|
United States47024 Posts
On March 28 2014 04:31 Sufficiency wrote:
When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions.
Even statistical analyses on NEJM won't utilize models that complicated.
That the complex model is impractical doesn't make the simplified one useful.
|
On March 28 2014 04:33 xes wrote:Show nested quote +On March 28 2014 04:31 Sufficiency wrote:On March 28 2014 04:21 TheYango wrote:On March 28 2014 02:19 Goumindong wrote:On March 28 2014 01:47 TheYango wrote:
It's also just more complex than that. Champion A might be a common pick in teamcomps that are good against champion B, but that doesn't mean that picking champion A in a vacuum will be good against champion B.
It's just muddled in general because you're basically using the working assumption that a team's strength is the sum of its components, when that's not remotely close to being true in this game. The problem is that the number of possible compositions and interaction terms will actually eat our degrees of freedom and make the calculation take way too long. In this case, assuming away those issue, even if you don't believe it, is probably the best option. So long as you state the caveat Except in this case it assumes away so much as to no longer be useful for deriving all but the most obvious conclusions. Personally, I've always had a very poor opinion of trying to analyze drafts through statistics (in both League and DotA) despite being a proponent of applying it to other aspects of the game. The assumptions that need to be made to analyze anything remove all validity they have. Drafting is simply a much more organic process than you can attempt to model with only a basic understanding, and the only people I've seen who really adhere to such statistical analyses on drafts are those with too poor of an understanding of the game to do otherwise. Put simply, unless you can come up with a vastly better model than what's available, the oversimplifications currently available are so inadequate as to be useless. When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions. Even statistical analyses on NEJM won't utilize models that complicated. Please don't drop the "I am a medical researcher" line. It is reserved for Guomin "I am an economist" Dang.
I am not an economist. Therefore what I am doing is an oversimplification that is so inadequate as to be useless.
|
On March 28 2014 04:39 TheYango wrote:Show nested quote +On March 28 2014 04:31 Sufficiency wrote:
When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions.
Even statistical analyses on NEJM won't utilize models that complicated.
That the complex model is impractical doesn't make the simplified one useful.
Out of curiosity, are you a mathematician by training? Because you sound like one and you are being really out of touch here.
|
|
|
Baa?21248 Posts
On March 28 2014 04:41 Sufficiency wrote:Show nested quote +On March 28 2014 04:33 xes wrote:On March 28 2014 04:31 Sufficiency wrote:On March 28 2014 04:21 TheYango wrote:On March 28 2014 02:19 Goumindong wrote:On March 28 2014 01:47 TheYango wrote:
It's also just more complex than that. Champion A might be a common pick in teamcomps that are good against champion B, but that doesn't mean that picking champion A in a vacuum will be good against champion B.
It's just muddled in general because you're basically using the working assumption that a team's strength is the sum of its components, when that's not remotely close to being true in this game. The problem is that the number of possible compositions and interaction terms will actually eat our degrees of freedom and make the calculation take way too long. In this case, assuming away those issue, even if you don't believe it, is probably the best option. So long as you state the caveat Except in this case it assumes away so much as to no longer be useful for deriving all but the most obvious conclusions. Personally, I've always had a very poor opinion of trying to analyze drafts through statistics (in both League and DotA) despite being a proponent of applying it to other aspects of the game. The assumptions that need to be made to analyze anything remove all validity they have. Drafting is simply a much more organic process than you can attempt to model with only a basic understanding, and the only people I've seen who really adhere to such statistical analyses on drafts are those with too poor of an understanding of the game to do otherwise. Put simply, unless you can come up with a vastly better model than what's available, the oversimplifications currently available are so inadequate as to be useless. When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions. Even statistical analyses on NEJM won't utilize models that complicated. Please don't drop the "I am a medical researcher" line. It is reserved for Guomin "I am an economist" Dang. I am not an economist. Therefore what I am doing is an oversimplification that is so inadequate as to be useless.
I'm really glad you came around to this point of view. Let's hope this opens a new chapter in our relationship.
|
On March 28 2014 04:42 Sufficiency wrote:Show nested quote +On March 28 2014 04:39 TheYango wrote:On March 28 2014 04:31 Sufficiency wrote:
When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions.
Even statistical analyses on NEJM won't utilize models that complicated.
That the complex model is impractical doesn't make the simplified one useful. Out of curiosity, are you a mathematician by training? Because you sound like one and you are being really out of touch here.
I'm sorry big bad Yango man is hurting your feelings. But I will have to reluctantly agree with Guomintang and point out that your more causative bend on the statistics is not very useful.
Pentakill data was cool and good, because it is extremely applicable. The champion matchup and lategame/earlygame ones are not (they're useful to not, but not good as models).
|
|
|
|
|
|
|
United States15536 Posts
|
On March 28 2014 04:53 xes wrote:Show nested quote +On March 28 2014 04:42 Sufficiency wrote:On March 28 2014 04:39 TheYango wrote:On March 28 2014 04:31 Sufficiency wrote:
When there are 10 players in a game (i.e. 10 variables), analyzing all possible interactions (even just for 2 way interactions) is quite difficult, regardless of sample size. Very few people go for higher level interactions.
Even statistical analyses on NEJM won't utilize models that complicated.
