|
On March 16 2012 19:08 neoghaleon55 wrote:Show nested quote +On March 16 2012 18:56 aebriol wrote:
Just one question: for how long back does the sample size go?
It would be relevant to look maybe 2-3 months back, but ... patches etc, will really mess with the statistics. ZvT was damn near impossible for Z for a while here and there - not really the case right now. Ok I need to address this question because people want an answer. The statistics are indeed spanning the entire career of a pro-gamer. In DRG's case it spans all the way back to 2011 GSTL S1. A counterpoint to the whole "wow so it totally doesn't apply anymore" is that we are taking the CURRENT top 20 or so koreans by ELO. Obviously, if someone hasn't done well recently, their ELO will drop. You don't see fruitdealer or jinro being talked about in these group (even though they are top of the line in their time), because their ELO has fallen off and their data are no longer relevant. People with very high ELO tend to perform well in all matchups most recently. Yes, the treatment of these statistics isn't perfect, I agree that we need better data. But no better data is available at this point for the tippiest top of koreans (most of which don't even ladder or keep their ladder ID hidden), TLPD is well respected and it's there...might as well use it. Also none of us have any idea how much the ghost patch will affect the matchup at the highest level of play. You can speculate all you want, but there's no data to say anything. We'll just have to wait and see.
Ghost nerf is nothing compared to the bug that let drones not attack as fast as scv and probes, leading to the constant marine scv all in fun and joy of early last year.
Which is why I don't think it's really relevant to look at their whole career ...
Instead, just looking at the last 2 GSL, including up and down matches, and check which race wins which matchup in every game (not just series but each), would give an overall better picture.
|
So what is point of this thread? Whine and qq some much imba that we start nerfing races because the top players on TLPD are too good? =D
|
Trade off is GOMTV's sample size is less than or very close to that magical number of N=20.
So... you found 3 out of FIFTY zergs that has less than 50% win rate vsT... and now you're making a conclusion based on that?
I can name you three more
Losira.
Coca and
Seal
All have winrates in the 40 percents
Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ...
I stopped looking pass top 80 ELO...
it was too depressing.
|
I don't understand why you think ZvT is DRG's worst matchup? Looking solely at percentages makes no sense when DRG has played 40 more games of ZvT than ZvZ, and 10 more games of ZvT than ZvP, those percentages are understandably lower. That comparison actually made no sense lol, I hope someones already brought that up.
More games equates to more losses and a lower percentage.
|
On March 16 2012 20:27 neoghaleon55 wrote:Show nested quote +On March 16 2012 20:25 Big J wrote:
What I wonder is, do you have a conclusion you draw from these stats, or did you only want to show some facts for the statistically interested (like me).
I mean, do you want to argue balance? Which is what most people assume from threads like this.
Do you want to argue skill?
Do you want to argue Zerg weaknesses/Terran strengths? (slash want people to argue about why this winrate disparity exists) I'm arguing that nobody has any idea what they're doing in ZvT. I mean, when the best ZvT player is absolutely Terrible at ZvT compared to his other matchups...you can't help but feel something's wrong. In contrast, everyone knows how to play TvZ...if you're GSL caliber, your TvZ tends to not fall below 50% winrate. I have no idea why this is or how to fix it.
but that is old data. Most people will agree that the TvZ balancing was Terranfavored at those times, the question is if this still holds true, which we can't know.
Also your stats absolutly don't match with that statement (old data, but talking about current TvZ).
|
On March 16 2012 20:47 neoghaleon55 wrote:Trade off is GOMTV's sample size is less than or very close to that magical number of N=20. Show nested quote +So... you found 3 out of FIFTY zergs that has less than 50% win rate vsT... and now you're making a conclusion based on that? I can name you three more Losira. Coca and Seal All have winrates in the 40 percents Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ... I stopped looking pass top 80 ELO... it was too depressing.
That wasn't my point, the point is, how does "I've found X number of zergs not over 50% win rate" equal to proofs for anything from a statistical point of view?
|
On March 16 2012 20:50 Fubi wrote:Show nested quote +On March 16 2012 20:47 neoghaleon55 wrote:Trade off is GOMTV's sample size is less than or very close to that magical number of N=20. So... you found 3 out of FIFTY zergs that has less than 50% win rate vsT... and now you're making a conclusion based on that? I can name you three more Losira. Coca and Seal All have winrates in the 40 percents Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ... I stopped looking pass top 80 ELO... it was too depressing. That wasn't my point, the point is, how does "I've found X number of zergs not over 50% win rate" equal to proofs for anything from a statistical point of view?
