[D] Why divisions aren't comparable - Page 3

On August 24 2010 06:42 RivalryRedux wrote:


It seems like the game not only looks at your opponents MMR compared to your points but also factors in how strong your opponent is relative to your entire division (MMR wise) before giving/taking away points. So the stronger your division, the harder you would to work for your points and vise versa. It applies to all divisions and not just Medic Mu, I just bring that one up a lot because it's fairly certain that it's a much stronger than average division. Blizzards Rankings have backed this up by being much more favorable towards players in that division compared to how their point totals stack up on a site like SC2ranks.

It's also an alternative way of thinking about why many players here get a ton of points while they're in platinum. It's because they're diamond bound and they're playing people who CRUSH the MMR of the other players in whatever platinum division they're in.

On August 24 2010 08:52 theSAiNT wrote:
This WOULD make sense if distributions of points were similar across divisions. However, they don't look like they are. For example, my division has a points range of 1000 to 200 while a friend's has a top end of 800. There are divisions where the top end is much higher and I'm sure some where the top end is lower.

Also, most matches are played with players OUTSIDE of your division anyway. So why should it be easier or harder to earn points based on your division?

I'm not convinced.

On August 24 2010 06:31 nickwtf wrote:
I think its obvious that rank (1-100) between divisions is mostly irrelevant, but I don't see how any of this proves that points aren't directly comparable between divisions. How would the division you are in have any effect on how many points you gain/lose in a game?

On August 24 2010 09:56 Vokasak wrote:
Given enough time, the divisions would theoretically all even out, more or less, give or take. Give the system some time to work it's magic. There are people out there who are still just now buying the game and doing placement matches still.

On August 24 2010 10:36 OTIX wrote:
The only example in the OP that doesn't seem to fit the standard MMR model is the guy in Talematros Eta, there is no way that he only gained 7 points from you when the others all gained 12-14. However if he did not have any bonus points when he met you then all the numbers fit quite well.

On August 24 2010 12:03 l90 Proof wrote:
This data is consistent with a post I read from someone who was informally aproached about helping to design Blizz's ladder system initially. Too lazy to look it up, but tldr version is:

1) Your points/ELO(MMR) ranking are separate. Points will inflate if you play more (use more bonus points), ELO reflects only win/loss vs strength of opposition & ELO is invisible.
2) ELO roughly determines your league (maybe even division?) but moving up/down in leagues is limited by your Sigma (variance)
3) Therefore it si possible that a recently placed or rapidly improving plat player would have a higher ELO score than a high points diamond player, but because of his correspondingly high sigma he hasn't moved up in leagues yet (thus you see him now in Diamond, he played enough games that the system became more "sure" he was proper to place in diamond)

Takeaways:
A 400 diamond player is not necessarily better than a 200 diamond player, though both are better than all plat players that have lower Sigma scores (a few outlier plats haven't been moved up yet and might be better than our diamond player). The only way to see how you are doing (in ELO/MMR sense) is to be matched directly against the player

On August 24 2010 11:16 RivalryRedux wrote:


Both you and Teddyman have brought up the dude from Talematros getting +7 and I'm not really sure where you're getting that from (if I said that somewhere it was a mistake). I only remember posting that the guy from Medic Mu received only +7 for beating me which was the most inconsistent with other players.

If you ignore the points gained by my opponents and just look at the points I'm losing it's not consistent with the idea that MMR will converge to points equally across all divisions unless there has been some crazy change in a players play.

Loss (-6) vs 648d(+0) (110-87 66th in Medic Mu)
Loss (-12) vs 737d(+2) (112-90, 8th in Duke Xi)

After 200 games you would think that they would have converged to MMR pretty accurately and yet I'm losing HALF the points to the guy that's -100 of the other. I looked at other numbers from players that I would consider to be in a weaker division and it seemed consistent that they would gain more points for similar wins than someone in Medic Mu of the same point rating would gain. Later I might go through and try to find some more examples if the numbers I'm showing aren't convincing enough.

On August 24 2010 09:33 RivalryRedux wrote:


Distributions don't have to be similar because it's all relative to your own division. A high top end is just an indication that the top player/players in that division are far apart from the low player/players in their own division. If you put 99 bronze players in a diamond division with me I would expect to have the highest top end of points in the world similarly to if you put me in bronze division with the 99 bronze players.

The system is designed for local competition and making points translate directly across all divisions would contradict that. Ideally you could say that Blizz should have you playing people from your own division more often but playing only people from your division would lead to more one sided matches than otherwise.

On August 24 2010 06:14 SichuanPanda wrote:
It's just like WC3 Blizzard has a hidden system that we will never ever see and it makes little to no sense. They then explain the system to us in terms that aren't entirely accurate.

On August 24 2010 07:37 FearMe.UK wrote:
The divisions system was a horrible idea. They should resort to an actual ladder like in WC3.

On August 24 2010 09:17 Kpyolysis32 wrote:
A little tale for you all that hints at a couple of interesting ladder mechanics: When the game first came out, I decided to Nuke rush all of my placement matches, and I won 4 out of 5 and ended up in mid Gold (I am- and was- a Diamond level player). I went on a gigantic winning streak because of my low placement and ended up with an insane MMR, I actually got paired up against people who were around 450 ELO Diamond (even if their MMR was lower than their ELO, it still meant that they had an absurd MMR compared to mine, and remember, 450 ELO was a lot higher when the game first came out because of point inflation). However, it didn't promote me until I lost a game, after which it immediately promoted me to something insane up in Platinum (around 900, IIRC, which was ridiculously high when the game first came out). This demonstrated a couple of theories I've heard, which are 1- that you won't promote until your W/L ratio over any stretch of time is 50/50 (I've also heard this as your MMR stabilizes, but I've never heard any details on what counts as 'stable' ) and 2- that you can't promote multiple divisions at a time.

It's very possible that the Plat guy was actually a very good player, but was on a large winning streak and didn't promote because of it (as predicted by theory 1 above), so his MMR was far higher than his ELO would indicate.

I'm now considering getting together a bunch of information, then asking people on the xkcd math forums if they can help figure out exactly how this ladder works. Hm..

On August 21 2010 12:32 RivalryRedux wrote:
Blizzard's top 200 seems to have been a conformation of this. I don't have the old ladder points from the day of their top 200 but I remember hearing that were some significant changes such as with Dayvie (who dropped from near 1st to 49th IIRC). There is almost NO WAY that it unused bonus points bringing Dayvie up on the point comparison when almost all of the players above him were active enough to use all of their bonus points. What I would imagine blizzard used when they were ranking players was their hidden Match Making Rating.