It is crystal-clear from how the system behaves that it does per-matchup MMR and that the MMR that is assigned to and displayed for your account is an average of your matchup mmrs, so why not display your mmr for each matchup?
Just the other night I played a zerg vs protoss in which my opponent was 300 MMR below me on the ladder but he did a proxy Shield battery rush and I demolished him and he got -50 to his mmr. For those of you who don’t know, I have beaten partings proxy Shield battery Rush, the same one that he successfully used in code s. So, when some NA protoss 300 MMR lower than me tries to do it I practically slept through the game. Anyway, when someone is 300 MMR lower than you and they lose to you they should get something like -2 or maybe -5, but -50? The only possible explanation here is that the system has a separate MMR for his protoss vs zerg and that his average MMR is much lower than his protoss vs Zerg, presumably because he’s boosted by proxy Shield battery rushes. That could explain the situation.
Regardless, it would be nice to be able to see your per matchup MMR rating. I think it’s a fantastic idea which brings up the question of why they haven’t done it. I suspect they probably discounted that idea due to the fact that Starcraft 2 as a rock paper scissors design and balance, and it would become abundantly clear if they were to reveal per match up MMR.
That’s very interesting and that is the most plausible reason of why somebody with 300 less MMR than you would drop that much (i guess normally he should have lost 5 points).
Now if they have calculated the Per-MU MMR of each player, why they do not take the further step to match players based on this much more precise measurement (Per-MU MMR)? That would eliminate the imbalances of having the General winrate at 50% (+/- 2-3) and unbalanced MU’s of the type:
ZvP 70%
ZvT 30%
ZvZ 52% ?
Much of the whine is caused by this MU-imbalance, i guess if the majority had a near 50% winrate across all MU then the frustration and whine would decrease.
The points that are rewarded to you after the game are scaled relative to whether or not you were favored to win. If you are favored to win and then you win, then it gives you a small increase in points. If you are favored to win and you lose, then it gives you a large decrease in points. The only way you can get a huge decrease in points is if you are hugely favored to win and then you lose. This cannot be the case however since the player was 300 MMR lower than I am, unless the system uses per match up MMR and his PvZ was actually enormously higher than mine. His average mmr was lower than my average because his PvT and PvP are abysmal compared to his PvZ, which would make sense since he does proxy shield battery rushes.
I have published stats, about the top 250 players, on numerous occasions that proved beyond doubt that protoss is very good against Terran. However, each time that I have done that I have also pointed out that Terran is twice as good vs Zerg than protoss is vs Terran. If not for PVP being a fustercluck, protoss would have the highest average ELO across all matchups. Tvt is much more stable than PvP and Terran is really freaking good against Zerg thus Terran has the highest average ELO in the top 250 players.
That’s an excellent observation. Sometimes players do have non 50/50 winrates for certain matchups and not for others and one way you could explain that is by saying that the system uses a single MMR ranking for all matchups and that being good at one match up and having a higher win rate in that matchup forces you to play against stronger players in other matchups that’s giving you a below 50% win rate in those matchups. However, that doesn’t explain my observation here in this thread. The system having ratings per matchup can explain both scenarios. If players have a non 50/50 winrate for a matchup then it just means their ranking is shifting for that matchup.
Maybe he was a M/GM that was climbing the ladder and he was under provisional MMR (where the points lost/found) are exagerated? Still this my hypothesis is weak because that higher-up that protoss should have been well beyond the number of games for having the Provisional MMR still active.
I think per-MU MMR is interesting … perhaps you could also include total number of games and performance in the last 3 month or some other key figures in MMR. Of course somebody who is over 80 times master perhaps is better than a one time lucky master. There is another forum (matchmaking should be related to stats partly) if you are interested in alternative ideas for MMR.