EPT[Champion] Miss Fortune - Page 10
| Forum Index > LoL Strategy |
|
krndandaman
Mozambique16569 Posts
| ||
|
ZERG_RUSSIAN
10417 Posts
So if you do the math on that, 617/.1503 = ~4105 which is about 4000 ish But yeah, that's why monthly + diamond is a better combination for stats analysis, it's been pretty well-explained over the past two pages by multiple people | ||
|
ZERG_RUSSIAN
10417 Posts
Literally no way to know based on what we have, but it might explain some of that. [edit] I just looked at it and you can't even see a yellow bar next to him indicating what percentage of games he's played in. Perhaps you should just listen to us if you can't interpret that graph by just looking at it. [/edit] | ||
|
SniperVul5
Canada166 Posts
On June 16 2014 10:36 krndandaman wrote: how can you see that there are 4000 games played in challenger? I don't see it anywhere on that page. also the list is almost equally trolly on every single league. I just used challenger as an example since it was the highest league and most troll. edit: I changed it to monthly for challenger and urgot is still #1? As a graduate student in epidemiology and biostatistics I'm appalled at the apparent lack of basic statistical inference from ZERG_RUSSIAN. The premise of your argument is that the higher the soloq win rate percentage equates to greater ability of the champion, which can't be further from the truth. Even for statistically significant results, there is still the possibility that it may be due to chance. Setting aside the fact that a large sample size can make any marginal difference significant, the question of what sample size would be appropriate given a marginal error would be acceptable to predict the strength of the champion is the true question. Even if we believe the numbers as they are, there are many confounders for why numbers can be skewed 1. The players using the champion are simply better in every way (people like WT are "outliers" that can make any champion look good) 2. If a champion is picked more often, then the law of large numbers suggest that the win rate will regress towards the mean, which is ~50% given the match maker. MF win rates being higher than other ADs may simply be the product of less games played overall compared to the other ADs, which may make her win rate appear higher. 3. The higher win rates may be contributed by lower tiers (bronze, silver, gold, etc) as opposed to diamond/ challenger. You would have to stratify the win rates based on tiers, and also calculate appropriate sample sizes, control for region, times played (patches). Lower elo players may prefer MF more, and given how they are less likely to know how to play against it (think season 2 AOE compositions), could boost the win rates. This is very likely given that more people are in lower elos. 4. Chance. We simply cannot ignore that the people that played MF got lucky in soloq, that even with thousands of games, people may have lucked out more than when they tryhard on another AD. Setting aside the argument of psychological impacts of playing different ADs (based on meta or non-meta picks), chance alone can still affect win rate. These arguments don't even add factors such as psychological effects, team compositions or win rates of individual players (of even if they main AD or not). There are more reasons but I can write you a 20 page paper on the entire topic of confounding but ZERG_RUSSIAN may not understand it. The premise of whether an AD is good hasn't truly come from pure soloq statistics anyways, but more about number crunching based on trading, ability to position, teamfight contributions, late game potential and so on. Just looking at statistics alone isn't enough evidence to prove that the champion is good or bad. Think about it, when was the last time we thought a champion was good purely based on statistics. Most people would side with "we think X is good since the pros feel that the champion has more utility, carry potential, for Y reasons." Lastly, isn't observational data purely speculative as to the true meaning behind the numbers? I can interpret the lower poppy win rate to players not understanding how to play the champion, and not due to the the fact that it may be a poor champ. You need more controlled environments to come up with more definitive conclusions. Mabye it's ZERG_RUSSIAN who should take some statistics courses, and stop pretending that he actually knows anything about it. | ||
|
ZERG_RUSSIAN
10417 Posts
