|
|
On January 05 2015 05:53 Sufficiency wrote:Show nested quote +On January 05 2015 05:49 wei2coolman wrote:On January 05 2015 05:45 Sufficiency wrote:On January 05 2015 05:36 wei2coolman wrote:On January 05 2015 05:26 Sufficiency wrote:On January 05 2015 05:22 wei2coolman wrote:On January 05 2015 05:19 Sufficiency wrote:On January 05 2015 00:50 Goumindong wrote:On January 04 2015 16:32 Sufficiency wrote:On January 04 2015 16:30 Goumindong wrote:
[quote]
So you're saying that, as of champion select finishing, no team composition has an advantage over another? That, as of champion select, no team of players in ranked has an advantage over any other?
An easy example. There are 5 players on team A all playing their mains and team B has 5 AD mains. Does team A or B have a better chance to win given the players mixed role mmr is the same? I would suggest that team a has a better win chance and that they will acquire gold because of that. What if team a has good counter picks and so easily wins lane. Does that effect the gold acquisition? Of course all that does.
That is why it's hard to say that the win % at a gold difference is the impact. That doesn't mean we can't generalize from very specific summary stats but it does mean that the value we are looking at is a mix of the indication and the impact with no clear way to disentangle the two.
If we don't care about disentangling them then you should be good to go. But you should still disentangle the effect of the dragons. I totally agree with what you said. But I also think we need to disentangle the effect of visiting TL. Or maybe the effect of being born in June. Or the effect of left-handedness. Or the effect of playing the game during the night instead of during the day. Or maybe they are Koreans! I think your suggested model is deeply flawed because it does not account for any of the aforementioned factors above. All of those above have deep, profound impact on gold acquisition. I can tell you quite simply why I think dragons are correlated with winning. Why would being left handed or visiting TL?(those effects, if they legitimately make you a better player would be nullified by MMR adjustments and so should have no effect). I can tell you quite simply how random chance gives some teams advantages over others at the same mmr (basically some teams will randomly have more people in their best role. Some teams will have people whose play styles mesh etc). But no such explanations exist for why those things correlate with individual game win probability. If you don't care that is fine, but you're supposedly refuting an "impact" statement and not stating that a gold lead is more indicative of a higher win rate. Let me ask you a different question: do you think smoking causes cancer? Because I can give you the argument that smoking does NOT cause cancer using what you just wrote. "Smoking leads to increased chance of lung cancer" is a very different statement than "Smoking causes cancer." Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. Do you have Pearson's R value to back that up? Also, less free objective gold on maps means you'd have to adjust scaling of gold values worth in predicting winners in game. Essentially comparing two different currencies, without taking scaling into account. Yes there are two different currencies, I am just arguing that gold in 4.21 is "worth more", not less, by looking at chance of winning after stratifying gold lead into the same levels/groups. and that's the problem. On January 05 2015 05:46 Sufficiency wrote:
Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them. Actually that it is very scientific. That's why it's so hard to prove things in Science, because the rigors of standards to come to a conclusion is very high. So while you may call it "anti-science", it is actually very scientific. No my argument about smoking is not scientific at all. A scientific theory needs to be falsifiable.
I mean the idea (Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible. ) is what I was talking about.
Of course that's not to say I think smoking doesn't cause health problems such as cancer, but from a pure science point of view, it's very hard to say it as such. That's also not to say that actuaries don't fucking love their correlations, hell entire companies base their billion dollar incomes off of it.
|
Cayman Islands24199 Posts
a statement of impact like"gold lead is more important to winning" is a causal statement.
|
On January 05 2015 06:02 oneofthem wrote:
a statement of impact like"gold lead is more important to winning" is a causal statement.
I don't think it is causal, nor did I mention the word "cause", or "causal" anywhere. I certainly did not think "causal" when I wrote it since it's observational data, nor did I try to phrase it in a certain way that may make a reader think that it's "causal".
|
At least this is definitely a causal statement:
On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful.
|
On January 05 2015 06:07 Prog wrote:At least this is definitely a causal statement: Show nested quote +On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful.
