I know there are those on these forums who fervently believe (as in have ‘faith’) in the current MMR MatchMaker™ as a good arbiter of a players skill; just as there are others who have serious doubts about the use of MMR and the MatchMaker™ to derive individual skill in a team-based game.
Please hear me, this post is not about that.
One of the explanations of how MMR/SR works in Competitive includes a blurb about how at the top end of the scale they have to reduce the amount of SR award for wins to keep players from reaching 500SR. While the policy behind it is reasonable because their ranking system requires absolute limits (0 - 5000) to do the appropriate calculations, I believe that their is a more fundamental problem that is being glossed over.
Problem: Setting the expectation that you always gain SR when you win a match creates a situation that is at odds with how the ELO system was designed to function.
How can I make that claim? In ranked chess matches it is entirely possible for a highly ranked player to beat a lower ranked player and find themselves with a lower ELO even after a win. Why? Because as a measure of skill, it does not reflect skill for a Grandmaster at chess to beat someone who has only played for a few days.
If Blizzard wants to fix the MMR system, they should adopt a policy where you only gain rank if you beat someone higher than you already are.
I could go on, but this is the essential point I wanted to put out there for discussion.
This won’t make sense in a team game like this here, because you are matched based on average SR. Thus, one player might have a significantly higher SR, but the average will bring it down by matching another player with significantly lower SR (i.e. Plat and Silver players in a team).
Jeff did mention a long time ago that zero and opposite-direction MMR movements were possible (i.e. gaining MMR on a loss and losing MMR on a win). This is one of (and one of the worst, IMO) reasons for differences other than scale between MMR and SR – SR always goes up on a win and always goes down on a loss, by some minimum amount, regardless of if any underlying MMR change happens.
I believe however that that was back when the older stronger performance-based modifiers were still in place… the current performance based adjustments are a fraction of that, and don’t apply above Platinum. Also, of course, it’s not possible for a GM to play against a Bronze (low Elo) or even a Gold (starting Elo). They do (rarely, in the dead of night) match against Diamond, or so I’ve heard which is kind of a ridiculous difference but not GM vs. novice ridiculous.
In any case, in chess and in other Elo-like systems you don’t have negative or zero movement merely for outranking your opponent. You don’t get to zero movement until you have a dramatic difference in rank (and get the expected outcome, of course; an upset would involve dramatic movement.)
The claim that chess Elo has negative movement on wins is also a little questionable to me… looking at the wikipedia it’s not self-evident how that would happen, and “omnicalculator” shows that e.g. Peak Gary Kasparov v. Novice (2851 v 700) results in 0 movement when Gary wins, not negative movement. Which intuitively makes sense. Winning against a lesser opponent doesn’t prove anything either way, while losing or tying to them would be notable and would hurt hypothetical Gary’s ranking.
In short, I believe that the system as it exists works extremely similarly to chess Elo in this respect. (It presumably also works differently in some respects, since it is stored in values alleged to represent standard deviations).