On June 16 2014 13:14 SniperVul5 wrote: + Show Spoiler + On June 16 2014 10:36 krndandaman wrote: how can you see that there are 4000 games played in challenger? I don't see it anywhere on that page. also the list is almost equally trolly on every single league. I just used challenger as an example since it was the highest league and most troll. edit: I changed it to monthly for challenger and urgot is still #1? As a graduate student in epidemiology and biostatistics I'm appalled at the apparent lack of basic statistical inference from ZERG_RUSSIAN. The premise of your argument is that the higher the soloq win rate percentage equates to greater ability of the champion, which can't be further from the truth. Even for statistically significant results, there is still the possibility that it may be due to chance. Setting aside the fact that a large sample size can make any marginal difference significant, the question of what sample size would be appropriate given a marginal error would be acceptable to predict the strength of the champion is the true question. Even if we believe the numbers as they are, there are many confounders for why numbers can be skewed 1. The players using the champion are simply better in every way (people like WT are "outliers" that can make any champion look good) 2. If a champion is picked more often, then the law of large numbers suggest that the win rate will regress towards the mean, which is ~50% given the match maker. MF win rates being higher than other ADs may simply be the product of less games played overall compared to the other ADs, which may make her win rate appear higher. 3. The higher win rates may be contributed by lower tiers (bronze, silver, gold, etc) as opposed to diamond/ challenger. You would have to stratify the win rates based on tiers, and also calculate appropriate sample sizes, control for region, times played (patches). Lower elo players may prefer MF more, and given how they are less likely to know how to play against it (think season 2 AOE compositions), could boost the win rates. This is very likely given that more people are in lower elos. 4. Chance. We simply cannot ignore that the people that played MF got lucky in soloq, that even with thousands of games, people may have lucked out more than when they tryhard on another AD. Setting aside the argument of psychological impacts of playing different ADs (based on meta or non-meta picks), chance alone can still affect win rate. These arguments don't even add factors such as psychological effects, team compositions or win rates of individual players (of even if they main AD or not). There are more reasons but I can write you a 20 page paper on the entire topic of confounding but ZERG_RUSSIAN may not understand it. The premise of whether an AD is good hasn't truly come from pure soloq statistics anyways, but more about number crunching based on trading, ability to position, teamfight contributions, late game potential and so on. Just looking at statistics alone isn't enough evidence to prove that the champion is good or bad. Think about it, when was the last time we thought a champion was good purely based on statistics. Most people would side with "we think X is good since the pros feel that the champion has more utility, carry potential, for Y reasons." Lastly, isn't observational data purely speculative as to the true meaning behind the numbers? I can interpret the lower poppy win rate to players not understanding how to play the champion, and not due to the the fact that it may be a poor champ. You need more controlled environments to come up with more definitive conclusions. Mabye it's ZERG_RUSSIAN who should take some statistics courses, and stop pretending that he actually knows anything about it. No, I'm not equating winrate with champion ability. I'm equating it with pick strength. As in, players who pick MF tend to win more games than players who pick other ADCs. Either way you look at this, it's in my favor: A) Better players pick MF more often. Sure. Why are these better players picking MF right now and winning with her? Win. B) Many different types of players are picking MF, and not just better players. Okay, then her winrate is just high. Win. Let me address your points, though. + Show Spoiler + Point 1: With a large enough sample size this should not occur. Even if it does, it kinda proves my point. Better players pick her more often why? Because she's bad and they want a challenge? I don't think so. Point 2: A regression towards the mean is seen in the data right now anyway and she's still got the highest winrate. The amount of games we're talking about here with MF picked in diamond alone over the past month is close to 60000. Also, your "regression towards the mean" argument assumes a normal curve of winrate on all champs, which is not true. Some champs win more than others. See Goumindong's post below for a more eloquent explanation. Point 3: WE DID THAT. READ. Point 4: Our sample size is close to 60000. It's not just a couple thousand. It's sixty-fucking-thousand on a conservative estimate (I rounded down when adding the numbers for this estimate. If you want to check it, just go to the link at the bottom and add it yourself.). The amount of influence "chance" has over this goes down in relation to the sample size. Our sample size is FUCKING HUGE. Basically, your argument comes down to our sample size not being large enough to make any inferences. I disagree and I'd be willing to bet a month ban on having enough games to make a conclusive statement at 99%. Our sample size for MF games played in diamond is something like 60000 games. Again, sixty-fucking-thousand. On a conservative estimate for JUST diamond. (By the way, I just went and added it, it's actually 68142 games.) For reference: + Show Spoiler + http://www.lolking.net/charts?region=all&type=bottom-matchup&range=monthly&map=sr&queue=1x1&league=diamond Tell me that 21344 Lucian vs MF matchups isn't large enough to make a conclusive statement about who tends to win the matchup? And that we can't take the aggregate of all that data and be confident about her winrate? With numbers like this we can be extremely confident that our results are NOT due to chance. And again, my point is that she wins the most games when picked. If you want to define how good a champ is in some other way, then by all means, go ahead, and I'll concede to you that she has her weaknesses. But if you just want to pick the ADC that statistically wins the highest percentage of games in solo queue, it's MF. And FYI I'm a fucking doctoral student in clinical psychology. On top of that, I'm one of those diamond players that plays MF. Don't credential drop on me and try to tell me I don't know stats. (Edited for clarity and to remove most of the reactive statements.) | ||
|
Goumindong
United States3529 Posts
On June 16 2014 13:14 SniperVul5 wrote: As a graduate student in epidemiology and biostatistics I'm appalled at the apparent lack of basic statistical inference from ZERG_RUSSIAN. As a graduate student in biostatistics you should know pretty intuitively that a 60% win rate with 1000 trials would reject a 50% win rate even with a p value threshold corrected for multiple tests, you would certainly know that it was the case for the tens of thousands of MF games played each week/month in diamond. But biostats is easy stats so maybe that isn't true. To reiterate, the variance on a (1000 binomial distributions with a p=.5) /1000 (I.E 1000 coin flips averaged) is .025% So about 1% of the time we would expect that in 1000 trials, a champion who had a win rate of exactly 50% would display a win rate over 50.5%. Lets correct for multiple champions. Rather we want a p-value such that we would find 1 champion of 100 total over that value supposing. The 99.99% for 100 tests at 1000 trails corresponds to the 99.9% for 1 test at 1000 trials. Which is 3.62 Z value or about 50.57%. Of course we don't believe that champion win rates are precisely 50%, but we do observe that, outside of meta shifts/buffs which effect the champion, most champions do not see much more than 1-2% variation in their win rate across days which is pretty consistent with relatively constant champion powers. 1. The players using the champion are simply better in every way (people like WT are "outliers" that can make any champion look good) Unlikely. 2. If a champion is picked more often, then the law of large numbers suggest that the win rate will regress towards the mean, which is ~50% given the match maker. MF win rates being higher than other ADs may simply be the product of less games played overall compared to the other ADs, which may make her win rate appear higher. This is not true. I cannot even begin to reason why someone would think this was true. The negation is obvious. Suppose we have a champion which straight wins every game its played in. It does this because its so OP its literally impossible to lose. As the champion is picked more often its win rate continues to be 100%. OK so 100% is a degenerate distribution, but it holds true for any champion whose power implies a non-50% win rate. Note that the existence of champions who have above average win rates does not imply that some summoners must be at levels where they have higher win rates and are playing that champion. This is because not every summoner gets their champion each game even if they get their role the majority of the time. Just looking at statistics alone isn't enough evidence to prove that the champion is good or bad. Fine, then our theory is "Miss Fortune is good because we have reason to believe so by looking at her kit" and we go and test our theory and we cannot reject it. And we have an alternate theory "Miss Fortune is average or below average" and we go and test that theory and holy christ we reject it pretty strongly. Now, it could be wrong. But the likelihood is not on your side to suggest it. We're way [i]way[i] past a 50/50 bet here even if a 50/50 bet is probably the p value we should be looking at. After all, if i am picking champions(or thinking about picking up a champion), so long as i pick champions which have a >50% chance of "winning >50% of the time" I will on average win >50% of the time and that is really what we're after isn't it? Lastly, isn't observational data purely speculative as to the true meaning behind the numbers? No. An example. Suppose i want to know how tall my door is. I go and get a tape measure and i record the height of my door. I didn't build my door so i guess i don't know the true meaning behind the number? What if i measure 1000 times and then average the values? I would have a pretty good estimate of the height of my door. OK this is a bit remedial, because here we are measuring and not hypothesis testing. We know the model and its very simple. But the differences between observational data and what we just did with my door isn't that far off. The main difference is that we're testing and not measuring(so