No proof.
I just find it interesting.
|
On March 16 2012 20:47 neoghaleon55 wrote:Trade off is GOMTV's sample size is less than or very close to that magical number of N=20. Show nested quote +So... you found 3 out of FIFTY zergs that has less than 50% win rate vsT... and now you're making a conclusion based on that? I can name you three more Losira. Coca and Seal All have winrates in the 40 percents Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ... I stopped looking pass top 80 ELO... it was too depressing.
The terran players you mentioned earlier, Yoda and TheBest, have played a combined total of 41 games against zerg, most of them played before august 2011. Do those matter in any way?
From what i can see both ASD and Boxer are between 70-80 ELO both with horrible TvZ winrates in the 30 percents.
This thread is exactly like the "Where have all the terrans gone?" thread. Ignoring anything that goes against your argument, misrepresenting statistics, and exaggerating. All posted by a OP with an agenda and obviously the race icon of the "underperforming" side of the argument.
Man i really hate threads like these, but i just can't stay away.
|
in my opinion the reason behind both this and also why terran has always won more than 50% of their tvz in the history of sc2 is simple:
terran can't just 'randomly' die. there are a lot of things that just outright kill zerg very early while terran can always just lift up and stuff and stay in the game. for that reason even if terran fails a cheese or gets cheesed and 'loses' (lifts up) his nat the game still can go on for a long time after making it possible to come back.
in other words terran has more chances to allow them to win by being better because they can always play a longer game even if they get significantly behind while it's easy to get in a situation where you just outright die as zerg vs terran even if you are the way better player.
tl;dr terran is way safer compared to zerg
|
On March 16 2012 20:58 karpo wrote:
From what i can see both ASD and Boxer are between 70-80 ELO both with horrible TvZ winrates in the 30 percents.
Man i really hate threads like these, but i just can't stay away.
Hey thanks.
I can sleep at night now.
I love boxer... so sad he's on the same plane as asd....
|
Well, it's so hard for a zerg to kill a terran, so as a zerg you have to win the game many times over - I think that influences stats at the highest levels where it's easier to make a mistake as zerg and lose than as terran. You see that DRG and Nestea try to close a lot of their games vs Terran early, since I don't think they're as confident in late game, since it's so fragile. So it's hard to be consistent when you cheese a lot and are punished for your mistakes.
Maybe it's just about volatility and not balance though. Like, a good TvZ player can always beat lesser competition, while a good ZvT player still often loses to random stuff.
|
On March 16 2012 20:47 neoghaleon55 wrote:
I can name you three more
Losira.
Coca and
Seal
All have winrates in the 40 percents
Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ...
I stopped looking pass top 80 ELO...
it was too depressing.
Also fOrGG 33.33% win rate versus zerg, ELO 35.
Nada 48% win rate, ELO 51.
Puma, 43% win rate, ELO 56.
Rain, sub 50 % win rate, ELO 59.
fOrGG and Rain have 12 and 18 matches against Z though. Am i doing this right?
|
On March 16 2012 12:51 HyperionDreamer wrote:Show nested quote +On March 16 2012 12:18 ETisME wrote:
I am sorry but you got the stats wrong.
The over 20 requirement is for cluster analysis, something that you aren't doing because you are not trying to make any clusters out from the data set.
The over 20 thing you talked about is just for normal hypothesis testing, which you aren't doing.
You need to calculate out the optimal minimal sample size based upon your confidence interval etc
in short, you need to calculate out a sample size that truely represent the population. Merely 50 games out of his entire ZvT history does not make sense Yep. The study cited in the OP pertains to a specific type of stats testing, called cluster analysis. Maybe read up on it a bit before you cite it as valid, OP. You're talking about simple testing for type 1/2 statistical errors, so you would need a much larger sample size. I did a post a while ago doing rigid scientific statistical analysis on korean matchup percentages, and I think even a sample size of ~200 games rendered a ~7% difference statistically irrelevant. http://en.wikipedia.org/wiki/Cluster_analysisEdit: It was a sample size of 130, and an ~8% statistical difference. This was rendered statistically insignificant using standard p-level analysis. Here's the link to my analysis. http://www.teamliquid.net/forum/viewmessage.php?topic_id=317114¤tpage=12#226
Just to quote this again to show why your sample size is way too small to be statistically significant.
Let me break it down even more for you:
can you do a survey containing 20 people out of a population of 1 billion people and claim the outcome represent the whole population?
