(seemingly) illogical ratings changes

I've got a rating of around 800-900 with just over 2000 games, played over a period of around 10 years. I play mostly 3v3 / 4v4

I've started noticing that when I WIN a game teamed up with players who have fewer games (Still non-grey) and a higher rating than me their rating increases MORE than mine does.
Similarly if I lose a game teamed up with players with a higher rating than me then my rating decreases more than theirs does.

These two together seem to not make sense to me. I get that a high number of games would in theory mean my rating is more accurate and therefore shouldn't be influenced as much by a single game, but then by the same token my rating shouldn't decrease as much when I lose, especially when my team were ranked more highly than me.

It's almost as though number of games is being taken into account as a sort of hidden addition to the rating.

The trouble is imagine I always play completely balanced games and win 50% my rating is going to go down over time.

I don't know whether this is relevant but about 1/3 of my games are played with the same two friends who are probably worse than me but have higher ratings, and unlike me they never play with anyone else.

I can share examples if anyone is actually interested.

Chris

Your shown rating has no impact in trueskill calculations which is what determines rating gain and loss.

Your rating is a combination of two values, mu and sigma. Mu is the top of your individual rating bell curve and therefore your most likely level of play. Sigma is how much you deviate from this level of play. Your shown rating is mu-3*sigma, or the level the system is 99.7% confident you can at least play at.

And so as you might expect, your sigma adjusts depending on the quantity and quality of data the system has to go off of. Play many 1v1s, the system is more confident of your general level and so you gain/lose less rating. Play an 8v8 once a month and the system adjusts you quite extremely relative to the 1v1 player.

So yes, the system is taking game count into account indirectly as it impacts sigma. This can also impact the confidence the system has in the rating of your friends if they only play with you.

If you play and lose 50/50 against players at your level of mu and sigma, your rating would hardly move up or down at all.

This is all said looking at a 1v1 scenario, but teamgames operate in the same way. Mu is simply added up for a “sum of team competency” and then point spread once game result is known is determined via individual sigma. It’s slightly more complicated than that but it’s a good enough explanation.

Does the rating system take account of how often you are playing when determining variation (sigma) I. E looking at frequency of games rather than just total? I would expect greater variation if you play irregularly (as you derust).

No it doesn't.

couple of FAQs that you might like :
https://forum.faforever.com/topic/274/how-can-i-increase-my-faf-rating
https://forum.faforever.com/topic/273/what-are-my-faf-ratings

@FtXCommando said in (seemingly) illogical ratings changes:

Mu is the top of your individual rating bell

Mu is the middle of your bell curve

@Reckless_Charger said in (seemingly) illogical ratings changes:

Does the rating system take account of how often you are playing when determining variation (sigma) I. E looking at frequency of games rather than just total? I would expect greater variation if you play irregularly (as you derust).

Technically if you were a good player but are de-rusting, as you win and lose games you are likely to get a larger sigma as your range of level of play will be increased as you get lucky and kill higher level opponents and make stupid mistakes against weaker opponents.
The system doesn't directly look at your frequency of games, but I expect there is some correlation between sigma and how long it's been since a player last played enough for there rating to hone in correctly.