This is drawn in the following way:
I controlled gold lead to a stratum (500-1000) then compared win rates for teams leading 500-1000 gold on patch 4.21 and 4.19. The finding is that win rate is higher on 4.21 in this stratum after controlling for gold lead.
|
On January 05 2015 06:16 Sufficiency wrote:Show nested quote +On January 05 2015 06:07 Prog wrote:At least this is definitely a causal statement: On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. This is drawn in the following way: I controlled gold lead to a stratum (500-1000) then compared win rates for teams leading 500-1000 gold on patch 4.21 and 4.19. The finding is that win rate is higher on 4.21 in this stratum after controlling for gold lead.
None of which accounts for how much dragon is worth (new currency).
|
On January 05 2015 06:20 wei2coolman wrote:Show nested quote +On January 05 2015 06:16 Sufficiency wrote:On January 05 2015 06:07 Prog wrote:At least this is definitely a causal statement: On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. This is drawn in the following way: I controlled gold lead to a stratum (500-1000) then compared win rates for teams leading 500-1000 gold on patch 4.21 and 4.19. The finding is that win rate is higher on 4.21 in this stratum after controlling for gold lead. None of which accounts for how much dragon is worth (new currency).
So, really, you want him to do what? Put a little asterisk by it? Like, you just seem to be being obstinate over grammar because you don't agree with the implications of his statistics.
|
On January 05 2015 06:27 cLutZ wrote:Show nested quote +On January 05 2015 06:20 wei2coolman wrote:On January 05 2015 06:16 Sufficiency wrote:On January 05 2015 06:07 Prog wrote:At least this is definitely a causal statement: On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. This is drawn in the following way: I controlled gold lead to a stratum (500-1000) then compared win rates for teams leading 500-1000 gold on patch 4.21 and 4.19. The finding is that win rate is higher on 4.21 in this stratum after controlling for gold lead. None of which accounts for how much dragon is worth (new currency). So, really, you want him to do what? Put a little asterisk by it? Like, you just seem to be being obstinate over grammar because you don't agree with the implications of his statistics.
No, I expect him to draw a clear and reasonable conclusion such as Yango stated.
On January 04 2015 23:26 TheYango wrote:Show nested quote +On January 04 2015 18:41 Sufficiency wrote:On January 04 2015 17:37 Zess wrote:
Sufficiency, you still haven't given any answer on how you expect to address the alternative hypothesis:
Gold leaders are harder to gain in 4.21 because the dragon doesn't give gold, so gold lead now is a better indicator of who is the better team, and thus better teams get more gold and then win more, rather than gold lead being more important and allowing a team to win more.
Right now all you've shown is that gold lead is a strong indicator of the better team (because the better team is the one that wins right?) and not "Consequently, once you do get a gold lead, it is actually a lot more valuable" I showed that gold lead is a good predictor of winning the game. With the 4.21 changes, it gold lead has become a better predictor if we keep the level of gold lead to be the same (e.g. 500-1000 lead). You can argue all you want that it's actually better team -> more gold -> win; at the end of the day you cannot directly measure how good a team is. The underlying causal relationship is really not that important. Then you should spell that out clearly, because on TL, Reddit, and your blog you make the vague, generalizing statement that a gold lead in 4.21 is "more important". Which, either deliberately or accidentally, will be read as you implying that causal relationship, because people will interpret it as "more important for winning a game", not "more important for predicting the winner of a game".
|
Watching DL play Graves, noticed he favor Youmuu's as second item vs Statik Shiv, both are pretty comparable to item timing (200 gold difference). Statik shiv's 100 magic damage + Graves burst almost seems to good to give up, but the arpen+AD from ghostblade might actually workout to do more damage.
|
Not to mention that meta was changed and it was made easier to close out games if you can get a lead.Also champion picks have gone even more toward mid game oriented.
|
Every graves does this. I think they already built ghostblade 2nd before iem.
|
It is the same as lucian.You get shiv if there is a lot of tankyness you need to go through if there isn't you get ghostblade and oneshot people with auto/q/r.
