|
On May 02 2013 03:25 BronzeKnee wrote:Show nested quote +On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm reject the statistics. There is a very simple argument for only using professional games and ignoring ladder. Ladder automatically correct win rates to 50%. Thus if the game favors Zerg, a medicore Zerg player will play a good Terran, yet the ladder win rates will not show this entirely. Using professional games shows how the game is played at the highest level. While a 4 Gate might seem overpowered in Gold, a Platinum player might have the skills to hold it easy. And a Gold player could learn those skills, and then the 4 Gate doesn't seem overpowered. Thus the only way to balance the game is at the top, everyone below simply needs to learn the skills necessary to get to the top before they can complain about balance. Map balance is related to racial balance. The statistics show that with the current balance and maps, Terran has an advantage over Zerg. Perhaps it is balance, perhaps it is the map. Unfortunately, these two variables are intertwined and can be difficult to seperate. In other words, since Starcraft games have to be played on a map, and strategies are developed based on the map and the strengths and weaknesses of each race the go hand in hand. Remember the Stephano 200 Roach push in PvZ? It lead to the creation of a map pool where thirds were easy to defend because it was easy to deny a Protoss third with that push on some maps. Furthermore because strategies are developed over a period of time, you can't just use random maps as the independent in determining balance unless tournaments began using random maps (meaning people wouldn't be able to plan strategies for maps and would have adjust on the fly). Finally, the methodology used is fine. Looking at the win percentages of each race in professional tournaments and comparing them is very useful. Are there uncontrolled variables? Of course. Player skill and latency are huge problems that would be very difficult to control for (though we did control for player skill by picking from tournament games). Also differing map pools between tournaments might lead to different win rates. However, with a large enough sample size, latency and player skill be should evenly effect all races and it is probably worth controlling l for the map pool to some extent.
My methods knowledge is't great but I think there might be a way to run the data in an improved fashion. You use a probit model with the dependent variable as Winning and the independent variable as say being terran (0 if zerg, 1 if terran). I'm not sure how you can control for player skill. I don't know if there are a few individuals that play enough games such that you can run a fixed effects model to take out the individual level characteristics. But you might be able to do something along the lines of using a skill gap, that is the difference in overall win percentage between the players. I'm not sure how best to implement that. Then you control for map, use a dummy variable for each map to take out the map effect. Then control for game length.
What you're trying to do is take the effect of everything else out of the impact of being terran on winning in a TvZ format. If someone wants to provide me the data set, I'll run the model.
Again, my methods aren't great, still learning, but if anyone has any ideas on improvement, that would be good. Then if we can get the data we can improve on our understanding of the match up.
|
Zergs won the first 2 major tourneys, they can't complain much tbh
|
On May 02 2013 03:37 Eventine wrote:Show nested quote +On May 02 2013 03:25 BronzeKnee wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm reject the statistics. There is a very simple argument for only using professional games and ignoring ladder. Ladder automatically correct win rates to 50%. Thus if the game favors Zerg, a medicore Zerg player will play a good Terran, yet the ladder win rates will not show this entirely. Using professional games shows how the game is played at the highest level. While a 4 Gate might seem overpowered in Gold, a Platinum player might have the skills to hold it easy. And a Gold player could learn those skills, and then the 4 Gate doesn't seem overpowered. Thus the only way to balance the game is at the top, everyone below simply needs to learn the skills necessary to get to the top before they can complain about balance. Map balance is related to racial balance. The statistics show that with the current balance and maps, Terran has an advantage over Zerg. Perhaps it is balance, perhaps it is the map. Unfortunately, these two variables are intertwined and can be difficult to seperate. In other words, since Starcraft games have to be played on a map, and strategies are developed based on the map and the strengths and weaknesses of each race the go hand in hand. Remember the Stephano 200 Roach push in PvZ? It lead to the creation of a map pool where thirds were easy to defend because it was easy to deny a Protoss third with that push on some maps. Furthermore because strategies are developed over a period of time, you can't just use random maps as the independent in determining balance unless tournaments began using random maps (meaning people wouldn't be able to plan strategies for maps and would have adjust on the fly). Finally, the methodology used is fine. Looking at the win percentages of each race in professional tournaments and comparing them is very useful. Are there uncontrolled variables? Of course. Player skill and latency are huge problems that would be very difficult to control for (though we did control for player skill by picking from tournament games). Also differing map pools between tournaments might lead to different win rates. However, with a large enough sample size, latency and player skill be should evenly effect all races and it is probably worth controlling l for the map pool to some extent. My methods knowledge is't great but I think there might be a