Here we go.
My raw results are here:
LINK
Feel free to peruse that, try to reproduce my results if you want, and most importantly, if you want this sort of ranking set up for your own character so you see who's who within your main, you can find all that info here.
----------
PURPOSE
----------
The main idea here is to determine the skill distribution among characters. This should give, as objectively as is possible, a display of which characters are "popular" at various skill levels.
If a character is simply better than another, one would expect the shape of these curves to be the same, but one curve to be higher than another at any given rank, vertically in the plots I will make.
If a character is simply more popular than another, one would expect the curve at any particular skill level to stretch out to higher ranks, horizontally in the plots I will make.
Of course, the two ideas are difficult to separate. We will therefore search for convenient reference points in the data to compare one character to the next.
-----------
METHODS
-----------
I created a Python script (provided in the zip file above) to use Ankoku's tournament results .txt exports and used Ankoku's scoring system to determine how much each player scored in each tournament for 1. the past 13 months of complete data (January 2009 through January 2010), and 2. the past 6 months (August 2009 to January 2010) of complete data.
These players and results were binned into the characters they represented. Within each character, players with identical names are automatically combined, and for each of Meta Knight, Snake, Diddy Kong, Falco, and Marth player, I manually went through the lists to find names that appear to be of the same player and combined them.
I admit that I do not know the tournament scene as well as I might; it is possible that I failed to merge some results under separate player names that should have been together. I invite people to inspect my .xls files in the .zip above to verify I didn't miss anyone.
It should be noted that no integer number of months would exactly reproduce Ankoku's current score per character, but 6 months came pretty close. Either Ankoku's export wasn't a complete represenation of his data, or the method of treating odd cases was different from mine. I have no reason to believe the errors are systematic, so the results should not be biased either way.
----------
RESULTS
----------
Above, I show, for each of the past 6 and 13 months, how many points each player scored with a given character as a function of their rank within that character, at two different levels of zoom for each data set, for each of the five best overall performing characters.
(Note: I know I accidentally changed the rank scales between 13 to 6 months. It's probably still fair, though, since 13 months has more people in it than 6. If this really bugs someone, I'll go through the effort to fix it and reupload.)
Below, I show histograms: they show how many PEOPLE scored between X and Y points. At each successive bin, the point values are doubled. These histograms should be better at displaying how the tournament scecne looks for the medium-low skill levels.
Fun fact:
In the past 6 months, Ally is the 9th highest ranked Meta Knight. (Overswarm is 10th).
----------------------------
ANALYSIS AND CONCLUSIONS
----------------------------
At this point this report becomes slightly more subjective.
Looking at the top few spots, it is obvious that the top player of each of Meta Knight, Snake, and Diddy Kong are outliers and do not fit their characters' overall trends.
For Diddy, Marth, and Falco, except for the top 3 players, those characters appear to be farily close in tournament performance from one to the next, though it appears Falco wasn't as good a year ago as he has been in the past half year. Snake has been better than those three, but the degree by which Meta Knight is better than Snake at all ranks is much higher than the amount by which Snake is better than the others.
This much could have been guessed just by Ankoku's summed data in his ranking list, but this does show that Meta Knight isn't worse than other character(s), in which case his high overall performance would require a very, very long "plateau" at a medium-high skill level, much longer than for Snake, to make up for the point difference observed.
As for popularity versus character goodness, first let us look at the 13 month data again.
There is a "kink" at around rank 6 for both Meta Knight and Snake, indicating the rank at which the players drop off from being high level players to being the next level down, indicating equal popularity. Throughout this zone, each MK player scored more than 2x as many points as each Snake. At 13 months, the rest of the data is pretty smooth and does not lend itself well to differentiating between skill and popularity; suffice it to say that MK's dominance does not disappear at any level, although some of the other characters do cross each other from rank to rank.
In the 6 month data, there are two usable kinks in Snake's data, and one in Meta Knight's data. For high level play, MK is about 2x better a character than Snake and slightly more popular among players at that level. It is not immediately obvious where to link up the second kink, but the two extremes of justifiable locations both say that MK is more popular, and differ in saying whether at medium-high level of play MK continues to be a 2x better a character at 1.5x the popularity or if instead MK is 3x as popular as Snake at that level.
Without a specific reason to believe that the selection of Snake or MK changes greatly between high and medium-high play, I'm inclined to believe that MK simply continues to be slightly more popular but 2x as good as Snake, and 3-5x as good as other characters.
Looking at the tails of the rank plots, it is clear that the amount by which MK has been outperforming other characters at ranks up to 27 or so has increased dramatically between 13 months ago and 6 months ago. For the next 10 ranks it's less pronounced, then it becomes roughly equivalent to the 13 months ago level.
Now let us compare the 13 month and 6 month histograms. The amount by which MK is more popular than other characters at lower levels doesn't appear to have changed much, but at medium and medium-high levels of play it does; his increased level of dominance appears to come primarily from there.
A side note: there are a lot more players in the 13 month data than in the 6 month data; a bit less than half as many, it appears. This would be consistent with both the following theories:
1. At every month, the set of players that make it to the top 8s in some tourney is selected in such a way that the players who have earned points in previous months are preferred NOT to hog the same spots again, so that the total number of unique players is linear in time.
2. The number of competitive players (that is, those who play brawl and actually have a shot at getting top 8 or better) has gone down by more than 50% in the past 13 months.
The first theory is obviously wrong. I can't prove or disprove the second with the data I have right now, but it is clear that the competitive Brawl community has been shrinking by some amount.
So, how much better is MK than other top characters? Let me finish with one more graph.
