[GSL] Race Winrate Graphs - Page 6

On October 14 2011 03:17 dschneid wrote:
There's a huge lack of stats knowledge here and it's sad that these graphs get put on display when people don't know how to read them. For people who said sample size doesn't matter because these are win percentages, it doesn't matter that percentages are normalized for sample size, sample size still effects how much weight outliers hold as well as the general significance of the number.

Put some confidence intervals around these, make note of possible selection bias, and tell me the overall sample sizes (a few hundred at best? That's not that much in the realm of stats)...

Bottom line you can only make a few inferences from these but no sound conclusions.

On October 14 2011 03:17 dschneid wrote:
There's a huge lack of stats knowledge here and it's sad that these graphs get put on display when people don't know how to read them. For people who said sample size doesn't matter because these are win percentages, it doesn't matter that percentages are normalized for sample size, sample size still effects how much weight outliers hold as well as the general significance of the number.

Put some confidence intervals around these, make note of possible selection bias, and tell me the overall sample sizes (a few hundred at best? That's not that much in the realm of stats)...

Bottom line you can only make a few inferences from these but no sound conclusions.

On October 14 2011 02:57 Wuster wrote:


Controversially, you could say that MC has dragged Protoss as a whole down. Reason being, Korea is a very copy-cat scene (the only real exception is the tank/marine vs mech split for TvT). And MC always had a really timing/micro heavy style, so people who weren't as sharp as him couldn't pull it off and eventually when people learned timing defenses, then those other Protoss' disappeared too (Alicia comes to mind - I still best remember him for straight up killing MVP when poking with a 3-gate expo).

Maybe this is reading too much into MC's fall from grace and yes I'm kind of ripping off what Wolf said, but players sure like to follow the example of their best players (how many NA zergs only respect macro ala Idra?). And Genius/Hongun, who never played like MC, are the two most consistent Protoss players (as in Code S every season knock on wood).

I think the huge swings in PvX are weird; how the hell do you balance a race that goes to 60% win rates then down to 30% winrates month to month?

Some of it though is caused by patch changes, like the Roach range buff in Open 2, Fungal buff in May all caused PvZ to tank immediately, but they more or less bounced back the season after.

Obviously, PvT hasn't bounced back from August yet; but 1-1-1 isn't a patch change so that's not too surprising. PvZ is starting to come back though, so that's a good sign.

I'm curious what meta/patch changes happened in PvT between January and December to cause Protoss to go from 2 in the final 4! MC FF's imba! Hongun's all-ins too good! to Protoss barely winning.



Inca, Genius, Hongun and Anypro have all made deep GSL runs (only Genius has missed on a Ro4 on that list). It's not like there are a lot of non-Nestea zergs that make GSL finals either (just Fruitdealer and Losira).

On October 14 2011 03:39 dschneid wrote:
It's fine to hypothesize, I won't even argue against it, that's why I said it was fine to infer but that people shouldn't be drawing conclusions and clamoring for changes etc when this data set is not at all representative of the daily match ups they face (unless you're a pro, this doesn't even apply to you, what you'd want to see is the data for win rates on your basic ladder, split out by league and server).

Furthermore, you can't really account for the selection bias. Everyone seems to be jumping to the conclusion that the best players play T because it's the best race? Maybe not, maybe it's the hardest and they like a challenge. I don't know as i'm not pro and I've never talked with one. Either way you have to be careful.

On October 14 2011 02:57 Wuster wrote:


Controversially, you could say that MC has dragged Protoss as a whole down. Reason being, Korea is a very copy-cat scene (the only real exception is the tank/marine vs mech split for TvT). And MC always had a really timing/micro heavy style, so people who weren't as sharp as him couldn't pull it off and eventually when people learned timing defenses, then those other Protoss' disappeared too (Alicia comes to mind - I still best remember him for straight up killing MVP when poking with a 3-gate expo).

Maybe this is reading too much into MC's fall from grace and yes I'm kind of ripping off what Wolf said, but players sure like to follow the example of their best players (how many NA zergs only respect macro ala Idra?). And Genius/Hongun, who never played like MC, are the two most consistent Protoss players (as in Code S every season knock on wood).