way to run the data in an improved fashion. You use a probit model with the dependent variable as Winning and the independent variable as say being terran (0 if zerg, 1 if terran). I'm not sure how you can control for player skill. I don't know if there are a few individuals that play enough games such that you can run a fixed effects model to take out the individual level characteristics. But you might be able to do something along the lines of using a skill gap, that is the difference in overall win percentage between the players. I'm not sure how best to implement that. Then you control for map, use a dummy variable for each map to take out the map effect. Then control for game length. What you're trying to do is take the effect of everything else out of the impact of being terran on winning in a TvZ format. If someone wants to provide me the data set, I'll run the model. Again, my methods aren't great, still learning, but if anyone has any ideas on improvement, that would be good. Then if we can get the data we can improve on our understanding of the match up. It's doable actually, You simply perform some qualitive based research and find a very clear link between skill factors and run them as criteria against a current pool of players and rank them skillwise then just pair up players against each other with the closest skill cap and run some quantative (statistical) research on those groups only. That would actually mean something.
|
On May 02 2013 03:27 Integra wrote:Show nested quote +On May 02 2013 03:22 Alryk wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm or reject the statistics. That sc2 statistics guy did the exact same thing for about 2 years and nobody complained. And obviously we aren't looking at gold etc. statistics because it takes too much work. The majority of people however are interested in balancing the game at the pro level, not for bronze players. And while you don't want it to be impossible for a bronze zerg to win, it doesn't make sense to make changes FOR them. Badly designed maps are a fact of life - there's no way around them. The statistics aren't trying to show that. They're simply showing game balance as is. If you notice, the OP didn't actually bring his opinions into the actual statistics. For what they're intended to represent, they work well. Are they perfect? no not really. But pointing out the things you did makes no sense; the survey doesn't aim to address map balance. Also, the low vs high latency games are so few and far between that they're drowned out by normal games. Things like that. I don't really feel like explaining all of statistics though lol. You point out bias, but it isn't relevant bias. It's not something they can prevent. How would you gather statistics that determined how good a map was? That's almost entirely subjective. So by other words, it could mean ANYTHING, and thus is just useless statitsics with no real intent or clear focus or aim. @BronzeKnee; thank you for proving my point regarding speculation and making their own shit up regarding what it it could mean.
Actually, what they literally mean is simple: zergs are the least winning race right now. Period. Do we know why? No, but that doesn't matter right now. The important thing is to watch and see if this trend continues or changes. If win rates are different next month or 6 months from now, then there is no cause for alarm. However, if this trend keeps up, THEN we need to look for explanations.
|
Please explain your process :D
I'd love to know: a: which tournaments b: total number of games per race/match-up
|
On May 02 2013 03:44 tili wrote: Please explain your process :D
I'd love to know: a: which tournaments b: total number of games per race/match-up
He kind of explained (a) in the first post. And (b) is shown in the bottom right square.
|
On May 02 2013 03:44 LoveBuzz wrote:Show nested quote +On May 02 2013 03:27 Integra wrote:On May 02 2013 03:22 Alryk wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm or reject the statistics. That sc2 statistics guy did the exact same thing for about 2 years and nobody complained. And obviously we aren't looking at gold etc. statistics because it takes too much work. The majority of people however are interested in balancing the game at the pro level, not for bronze players. And while you don't want it to be impossible for a bronze zerg to win, it doesn't make sense to make changes FOR them. Badly designed maps are a fact of life - there's no way around them. The statistics aren't trying to show that. They're simply showing game balance as is. If you notice, the OP didn't actually bring his opinions into the actual statistics. For what they're intended to represent, they work well. Are they perfect? no not really. But pointing out the things you did makes no sense; the survey doesn't aim to address map balance. Also, the low vs high latency games are so few and far between that they're drowned out by normal games. Things like that. I don't really feel like explaining all of statistics though lol. You point out bias, but it isn't relevant bias. It's not something they can prevent. How would you gather statistics that determined how good a map was? That's almost entirely subjective. So by other words, it could mean ANYTHING, and thus is just useless statitsics with no real intent or clear focus or aim. @BronzeKnee; thank you for proving my point regarding speculation and making their own shit up regarding what it it could mean. Actually, what they literally mean is simple: zergs are the least winning race right now. Period. Do we know why? No, but that doesn't matter right now. The important thing is to watch and see if this trend continues or changes. If win rates are different next month or 6 months from now, then there is no cause for alarm. However, if this trend keeps up, THEN we need to look for explanations. OR until somene gives us another size or context sample, the manipulation of the data alone is enough to change the outcome of the statistics.
