Datamined Game Win/Loss Statistics

On July 16 2010 04:11 DeckTech wrote:
Nice!
Any idea how many players there are in the diamond league in total?

On July 16 2010 04:00 jamesr12 wrote:
Sweet work, do you think you would be able to make a server ladder? What about Mu data that is the most interesting to me.

Fantastic post! Is it possible to get some stats on the average rating?

Also, what is the difference between

Total average player win %
and Total player win %

Any idea how many players there are in the diamond league in total?

On July 16 2010 04:16 Chriamon wrote:
I think some of this data is going to be misinterpreted. Having similar overall win % doesn't mean much. You should only count games that are vs other diamond players, and go by specific matchups as well.

On July 16 2010 04:58 EccoEcco wrote:


While I'll agree that this isn't end all be all data (Certainly, the sample size is still small), actual data from aggregate results is a lot more significant than the anecdotal evidence that's been dominating the discussion thus far.

A little induction would suggest it means each race has a success rate against the other two races that averages out equally.

(TvP + TvZ)/2 = (PvT + PvZ)/2 = (ZvT + ZvP)/2 with a deviation of <2% at both diamond level and general play.

That means that if Terran were truly overpowered against Zerg and dominating that match-up, then Terran would have to be equally underpowered against Protoss to generate such a close winning percentage to the other 2 races. Until that result is ruled out, its a possibility and Rock>Paper>Scissor is not the intended balance, but I for one, would be surprised if that were the case. Particularly because of how consistent the numbers are at both the general and top level.

On July 16 2010 04:36 3clipse wrote:
Nice stuff. Confirms the change I thought occurred- Diamond is now top 10% rather than 20%.

On July 16 2010 05:11 ashaman771 wrote:
What site is this information coming from??

On July 16 2010 04:20 Logo wrote:
Any chance you could release the program and/or run the same on US and Asia?

Even better if you make it an open source program on something like github or sourceforge so more people can pitch in on it.

what website is this data from?

On July 16 2010 05:22 Excalibur_Z wrote:
Silu, thanks for the data. Could you parse through the leagues to determine the number of players in each league? I'd like to update my ladder analysis post with confirmed results beyond the diamond level.

On July 16 2010 05:38 Silu wrote:


Hmm, you mean just the amount of players in each of Bronze/Silver/Gold/Plat/Dia? From the EU data I gathered already, this is the distribution:

Bronze: 1424
Silver: 1579
Gold: 1795
Platinum: 584
Diamond: 600


Probably very skewed from the very low amount of games played and the vast majority of people going in Bronze/Silver/Gold from placement matches.

On July 16 2010 04:14 jamesr12 wrote:


350 just add them up

Other interesting things:

350/3151 = 11% so diamond leauge is around 11% of all players

19933 total games played in diamond leauge, that is 28.475 games per player in diamond.
note: 2 players in each game

On July 16 2010 05:38 Silu wrote:


Hmm, you mean just the amount of players in each of Bronze/Silver/Gold/Plat/Dia? From the EU data I gathered already, this is the distribution:

Bronze: 1424
Silver: 1579
Gold: 1795
Platinum: 584
Diamond: 600


Probably very skewed from the very low amount of games played and the vast majority of people going in Bronze/Silver/Gold from placement matches.

+ Show Spoiler +

There's more to it than that because it assumes that all races are rated equally. The matchmaking system works to ensure 50% win ratios. If we want to use this kind of information it has to be based on things like average player ratings by race or some aggregate stat like average player rating/# of games

On July 16 2010 04:14 jamesr12 wrote:


350 just add them up

Other interesting things:

350/3151 = 11% so diamond leauge is around 11% of all players

19933 total games played in diamond leauge, that is 28.475 games per player in diamond.
note: 2 players in each game

On July 16 2010 06:34 Gescom wrote:
Scary how balanced these numbers look.

On July 16 2010 05:16 Logo wrote:


There's more to it than that because it assumes that all races are rated equally. The matchmaking system works to ensure 50% win ratios. If we want to use this kind of information it has to be based on things like average player ratings by race or some aggregate stat like average player rating/# of games

On July 16 2010 07:12 Hinni wrote:


This.

