|
On February 25 2013 00:19 Gimpb wrote:Show nested quote +On February 24 2013 23:48 bananafone wrote: hardly relevant considering the small samplesize. when you get 1000 games you can start talking about tendencies. 150 games however is nothing. The sample size isn't so small that it can be discounted. It's borderline significant based on standard statistical methods. For example, let's say you wanted to know the chance that the true TvZ win probability is >60%. pi = .6 standard error = .0722 Z-obs = 1.329 prob value = 90.8% Read: There is a 90.8% chance that the true TvZ win percentage is greater than 60% Personally, I see the rapid development of strategies as a much bigger detractor than the sample size.
Exactly, people see "small" sample sizes and assume they are flawed, but unless your sample size is just a few games, statistics can be very revealing, as anyone who has studied statistics will know. 150 games is more than enough.
But as you said the rapid development of strategies combined with balance changes by Blizzard and the time (or lack thereof) that each player has spent with HOTS is likely what is causing the statistics to be unstable.
With time, we'll see what happens, but after looking at the buffs and nerfs each race received, I would be shocked if Zerg wasn't the weakest after some time, assuming Blizzard makes no balance changes. While many of their units received buffs, the Infestor carried Zerg and I don't think these buffs make up for the nerf that the Infestor received.
I am also struggling vs Terran as a Protoss, I am apparently missing something...
|
This topic has discussed how its content is flawed therefore not worth taking seriously. It begins with the fact that the author himself didn't even understand the difference between winrates and maprates, after that you see that the sample size is way to small and to top it off, these games where played on old patches and with strategy's that are no longer viable.
What is it that you want to discuss again?
While the data may no longer be relevant to the current patch it is by no means useless. It does shed some light on why blizzard made such strong nerfs to hellbat strategies. It also confirms what most of us have said over the course of the beta. Essentially that:
Protoss is strong against Terran. Protoss is somewhat strong against Zerg Terran is really strong against Zerg.
I have the feeling that most people don't like this because they don't like their success in beta to be discounted as racial imbalance. Just like zerg didn't like to admit to it on WOL prior to the infestor nerfs.
It's the beta...it's going to be imbalanced. This is just a little look back at how things look so far prior to launch.
I think what we can gleam from this is that Terran is likely going to need some help v Toss as it's still pretty Toss favored and it may get worse with the nerfs to hellback strats. We'll have to wait and see on ZVT since it's just changed a bit, and there may be a case for a slight buff to zerg anti-air late game VS both races.
|
Because Zergs are busing dominating WoL, so they haven't really swapped over yet =/
But in all honesty, Zerg's strength comes through the ability to be "safe" because of their production and how they can crush anything short of dedicated attacks. Recall and Speedvacs nullify a lot of that to an extent and let harassment happen much more frequently/earlier than the MU used to dictate.
As far as PvT goes, I think Terrans just need to learn new builds. The old 1 Rax doesn't work anymore, but for the most part, standard Toss translates OK and now the Terran needs to worry about varied all ins, while they themselves don't have anything very new to pressure the Toss early game.
|
Interesting, I thought PvT would be in favor of terran (My weakest main mu is pvt, my strongest offrace mu is tvp and those of my friends who play terran/random has tvp as their strongest).
But then again, we're just diamond scrubs :p
|
On February 25 2013 01:50 Supah wrote: The old 1 Rax doesn't work anymore, but for the most part, standard Toss translates OK and now the Terran needs to worry about varied all ins, while they themselves don't have anything very new to pressure the Toss early game.
Even as a Protoss player, it makes me angry that Terran has nothing it can pressure with the Protoss with early because of the Nexus Cannon. All of the old barracks play is completely useless and it has ruined the variety of the game. Sure, they can still use Banshees, Hellion/Widow Mine Drops and Reaper to harass, but it isn't that difficult to stop honestly.
If there is no way to do damage in the early game, why doesn't Blizzard just start out both races with two bases and some basic structures? What really is the purpose?
|
supercool tnx i hope you keep track of this and next week/month do another one
|
On February 25 2013 01:49 BronzeKnee wrote:Show nested quote +On February 25 2013 00:19 Gimpb wrote:On February 24 2013 23:48 bananafone wrote: hardly relevant considering the small samplesize. when you get 1000 games you can start talking about tendencies. 150 games however is nothing. The sample size isn't so small that it can be discounted. It's borderline significant based on standard statistical methods. For example, let's say you wanted to know the chance that the true TvZ win probability is >60%. pi = .6 standard error = .0722 Z-obs = 1.329 prob value = 90.8% Read: There is a 90.8% chance that the true TvZ win percentage is greater than 60% Personally, I see the rapid development of strategies as a much bigger detractor than the sample size. Exactly, people see "small" sample sizes and assume they are flawed, but unless your sample size is just a few games, statistics can be very revealing, as anyone who has studied statistics will know. 150 games is more than enough. But as you said the rapid development of strategies combined with balance changes by Blizzard and the time (or lack thereof) that each player has spent with HOTS is likely what is causing the statistics to be unstable. With time, we'll see what happens, but after looking at the buffs and nerfs each race received, I would be shocked if Zerg wasn't the weakest after some time, assuming Blizzard makes no balance changes. While many of their units received buffs, the Infestor carried Zerg and I don't think these buffs make up for the nerf that the Infestor received. I am also struggling vs Terran as a Protoss, I am apparently missing something... Actually, anyone who has studied statistics would know that the individual samples would need to be completely randomly selected to draw conclusions based upon them. Unfortunately, that is not the case here.