|
Id say buff the swarmhost a bit since it's underused and then sit back and wait. Possibly some buff that makes it more viable vs terran so it wont tip the scale vs toss.
|
1561 Posts
|
On May 02 2013 03:43 Integra wrote:Show nested quote +On May 02 2013 03:37 Eventine wrote:On May 02 2013 03:25 BronzeKnee wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm reject the statistics. There is a very simple argument for only using professional games and ignoring ladder. Ladder automatically correct win rates to 50%. Thus if the game favors Zerg, a medicore Zerg player will play a good Terran, yet the ladder win rates will not show this entirely. Using professional games shows how the game is played at the highest level. While a 4 Gate might seem overpowered in Gold, a Platinum player might have the skills to hold it easy. And a Gold player could learn those skills, and then the 4 Gate doesn't seem overpowered. Thus the only way to balance the game is at the top, everyone below simply needs to learn the skills necessary to get to the top before they can complain about balance. Map balance is related to racial balance. The statistics show that with the current balance and maps, Terran has an advantage over Zerg. Perhaps it is balance, perhaps it is the map. Unfortunately, these two variables are intertwined and can be difficult to seperate. In other words, since Starcraft games have to be played on a map, and strategies are developed based on the map and the strengths and weaknesses of each race the go hand in hand. Remember the Stephano 200 Roach push in PvZ? It lead to the creation of a map pool where thirds were easy to defend because it was easy to deny a Protoss third with that push on some maps. Furthermore because strategies are developed over a period of time, you can't just use random maps as the independent in determining balance unless tournaments began using random maps (meaning people wouldn't be able to plan strategies for maps and would have adjust on the fly). Finally, the methodology used is fine. Looking at the win percentages of each race in professional tournaments and comparing them is very useful. Are there uncontrolled variables? Of course. Player skill and latency are huge problems that would be very difficult to control for (though we did control for player skill by picking from tournament games). Also differing map pools between tournaments might lead to different win rates. However, with a large enough sample size, latency and player skill be should evenly effect all races and it is probably worth controlling l for the map pool to some extent. My methods knowledge is't great but I think there might be a way to run the data in an improved fashion. You use a probit model with the dependent variable as Winning and the independent variable as say being terran (0 if zerg, 1 if terran). I'm not sure how you can control for player skill. I don't know if there are a few individuals that play enough games such that you can run a fixed effects model to take out the individual level characteristics. But you might be able to do something along the lines of using a skill gap, that is the difference in overall win percentage between the players. I'm not sure how best to implement that. Then you control for map, use a dummy variable for each map to take out the map effect. Then control for game length. What you're trying to do is take the effect of everything else out of the impact of being terran on winning in a TvZ format. If someone wants to provide me the data set, I'll run the model. Again, my methods aren't great, still learning, but if anyone has any ideas on improvement, that would be good. Then if we can get the data we can improve on our understanding of the match up. It's doable actually, You simply perform some qualitive based research and find a very clear link between skill factors and run them as criteria against a current pool of players and rank them skillwise then just pair up players against each other with the closest skill cap and run some quantative (statistical) research on those groups only. That would actually mean something.
My fear in that scenario is sample size. It's not even difficult to rank them, we can use the TL ranking systems to do so. But if we constrain our sample to say only the top 5 T and Z players, I don't know if we'll have large amount of samples, especially given the degrees of freedom I want to use to up control for maps. Again, it's all conjecture without seeing the data.
Continuing with thinking process, I think using the close skill level, we'll quickly jump down on the actual skill. It's easier to find a similarly skilled players at lower levels of competition. There is a possibility that different races have different skill caps. Potentially, this method could be more vulnerable to challenges than a less than perfect skill gap measurement.
|
|
On May 02 2013 03:57 Emzeeshady wrote:Show nested quote +On May 02 2013 03:41 tuho12345 wrote: Zergs won the first 2 major tourneys, they can't complain much tbh What kind of flawed logic is this. Since 2 or 3 Zergs had a really good tournament the game is balance?
Zerg also won the first 2 gsl, I don't think there's too much debate over the terran imbalance during that time. Speaking of small sample size...
|
On May 02 2013 03:54 Eventine wrote:Show nested quote +On May 02 2013 03:43 Integra wrote:On May 02 2013 03:37 Eventine wrote:On May 02 2013 03:25 BronzeKnee wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm reject the statistics. There is a very simple argument for only using professional games and ignoring ladder. Ladder automatically correct win rates to 50%. Thus if the game favors Zerg, a medicore Zerg player will play a good Terran, yet the ladder win rates will not show this entirely. Using professional games shows how the game is played at the highest level. While a 4 Gate might seem overpowered in Gold, a Platinum player might have the skills to hold it easy. And a Gold player could learn those skills, and then the 4 Gate doesn't seem overpowered. Thus the only way to balance the game is at the top, everyone below simply needs to learn the skills necessary to get to the top before they can complain about balance. Map balance is related to racial balance. The statistics show that with the current balance and maps, Terran has an advantage over Zerg. Perhaps it is balance, perhaps it is the map. Unfortunately, these two variables are intertwined and can be difficult to seperate. In other words, since Starcraft games have to be played on a map, and strategies are developed based on the map and the strengths and weaknesses of each race the go hand in hand. Remember the Stephano 200 Roach push in PvZ? It lead to the creation of a map pool where thirds were easy to defend because it was easy to deny a Protoss third with that push on some maps. Furthermore because strategies are developed over a period of time, you can't just use random maps as the independent in determining balance unless tournaments began using random maps (meaning people wouldn't be able to plan strategies for maps and would have adjust on the fly). Finally, the methodology used is fine. Looking at the win percentages of each race in professional tournaments and comparing them is very useful. Are there uncontrolled variables? Of course. Player skill and latency are huge problems that would be very difficult to control for (though we did control for player skill by picking from tournament games). Also differing map pools between tournaments might lead to different win rates. However, with a large enough sample size, latency and player skill be should evenly effect all races and it is probably worth controlling l for the map pool to some extent. My methods knowledge is't great but I think there might be a way to run the data in an improved fashion. You use a probit model with the dependent variable as Winning and the independent variable as say being terran (0 if zerg, 1 if terran). I'm not sure how you can control for player skill. I don't know if there are a few individuals that play enough games such that you can run a fixed effects model to take out the individual level characteristics. But you might be able to do something along the lines of using a skill gap, that is the difference in overall win percentage between the players. I'm not sure how best to implement that. Then you control for map, use a dummy variable for each map to take out the map effect. Then control for game length. What you're trying to do is take the effect of everything else out of the impact of being terran on winning in a TvZ format. If someone wants to provide me the data set, I'll run the model. Again, my methods aren't great, still learning, but if anyone has any ideas on improvement, that would be good. Then if we can get the data we can improve on our understanding of the match up. It's doable actually, You simply perform some qualitive based research and find a very clear link between skill factors and run them as criteria against a current pool of players and rank them skillwise then just pair up players against each other with the closest skill cap and run some quantative (statistical) research on those groups only. That would actually mean something. My fear in that scenario is sample size. It's not even difficult to rank them, we can use the TL ranking systems to do so. But if we constrain our sample to say only the top 5 T and Z players, I don't know if we'll have large amount of samples, especially given the degrees of freedom I want to use to up control for maps. Again, it's all conjecture without seeing the data. Continuing with thinking process, I think using the close skill level, we'll quickly jump down on the actual skill. It's easier to find a similarly skilled players at lower levels of competition. There is a possibility that different races have different skill caps. Potentially, this method could be more vulnerable to challenges than a less than perfect skill gap measurement. no it is doable. What I was talking about is a proven metholodgy to measure performance in sports and its very reilable, even from small samples given the research is done properly,
And regarding skill cap for various races: simply do mirror games for starters and then do comparative analysis with mixed MU's, find the differences, explain them, find way to quanitifie and map them out, should not be a problem either.
|
On May 02 2013 03:59 Integra wrote:Show nested quote +On May 02 2013 03:54 Eventine wrote:On May 02 2013 03:43 Integra wrote:On May 02 2013 03:37 Eventine wrote:On May 02 2013 03:25 BronzeKnee wrote:On May 02 2013 03:05 Integra wrote:On May 02 2013 02:52 Alryk wrote:On May 02 2013 02:45 synd wrote: I don't trust this statistics at all Any reasoning for that lol? 52/48 is almost definitely within margin of error, although we don't know exactly what the margin of error is. I don't know if including qualifiers was a good idea because of the potential for amateur vs pro matches, but there should be so few of them included that they get drowned out in the sample anyways. Kind of interesting statistics, and I'm not too too surprised by what I see. Even 55/45 is potentially close to margin of error, although in this case it's unlikely. I hope we see buffs to zerg and not nerfs to whatever is necessary - it's usually a much more entertaining way of doing the game. Because it's bullshit, nor sample size, how relevant the sample size or that the numbers actually are showing "balance" are determined or properly explained.How do we know that the sample size is represental for the game a a whole, how accurate does it reflect the MU's on other leagues such as gold, plat, diamond etc, How does it compare to different regions, like EU, NA and KR etc. How do we know that these numbers only relate to race balance issues, it could just as well showing bad designed maps, people with actual lower skill meeting other players with better skill or just the fact that certain players are better at playing on high latency and that just happens to be Terran players, or maybe Terran players ARE easier to play during latency compared to the other races. No methodlogy for data collecting has been determined and no attempt has been made to actually isolate the important variables for proving if one race is better than another, we don't even know what those variables are People will only start to speculate and make their own shit up as of why this has happened, and depeding of their personal bias they will either confirm reject the statistics. There is a very simple argument for only using professional games and ignoring ladder. Ladder automatically correct win rates to 50%. Thus if the game favors Zerg, a medicore Zerg player will play a good Terran, yet the ladder win rates will not show this entirely. Using professional games shows how the game is played at the highest level. While a 4 Gate might seem overpowered in Gold, a Platinum player might have the skills to hold it easy. And a Gold player could learn those skills, and then the 4 Gate doesn't seem overpowered. Thus the only way to balance the game is at the top, everyone below simply needs to learn the skills necessary to get to the top before they can complain about balance. Map balance is related to racial balance. The statistics show that with the current balance and maps, Terran has an advantage over Zerg. Perhaps it is balance, perhaps it is the map. Unfortunately, these two variables are intertwined and can be difficult to seperate. In other words, since Starcraft games have to be played on a map, and strategies are developed based on the map and the strengths and weaknesses of each race the go hand in hand. Remember the Stephano 200 Roach push in PvZ? It lead to the creation of a map pool where thirds were easy to defend because it was easy to deny a Protoss third with that push on some maps. Furthermore because strategies are developed over a period of time, you can't just use random maps as the independent in determining balance unless tournaments began using random maps (meaning people wouldn't be able to plan strategies for maps and would have adjust on the fly). Finally, the methodology used is fine. Looking at the win percentages of each race in professional tournaments and comparing them is very useful. Are there uncontrolled variables? Of course. Player skill and latency are huge problems that would be very difficult to control for (though we did control for player skill by picking from tournament games). Also differing map pools between tournaments might lead to different win rates. However, with a large enough sample size, latency and player skill be should evenly effect all races and it is probably worth controlling l for the map pool to some extent. My methods knowledge is't great but I think there might be a way to run the data in an improved fashion. You use a probit model with the dependent variable as Winning and the independent variable as say being terran (0 if zerg, 1 if terran). I'm not sure how you can control for player skill. I don't know if there are a few individuals that play enough games such that you can run a fixed effects model to take out the individual level characteristics. But you might be able to do something along the lines of using a skill gap, that is the difference in overall win percentage between the players. I'm not sure how best to implement that. Then you control for map, use a dummy variable for each map to take out the map effect. Then control for game length. What you're trying to do is take the effect of everything else out of the impact of being terran on winning in a TvZ format. If someone wants to