|
On January 05 2015 06:31 wei2coolman wrote:Show nested quote +On January 05 2015 06:27 cLutZ wrote:On January 05 2015 06:20 wei2coolman wrote:On January 05 2015 06:16 Sufficiency wrote:On January 05 2015 06:07 Prog wrote:At least this is definitely a causal statement: On January 05 2015 05:26 Sufficiency wrote:
Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. This is drawn in the following way: I controlled gold lead to a stratum (500-1000) then compared win rates for teams leading 500-1000 gold on patch 4.21 and 4.19. The finding is that win rate is higher on 4.21 in this stratum after controlling for gold lead. None of which accounts for how much dragon is worth (new currency). So, really, you want him to do what? Put a little asterisk by it? Like, you just seem to be being obstinate over grammar because you don't agree with the implications of his statistics. No, I expect him to draw a clear and reasonable conclusion such as Yango stated. Show nested quote +On January 04 2015 23:26 TheYango wrote:On January 04 2015 18:41 Sufficiency wrote:On January 04 2015 17:37 Zess wrote:
Sufficiency, you still haven't given any answer on how you expect to address the alternative hypothesis:
Gold leaders are harder to gain in 4.21 because the dragon doesn't give gold, so gold lead now is a better indicator of who is the better team, and thus better teams get more gold and then win more, rather than gold lead being more important and allowing a team to win more.
Right now all you've shown is that gold lead is a strong indicator of the better team (because the better team is the one that wins right?) and not "Consequently, once you do get a gold lead, it is actually a lot more valuable" I showed that gold lead is a good predictor of winning the game. With the 4.21 changes, it gold lead has become a better predictor if we keep the level of gold lead to be the same (e.g. 500-1000 lead). You can argue all you want that it's actually better team -> more gold -> win; at the end of the day you cannot directly measure how good a team is. The underlying causal relationship is really not that important. Then you should spell that out clearly, because on TL, Reddit, and your blog you make the vague, generalizing statement that a gold lead in 4.21 is "more important". Which, either deliberately or accidentally, will be read as you implying that causal relationship, because people will interpret it as "more important for winning a game", not "more important for predicting the winner of a game".
The conclusion was clear and reasonable. I can't convince everyone that it's clear and reasonable, just like even the best scientist can't convince everyone that global warming is real.
|
On January 05 2015 06:54 nafta wrote:
It is the same as lucian.You get shiv if there is a lot of tankyness you need to go through if there isn't you get ghostblade and oneshot people with auto/q/r.
I would think it would be the other way around due to the arpen on youmou's, and 3rd item lw. I don't play ADC so no real input other than just casual interest.
|
On January 05 2015 05:46 Sufficiency wrote:
Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them.
Actually this is strikingly close to what happens in real life, though. The only difference is that the MMR factor doesn't cause smoking. It causes cancer and smoking increases the likelihood of developing cancer. So, in a way, Eloboosting is like smoking in that it increases the likelihood of cancer.
The last part was a joke, but for real, if you just changed the "MMR factor causes smoking and lung cancer and it is the sole cause of both" part to "MMR factor causes lung cancer; smoking increases the likelihood of someone developing cancer but is not the sole cause" it would be pretty accurate.
And I think MMR factor, in this case, would be malfunction in the DNA reproduction process affecting apoptosis.
|
On January 05 2015 06:58 ZERG_RUSSIAN wrote:Show nested quote +On January 05 2015 05:46 Sufficiency wrote:
Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them. Actually this is strikingly close to what happens in real life, though. The only difference is that the MMR factor doesn't cause smoking. It causes cancer and smoking increases the likelihood of developing cancer. So, in a way, Eloboosting is like smoking in that it increases the likelihood of cancer. The last part was a joke, but for real, if you just changed the "MMR factor causes smoking and lung cancer and it is the sole cause of both" part to "MMR factor causes lung cancer; smoking increases the likelihood of someone developing cancer but is not the sole cause" it would be pretty accurate. And I think MMR factor, in this case, would be malfunction in the DNA reproduction process affecting apoptosis.
Technically all cancer's are caused by the malfunction in the DNA reproduction process. Question is whether or not studies can prove that smoking leads to it. MMR factor would have to be some other external environmental cause or perhaps genetic predisposition.
|
On January 05 2015 05:36 wei2coolman wrote:Show nested quote +On January 05 2015 05:26 Sufficiency wrote:On January 05 2015 05:22 wei2coolman wrote:On January 05 2015 05:19 Sufficiency wrote:On January 05 2015 00:50 Goumindong wrote:On January 04 2015 16:32 Sufficiency wrote:On January 04 2015 16:30 Goumindong wrote:On January 04 2015 16:07 Sufficiency wrote:On January 04 2015 15:54 Goumindong wrote:On January 04 2015 15:08 Sufficiency wrote:
[quote]
If I give you the data can you analyze it "properly" according to your standard, then? Probably. It's been a while since I've had to work with raw data like that unfortunately. So I will have to relearn the code/programs. But if you want the short version you can do easily. Just do exactly what you're doing right now but for the new patch also separate out the data by dragons. So that you have 0-500 gold diff (has dragon) 0-500 gold diff (other team has dragon) and 0-500 (no dragons). The no dragons version is essentially the "even comparison of pre patch gold impact". Which, while its not the actual impact, will tell us whether or not the impact went appreciably up or down when compared to the prepatch value. Basically you're trying to fight against the alternate explanation; " the better team gets more gold because it's better, the gold helps but they were already better". With regards to the patch changes we expect that lower gold on this explanation increases the win rate more. "Now that dragons give no gold you have to be even better to get a big gold lead and so the same amount of gold lead indicates a better team and so higher win rate" Moreover making the big table gives us the information for the game that we really care about. Since we can count dragons as well as gold and counting both lets us be precise in the summary stats and get a better prediction after observing the first 10 minutes. Another option which has the reverse causation issues moreso but will give you a potentially better answer to the effect on dragon and gold is to do a logistic regression with the variables "gold diff, dragon dummies, gold diff * dragon dummies". Which would separate out the effect of gold the effect of dragons and the effect of dragons on gold. The difference between the gold effects from patch 1 to patch 2 would indicate a change in the impact of gold (I think that the "better team" effect would difference out, but not entirely sure) and you could then see better how the impact changes with dragons. I am super busy next week though so let me know if you can't run those. I believe the team wins more because they have visit Team Liquid at least once a week. I think I should also include a variable of how many times they visited Team Liquid during the last 72 hours. The gold and the player skills help, but ultimate it's because they visit Team Liquid that they are better players, and then because they are better players they get more gold. So you're saying that, as of champion select finishing, no team composition has an advantage over another? That, as of champion select, no team of players in ranked has an advantage over any other? An easy example. There are 5 players on team A all playing their mains and team B has 5 AD mains. Does team A or B have a better chance to win given the players mixed role mmr is the same? I would suggest that team a has a better win chance and that they will acquire gold because of that. What if team a has good counter picks and so easily wins lane. Does that effect the gold acquisition? Of course all that does. That is why it's hard to say that the win % at a gold difference is the impact. That doesn't mean we can't generalize from very specific summary stats but it does mean that the value we are looking at is a mix of the indication and the impact with no clear way to disentangle the two. If we don't care about disentangling them then you should be good to go. But you should still disentangle the effect of the dragons. I totally agree with what you said. But I also think we need to disentangle the effect of visiting TL. Or maybe the effect of being born in June. Or the effect of left-handedness. Or the effect of playing the game during the night instead of during the day. Or maybe they are Koreans! I think your suggested model is deeply flawed because it does not account for any of the aforementioned factors above. All of those above have deep, profound impact on gold acquisition. I can tell you quite simply why I think dragons are correlated with winning. Why would being left handed or visiting TL?(those effects, if they legitimately make you a better player would be nullified by MMR adjustments and so should have no effect). I can tell you quite simply how random chance gives some teams advantages over others at the same mmr (basically some teams will randomly have more people in their best role. Some teams will have people whose play styles mesh etc). But no such explanations exist for why those things correlate with individual game win probability. If you don't care that is fine, but you're supposedly refuting an "impact" statement and not stating that a gold lead is more indicative of a higher win rate. Let me ask you a different question: do you think smoking causes cancer? Because I can give you the argument that smoking does NOT cause cancer using what you just wrote. "Smoking leads to increased chance of lung cancer" is a very different statement than "Smoking causes cancer." Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. Do you have Pearson's R value to back that up? Also, less free objective gold on maps means you'd have to adjust scaling of gold values worth in predicting winners in game. Essentially comparing two different currencies, without taking scaling into account.