You get ranked by the match making system, so that you will have around 50% wins (for obvious reasons higher for top players). So the results just show that the match making system works. If one race would be op, all players of this race would be higher ranked so that they will fight stronger enemies and should still have 50% win rate.
It's very difficult to get 'correct' results. There are too many unknowns I think. You could assume that the average (diamond) player of each race is equally strong (which blizzard is doing I think if I interpret their interviews correctly but obviously this assumption is not very strong).

I think its interesting that around 10.5% of all toss players are in diamond league, around 9.5% of all terran players and 11.5% of all zerg players are in diamond league (6.5% for random). So this could either mean that it's easier for zerg players to get into diamond league or that the average zerg player is stronger then the average player of the other races.

On July 16 2010 07:27 Silu wrote:


Good point. I wouldn't read that much into the win percentages either from a race vs race perspective; nevertheless it still tells something.

In any case, I added the US stats. The league distribution is clearly more sensible, probably because US has been up for a couple of days longer. The race and race&league distribution looks to be quite similar.

On July 16 2010 07:34 Excalibur_Z wrote:


So based on the initial US numbers, we have a distribution of approximately 9/18/23/23/27

A bit closer to my 10/15/20/25/30 estimate. Let's see how this evolves in the coming days =)

On July 16 2010 07:45 Liquid`Jinro wrote:

I don't know this for a fact, but I've heard from several people in gold that then got immediately promoted to diamond - I wonder if this could have something to do with the European equality between plat/diamond....

(I'm still stuck in gold at 20-2 or 22-2 or whatever, while someone else got promoted to diamond at 17-11 or something like that - I really do not get the system, I guess there's a minimum game requirement?).

On July 16 2010 07:48 CauthonLuck wrote:
There seems to be a minimum loss requirement, as little sense as that makes. I haven't seen anyone get promoted to diamond with any less than 8 losses, regardless of having 90% win rates through 30+ games.

On July 16 2010 07:53 EliteAzn wrote:
There is data missing from this...data

Anyways...using good 'ol Excel...I managed to figure out some things....

From this data (US DATA), you have a total of 70959 wins, 68312 losses. There are 2647 losses missing. This makes no sense because there's a winner and a loser in every game. So if we assume the total avg. win percent of everyone INCLUDING people still playing placement matches is 50% and we keep the win % of the leagues, the win percentage of these people (placement match people) has to be 47%....may makes sense, may not.

*please put this in OP, assuming numbers my numbers are correct...*

On July 16 2010 09:10 vica wrote:


Ties. Sometimes, in games where my opponent quits early, like 2 minutes in, before any action happened, I check my match history and it doesn't update as a win or lose until a day later.

But that number seems a bit too high. There are several possibilities.

1. Wins update faster than losses.
2. Data is missing from servers
3. Disconnects count as win and tie, but appears as a -0 loss
4. Placement matches do not count towards ladder ranking, so if you are in placement and you've played against someone in ladder, your loss is not recorded, but his win is. Or something of the sort. Seems the most plausible.

On July 16 2010 04:14 jamesr12 wrote:



350/3151 = 11% so diamond leauge is around 11% of all players

On July 16 2010 09:10 vica wrote:


Ties. Sometimes, in games where my opponent quits early, like 2 minutes in, before any action happened, I check my match history and it doesn't update as a win or lose until a day later.

But that number seems a bit too high. There are several possibilities.

1. Wins update faster than losses.
2. Data is missing from servers
3. Disconnects count as win and tie, but appears as a -0 loss
4. Placement matches do not count towards ladder ranking, so if you are in placement and you've played against someone in ladder, your loss is not recorded, but his win is. Or something of the sort. Seems the most plausible.

On July 16 2010 09:41 Najda wrote:


A tie wouldn't result in a win for one player though. Your other 4 points seem reasonable, I think that it has to do with people still being in placement, because I didn't see anything in the OP about that.

On July 16 2010 11:04 Sentient wrote:
I do not think these numbers are useful.

The matchmaking system tries to push everyone to a 50% win rate. If there is an imbalance in the game, the player with the stronger race will start playing harder opponents of weaker races until his win rate is 50%. This negative feedback ensures that the races will always appear balanced. The win rates need to be weighted by the underlying matchmaking parameters, and we don't have access to that information.

Props nonetheless for putting this together. It's pretty cool to see .

On July 16 2010 11:15 Excalibur_Z wrote:


That doesn't make any sense whatsoever. Players are matched against other players, regardless of what race they play.

On July 16 2010 09:41 Najda wrote:


A tie wouldn't result in a win for one player though. Your other 4 points seem reasonable, I think that it has to do with people still being in placement, because I didn't see anything in the OP about that.

On July 16 2010 13:11 Sentient wrote:

Right, but if a person gives themselves a handicap, the system will generally find an equilibrium where they are at 50%. It doesn't matter what you do as long as you are consistent. You could decide to only ever build one extractor, and you will keep losing until it finds truly terrible players where you are at 50%. You would never conclude that building only one extractor is just as good as building two, but you would still have a 50% win rate with it.

On July 16 2010 13:22 happyness wrote:


Another possibility is that the bottom bronze players who have only played a few games haven't gone on the forums ( and may not at all)

On July 16 2010 10:22 Wr3k wrote:
Can we see some matchup specific win %'s please?

ZvT=?
PvT=?
PvZ=?

On July 16 2010 13:44 MangoTango wrote:


You don't really understand how this data is compiled, do you? It's done by looking up the win-loss ratio of every player, assuming that they play exclusively their "most played" race, then totaling them up. Not individual matchups.

And if I'm not mistaken, there's only a portion of each player's matches that's kept and displayed on the website. Hence why anything statistics of MU would not be complete.

On July 16 2010 17:49 Multis wrote:
T seems to be pretty imbalanced on metalopolis and desert oasis. On the other hand, random has 59% on metalopolis also, so I wonder if any real conclusions can be drawn from these stats at all?

On July 16 2010 17:49 Multis wrote:
T seems to be pretty imbalanced on metalopolis and desert oasis. On the other hand, random has 59% on metalopolis also, so I wonder if any real conclusions can be drawn from these stats at all?

On July 17 2010 00:49 Glacierz wrote:
Can someone explain to me why every single race has over 50% on kulas? How is that possible?

Zerg on Kulas Ravine : 54.56%
Zerg on Metalopolis : 49.93 %
Zerg on Scrap Station : 56.45 %
Zerg on Desert Oasis : 52.87 %
Zerg on Steppes of War : 52.23 %
Zerg on Lost Temple : 48.47 %
Zerg on Blistering Sands : 54.94 %

On July 17 2010 00:58 Logo wrote:


Diamond League has a >50% win rate as a whole because they kill lower level leagues.



Did no one think to point this out, this is mind boggling.

The ONLY sub 50% win rate for any race on any map in diamond is Zerg in Metalopolis and LT! That's just nuts for the 'most zerg favored map'.

I wonder what kind of bias there is. Are inferior zerg diamond players thumbing down everything they can so they play Metaloplis and LT more often because those maps are 'zerg favored'? If they're doing this while stronger zerg players leave everything thumbed up it could account for the %s.

On July 17 2010 00:56 Liquid`Jinro wrote:

How exactly do you find a 51% win ratio to be imbalanced (desert oasis), especially when the other races have higher winrates on that map?

EDIT: Oh I see, EU stats are different from the US stats. Pretty big difference too.

Oh right, but in time most people in diamond should also get 50% as they get matched against one another... I think theres not enough data history here for anything conclusive as a lot of these high win %s are from climbing up the ladder while the ladder is still settling players.

On July 17 2010 01:30 Glacierz wrote:


Oh right, but in time most people in diamond should also get 50% as they get matched against one another... I think theres not enough data history here for anything conclusive as a lot of these high win %s are from climbing up the ladder while the ladder is still settling players.

On July 17 2010 01:44 Paramore wrote:
This data is completely useless.. It just proves that Diamond players across all races play more non-diamond players than other Diamond players, otherwise there wouldn't be 50+% for every single race across every single map (with the exception of 2 zerg maps) .

On July 17 2010 01:40 Logo wrote:


No.

Long Version:
Diamond comprises the top 10% of the ladder, even more it extends up to the top 1 player. Players at the top are statistically more likely to play someone below them than above them given the possible range in match making. It sounds obvious but think about what it means. When you account for people being offline, in game, or not playing then the top 10 players are more likely to play the top 11-50 than they are each other, the top 11-50 are more likely to play 51-200 than they are to play the top 1-50 and so on. It's only the middle part of the bell-curve of player skill that you're statistically likely to have opponents that are on average exactly your skill level. You have to remember that the # of people with say 400 Diamond points is significantly more than the # of people with 500 diamond points, but those groups of players are still close enough to be matched.

So what happens is players play people that they're slightly favored to beat (say 57% likely as the racial win %s for diamond show) because when searching for a game it's more likely for a player at the top to match a player slightly below them.

To put it another way, even though behind the scenes it's the same effect, players in platinum may play vs diamond players as their rating streaks up. On the other end there's no one better than diamond to come down and play diamond.

If you looked at map win %s of Platinum or Gold they'd be much closer to 50%. If you looked at map win %s of Bronze they'd average slightly below 50%.

This sounds like it's largely about overall win % and it is. If diamond players are winning 57% of their games how could they possible have an average of 50% win rate on the maps. That'd be mathematically impossible.

Now the larger the bell curve (more players that play) the less the bell curve dips off. So right now it's a 7% difference for top players. Come release that may shrink down to 1%-5%, but it'll still exist.

On July 17 2010 01:50 Logo wrote:


That's not at all what it proves or says.

Every diamond vs diamond game drives the rating towards 50% win. Every diamond vs non-diamond game drives it away from that 50% by some unspecified amount based on how likely a diamond is to beat a non-diamond.

Here's an example... If a diamond pool has 2000 games vs diamond players and 1000 games vs non-diamond players with a 75% win rate then the pool's win % is 55% while 2/3rds of the games were played diamond vs diamond.

If the win rate vs non-diamonds is higher than 75% then you could have 57% with an even higher % of games being diamond vs diamond.

On July 17 2010 01:56 MattDamon wrote:
I don't know how much this means, because I thought the pros agreed that Metalopolis is a zerg favored map, and that Kulas is a pretty bad map for zerg,


Zerg on Metalopolis : 49.93 %
Zerg on Kulas Ravine : 54.56%


There might be too many bronze players messing up the stats to find anything useful

On July 17 2010 01:56 MattDamon wrote:
Zerg on Metalopolis : 49.93 %
Zerg on Kulas Ravine : 54.56%

On July 17 2010 01:50 Logo wrote:


That's not at all what it proves or says.

Every diamond vs diamond game drives the rating towards 50% win. Every diamond vs non-diamond game drives it away from that 50% by some unspecified amount based on how likely a diamond is to beat a non-diamond.

Here's an example... If a diamond pool has 2000 games vs diamond players and 1000 games vs non-diamond players with a 75% win rate then the pool's win % is 55% while 2/3rds of the games were played diamond vs diamond.

If the win rate vs non-diamonds is higher than 75% then you could have 57% with an even higher % of games being diamond vs diamond.

On July 17 2010 05:15 Paramore wrote:
Well, neither of us know what the frequency of diamond vs diamond and diamond vs non-diamond.. so that information becomes even more ambiguous and useless.... it could even be possible that non-diamond win vs diamond more often than diamond win vs non-diamond lol... I'm not saying that is the case, but we actually don't know anything from this information.

On July 17 2010 05:57 Silu wrote:


.......???

So you know that across all leagues, a race has 50% win. You know that diamond players of that race have ~57% win. And you say that from this we can't say if dias win vs non-dias more often than not? Please elaborate how the percentage discrepancy is possible if dias don't win against non-dias more often than not. Or tell me what I misunderstood.

Anyhow, the main point here is the dia race avg ratings and dia map win percentages... As that's the only stuff with a really sufficient sample size. Plus the only things that people (other than the "0mg ter4n has 80% win r31t agenst all and 73% of dias is ter4n" folk) didn't already know