provide me the data set, I'll run the model. Again, my methods aren't great, still learning, but if anyone has any ideas on improvement, that would be good. Then if we can get the data we can improve on our understanding of the match up. It's doable actually, You simply perform some qualitive based research and find a very clear link between skill factors and run them as criteria against a current pool of players and rank them skillwise then just pair up players against each other with the closest skill cap and run some quantative (statistical) research on those groups only. That would actually mean something. My fear in that scenario is sample size. It's not even difficult to rank them, we can use the TL ranking systems to do so. But if we constrain our sample to say only the top 5 T and Z players, I don't know if we'll have large amount of samples, especially given the degrees of freedom I want to use to up control for maps. Again, it's all conjecture without seeing the data. Continuing with thinking process, I think using the close skill level, we'll quickly jump down on the actual skill. It's easier to find a similarly skilled players at lower levels of competition. There is a possibility that different races have different skill caps. Potentially, this method could be more vulnerable to challenges than a less than perfect skill gap measurement. no it is doable. What I was talking about is a proven metholodgy to measure performance in sports and its very reilable, even from small samples given the research is done properly, And regarding skill cap for various races: simply do mirror games for starters and then do comparative analysis with mixed MU's, find the differences, explain them, find way to quanitifie and map them out, should not be a problem either.
Okay. I don't know the methodology to do that... So I'm just sounding out concerns without knowing what that model would do.
|
Pie charts are probably the least efficient and lease useful method of presenting information.
Regardless, I have a hard time believing these statistics simply because of he lack of references or major tournaments that went into this calculation. Sample size feels small to me too, but I don't know whether it actually is or not as I am not a statistician.
|
On May 02 2013 04:17 Butterednuts wrote: Pie charts are probably the least efficient and lease useful method of presenting information.
Regardless, I have a hard time believing these statistics simply because of he lack of references or major tournaments that went into this calculation. Sample size feels small to me too, but I don't know whether it actually is or not as I am not a statistician.
Given the number of tournaments that were said to be included in the data set (I don't know what other tournament you would add into the model), we're more likely looking at the results of the population data instead of a sample drawn from the population. In that case, there's no reason to be concerned with sample size.
Also, if we can consider the data set to be the entire population, we shouldn't even worry about a confidence interval, the results are the results. Though, given that it's constrained to one month, you can make the legitimate argument that it's not the population set.
|
I noticed some vital matchup information was missing in the opening post. I took the liberty of creating some extra pie charts to further our understanding of the MU balance within HotS. Please add it to the OP.
http://i.imgur.com/vecxYX4.jpg
In all seriousness though, these kind of 'statistics' threads are silly.
|
to me this seems pretty balanced.
|
Using the google spreadsheet posted in some places I also looked at that data. While the TvZ winrate isn't completely off-center due to WCS europe, it does have a very significant influence. After deleting that row it suddenly was alot more even (and result was that you should nerf toss).
And WCS europe qualifiers (same for NA btw) is a bit hard to justify to use imo, since a larger number of zerg players than terran players were already invited. But regardless if you should use it or not, if a single tournament can make such a large shift the main conclusion is that sample size is too low to make accurate observations. Aditionally all races are still experimenting with the new units, so for sure you don't want large balance changes.
Finally while ladder is in principle always 50% win rate due to matchmaking, it is also true that in every game the strongest items, race, weapon, etc always seems to attract alot of people. And looking at SC2ranks is it absolutely not that terran has the most players, actually the opposite, it has the least players. While zerg in general for gold and higher leagues has most players.
|
On May 02 2013 03:22 Eventine wrote: always nice to see people with demands and complaints on the data and a lack of commitment to actually provide "better" data or analysis.
Probably because:
"all pro data" = too small sample size "all ladder data" = way too noob "all tourny data" = korea vs foreign skew "all top ladder" = too noob/cross region issues "all top korea" = terran-culture/too small sample/(variable complaint here)
In short, I don't think the SC2 community will ever agree on any statistical analysis.
|
|
|
|