No, you need around ~10% of the population to be a good representative of the population (not to mention the different sampling techniques) and if you want, you can do an intensive calculation for calculating the optimal sample size using the desired confidence level
I think you got some of the very fundermental statisic wrong/mixed up.
|
On March 16 2012 21:06 karpo wrote:Show nested quote +On March 16 2012 20:47 neoghaleon55 wrote:
I can name you three more
Losira.
Coca and
Seal
All have winrates in the 40 percents
Here's your homework...go find me a Terran (with 20 or more GSL games) on TLPD that has a less than 50% winrate in TvZ...
I stopped looking pass top 80 ELO...
it was too depressing. Also fOrGG 33.33% win rate versus zerg, ELO 35. Nada 48% win rate, ELO 51. Puma, 43% win rate, ELO 56. Rain, sub 50 % win rate, ELO 59. fOrGG and Rain have 12 and 18 matches against Z though. Am i doing this right?
I said top 50...
and forGG played a total of 12 games vs Zerg...not enough, sir.
|
On March 16 2012 21:11 ETisME wrote:Show nested quote +On March 16 2012 12:51 HyperionDreamer wrote:On March 16 2012 12:18 ETisME wrote:
I am sorry but you got the stats wrong.
The over 20 requirement is for cluster analysis, something that you aren't doing because you are not trying to make any clusters out from the data set.
The over 20 thing you talked about is just for normal hypothesis testing, which you aren't doing.
You need to calculate out the optimal minimal sample size based upon your confidence interval etc
in short, you need to calculate out a sample size that truely represent the population. Merely 50 games out of his entire ZvT history does not make sense Yep. The study cited in the OP pertains to a specific type of stats testing, called cluster analysis. Maybe read up on it a bit before you cite it as valid, OP. You're talking about simple testing for type 1/2 statistical errors, so you would need a much larger sample size. I did a post a while ago doing rigid scientific statistical analysis on korean matchup percentages, and I think even a sample size of ~200 games rendered a ~7% difference statistically irrelevant. http://en.wikipedia.org/wiki/Cluster_analysisEdit: It was a sample size of 130, and an ~8% statistical difference. This was rendered statistically insignificant using standard p-level analysis. Here's the link to my analysis. http://www.teamliquid.net/forum/viewmessage.php?topic_id=317114¤tpage=12#226 Just to quote this again to show why your sample size is way too small to be statistically significant. Let me break it down even more for you: can you do a survey containing 20 people out of a population of 1 billion people and claim the outcome represent the whole population? No, you need around ~10% of the population to be a good representative of the population (not to mention the different sampling techniques) I think you got some of the very fundermental statisic wrong/mixed up.
Please go read the opening post again under question 1.
I made it very simple for everyone to understand
with pictures, too!
|
So its DRG's worst matchup, everybody as one and honestly, his is still well over 50% so I think he is going to be just fine....
And futhermore, if you disregard the top zerg and the top terran, the win rates are pretty close to each other for the remaining top 5. I see nothing wrong here, what exactly is eating at you OP?
|
On March 16 2012 21:15 neoghaleon55 wrote:Show nested quote +On March 16 2012 21:11 ETisME wrote:On March 16 2012 12:51 HyperionDreamer wrote:On March 16 2012 12:18 ETisME wrote:
I am sorry but you got the stats wrong.
The over 20 requirement is for cluster analysis, something that you aren't doing because you are not trying to make any clusters out from the data set.
The over 20 thing you talked about is just for normal hypothesis testing, which you aren't doing.
You need to calculate out the optimal minimal sample size based upon your confidence interval etc
in short, you need to calculate out a sample size that truely represent the population. Merely 50 games out of his entire ZvT history does not make sense Yep. The study cited in the OP pertains to a specific type of stats testing, called cluster analysis. Maybe read up on it a bit before you cite it as valid, OP. You're talking about simple testing for type 1/2 statistical errors, so you would need a much larger sample size. I did a post a while ago doing rigid scientific statistical analysis on korean matchup percentages, and I think even a sample size of ~200 games rendered a ~7% difference statistically irrelevant. http://en.wikipedia.org/wiki/Cluster_analysisEdit: It was a sample size of 130, and an ~8% statistical difference. This was rendered statistically insignificant using standard p-level analysis. Here's the link to my analysis. http://www.teamliquid.net/forum/viewmessage.php?topic_id=317114¤tpage=12#226 Just to quote this again to show why your sample size is way too small to be statistically significant. Let me break it down even more for you: can you do a survey containing 20 people out of a population of 1 billion people and claim the outcome represent the whole population? No, you need around ~10% of the population to be a good representative of the population (not to mention the different sampling techniques) I think you got some of the very fundermental statisic wrong/mixed up. Please go read the opening post again under question 1. I made it very simple for everyone to understand with pictures, too!