No. He is saying he doesn't care about the fact that there is less free gold on the map. He is only making a prediction and making no other statement. He just isn't saying that when he makes his claim.
Also sufficiency "omitted variable bias" is a big fucking deal in science. Dragons are not hidden and can be observed. They correlate both with winning and with gold lead and so if you are running stats on win% by gold lead stratum you will bias your results.
This is it is not same as saying that there is a hidden MMR that matters because mmr neither correlates with cancer or with smoking. (Though if it did and it was exogenous you would want to include it)
But you do want to include dragons because it does correlate with both. Seriously you can do the math yourself to show it in 1 minute. Additionally it should be noted that while most "your model is bad" knocks are IV complaints, omitted variable model specification complaints are very common.
|
On January 05 2015 05:53 Sufficiency wrote:Show nested quote +On January 05 2015 05:49 wei2coolman wrote:On January 05 2015 05:45 Sufficiency wrote:On January 05 2015 05:36 wei2coolman wrote:On January 05 2015 05:26 Sufficiency wrote:On January 05 2015 05:22 wei2coolman wrote:On January 05 2015 05:19 Sufficiency wrote:On January 05 2015 00:50 Goumindong wrote:On January 04 2015 16:32 Sufficiency wrote:On January 04 2015 16:30 Goumindong wrote:
[quote]
So you're saying that, as of champion select finishing, no team composition has an advantage over another? That, as of champion select, no team of players in ranked has an advantage over any other?
An easy example. There are 5 players on team A all playing their mains and team B has 5 AD mains. Does team A or B have a better chance to win given the players mixed role mmr is the same? I would suggest that team a has a better win chance and that they will acquire gold because of that. What if team a has good counter picks and so easily wins lane. Does that effect the gold acquisition? Of course all that does.
That is why it's hard to say that the win % at a gold difference is the impact. That doesn't mean we can't generalize from very specific summary stats but it does mean that the value we are looking at is a mix of the indication and the impact with no clear way to disentangle the two.
If we don't care about disentangling them then you should be good to go. But you should still disentangle the effect of the dragons. I totally agree with what you said. But I also think we need to disentangle the effect of visiting TL. Or maybe the effect of being born in June. Or the effect of left-handedness. Or the effect of playing the game during the night instead of during the day. Or maybe they are Koreans! I think your suggested model is deeply flawed because it does not account for any of the aforementioned factors above. All of those above have deep, profound impact on gold acquisition. I can tell you quite simply why I think dragons are correlated with winning. Why would being left handed or visiting TL?(those effects, if they legitimately make you a better player would be nullified by MMR adjustments and so should have no effect). I can tell you quite simply how random chance gives some teams advantages over others at the same mmr (basically some teams will randomly have more people in their best role. Some teams will have people whose play styles mesh etc). But no such explanations exist for why those things correlate with individual game win probability. If you don't care that is fine, but you're supposedly refuting an "impact" statement and not stating that a gold lead is more indicative of a higher win rate. Let me ask you a different question: do you think smoking causes cancer? Because I can give you the argument that smoking does NOT cause cancer using what you just wrote. "Smoking leads to increased chance of lung cancer" is a very different statement than "Smoking causes cancer." Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. Do you have Pearson's R value to back that up? Also, less free objective gold on maps means you'd have to adjust scaling of gold values worth in predicting winners in game. Essentially comparing two different currencies, without taking scaling into account. Yes there are two different currencies, I am just arguing that gold in 4.21 is "worth more", not less, by looking at chance of winning after stratifying gold lead into the same levels/groups. and that's the problem. On January 05 2015 05:46 Sufficiency wrote:
Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them. Actually that it is very scientific. That's why it's so hard to prove things in Science, because the rigors of standards to come to a conclusion is very high. So while you may call it "anti-science", it is actually very scientific. No my argument about smoking is not scientific at all. A scientific theory needs to be falsifiable.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them.
