number of plays:
Treshold Mean
20 2,57
50 8,21
100 11,98
500 18,48
The mean grows slowly even when applying quite strict filters.
Weighting the mean by number of plays, we get 13.
I believe that the most interesting parameter of this set is the median Iso level calculated on the "gameplays" population. It returns 15. This means that half of the games played on MF were played by players with a level lower than 15.
When I calculated the proportion of games played by players with level above a treshold of 30 (where I arbitrarily drew the line for "competitivish" players), I got 18%.
The player mode is an exhilarating -54, most likely due to how Trueskill makes players with very few (losing) plays converge.
I also made a couple simple distribution histograms for players, plays, and play density on Iso level:
I scaled plays number to get both histograms on the same picture.
The player distribution is frankly multimodal, with peaks at -54, -33 and 0. This is probably due to artifacts from the Trueskill algorythm (players with one or two losing games might converge here?). Anyway, the curve is quite far to the left.
The plays distribution looks nearly normal, I haven't had time to try and model it, but I might try to do that.
The big difference in the curve locations inspired me to do another histogram:
It seems that playing more dominion makes you better at it.
But also, it makes your rating better by lowering your sigma.
I suspect this "sigma effect" might be more pronounced around the level interval between 0 and 20, while skill matters more for higher levels. The far right region gets more erratic due to low player density.
So, conclusions.
It was fun to make this simple analysis. It seems like there's plenty of people who play a lot of 2p online dominion while not being very good at it, I guess.
I'd write more, but I really have to go.
Extra chart: