|
On September 03 2010 21:45 MrCeeJ wrote:
Without knowing what race the opponent was the % are a bit distorted. Given that there are more Terrans than the others more than 1/3 of the games in the Terran section will be mirror matches where as less than 1/3 of the zergs are.
Also games played against a race that is strong on this map will be biased against, whilst games played against races that are weak on the map will be biased towards the current race.
These two anomalies make it very hard to come to any conclusions about map balance when the different % are so close to each other.
If you look at the numbers being as relative and not absolute, neither of these matter that much. The idea is to gauge what the maps are like when compared to each other for one race. You're correct that 56% for Zerg doesn't quite mean the same as 56% for Terran (the latter being a bigger sign of "imba" in this case, since they still manage a big win percentage even with relatively more mirror matches included in it). However the numbers are very comparable among a single race.
Perhaps taking the top 1000 of each race and looking at a fixed number of their games would give you a better 'source' pool, but you still have the problem that their opponents are of unevenly distributed race compositions..
This would not change anything at all. More samples is better, it doesn't matter that some races have more samples than others. It would be different if I could pick actual GAMES with known matchups (better yet, games among the top players, not against lower ones), not just half-games knowing only one side (which among other things leads to some games being counted "twice", but for the purpose of the evaluation this doesn't matter either).
For the record, the percentages are as follows in EU Diamond:
35.0% Protoss
30.2% Terran
25.7% Zerg
|
A lot of protoss can do effective cheese on DO too. A lot of place to hide a proxy 2 gate etc.
|
On September 03 2010 20:34 StupidFatHobbit wrote:
A very large number of zergs have Kulas X'd out. If you pick zerg and don't X out kulas you will get a ton of zvt on it because practically none of the terrans X it out.
It depends how biased opposing terrans are. If they want to take fair maps out (blistering, meta) to only have T favoring maps in pool, then sure you'll get tons of kulas. Against non-nerd terrans you should get pretty fair share of step LT meta blistering DQ xnc. I see no fair point for terrans to thumb down any of those. DO is crap, and I can understand some diss scrap. It's beyond me why so many think steppes is fine and scrap isn't though. Steppes just encourages all sort of impulsive pushes with zero reacting time and thus is pretty shitty map for macroing.
Seems DQ is pretty much thumbed down by z too. Might join the anti-kulas/DO/DQ movement myself, those maps just are bad.
|
Either the results of this aren't "statistically significant" (56%>53%) or I've been making some poor assumptions. Apparently Terran > Zerg on 4 player maps? This is news to me.
|
Is down voting maps you don't like common? I always play with all the maps on even though I really hate some of them. I like to have replays, win or loose, on maps I don't like since I feel like I learn more from them somehow.
|
These statistics aren't even good to analyze the maps. It should be possible to remove mirror matchups, because they put all win rates closer to 50% than they really are. Expect most win rates to be much further from 50% than these data gives the impression of, specially for protoss and terran, where the chance of mirror matchup is higher due to more people playing them.
|
I did some quick stats on that data, and most of the differences aren't statistically significant, if you just compare within races (using the average win rate for each race). From that it seems:
For terrans, your win rate will be highest if you pick lost temple and maybe metalopolis. Stay away from blistering sands and probably scrap station as well. This seems to be less important for the top 5000 than for the general diamond population.
For zerg, pick scrap station and blistering sands, avoid delta quadrant like the pague, and also steppes of war and probably kulas ravine as well. Again, things seem more even for the top 5k.
For protoss if you avoid any maps, let them be scrap station and maybe lost temple.
Random players, just keep doing what you do.
I also looked to see how win rate correlated with map choice. You'd expect to see that players pick maps that help their races out. This is certainly true for terran and a bit for zerg, and especially among the top 5k, where there are strong positive relationships. However, protoss of both experience levels are doing a bad job of picking maps! For them, performance is negatively correlated with map preference. So protoss players, think twice about which maps you vote down! Protoss have been playing a lot of games on maps they aren't good at. Protoss players, stop voting down DO and stop playing scrap station and lost temple!
|
Interesting stat for Protoss on Blistering Sands. People have been calling it toss' backyard recently, but for the very top players, it doesn't seem to be all that great for toss. But for diamond as a whole, its one of their best maps.
I assume the very best are just capable of defending against a 4 warpgate push that abuses the destructible rocks?
|
What your data is telling me is that an average of 50% winrate for all matchups on every map. Blizzard probably has the same data and is probably why we're not seeing such drastic balance changes. Atleast, on EU server.
Thanks for the time and effort for that.
|
Call me ignorant, but I'm of the party that believes that if theory can deem a map imbalanced without a doubt, the statistics take second banana. Terran obviously has some racial advantages on Kulas, but that doesn't mean that they've figured out how to abuse it (especially across all Diamond players). To trust the statistics over theory in this situation would be to assume that everyone plays perfectly and takes advantage of every advantage. I don't.
|
it's hard to read this data correctly imo, without the 'other side' of the matchups you're kind of left with one conclusion: higher ranked players are better than lower ranked players
which is nice and all as it suggests the ranking system is working, but very non-revealing when it comes to actual map balance imo
|
This thread is awesome. I was just thinking today I would like to see some statistics like this.
It sucks that you can't see specific matchups, but it is some pretty interesting information.