|
The stats aren't perfect obviously, but since they confirm what I suspected (ZvP isn't as bad as people say, ZvT is much more problematic) I'll just assume they are :p
|
On February 25 2013 01:57 JDub wrote:Show nested quote +On February 25 2013 01:49 BronzeKnee wrote:On February 25 2013 00:19 Gimpb wrote:On February 24 2013 23:48 bananafone wrote: hardly relevant considering the small samplesize. when you get 1000 games you can start talking about tendencies. 150 games however is nothing. The sample size isn't so small that it can be discounted. It's borderline significant based on standard statistical methods. For example, let's say you wanted to know the chance that the true TvZ win probability is >60%. pi = .6 standard error = .0722 Z-obs = 1.329 prob value = 90.8% Read: There is a 90.8% chance that the true TvZ win percentage is greater than 60% Personally, I see the rapid development of strategies as a much bigger detractor than the sample size. Exactly, people see "small" sample sizes and assume they are flawed, but unless your sample size is just a few games, statistics can be very revealing, as anyone who has studied statistics will know. 150 games is more than enough. But as you said the rapid development of strategies combined with balance changes by Blizzard and the time (or lack thereof) that each player has spent with HOTS is likely what is causing the statistics to be unstable. With time, we'll see what happens, but after looking at the buffs and nerfs each race received, I would be shocked if Zerg wasn't the weakest after some time, assuming Blizzard makes no balance changes. While many of their units received buffs, the Infestor carried Zerg and I don't think these buffs make up for the nerf that the Infestor received. I am also struggling vs Terran as a Protoss, I am apparently missing something... Actually, anyone who has studied statistics would know that the individual samples would need to be completely randomly selected to draw conclusions based upon them. Unfortunately, that is not the case here.
That was certainly not my experience when I did psychological testing in college, you have to control variables to make a good assessment, and thus it can't be completely random. Maybe if you are testing water in a river or something sure, you'd want random samples... but in this case you can only use what you have.
In other words, if you wanted to find out whether college aged males had better reflexes than high school aged males, you control the variables of age and sex. There would be be no point to testing females, or post college or pre-high school males. The same can be said here, we are controlling for high level play by looking at tournaments.
Making assessments from completely random HOTS games is terrible idea if that is what you are suggesting. The winrates in the lower leagues probably look a lot different, but as players get better and learn new skills, they get into the higher leagues and begin to experience issues that higher level players have. Also, higher level players have solved many of the issues that player experience in lower leagues.
Thus we should control the variable of skill as much as possible, by only looking at tournaments and high level play.
Certainly, if there were more tournaments out there they should be included, but this is fine for what it is.
|
I didn't really watch many of the pro games lately except Axiom vs IM, and I was wondering if most of the TvZ's were won by Terran doing Hellbat early game stuff?
Like is the recent patch with the Hellions unable to transform right aways going to help that?
|
On February 25 2013 02:02 BronzeKnee wrote:Show nested quote +On February 25 2013 01:57 JDub wrote:On February 25 2013 01:49 BronzeKnee wrote:On February 25 2013 00:19 Gimpb wrote:On February 24 2013 23:48 bananafone wrote: hardly relevant considering the small samplesize. when you get 1000 games you can start talking about tendencies. 150 games however is nothing. The sample size isn't so small that it can be discounted. It's borderline significant based on standard statistical methods. For example, let's say you wanted to know the chance that the true TvZ win probability is >60%. pi = .6 standard error = .0722 Z-obs = 1.329 prob value = 90.8% Read: There is a 90.8% chance that the true TvZ win percentage is greater than 60% Personally, I see the rapid development of strategies as a much bigger detractor than the sample size. Exactly, people see "small" sample sizes and assume they are flawed, but unless your sample size is just a few games, statistics can be very revealing, as anyone who has studied statistics will know. 150 games is more than enough. But as you said the rapid development of strategies combined with balance changes by Blizzard and the time (or lack thereof) that each player has spent with HOTS is likely what is causing the statistics to be unstable. With time, we'll see what happens, but after looking at the buffs and nerfs each race received, I would be shocked if Zerg wasn't the weakest after some time, assuming Blizzard makes no balance changes. While many of their units received buffs, the Infestor carried Zerg and I don't think these buffs make up for the nerf that the Infestor received. I am also struggling vs Terran as a Protoss, I am apparently missing something... Actually, anyone who has studied statistics would know that the individual samples would need to be completely randomly selected to draw conclusions based upon them. Unfortunately, that is not the case here. That was certainly not my experience when I did psychological testing in college, you have to control variables to make a good assessment, and thus it can't be completely random. Maybe if you are testing water in a river or something sure, you'd want random samples... but in this case you can only use what you have. In other words, if you wanted to find out whether college aged males had better reflexes than high school aged males, you control the variables of age and sex. There would be be no point to testing females, or post college or pre-high school males. The same can be said here, we are controlling for high level play by looking at tournaments. Making assessments from completely random HOTS games is terrible idea if that is what you are suggesting. The winrates in the lower leagues probably look a lot different, but as players get better and learn new skills, they get into the higher leagues and begin to experience issues that higher level players have. Also, higher level players have solved many of the issues that player experience in lower leagues. Thus we should control the variable of skill as much as possible, by only looking at tournaments and high level play. Certainly, if there were more tournaments out there they should be included, but this is fine for what it is.
