Currently best \ worst versus doesn't really indicate which hero can be a good counterpick for other hero - just because it is affected by how good a particular hero is (if you notice, both statistics list for all heroes almost always the same heroes being the best \ the worst heroes in game). Instead of actual win ratio, it would be nice if linkage disequilibrium (D) would be measured (some kind of difference between the joint win\lose rate of the two heroes and the product of how good those heroes are);
For example to calculate stats for broodmother <-> necrolyte:
D = (0,3384)^2-0,3982*0,4501 = -0,0647, where 33,84% is BM win rate against Necrolyte, 39,82% is overall BM win rate and 45,01% is overall Necro lose rate; in this case the negative value of D means that necro is a good pick versus brood
Another example: both AM and meepo are heroes with low win % ratio and both of them have each other in their best versus, so which heroes "counterpicks" which from this pair ? simple calculations of D shows that despite AM being a little better hero overall (41,14% vs 40,23%), meepo is actually a good hero versus AM
Ofc those statistics for each pair of heroes could be presented by a single matrix.
PS. there is something wrong with heroes win\lose counts anyway - for example am and meepo had 9267 games, meepo won 4635 and am won 4435 (4435+4635 != 9267). My guess is that 9267 includes matches with meepo and am playing on the same side (which is incorrect).
Currently best \ worst versus doesn't really indicate which hero can be a good counterpick for other hero - just because it is affected by how good a particular hero is (if you notice, both statistics list for all heroes almost always the same heroes being the best \ the worst heroes in game). Instead of actual win ratio, it would be nice if linkage disequilibrium (D) would be measured (some kind of difference between the joint win\lose rate of the two heroes and the product of how good those heroes are);
For example to calculate stats for broodmother <-> necrolyte:
D = (0,3384)^2-0,3982*0,4501 = -0,0647, where 33,84% is BM win rate against Necrolyte, 39,82% is overall BM win rate and 45,01% is overall Necro lose rate; in this case the negative value of D means that necro is a good pick versus brood
Another example: both AM and meepo are heroes with low win % ratio and both of them have each other in their best versus, so which heroes "counterpicks" which from this pair ? simple calculations of D shows that despite AM being a little better hero overall (41,14% vs 40,23%), meepo is actually a good hero versus AM
Ofc those statistics for each pair of heroes could be presented by a single matrix.
PS. there is something wrong with heroes win\lose counts anyway - for example am and meepo had 9267 games, meepo won 4635 and am won 4435 (4435+4635 != 9267). My guess is that 9267 includes matches with meepo and am playing on the same side (which is incorrect).