Interestingly, if you remove your statement of bias that "its exact composition and form we do not know, nor can we measure it" and replace it with "we don't know what it is or HOW to measure it yet" and change some of the "believes" to "hypothesize" it becomes a lot more scientific because it's suddenly falsifiable.
Like I'm sure that something like... oh, say, INTELLIGENCE would be this elusive thing we seek...
|
On January 05 2015 07:04 wei2coolman wrote:Show nested quote +On January 05 2015 06:58 ZERG_RUSSIAN wrote:On January 05 2015 05:46 Sufficiency wrote:
Anyway, my point is, criticizing an observational study by pointing out a hidden, unobservable, unverifiable confounding factor is deeply fallacious and anti-science, especially when a controlled experiment is not feasible.
As an example, I can argue the following:
All epidemiological data (i.e. pretty much all observational studies) shows a strong link between smoking and lung cancer - i..e smoking X number of cigarettes per day increases your risk of being diagnosed with lung cancer by Y%. But does it mean cigarettes cause lung cancer?
No, it does not. As a matter of fact, we believe that smoking is caused by the "MMR" factor within an individual - its exact composition and form we do not know, nor can we measure it. However, we believe the "MMR" factor causes smoking and lung cancer and it is the sole cause of both. As a result, one may see correlation between smoking and lung cancer, but smoking is not a cause of it. We do not know exactly what this "MMR" factor is or how to measure it, but we constructed in our heads since the tobacco companies paid us millions of dollars to defend them. Actually this is strikingly close to what happens in real life, though. The only difference is that the MMR factor doesn't cause smoking. It causes cancer and smoking increases the likelihood of developing cancer. So, in a way, Eloboosting is like smoking in that it increases the likelihood of cancer. The last part was a joke, but for real, if you just changed the "MMR factor causes smoking and lung cancer and it is the sole cause of both" part to "MMR factor causes lung cancer; smoking increases the likelihood of someone developing cancer but is not the sole cause" it would be pretty accurate. And I think MMR factor, in this case, would be malfunction in the DNA reproduction process affecting apoptosis. Technically all cancer's are caused by the malfunction in the DNA reproduction process. Question is whether or not studies can prove that smoking leads to it. MMR factor would have to be some other external environmental cause or perhaps genetic predisposition.
Depends how you're looking at it. If you start off by defining MMR factor as cell death malfunction (which by your own admission causes cancer), then you just need to alter your definition of smoking. If you define smoking as causing cancer then you have to alter your definition of MMR.
Personally I think that the second one is illogical because it makes the assumption that smoking causes cancer by definition, which needs to be proven, but whatever floats your boat  .
[edit]just reread, that's unclear what i'm saying--basically, the bolded part of your post is what I think conflicts with your admission that MMR factor causes cancer.[/edit]
|
On January 05 2015 07:05 Goumindong wrote:Show nested quote +On January 05 2015 05:36 wei2coolman wrote:On January 05 2015 05:26 Sufficiency wrote:On January 05 2015 05:22 wei2coolman wrote:On January 05 2015 05:19 Sufficiency wrote:On January 05 2015 00:50 Goumindong wrote:On January 04 2015 16:32 Sufficiency wrote:On January 04 2015 16:30 Goumindong wrote:On January 04 2015 16:07 Sufficiency wrote:On January 04 2015 15:54 Goumindong wrote:
[quote]
Probably. It's been a while since I've had to work with raw data like that unfortunately. So I will have to relearn the code/programs.
But if you want the short version you can do easily. Just do exactly what you're doing right now but for the new patch also separate out the data by dragons. So that you have 0-500 gold diff (has dragon) 0-500 gold diff (other team has dragon) and 0-500 (no dragons). The no dragons version is essentially the "even comparison of pre patch gold impact". Which, while its not the actual impact, will tell us whether or not the impact went appreciably up or down when compared to the prepatch value.
Basically you're trying to fight against the alternate explanation; "the better team gets more gold because it's better, the gold helps but they were already better". With regards to the patch changes we expect that lower gold on this explanation increases the win rate more. "Now that dragons give no gold you have to be even better to get a big gold lead and so the same amount of gold lead indicates a better team and so higher win rate"
Moreover making the big table gives us the information for the game that we really care about. Since we can count dragons as well as gold and counting both lets us be precise in the summary stats and get a better prediction after observing the first 10 minutes.