On March 16 2012 20:04 Fubi wrote:
Exactly, luck (or chance) is involved. And you CAN quantify luck, because luck = chance = variance, therefore, that is why standard deviation, confidence interval, margin of errors, etc, exists in statistics.
Here, I'll use YOUR graphs to help you understand
look at your own graphs that you just made: even at n = 32, there is still a fairly large chance that you can get 14/32 (=~44%) heads or 18/32 (=~56%) heads (+/- 2 from normal), simply due to random chance.
So, say we flip 32 times, and I get 12 heads, and you get 20 heads, that is a ~12% difference. BUT, you can NOT prove whether or not this ~12% difference is due to our skills at flipping, or simply due to chance. You have to do variance analysis with a confidence interval to prove the significance of this statistical test
*edit: and btw, you made that statistical analysis, and based on that, you made a statistical claim, therefore, the onus of proof is on you, not on me, hence why you need to be the one to do it if you want to prove that your stats has any statistical significance, not me.
So what about this? You thanked him for his input but it seems you still reference your OP whereas this was proven to not be very reliable.
|
On March 16 2012 21:16 CaptainCrush wrote:
So its DRG's worst matchup, everybody as one and honestly, his is still well over 50% so I think he is going to be just fine....
And futhermore, if you disregard the top zerg and the top terran, the win rates are pretty close to each other for the remaining top 5. I see nothing wrong here, what exactly is eating at you OP?
I'm wondering why there's that 5-6% difference between ZvT and TvZ at the tiptop level.
and why nobody in the Top 50 ELO has problems with TvZ
but plenty of zergs (more than half) has problems with ZvT....
User was temp banned for this post.
|
On March 16 2012 21:18 karpo wrote:Show nested quote +On March 16 2012 21:15 neoghaleon55 wrote:On March 16 2012 21:11 ETisME wrote:On March 16 2012 12:51 HyperionDreamer wrote:On March 16 2012 12:18 ETisME wrote:
I am sorry but you got the stats wrong.
The over 20 requirement is for cluster analysis, something that you aren't doing because you are not trying to make any clusters out from the data set.
The over 20 thing you talked about is just for normal hypothesis testing, which you aren't doing.
You need to calculate out the optimal minimal sample size based upon your confidence interval etc
in short, you need to calculate out a sample size that truely represent the population. Merely 50 games out of his entire ZvT history does not make sense Yep. The study cited in the OP pertains to a specific type of stats testing, called cluster analysis. Maybe read up on it a bit before you cite it as valid, OP. You're talking about simple testing for type 1/2 statistical errors, so you would need a much larger sample size. I did a post a while ago doing rigid scientific statistical analysis on korean matchup percentages, and I think even a sample size of ~200 games rendered a ~7% difference statistically irrelevant. http://en.wikipedia.org/wiki/Cluster_analysisEdit: It was a sample size of 130, and an ~8% statistical difference. This was rendered statistically insignificant using standard p-level analysis. Here's the link to my analysis. http://www.teamliquid.net/forum/viewmessage.php?topic_id=317114¤tpage=12#226 Just to quote this again to show why your sample size is way too small to be statistically significant. Let me break it down even more for you: can you do a survey containing 20 people out of a population of 1 billion people and claim the outcome represent the whole population? No, you need around ~10% of the population to be a good representative of the population (not to mention the different sampling techniques) I think you got some of the very fundermental statisic wrong/mixed up. Please go read the opening post again under question 1. I made it very simple for everyone to understand with pictures, too! Show nested quote +On March 16 2012 20:04 Fubi wrote:
Exactly, luck (or chance) is involved. And you CAN quantify luck, because luck = chance = variance, therefore, that is why standard deviation, confidence interval, margin of errors, etc, exists in statistics.
Here, I'll use YOUR graphs to help you understand
look at your own graphs that you just made: even at n = 32, there is still a fairly large chance that you can get 14/32 (=~44%) heads or 18/32 (=~56%) heads (+/- 2 from normal), simply due to random chance.
