|
On October 03 2010 13:26 Jibba wrote:Show nested quote +On October 03 2010 12:37 electronicJustice wrote:On September 29 2010 04:57 gospelwut wrote:
I'm sure this has been mentioned, so I apologize if it has.
It should be noted and always thought about that this is one (incomplete) data set. On its own, this yields no real conclusions or even correlations. To make any assumptions you would need other data sets including but not limited to: (a) distribution of races entered into said tournements (b) a metric for judging skill to further polish the numbers (there's no point in including non top players) and (c) some manner of standard deviation or statistical probability -- which is very difficult. Drawing any conclusions from this set of data alone is just as foolish as using Battle.net rankings as a metric of balance. You are correct that this would be meaningless without other data. Fortunately, that other data exists and is easily available, and the statistical techniques required to analyze the data are not hard. In order to get an idea of the ratio of Terran, Zerg, and Protoss players in these tournaments, lets use the ratio for the top 100 players in diamond. http://sc2ranks.com/stats/league/all/1/100/They are not the exact players, but those players are all probably of the caliber required to be competitive in these tournaments, and all we are looking for is an idea of how many players were from each race. It seems unlikely that this subsection gives bad results, since we see the same basic ratio that exists throughout the ladder. So, from that data we expect about 23% of the players to be zerg. If we assume that they are all equally competitive (assuming that one group is less competitive would automatically make the conclusion true), then the expected value for the proportion of zerg wins is 0.23. Now, we can conduct a standard one proportion z test as someone might learn about in any High-school statistics class. I believe most graphing calculators have this function built in (which is how I will do it) and a quick google search will acquaint you with it if you haven't heard of it. The P-value I get from the test is .0001026. This is very small. It means, basically, that we can be 99.9897 percent sure that zerg players could not have a win rate this small through random chance if the three races are equally balanced. You haven't defined what constitutes a balanced race, and you've ignored all external factors. So far you've been able to use numbers to find that a problem exists (Zerg does not win as often as it should in an ideal game state) but you can't recommend corrective measures from that because everything else is still unsolved. If the only map available was Scrap Station, what would the numbers then reflect?
You're correct, this only says that there is something causing zergs to win less than would be expected if everything was even. It doesn't necessarily have to be something built into the race. It is entirely possible that something else, like the map pool, is causing zerg to be less likely to win (unfortunately, I don't know of a site that has data on individual maps). Since we can't remove confounding variables like that, the races' win probabilities in that analysis include not only the inherent strength of the race but also things like map imbalance.
The data we would need to find a more specific answer or find something to change would be way harder to find, and it may not exist in a form we can access, but that doesn't mean we can't look at the data we do have.
|
United States22883 Posts
On October 03 2010 13:52 electronicJustice wrote:Show nested quote +On October 03 2010 13:26 Jibba wrote:On October 03 2010 12:37 electronicJustice wrote:On September 29 2010 04:57 gospelwut wrote:
I'm sure this has been mentioned, so I apologize if it has.
It should be noted and always thought about that this is one (incomplete) data set. On its own, this yields no real conclusions or even correlations. To make any assumptions you would need other data sets including but not limited to: (a) distribution of races entered into said tournements (b) a metric for judging skill to further polish the numbers (there's no point in including non top players) and (c) some manner of standard deviation or statistical probability -- which is very difficult. Drawing any conclusions from this set of data alone is just as foolish as using Battle.net rankings as a metric of balance. You are correct that this would be meaningless without other data. Fortunately, that other data exists and is easily available, and the statistical techniques required to analyze the data are not hard. In order to get an idea of the ratio of Terran, Zerg, and Protoss players in these tournaments, lets use the ratio for the top 100 players in diamond. http://sc2ranks.com/stats/league/all/1/100/They are not the exact players, but those players are all probably of the caliber required to be competitive in these tournaments, and all we are looking for is an idea of how many players were from each race. It seems unlikely that this subsection gives bad results, since we see the same basic ratio that exists throughout the ladder. So, from that data we expect about 23% of the players to be zerg. If we assume that they are all equally competitive (assuming that one group is less competitive would automatically make the conclusion true), then the expected value for the proportion of zerg wins is 0.23. Now, we can conduct a standard one proportion z test as someone might learn about in any High-school statistics class. I believe most graphing calculators have this function built in (which is how I will do it) and a quick google search will acquaint you with it if you haven't heard of it. The P-value I get from the test is .0001026. This is very small. It means, basically, that we can be 99.9897 percent sure that zerg players could not have a win rate this small through random chance if the three races are equally balanced. You haven't defined what constitutes a balanced race, and you've ignored all external factors. So far you've been able to use numbers to find that a problem exists (Zerg does not win as often as it should in an ideal game state) but you can't recommend corrective measures from that because everything else is still unsolved. If the only map available was Scrap Station, what would the numbers then reflect? You're correct, this only says that there is something causing zergs to win less than would be expected if everything was even. It doesn't necessarily have to be something built into the race. It is entirely possible that something else, like the map pool, is causing zerg to be less likely to win (unfortunately, I don't know of a site that has data on individual maps). Since we can't remove confounding variables like that, the races' win probabilities in that analysis include not only the inherent strength of the race but also things like map imbalance. The data we would need to find a more specific answer or find something to change would be way harder to find, and it may not exist in a form we can access, but that doesn't mean we can't look at the data we do have. Hopefully the SC2 TLPD will be a bit more robust in time, and we'll at least get better access to map information. With VODs and everything, could even include match length.
|
would someone make a fresh one or update this one? I'm curious how things went in the last 2-3 weeks
|
On October 08 2010 21:08 Geo.Rion wrote:
would someone make a fresh one or update this one? I'm curious how things went in the last 2-3 weeks
Well, if you ignore the small stuff and count only the Top Tier Tournaments then it looks something like this:
Terran: 0%
Protoss: 0%
Zerg: 100%
I think it's pretty clear which race needs to be nerfed.
|
On October 08 2010 21:22 OTIX wrote:Show nested quote +On October 08 2010 21:08 Geo.Rion wrote:
would someone make a fresh one or update this one? I'm curious how things went in the last 2-3 weeks Well, if you ignore the small stuff and count only the Top Tier Tournaments then it looks something like this: Terran: 0% Protoss: 0% Zerg: 100% I think it's pretty clear which race needs to be nerfed.
very funny + Show Spoiler +
i can follow the big tournaments, this thread was cool because all little tournaments were counted too, like 50 euro weekly stuff
|
this needs to get updated
|
On October 08 2010 22:30 Let it Raine wrote:
this needs to get updated
I concurr wholeheartedly.
|
Please can a mod update this thread OP. It would help to know how foreigners stand against each other by seeing who's been winning all the tournaments.
either that or incorporate all the information into the TLPD.
|
This is no longer useful since there has already been a patch which is probably why he isn't updating it. You really need to build a new list every patch for it to be meaningful.
|
Wow, Morrow should goto gsl!
|
this should be started for every patch IMO
|
On October 13 2010 06:22 Cyber_Cheese wrote:
this should be started for every patch IMO Yes please.
|
On October 12 2010 06:32 EnderCN wrote:
This is no longer useful since there has already been a patch which is probably why he isn't updating it. You really need to build a new list every patch for it to be meaningful.
On October 13 2010 06:22 Cyber_Cheese wrote:
this should be started for every patch IMO
On October 14 2010 23:11 wrgrbl wrote:Show nested quote +On October 13 2010 06:22 Cyber_Cheese wrote:
this should be started for every patch IMO Yes please.
Done.
http://www.teamliquid.net/forum/viewmessage.php?topic_id=160713
|
|
|
|
|
|
EPT