we're looking at error rates mainly, and we're checking against a model whose distribution we know) and that we have more variables. Observational data can be hard to interpret, especially if you don't know the underlying model, or if there are various effects which it doesn't make sense to talk about right now. This makes the statistics and interpretation more difficult, but not speculative. Another way to say it is that correlation does imply causation. Yes, yes, i know your into to stats teacher told you otherwise, but they were wrong and/or being overly simple. Correlation implies that f(x) exist, x(f) exists, f(g(x)) [or x(g(f))] exists or some combination exists. If none of those things existed then we would not have any correlation. When doing stats on observational data you're mainly trying to make sure that you're not seeing x(f) or f(g(x)) [you might know these as endogeneity/reverse causation and spurious correlation], and that if you're looking at the right f(x) [and getting f(x) looked at in a correct way so that x(f) if it exists as well is not mucking up the scales]. When doing stats in controlled environments, because you can know that your control is exogenous, you can be sure that only f(x) exists, and so you're only trying to find the right model(well if you care about that accuracy) One of the ways to informally ask yourself how strong your correlation results are with regards to causation is to simply look at the other functions. Do better players tend play MF? No, i doubt that. Though if they do we would have to ask why do only good players play MF and why they would not raise up in ranking. Is there a spurious correlation here? Does playing MF correspond with playing blue side and playing blue side corresponds to winning? Well, also probably no. Unless MF was such a contested pick that people were grabbing her first. But that isn't the case. | ||
|
ZERG_RUSSIAN
10417 Posts
Anyway, back to the matter at hand: My split pen build was dumb and I dominate lanes a lot harder with a standard armorpen/armor/(mp5/MR)/(flat ad/lifesteal) page. I still think sorc boots are worth it sometimes, though, but the buildup to them is a lot harder than the buildup to zerks. You can also run full flat AD reds but I think with the changes to armor runes and the large amount of penetration you build on MF, the rune penetration is worth it. Adjust your last hitting accordingly under tower (auto-tower shot-auto for casters for the first couple levels) or go AD reds if you don't want to. I have not done the math on it but I'm assuming based on experience. If someone can verify or show me why I'm wrong I would very much like to know because climbing diamond is going to be a little harder than blowing through plat and marginal advantages like runes are going to start to become more important. | ||
|
SniperVul5
Canada166 Posts
On June 16 2014 16:02 Goumindong wrote: As a graduate student in biostatistics you should know pretty intuitively that a 60% win rate with 1000 trials would reject a 50% win rate even with a p value threshold corrected for multiple tests, you would certainly know that it was the case for the tens of thousands of MF games played each week/month in diamond. But biostats is easy stats so maybe that isn't true. To reiterate, the variance on a (1000 binomial distributions with a p=.5) /1000 (I.E 1000 coin flips averaged) is .025% So about 1% of the time we would expect that in 1000 trials, a champion who had a win rate of exactly 50% would display a win rate over 50.5%. Lets correct for multiple champions. Rather we want a p-value such that we would find 1 champion of 100 total over that value supposing. The 99.99% for 100 tests at 1000 trails corresponds to the 99.9% for 1 test at 1000 trials. Which is 3.62 Z value or about 50.57%. Of course we don't believe that champion win rates are precisely 50%, but we do observe that, outside of meta shifts/buffs which effect the champion, most champions do not see much more than 1-2% variation in their win rate across days which is pretty consistent with relatively constant champion powers. Unlikely. This is not true. I cannot even begin to reason why someone would think this was true. The negation is obvious. Suppose we have a champion which straight wins every game its played in. It does this because its so OP its literally impossible to lose. As the champion is picked more often its win rate continues to be 100%. OK so 100% is a degenerate distribution, but it holds true for any champion whose power implies a non-50% win rate. Note that the existence of champions who have above average win rates does not imply that some summoners must be at levels where they have higher win rates and are playing that champion. This is because not every summoner gets their champion each game even if they get their role the majority of the time. Fine, then our theory is "Miss Fortune is good because we have reason to believe so by looking at her kit" and we go and test our theory and we cannot reject it. And we have an alternate theory "Miss Fortune is average or below average" and we go and test that theory and holy christ we reject it