2s Ratings Deflated (and proposed fix)

[dak1]

Cross post [was] at: us.battle.net/wow/en/forum/topic/11914282146
THE THREAD WAS JUST DELETED BY BLIZZARD. NO IDEA WHY.

Why are ratings in the 2s bracket this season so deflated?

References:
us.battle.net/api/wow/leaderboard/rankings/2v2
us.battle.net/api/wow/leaderboard/rankings/3v3

• There's substantially more people playing 2s:
  • 92,138 at rating 1000+ out of 242,013 in 2s
  • 31,920 at rating 1000+ out of 92,952 in 3s

• The top 3s rating is significantly higher
  • Top rating in 2s is 2135
  • Top rating in 3s is 2614

• 2s have lower average and median than 3s:
  • A rating of 1750 in 2s is rank 302 (top 0.3% over 1000)
  • A rating of 1750 in 3s is rank 5054 (top 15.8% over 1000)
  • There are 1231.8 points per player over 1000 in 2s
  • There are 1446.9 points per player over 1000 in 3s

My rating of a measly 1567 (I was 1800 in 2s last season) puts me at rank 2693 in 2s (top 2.9% of those over 1000, or the equivalent of Duelist). The equivalent rank in 3s (top 2.9%) would have a rank of 926 and a rating of 2043.

With so many people playing in 2s, mathematically there should be more points being fed into the higher rankings (the same phenomenon is seen in other ratings like ELO for Chess, and should be even more pronounced if wins give more points than losses). Yet for whatever reason, this clearly isn't the case.

This especially sucks for me, since I don’t have a consistent 3s or RBG team to play with (I know, oqueue, but playing with randoms from oqueue can be pretty painful). My cap for this week is 1905 (from a 1576 rating), compared to a cap of 2693 for the equivalent rank in 3s. Falling behind nearly 700 points a week makes a pretty significant difference pretty quickly (and it’s even more extreme with RBGs).

[Anzen]

The season started 1 week ago.

People only play 2's for cap so most likely have played a grand total of 10 wins in 2's this season.

People only really play 3's for rating so most likely have played to get rating.

And you wonder why rating is deflated at the moment.

Wait one month.

[dak1]

People only play 2's for cap so most likely have played a grand total of 10 wins in 2's this season.

Looking at :
us.battle.net/api/wow/leaderboard/3v3
us.battle.net/api/wow/leaderboard/2v2

Some quick calculations shows the following:

In 3s, there has been 52455 wins and 34015 losses for a total of 86470 games played by the top 1007 teams.
In 2s, there has been 38468 wins and 17797 losses for a total of 56265 games played by the top 1007 teams.

Also, about twice as many people have won more than 10 games in 2s than in 3s (51,261 in 2s vs 25,640 in 3s with a rating of 1200 or higher).

So while at the higher rankings more games have been played, the difference isn't that large (about 1.5 games played in 3s by the top teams for every 1 game played in 2s), and there is a much larger pool of players in 2s that likely makes up for that difference.

The number of games played may explain some of the ratings gap, but it doesn't seem to be able to explain all (most?) of it.

And even later in the season, 2s ratings still seem depressed compared to 3s (last season there was still a noticeable gap all the way to the last week).

- - - Updated - - -

Some more facts as I look into this more:

In 2s, out of players with a rating over 1000, there are 120,740,839 rating points divided between 98,016 players, for an average of 1231.8 points per player. There's 212,203,352 total rating points in 2s, which divided by 246,042 players comes out to about 862.5 points per player in 2s.

In 3s, out of players with a rating over 1000, there are 48,360,326 rating points divided between 33,423 players, for an average of 1446.9 points per player. There's 77,915,997 total rating points in 3s, which divided by 95,361 players comes out to about 817.1 points per player in 3s.

That means as a whole, 2s players actually play more games (more points per player in aggregate) even though there are more points being fed into 3s than into 2s (since 3s has a 215.1 higher average rating for players above a rating of 1000).

There's more games being played in 2s, more people in 2s, and yet lower average point totals for 2s players than for 3s players. These facts are completely at odds with one another. More players and more games necessitates that 2s would have more average points per player above certain thresholds unless there was something relatively deflating the system in 2s compared to 3s (or vice versa).

So basically, it looks like Blizzard is either intentionally deflating 2s ratings compared to 3s, or there's a huge bug that's only affecting 2s.

[PPN]

Without doing any maths i would just say:

Ppl Play 2s for cap and cause 2s suck more then ever These last 2 seasons no1 gives a fuck to try get high rating. For many classes ist anyway strictly not possible to get high rating cause of 2-3 faceroll lol comps who will beat the shit out of you anyway...So why even try.

[Terahertz]

2s is always deflated early on in the season. It happens and there's nothing you can really do about it.

[dak1]

2s is always deflated early on in the season. It happens and there's nothing you can really do about it.

Well, the deflation (or inflation) of the average rating across the entire bracket is due to the design of the system itself (there's not much players can really do to change that).

The ratings system is a zero-sum game (or effectively so close to it that it can be treated as such), minus some predictable inflation that was added recently. That means you can shuffle points around between teams, but the average rating across all players will remain the same (or about the same).

I guess my hope is to raise awareness. If the community it aware, is can decide if this is a good or bad thing, and if people don't like it (remember, there's over 200,000 people who play 2s, almost 3 times as many in any other bracket), then hopefully they can put pressure on Blizzard to change it.

I'm writing a quick little script to calculate the mean rating of all players (and players over a couple specific thresholds) in every bracket and will post the results in a few weeks. If I'm correct and this is effectively a closed system (which I think experience seems to suggest), then we'll see the mean rating in 2s and 3s stay about the same, except for the slight increase due to inflation. That should hopefully be sufficient evidence to convince most people that this effect is real, although I'm sure there will still be people that aren't very good at math who will still think otherwise.

[Deleted]

Dont know, but maybe it could be worth reposting on ArenaJunkies? not as active as official forums but still more than this.

+1 like your math