That the complex model is impractical doesn't make the simplified one useful. Out of curiosity, are you a mathematician by training? Because you sound like one and you are being really out of touch here. I'm sorry big bad Yango man is hurting your feelings. But I will have to reluctantly agree with Guomintang and point out that your more causative bend on the statistics is not very useful. Pentakill data was cool and good, because it is extremely applicable. The champion matchup and lategame/earlygame ones are not (they're useful to not, but not good as models).
If you wish, we can argue about the merits. No models are perfect, and there are always trade offs. A simple model for an exploratory analysis is very useful, in particular, before a more complex model is applied.
I have been avoiding discussing Guomintang's academic diarrhea. He is clearly trying to demonstrate how much he knows about statistics, but I am actually not that impressed.
|
Wow, what do I do against vayne top? Every time I face that I just sit under the turret and hope she wont eventually dive me and get free kills (she does).
|
United States47024 Posts
On March 28 2014 05:05 Sufficiency wrote:
If you wish, we can argue about the merits. No models are perfect, and there are always trade offs. A simple model for an exploratory analysis is very useful, in particular, before a more complex model is applied.
The exploratory analysis is a lot less useful than you make out because in general the amount of rigor needed for something to be established within this sphere is far less than in formal scientific investigation.
You need far less than a formal theory with significant data backing it up to convince a bunch of nerds on the internet playing video games of anything. Which in turn means the practical usefulness of such an approach is far less.
|
i took precalc as a high school freshman, put me in coach
it seems that for alpha > .05, an overcompensating beta will create a charlie foxtrot of poorly fitted regressions when a simple montecristo simulation would have sufficed to reveal an inherent meta/data dissonance
|
Baa?21248 Posts
On March 28 2014 05:18 Kyrie wrote:
i took precalc as a high school freshman, put me in coach
it seems that for alpha > .05, an overcompensating beta will create a charlie foxtrot of poorly fitted regressions when a simple montecristo simulation would have sufficed to reveal an inherent meta/data dissonance
Can we all pause a moment and talk about how amazing this post is?
|
On March 28 2014 05:12 TheYango wrote:Show nested quote +On March 28 2014 05:05 Sufficiency wrote:
If you wish, we can argue about the merits. No models are perfect, and there are always trade offs. A simple model for an exploratory analysis is very useful, in particular, before a more complex model is applied. The exploratory analysis is a lot less useful than you make out because in general the amount of rigor needed for something to be established within this sphere is far less than in formal scientific investigation. You need far less than a formal theory with significant data backing it up to convince a bunch of nerds on the internet playing video games of anything. Which in turn means the practical usefulness of such an approach is far less.
Actually a good exploratory factor analysis would turn me on pretty hard
|
On March 28 2014 05:21 Carnivorous Sheep wrote:Show nested quote +On March 28 2014 05:18 Kyrie wrote:
i took precalc as a high school freshman, put me in coach
it seems that for alpha > .05, an overcompensating beta will create a charlie foxtrot of poorly fitted regressions when a simple montecristo simulation would have sufficed to reveal an inherent meta/data dissonance Can we all pause a moment and talk about how amazing this post is?
i think he needs to add the Rivington Regression to make that model work more accurately for the target audience
|
On March 28 2014 05:12 TheYango wrote:Show nested quote +On March 28 2014 05:05 Sufficiency wrote:
If you wish, we can argue about the merits. No models are perfect, and there are always trade offs. A simple model for an exploratory analysis is very useful, in particular, before a more complex model is applied. The exploratory analysis is a lot less useful than you make out because in general the amount of rigor needed for something to be established within this sphere is far less than in formal scientific investigation. You need far less than a formal theory with significant data backing it up to convince a bunch of nerds on the internet playing video games of anything. Which in turn means the practical usefulness of such an approach is far less.
You are saying my model is an oversimplification so much as to be useless.
Unfortunately, doctors prescribe medicine based on statistical evidence that is even more oversimplified. Bankers invest money based on statistical models that barely make sense. Yet I am just here analyzing data from a video game.
|
On March 28 2014 05:22 Nos- wrote:Show nested quote +On March 28 2014 05:21 Carnivorous Sheep wrote:On March 28 2014 05:18 Kyrie wrote:
i took precalc as a high school freshman, put me in coach
it seems that for alpha > .05, an overcompensating beta will create a charlie foxtrot of poorly fitted regressions when a simple montecristo simulation would have sufficed to reveal an inherent meta/data dissonance Can we all pause a moment and talk about how amazing this post is? i think he needs to add the Rivington Regression to make that model work more accurately for the target audience
on the backside
|
On March 28 2014 05:23 Sufficiency wrote:Show nested quote +On March 28 2014 05:12 TheYango wrote:On March 28 2014 05:05 Sufficiency wrote:
If you wish, we can argue about the merits. No models are perfect, and there are always trade offs. A simple model for an exploratory analysis is very useful, in particular, before a more complex model is applied. The exploratory analysis is a lot less useful than you make out because in general the amount of rigor needed for something to be established within this sphere is far less than in formal scientific investigation. You need far less than a formal theory with significant data backing it up to convince a bunch of nerds on the internet playing video games of anything. Which in turn means the practical usefulness of such an approach is far less. You are saying my model is an oversimplification so much as to be useless. Unfortunately, doctors prescribe medicine based on statistical evidence that is even more oversimplified. Bankers invest money based on statistical models that barely make sense. Yet I am just here analyzing data from a video game.
Actually tho doctors prescribe medication based on a good deal of understanding of pharmacology BACKED by statistical analysis
idk about bankers
|
|
|
|
|
|
EPT
