Guild Wars 2

What I mean by this is, “does the MMR system in place work to bring players to a common value” based on rating differentials given and lost.

In general I’ve been fairly satisfied with my games and the season over all, but recently I had been duo queuing with a friend who was really unlucky in their placement games and so I decided to help them out. We are on quite a favorable win ratio, but I’m down about 100 rating because every Loss is around -20, and every win is about + 4-7. Obviously his volatility is higher right now so I can’t use his numbers (last game +55), but I’m wondering how much our team mates are getting in these scenarios. There have been many situations where I witness people doing things where I have to think to myself “omg that rotation is so bad, you really do not deserve this win” and the funny thing is they’ll probably get more rating than I will from the match, meaning I’ll probably see them again, hopefully on the other team.

I think some normalization is good otherwise you’d have people that shoot up to the top or bottom extremes, but I’m just curious if there are potential issues that could be caused by this, or if over enough time matchmaking/RNG should balance out the normalization issues from individual matches.

Anyways, tldr; is there a rating normalization problem in the game that can cause misplacements of people’s rank after their placement games?

To me it seems like you are playing against lower tier players. Let’s say yo are gold and he is bronze. You really should get rewarded the same for beating people that are below your level as he does for bearing people above his. Makes me think it is averaging the two ratings, or it is taking his vs yours.

Duo-queue puts extra stress on the MMR system because it forces a mismatched team. That definitely hurts things when it comes to finding the right balance.

What you are describing is that you are normally winning when expected to win (thus, small gains) and unexpectedly losing when expected to win (thus, relatively large drops), moving your own MMR down.

The teammates are getting whatever size adjustment their projected win/loss vs actual win/loss, and their volatility, predict. Probably slightly off what they would get without the duo queue, because that adds extra noise to the system, but close enough not to matter.

Anyway, individual games are not that significant overall. If someone is placed “wrong”, from whatever cause, they will fairly rapidly converge on the correct MMR in the absence of other problems.

It’s around 89 games to hit “spot on” in an ideal world with 5v5 matches, but the majority of movement happens in the first part of that — so your buddy could just play, like, 10 or 20 solo matches and get pretty close to their real MMR, and likewise, those players who you may have artificially lifted, they will get to the right place too.

PS: don’t forget they have just as good a chance of being queued against you, and getting an unexpected loss when they might have been predicted to win.

I Duo Q’d with someone who was 100 MMR lower than me. We compared changes to Rating after each match. When we lost I lost 2 pts more. When we won, I won less then he did. I did not think this would be the case, I thought the adjustment would be based on the teams average MMR. So even though I might have been more volatitle than my partner I’d then expect that if I lost more I would gain more. The outcome of the rating was puzzling because it would seem to indicate that my individual Rating was being factored in to compare to the other teams rating, not the average rating of the team I was on.

To exactly answer your question you can look up my thread on the rise and fall of a PvP player. I have not settled anywhere. I go from lower gold to platinum and back over and over again. I reached as high as rank 181 in early January and as low as lower teir gold. There is no settling, which implies obvious things about the systems ability to predict the actual outcome of a match based on supposedly being able to rate indivdual players.

So as I mentioned, I don’t have too much of an issue with what is happening for my personal rating, I’m just curious what people think about how a team based game should handle rating gains/losses.

SlippyCheeze you mentioned that my rating losses are large/gains are small for games that I’m expected to win, but am I expected to win more so than my other teammates? If we exclude volatility calculations, shouldn’t the chance of winning be the same across the board between you and your team? An individual can’t be more likely to win on the same team, no?

Let me pose a question, in a hypothetical match, we have two teams:
R: 2100 2100 1800 1800 1800
vs.
B: 2100 2100 1800 1800 1800

Very even match, let’s say Red Wins. What should the rank gains/losses be, ideally?
In my games where I am soloqing and playing regularly , I probably get around 11-15 rating gain/loss. So let’s say that is the normal amount of rank gain/loss if you have low rating deviation.

Should players Red 1-5 all gain between 11-15 rating?
I am not sure to be honest. And I can’t say for 100% certainty what is expected in the current system, but from observation of streams, it SEEMS like players 1-2 will gain some 2-7 rating, and I do not know what players 3-5 get. My personal opinion though, is that it is actually okay for rank gain/loss normalization is fine at this extreme end of the spectrum, otherwise, we’d eventually have large gaps in the ranking distribution.

If we were to apply this to a lower scale:
R: 1200 1200 900 900 900
vs.
B: 1200 1200 900 900 900

Same situation, Red wins, what would be considered the FAIR rank distribution?

Secondary question, if there IS rank normalization going on is it an issue? Or are these individual games balanced out by expected win/loss ratios for players that might gain or lose more than they should?

Edit: SlippyCheeze.5483, you did address some of the things that I have asked, just thought I’d hear a bit more from you and the others on the forum with more concrete examples. And yeah, duo-queue does introduce extra noise, but I think from a human/player enjoyability perspective, it is a net gain over solo-queue only.

Slightly off topic but just looking at some other numbers for discussion

Division/Tier Minimums
(If someone could help me fill out the rest of the values from ingame that’d be great)

Legend: T1: 1950 T2: 2150 T3: 2250
Platinum: T1: 1650 T2: 1750 T3: 1850
Gold: T1: 1350 T2: 1450 T3: 1550
Silver: T1: 1050 T2: 1150 T3: 1250
Bronze: T1: 0 T2: 850 T3: 950

From GW2 Efficiency:
Everybody
Minimum 257
Maximum 2,184
Mean 1,231
Median 1,232
0.5% of players 1,899
1% of players 1,848
10% of players 1,608
20% of players 1,479
50% of players 1,232
75% of players 1,030
90% of players 857
99% of players 594

What do we see here?
Of the people who have hooked in the gw2 API to their gw2 efficiency account:
A little below 10% of the players are in platinum
~50% of players are Silver (T2?) or below.
~25% of players are in Bronze.
Not even 1% of players are in legendary.

