Guild Wars 2

I’ve come out of matches where I gain +2MMR, while someone else on the team has gained +20. Flip this around, I can lose -30 from a match, while others on the same team lose only -5 or -4.

Why is this?

From my understanding, the matchmaker tries to put together equally rated teams. If this is the case, shouldn’t the MMR gains and losses from a match be more uniform across the board for all members of the team?

so there are two components that affect how much MMR you gain or lose from a match.

One is due to your current Rating Deviance, which is the Match Maker’s uncertainty about your “True” MMR, which, while represented by a number, behind the scenes, it is more of a range. The greater your rating deviance the more you lose or gain per match.

Now the SECOND component is where it gets tricky. This unfortunately is just my personal observations from watching other players, reading annecdotes, and my own play experience. This is where they take what your individual MMR is compared to the relative averages of the MMR between the teams, and determine how much you should gain or lose because of this.

To illustrate, let’s say the match maker made these two teams:
Red: 1900 1900 1200 1200 1200 :: avg. 1480
vs.
Blue: 1900 1900 1200 1200 1200 :: avg. 1480
This is arguably an even match, with the average mmr of the two teams being at 1480 each. Let’s assume both teams have low rating deviance from frequent play, and that at the end of the match Red Team wins. If I were to guess, the two red players with 1900 mmr would gain some rating under 10 points, while their three teammates would gain somewhere upwards of 17+ rating. On the other side we have the reverse effect, where Blue’s 1900 point players will lose more rating than their 1200 point compatriots.

While Rating Deviance is fine, one could argue that this second effect, one I would personally name as “rank normalization” (however if there is an official term, I’d like to be corrected), has no place in an environment that is supposed to distribute players to their appropriate skill rating.

Kokoro.8437:

From my understanding, the matchmaker tries to put together equally rated teams. If this is the case, shouldn’t the MMR gains and losses from a match be more uniform across the board for all members of the team?

You can get larger differences in the ratings of teammates because of:

• Group queue when the two players have a large rating difference.
• Low population – being at the extreme (top or bottom) of the ratings or playing at off-hours.
• Uncertain or volatile rating. Glicko2 will adjust ratings by a larger magnitude if it’s uncertain of a player’s rating. As that player plays more often and his or her outcomes are more predictable, that uncertainty goes down. It can go up if the player does not perform to expectation or has been inactive.

Exedore.6320:

You can get larger differences in the ratings of teammates because of:

• Group queue when the two players have a large rating difference.
• Low population – being at the extreme (top or bottom) of the ratings or playing at off-hours.

Volatile rating aside, why should these two affect the the MMR gains/losses of players on the same team by up to 10 times or even more?

Being at the top or bottom end of the ratings means nothing if the matches you are put in also have people at the top/bottom end of ratings on the other team.

And why should two players who have any difference in rating be rewarded differently for playing the same match?

Kokoro.8437:

Exedore.6320:

You can get larger differences in the ratings of teammates because of:

• Group queue when the two players have a large rating difference.
• Low population – being at the extreme (top or bottom) of the ratings or playing at off-hours.

Volatile rating aside, why should these two affect the the MMR gains/losses of players on the same team by up to 10 times or even more?

Being at the top or bottom end of the ratings means nothing if the matches you are put in also have people at the top/bottom end of ratings on the other team.

And why should two players who have any difference in rating be rewarded differently for playing the same match?

Rating gain or loss is based on the player’s rating relative to his or her opponent’s rating. Using a 1500 opponent as an example:
- A 1800 player is expected to win. He or she receives little for a win, but loses a lot in a loss.
- A 1200 player is expected to lose. He or she receives a lot for a win, but little for a loss.
- A 1500 player has a 50% chance to win. He or she will lose or gain the same amount.

Low population comes into play because the matchmaker expands rating range for a match the longer a player is in queue. That means a player 1900+ can end up in game where the average rating is 1800. The 1900 player won’t gain as much for a win as the 1800 players.

With group queue, you can have people of disparate ratings in the same match. If the 1200 and 1800 player from my example were in the same group, their average rating would be the same as the opponents’, but they would have different gains and losses.

Exedore.6320:

Rating gain or loss is based on the player’s rating relative to his or her opponent’s rating. Using a 1500 opponent as an example:
- A 1800 player is expected to win. He or she receives little for a win, but loses a lot in a loss.
- A 1200 player is expected to lose. He or she receives a lot for a win, but little for a loss.
- A 1500 player has a 50% chance to win. He or she will lose or gain the same amount.

Low population comes into play because the matchmaker expands rating range for a match the longer a player is in queue. That means a player 1900+ can end up in game where the average rating is 1800. The 1900 player won’t gain as much for a win as the 1800 players.

With group queue, you can have people of disparate ratings in the same match. If the 1200 and 1800 player from my example were in the same group, their average rating would be the same as the opponents’, but they would have different gains and losses.

That would make sense if, for example the 1,800 rating match was 9 1,800 players and one 1,900 player. (In which case, the team with the 1900 would be rated 1820, not 1800.)

However the reality is that its probably closer to 5 1800 players on Team A.
On team B its 3 1800, 1 1900 and 1700.

Both teams have an “average” of 1800. But the 1900 player in team B is expected to pick up the slack of the 1700 player, and yet still gets rewarded less, and punished more than the other players in the match. While on the other hand, the slacking 1700 player gets rewarded more, and punished less than the other players in the match.

I am of the opinion for my second example, that since both teams have an average of 1800, the reward and punishment for all players on both team A and B should be normalized.

Kokoro.8437:

I am of the opinion for my second example, that since both teams have an average of 1800, the reward and punishment for all players on both team A and B should be normalized.

That doesn’t work well. What you describe is the “Composite Teams” approach in this paper:
http://rhetoricstudios.com/downloads/AbstractingGlicko2ForTeamGames.pdf

Rating should be based on the class and build you play.

Exedore.6320:

Kokoro.8437:

I am of the opinion for my second example, that since both teams have an average of 1800, the reward and punishment for all players on both team A and B should be normalized.

That doesn’t work well. What you describe is the “Composite Teams” approach in this paper:
http://rhetoricstudios.com/downloads/AbstractingGlicko2ForTeamGames.pdf

Thanks for linking that thesis, was an interesting read.

Exedore.6320:

Kokoro.8437:

I am of the opinion for my second example, that since both teams have an average of 1800, the reward and punishment for all players on both team A and B should be normalized.

That doesn’t work well. What you describe is the “Composite Teams” approach in this paper:
http://rhetoricstudios.com/downloads/AbstractingGlicko2ForTeamGames.pdf

The end result in that experiment happened because the matches in that experiment were not made to be “even” in terms of MMR. Instead it was a complete random arena, where the matchmaking did not bother attempting to put together even teams.