So, say we flip 32 times, and I get 12 heads, and you get 20 heads, that is a ~12% difference. BUT, you can NOT prove whether or not this ~12% difference is due to our skills at flipping, or simply due to chance. You have to do variance analysis with a confidence interval to prove the significance of this statistical test
*edit: and btw, you made that statistical analysis, and based on that, you made a statistical claim, therefore, the onus of proof is on you, not on me, hence why you need to be the one to do it if you want to prove that your stats has any statistical significance, not me. So what about this? You thanked him for his input but it seems you still reference your OP whereas this was proven to not be very reliable.
Everyone has a voice and should be cordially respected.
But I don't have to take everyone on seriously.
|
On March 16 2012 21:18 neoghaleon55 wrote:Show nested quote +On March 16 2012 21:16 CaptainCrush wrote:
So its DRG's worst matchup, everybody as one and honestly, his is still well over 50% so I think he is going to be just fine....
And futhermore, if you disregard the top zerg and the top terran, the win rates are pretty close to each other for the remaining top 5. I see nothing wrong here, what exactly is eating at you OP? I'm wondering why there's that 5-6% difference between ZvT and TvZ at the tiptop level. and why nobody in the Top 50 ELO has problems with TvZ but plenty of zergs (more than half) has problems with ZvT....
So give us your honest opinion on why this is? I would really like you to just come out straight and say exactly what you think is the reason.
On March 16 2012 21:20 neoghaleon55 wrote:Show nested quote +On March 16 2012 21:18 karpo wrote:On March 16 2012 21:15 neoghaleon55 wrote:On March 16 2012 21:11 ETisME wrote:On March 16 2012 12:51 HyperionDreamer wrote:On March 16 2012 12:18 ETisME wrote:
I am sorry but you got the stats wrong.
The over 20 requirement is for cluster analysis, something that you aren't doing because you are not trying to make any clusters out from the data set.
The over 20 thing you talked about is just for normal hypothesis testing, which you aren't doing.
You need to calculate out the optimal minimal sample size based upon your confidence interval etc
in short, you need to calculate out a sample size that truely represent the population. Merely 50 games out of his entire ZvT history does not make sense Yep. The study cited in the OP pertains to a specific type of stats testing, called cluster analysis. Maybe read up on it a bit before you cite it as valid, OP. You're talking about simple testing for type 1/2 statistical errors, so you would need a much larger sample size. I did a post a while ago doing rigid scientific statistical analysis on korean matchup percentages, and I think even a sample size of ~200 games rendered a ~7% difference statistically irrelevant. http://en.wikipedia.org/wiki/Cluster_analysisEdit: It was a sample size of 130, and an ~8% statistical difference. This was rendered statistically insignificant using standard p-level analysis. Here's the link to my analysis. http://www.teamliquid.net/forum/viewmessage.php?topic_id=317114¤tpage=12#226 Just to quote this again to show why your sample size is way too small to be statistically significant. Let me break it down even more for you: can you do a survey containing 20 people out of a population of 1 billion people and claim the outcome represent the whole population? No, you need around ~10% of the population to be a good representative of the population (not to mention the different sampling techniques) I think you got some of the very fundermental statisic wrong/mixed up. Please go read the opening post again under question 1. I made it very simple for everyone to understand with pictures, too! On March 16 2012 20:04 Fubi wrote:
Exactly, luck (or chance) is involved. And you CAN quantify luck, because luck = chance = variance, therefore, that is why standard deviation, confidence interval, margin of errors, etc, exists in statistics.
Here, I'll use YOUR graphs to help you understand
look at your own graphs that you just made: even at n = 32, there is still a fairly large chance that you can get 14/32 (=~44%) heads or 18/32 (=~56%) heads (+/- 2 from normal), simply due to random chance.
So, say we flip 32 times, and I get 12 heads, and you get 20 heads, that is a ~12% difference. BUT, you can NOT prove whether or not this ~12% difference is due to our skills at flipping, or simply due to chance. You have to do variance analysis with a confidence interval to prove the significance of this statistical test
*edit: and btw, you made that statistical analysis, and based on that, you made a statistical claim, therefore, the onus of proof is on you, not on me, hence why you need to be the one to do it if you want to prove that your stats has any statistical significance, not me. So what about this? You thanked him for his input but it seems you still reference your OP whereas this was proven to not be very reliable. Everyone has a voice and should be cordially respected. But I don't have to take everyone on seriously.
So someone questions and gives you feedback on what's wrong with your statistical analysis, all you say is "thanks" and just keep on trucking replying to people with the same flawed argument? Great.
|
|
|
|
|
|
EPT