What does the rank distribution look like approximately from this data? From this data, does it seem like there is any normalization effect pulling most players of differing skill levels towards, with a few outliers on both extremes?

alemfi.5107:

Btw, gw2 efficiency only use the data of the people registered there?

Lucius.2140:

alemfi.5107:

Btw, gw2 efficiency only use the data of the people registered there?

Yes, it is not the complete database of players. Only anet has that.

alemfi.5107:

Lucius.2140:

alemfi.5107:

Btw, gw2 efficiency only use the data of the people registered there?

Yes, it is not the complete database of players. Only anet has that.

gw2efficiency is pulling data from 24,550 accounts for PvP League Rating statistics; don’t you think that’s a large enough sample size to deduce trends like OP is attempting?

Kreweless.2196:

alemfi.5107:

Lucius.2140:

alemfi.5107:

Btw, gw2 efficiency only use the data of the people registered there?

Yes, it is not the complete database of players. Only anet has that.

gw2efficiency is pulling data from 24,550 accounts for PvP League Rating statistics; don’t you think that’s a large enough sample size to deduce trends like OP is attempting?

The sample needs to be representative of the population too.

True at all, but i have one question:

My rating is 853 and i start to play with ma friend who has rating 950. U write here that ppl with high rating get less points to win a more to lose yes correct. But what is wrong that we win and I with lower lvl get +15 a him get +36. Other starnge thing when i start was bronze under 850 a got for loose more than for win its quite demotivated u loose u get -20 once i get -44 without decay a if i win +14-16.

So 1st question why better rated ppl if we played together get better rate points ?

2st Why loose its so hard that i must win 2 or 3 games to get rate up only at bronze…?

alemfi.5107:

So as I mentioned, I don’t have too much of an issue with what is happening for my personal rating, I’m just curious what people think about how a team based game should handle rating gains/losses.

SlippyCheeze you mentioned that my rating losses are large/gains are small for games that I’m expected to win, but am I expected to win more so than my other teammates? If we exclude volatility calculations, shouldn’t the chance of winning be the same across the board between you and your team? An individual can’t be more likely to win on the same team, no?

You can see the details of this over in the wiki and at wikipedia for the math behind it

Short answer is, yes, you are correct there — the prediction is for the team. Volatility doesn’t change the “should you win” part, it changes the “how far your MMR moves” part after the match.

alemfi.5107:

Very even match, let’s say Red Wins. What should the rank gains/losses be, ideally?
In my games where I am soloqing and playing regularly , I probably get around 11-15 rating gain/loss. So let’s say that is the normal amount of rank gain/loss if you have low rating deviation.

Should players Red 1-5 all gain between 11-15 rating?
I am not sure to be honest. And I can’t say for 100% certainty what is expected in the current system, but from observation of streams, it SEEMS like players 1-2 will gain some 2-7 rating, and I do not know what players 3-5 get. My personal opinion though, is that it is actually okay for rank gain/loss normalization is fine at this extreme end of the spectrum, otherwise, we’d eventually have large gaps in the ranking distribution.

Every player changes MMR based on their personal values, which include some profession-weighting, volatility, how many 3 day periods since they last played, how frequently they win or lose compared to prediction over many matches, etc.

Your MMR is personal, and changes to it are also personal. If you look at one single match, and only one single match, this can look very strange. If you look at it over 89 matches (picking as an example the average number for gliko-2 to converge in 5v5 with no changes in skill on either side) it averages out to the right thing.

alemfi.5107:

If we were to apply this to a lower scale:
R: 1200 1200 900 900 900 vs. B: 1200 1200 900 900 900

Same situation, Red wins, what would be considered the FAIR rank distribution?

Well, assuming nothing else, yes, this is far — both sides should have roughly equal skill levels, and thus roughly equal chances of winning.

alemfi.5107:

Edit: SlippyCheeze.5483, you did address some of the things that I have asked, just thought I’d hear a bit more from you and the others on the forum with more concrete examples. And yeah, duo-queue does introduce extra noise, but I think from a human/player enjoyability perspective, it is a net gain over solo-queue only.

TBH, I think people are way too focused on MMR, and certain aspects of it (like the placement matches) at the expense of fun. I’d encourage you, and everyone else, to play and enjoy things without worrying about the details of MMR, etc, so much.

I’m not saying that MMR is perfect, or that there are no bugs in the GW2 implementation, but rather, that over the course of your play your MMR will approximate your actual skill reasonably well — at a statistically significant number of games.

Don’t sweat the small stuff, like what happened after one match. Worry about where you are after the next 100.

Faux Play.6104:

Kreweless.2196:

alemfi.5107:

Lucius.2140:

alemfi.5107:

Btw, gw2 efficiency only use the data of the people registered there?

Yes, it is not the complete database of players. Only anet has that.

gw2efficiency is pulling data from 24,550 accounts for PvP League Rating statistics; don’t you think that’s a large enough sample size to deduce trends like OP is attempting?

The sample needs to be representative of the population too.

Was more curious than other thing. But yeah, theres probably a small bias because the people in gw2 efficiency may be more hardcore or pve hardcore, because they check comparatives, it was used for checking gear on raids, etc.

That a side, the statistics there do have some kind of normal distribution, perhaps a little skewed.

On another point it does show how much the pvp community have grown, if an incomplete data base have that many players when in December of 2015, 10k was the historical top (according to Anet).