That's exactly the issue with the data I think. There are too many uncontrolled variables to really speak of valid significant findings. That doesn't mean the data is completely useless of course. So I agree with you - it's fine for what it is.
|
Zerg playing hots like its wol in zvt imo. Zerg metagame is so slow.
|
On February 25 2013 01:57 MilesTeg wrote: The stats aren't perfect obviously, but since they confirm what I suspected (ZvP isn't as bad as people say, ZvT is much more problematic) I'll just assume they are :p
People don't like ZvP lategame. And in those tournaments the game ended usually before lategame. That's why we don't see it in these statistics. Learning your build to the lategame takes more time and that's why games will usually end rather quickly in tournaments if the game is new.
|
On February 25 2013 02:02 BronzeKnee wrote:Show nested quote +On February 25 2013 01:57 JDub wrote:On February 25 2013 01:49 BronzeKnee wrote:On February 25 2013 00:19 Gimpb wrote:On February 24 2013 23:48 bananafone wrote: hardly relevant considering the small samplesize. when you get 1000 games you can start talking about tendencies. 150 games however is nothing. The sample size isn't so small that it can be discounted. It's borderline significant based on standard statistical methods. For example, let's say you wanted to know the chance that the true TvZ win probability is >60%. pi = .6 standard error = .0722 Z-obs = 1.329 prob value = 90.8% Read: There is a 90.8% chance that the true TvZ win percentage is greater than 60% Personally, I see the rapid development of strategies as a much bigger detractor than the sample size. Exactly, people see "small" sample sizes and assume they are flawed, but unless your sample size is just a few games, statistics can be very revealing, as anyone who has studied statistics will know. 150 games is more than enough. But as you said the rapid development of strategies combined with balance changes by Blizzard and the time (or lack thereof) that each player has spent with HOTS is likely what is causing the statistics to be unstable. With time, we'll see what happens, but after looking at the buffs and nerfs each race received, I would be shocked if Zerg wasn't the weakest after some time, assuming Blizzard makes no balance changes. While many of their units received buffs, the Infestor carried Zerg and I don't think these buffs make up for the nerf that the Infestor received. I am also struggling vs Terran as a Protoss, I am apparently missing something... Actually, anyone who has studied statistics would know that the individual samples would need to be completely randomly selected to draw conclusions based upon them. Unfortunately, that is not the case here. That was certainly not my experience when I did psychological testing in college, you have to control variables to make a good assessment, and thus it can't be completely random. Maybe if you are testing water in a river or something sure, you'd want random samples... but in this case you can only use what you have. In other words, if you wanted to find out whether college aged males had better reflexes than high school aged males, you control the variables of age and sex. There would be be no point to testing females, or post college or pre-high school males. The same can be said here, we are controlling for high level play by looking at tournaments. Making assessments from completely random HOTS games is terrible idea if that is what you are suggesting. The winrates in the lower leagues probably look a lot different, but as players get better and learn new skills, they get into the higher leagues and begin to experience issues that higher level players have. Also, higher level players have solved many of the issues that player experience in lower leagues. Thus we should control the variable of skill as much as possible, by only looking at tournaments and high level play. Certainly, if there were more tournaments out there they should be included, but this is fine for what it is. I mean random selection within the target population. If the target is pro level HotS, you need the game selection to be a random selection with regards to the variables within the population, and any two samples should not be correlated (e.g. If two games were part of the same series, then they are not independent and you have faulty sampling).
My point isn't that the numbers are utterly useless. My point is that the mathematical analysis of probabilities requires randomly selected independent samples, which the HotS games so far are not. So the argument that the sample size is big enough is moot.
|
|
|
|