Another option which has the reverse causation issues moreso but will give you a potentially better answer to the effect on dragon and gold is to do a logistic regression with the variables "gold diff, dragon dummies, gold diff * dragon dummies". Which would separate out the effect of gold the effect of dragons and the effect of dragons on gold. The difference between the gold effects from patch 1 to patch 2 would indicate a change in the impact of gold (I think that the "better team" effect would difference out, but not entirely sure) and you could then see better how the impact changes with dragons.
I am super busy next week though so let me know if you can't run those. I believe the team wins more because they have visit Team Liquid at least once a week. I think I should also include a variable of how many times they visited Team Liquid during the last 72 hours. The gold and the player skills help, but ultimate it's because they visit Team Liquid that they are better players, and then because they are better players they get more gold. So you're saying that, as of champion select finishing, no team composition has an advantage over another? That, as of champion select, no team of players in ranked has an advantage over any other? An easy example. There are 5 players on team A all playing their mains and team B has 5 AD mains. Does team A or B have a better chance to win given the players mixed role mmr is the same? I would suggest that team a has a better win chance and that they will acquire gold because of that. What if team a has good counter picks and so easily wins lane. Does that effect the gold acquisition? Of course all that does. That is why it's hard to say that the win % at a gold difference is the impact. That doesn't mean we can't generalize from very specific summary stats but it does mean that the value we are looking at is a mix of the indication and the impact with no clear way to disentangle the two. If we don't care about disentangling them then you should be good to go. But you should still disentangle the effect of the dragons. I totally agree with what you said. But I also think we need to disentangle the effect of visiting TL. Or maybe the effect of being born in June. Or the effect of left-handedness. Or the effect of playing the game during the night instead of during the day. Or maybe they are Koreans! I think your suggested model is deeply flawed because it does not account for any of the aforementioned factors above. All of those above have deep, profound impact on gold acquisition. I can tell you quite simply why I think dragons are correlated with winning. Why would being left handed or visiting TL?(those effects, if they legitimately make you a better player would be nullified by MMR adjustments and so should have no effect). I can tell you quite simply how random chance gives some teams advantages over others at the same mmr (basically some teams will randomly have more people in their best role. Some teams will have people whose play styles mesh etc). But no such explanations exist for why those things correlate with individual game win probability. If you don't care that is fine, but you're supposedly refuting an "impact" statement and not stating that a gold lead is more indicative of a higher win rate. Let me ask you a different question: do you think smoking causes cancer? Because I can give you the argument that smoking does NOT cause cancer using what you just wrote. "Smoking leads to increased chance of lung cancer" is a very different statement than "Smoking causes cancer." Let me rephrase then: gold lead increases chance of winning when comparing 4.21 to 4.19. Therefore, gold lead is more impactful. Do you have Pearson's R value to back that up? Also, less free objective gold on maps means you'd have to adjust scaling of gold values worth in predicting winners in game. Essentially comparing two different currencies, without taking scaling into account. No. He is saying he doesn't care about the fact that there is less free gold on the map. He is only making a prediction and making no other statement. He just isn't saying that when he makes his claim. Also sufficiency "omitted variable bias" is a big fucking deal in science. Dragons are not hidden and can be observed. They correlate both with winning and with gold lead and so if you are running stats on win% by gold lead stratum you will bias your results. This is it is not same as saying that there is a hidden MMR that matters because mmr neither correlates with cancer or with smoking. (Though if it did and it was exogenous you would want to include it) But you do want to include dragons because it does correlate with both. Seriously you can do the math yourself to show it in 1 minute. Additionally it should be noted that while most "your model is bad" knocks are IV complaints, omitted variable model specification complaints are very common.
It would be really interesting if the number of dragons taken actually didn't have a significant effect on the game when compared to gold lead at given points--like, if when things were even, dragons mattered, but if there was an X amount of gold lead or higher, dragons had no effect on outcome of game.
|
|
|
|
|
|
EPT