I think the huge swings in PvX are weird; how the hell do you balance a race that goes to 60% win rates then down to 30% winrates month to month?

Some of it though is caused by patch changes, like the Roach range buff in Open 2, Fungal buff in May all caused PvZ to tank immediately, but they more or less bounced back the season after.

Obviously, PvT hasn't bounced back from August yet; but 1-1-1 isn't a patch change so that's not too surprising. PvZ is starting to come back though, so that's a good sign.

I'm curious what meta/patch changes happened in PvT between January and December to cause Protoss to go from 2 in the final 4! MC FF's imba! Hongun's all-ins too good! to Protoss barely winning.



Inca, Genius, Hongun and Anypro have all made deep GSL runs (only Genius has missed on a Ro4 on that list). It's not like there are a lot of non-Nestea zergs that make GSL finals either (just Fruitdealer and Losira).

On October 14 2011 03:36 Sabu113 wrote:


Right now I think the standard response would be that it's foolish to ignore the story around the stats. Most arguments about balance have observations about why the game is the way it is and stats merely augment the argument by pointing out something that seems abnormal.

Just to make this post/ thread a little productive, how do you feel about psych studies? They can have <50 obs in a study and still make definitive claims.

What assumptions do we have to make to make these stats into something worthwhile on their own? If we recorded the data in a different way to run a regression what lag should we throw onto a metagame variable.

On October 14 2011 01:29 Roxy wrote:
Very Interesting

It is a shame that all of the skilled players chose terran.. wish i could see some protoss innovate new strategies but alas all of the protoss players are scrubs

I guess they just pick terran because it is the hardest race to play and we all know koreans love high APM

On October 14 2011 03:47 NicolBolas wrote:


Stats that don't mean anything don't suddenly gain meaning just because you put some theorycraft next to them. Either the stats are real statistics that have actual value, or they are junk. There's no real middle-ground here.

On October 14 2011 03:17 dschneid wrote:
There's a huge lack of stats knowledge here and it's sad that these graphs get put on display when people don't know how to read them. For people who said sample size doesn't matter because these are win percentages, it doesn't matter that percentages are normalized for sample size, sample size still effects how much weight outliers hold as well as the general significance of the number.

Put some confidence intervals around these, make note of possible selection bias, and tell me the overall sample sizes (a few hundred at best? That's not that much in the realm of stats)...

Bottom line you can only make a few inferences from these but no sound conclusions.

On October 14 2011 03:17 dschneid wrote:
There's a huge lack of stats knowledge here and it's sad that these graphs get put on display when people don't know how to read them. For people who said sample size doesn't matter because these are win percentages, it doesn't matter that percentages are normalized for sample size, sample size still effects how much weight outliers hold as well as the general significance of the number.

Put some confidence intervals around these, make note of possible selection bias, and tell me the overall sample sizes (a few hundred at best? That's not that much in the realm of stats)...

Bottom line you can only make a few inferences from these but no sound conclusions.

On October 14 2011 03:03 SeaSwift wrote:


I like how you say that timing pushes have been figured out but Hongun has never played like MC and has remained a consistent Protoss player.

HongUn has ALWAYS relied on timing pushes. Be it 3gate Void Ray all-in, 4gate all-in, you name it. I think I can count the number of times he has expanded on one hand. He has just been lucky enough so far to face some relatively mediocre players or else beat people on their off-day (vs MC, for example. 3 4gates in a row...). He's the Protoss version of YuGiOh (or YuG1-1-1Oh), but more successful and lucky IMO.

And Genius is just a really solid all-round player, making up for his relatively poor mechanics (compared to MC/MVP et al) with being FAR ahead of his time in the metagame.

On October 14 2011 03:37 K3Nyy wrote:


I dunno if you realize it but timing attacks happen in a lot of games, not just MC's game.

TvZ, the 3 tank timing attack happens in a lot of the games. TvP, Stim/Ghost/2 medivac timings happen all the timing. ZvP, roach ling, etc.

The point is nobody maxes to 200/200 and then goes attack, that's not a macro game. Timing attacks always happens, not just in the Protoss race. The difference is that pretty much all the Protoss timings have been figured out.