It's good to know that everybody hates Desert Oasis as much as I do!
|
unfortunately other stats tell us theres less zergs in diamond than toss or terran.
http://sc2ranks.com/stats/race/eu/1
so if the majority of the games played were pvt or mirror then its not gonna tell you abt balance for zerg.
|
As a Zerg player all I know is that Scrap Station is my favourite map! The statistics say the map favours me, it sure feels like it when playing anyway!
|
Pretty sure that all of them are going to hover around 50% as the system is designed to match people up against someone they'll win against 50% of the time. People who can't hold a 50% win rate against even-matched players get bumped down into platinum. That's why you have such a disparity in race played and not in win %.
|
On September 03 2010 21:45 MrCeeJ wrote:
Without knowing what race the opponent was the % are a bit distorted. Given that there are more Terrans than the others more than 1/3 of the games in the Terran section will be mirror matches where as less than 1/3 of the zergs are.
Just because of you, and extensive boredom, here (at the bottom) are the numbers with the mirror match bias removed.
+ Show Spoiler [How does that work?] +Easy. We know the amount of mirror matches on EU Diamond (38/33/29 % for P/T/Z, Random players included - this is VERY slightly skewed due to Diamond players playing lower leagues, but as they generally don't play Bronze ones the effect is very very minor as the race distribution is very similar for the other leagues), and we know the expected result of a mirror match (50/50). So we can subtract 38/33/29 % from the total games, and half of what was subtracted from the wins.
And to reiterate: some people think the stats are flawed because it doesn't account for things like variable opponent skill or similar. But skillful and bad players affect all the stats in exactly the same way, as everyone, regardless of race, is equally likely to get a similarly skilled opponents compared to their own level, and thus cancel each other out. Yes, the matchmaking system tries to bring everyone's win percentage to 50%, but that's all the more reason to put significance on stats that show a deviation from that target (naturally Diamond players already have a win percentage over 50 by beating sub-Diamond players, as the average across all the maps for a certain race show here as well). Actually the "50% target win" principle also removes overall race imbalance completely from this kind of stats (which would manifest as an offset race-specific bias) because that can only be shown in the ratings/ranks and not the win percentages due to how the system works.
Some think that the results can't be interpreted because they include two matchups for each race - to offset this you can simply look for the results for a single map for all the races. For example, Scrap Station is by far the best Zerg map, and is the worst and second worst for T and P. Hence, Z is very good against both on that. On the other end we have Blistering Sands which is among the top Protoss maps - it is also among the top Zerg maps, but the worst by far Terran map. So, Protoss and Zerg have a very good time against Terran on that map, as the win percentage is relatively so high even with their other non-mirror matchup also having success on it.
Also people tend to dislike maps that are shitty in general (and by extension think they're bad for them) regardless of their race's success on it, which is why no one likes Desert Oasis no matter their race, so keep that in mind.
Hopefully that settles some logical fallacies that people tend to have when looking at stats like this. If you're going to claim bullshit, your main argument should probably be sample size  (though even that doesn't give that much leeway - think voting polls and their sample sizes)
The results without mirror match bias:
+ Show Spoiler +Terran on Lost Temple : 2813/4934 = 57.01%
Terran on Metalopolis : 2496/4501 = 55.45%
Terran on Kulas Ravine : 1713/3117 = 54.95%
Terran on Xel'Naga Caverns : 2606/4791 = 54.39%
Terran on Steppes of War : 2767/5134 = 53.89%
Terran on Desert Oasis : 1349/2514 = 53.65%
Terran on Delta Quadrant : 1686/3156 = 53.42%
Terran on Scrap Station : 1893/3637 = 52.04%
Terran on Blistering Sands : 2088/4175 = 50.01%
Zerg on Scrap Station : 2341/4097 = 57.13%
Zerg on Desert Oasis : 1359/2479 = 54.82%
Zerg on Blistering Sands : 2448/4476 = 54.69%
Zerg on Metalopolis : 2437/4573 = 53.29%
Zerg on Lost Temple : 2341/4444 = 52.67%
Zerg on Xel'Naga Caverns : 2366/4559 = 51.89%
Zerg on Steppes of War : 2336/4616 = 50.60%
Zerg on Kulas Ravine : 1235/2443 = 50.55%
Zerg on Delta Quadrant : 1424/2938 = 48.46%
Protoss on Desert Oasis : 1855/3415 = 54.31%
Protoss on Blistering Sands : 3337/6238 = 53.49%
Protoss on Xel'Naga Caverns : 3463/6512 = 53.17%
Protoss on Steppes of War : 3718/7000 = 53.11%
Protoss on Delta Quadrant : 2159/4071 = 53.03%
Protoss on Kulas Ravine : 2150/4067 = 52.86%
Protoss on Metalopolis : 3257/6164 = 52.83%
Protoss on Lost Temple : 3398/6608 = 51.42%
Protoss on Scrap Station : 2636/5128 = 51.40%
|
how come almost all the % are over 50? i think like 1 under. that cant add up?
|
yeah could someone explain why every race seems to have like 55% win? its probably basic math but i'm kind of missing how this is possible
|
Wow! protoss winrate is affected much less by map... Which should mean that the winrates for zerg and terran fluctuate EVEN more taking into consideration only zvt matchup, Scrap station Heavily fauvors zerg and lost temple heavily fauvors terran! thanks alot for these stats very interesting!
*was looking only on the stats without mirror bias.
|
they're taking some arbitrary X number of players, but those players don't necessarily only play games within those X number of players. so it doesn't need to add up to 100%.
basically it's not zero-sum, it'd only need to add up to 100% if they were counting every single player on the EU server.
|
|
|
|
|
|
EPT