pretty strongly. Now, it could be wrong. But the likelihood is not on your side to suggest it. We're way [i]way[i] past a 50/50 bet here even if a 50/50 bet is probably the p value we should be looking at. After all, if i am picking champions(or thinking about picking up a champion), so long as i pick champions which have a >50% chance of "winning >50% of the time" I will on average win >50% of the time and that is really what we're after isn't it? No. An example. Suppose i want to know how tall my door is. I go and get a tape measure and i record the height of my door. I didn't build my door so i guess i don't know the true meaning behind the number? What if i measure 1000 times and then average the values? I would have a pretty good estimate of the height of my door. OK this is a bit remedial, because here we are measuring and not hypothesis testing. We know the model and its very simple. But the differences between observational data and what we just did with my door isn't that far off. The main difference is that we're testing and not measuring(so we're looking at error rates mainly, and we're checking against a model whose distribution we know) and that we have more variables. Observational data can be hard to interpret, especially if you don't know the underlying model, or if there are various effects which it doesn't make sense to talk about right now. This makes the statistics and interpretation more difficult, but not speculative. Another way to say it is that correlation does imply causation. Yes, yes, i know your into to stats teacher told you otherwise, but they were wrong and/or being overly simple. Correlation implies that f(x) exist, x(f) exists, f(g(x)) [or x(g(f))] exists or some combination exists. If none of those things existed then we would not have any correlation. When doing stats on observational data you're mainly trying to make sure that you're not seeing x(f) or f(g(x)) [you might know these as endogeneity/reverse causation and spurious correlation], and that if you're looking at the right f(x) [and getting f(x) looked at in a correct way so that x(f) if it exists as well is not mucking up the scales]. When doing stats in controlled environments, because you can know that your control is exogenous, you can be sure that only f(x) exists, and so you're only trying to find the right model(well if you care about that accuracy) One of the ways to informally ask yourself how strong your correlation results are with regards to causation is to simply look at the other functions. Do better players tend play MF? No, i doubt that. Though if they do we would have to ask why do only good players play MF and why they would not raise up in ranking. Is there a spurious correlation here? Does playing MF correspond with playing blue side and playing blue side corresponds to winning? Well, also probably no. Unless MF was such a contested pick that people were grabbing her first. But that isn't the case. So these are great points and I apologize to ZERG_RUSSIAN (whoever you are) for reacting the way I did. It was inappropriate given the above average quality of discussion (setting aside the fact that I shouldn't act the way I did in any discussion). You are right in that biostatistics is easy stats because the focus is more towards predicting the biological mechanisms behind the numbers produced. For example we may look to attribute changes to wait times in the hospital to under-staffing, or poor productivity. Further tests would then be done comparing various hospitals across localities to better understand what is going on. Another example could be looking at the effect of newly developed drugs on decrease in systolic blood pressure and trying to link numerical differences to the drug modifications. Eventually we may look at odds ratios for developing X disease and think about how a new drug would affect one's odds compared to a reference. The statistics portion is not as strong with respect to the theory, which also explains why I didn't use hard number examples. However, the examples you posed are very interesting to think about and I'll keep them in mind in the future. With respect to your other point with observational data, naturally we'd like to then test what we see in a more controlled environment. What I was trying to say was that right now, we don't have that type of environment, and that given what we have it's more difficult to interpret whether the higher win-rate is attributed to MF being OP or the other functions you mentioned. Perhaps after looking into those, we'd have more confidence in our interpretation but I wasn't fully convinced at the time. I'm assuming from your discussion on the law of large numbers is that a truly strong champion will theoretically have >50% win rate regardless of number of trials, while average champions may fluctuate high/low but eventually reach ~50%. If MF truly is above average then I would see why it further supports your point over mine. Once again, thanks for your response and my apologies to ZERG_RUSSIAN for my rude demeanor. | ||
|
Goumindong
United States3529 Posts
On June 17 2014 14:20 SniperVul5 wrote: I'm assuming from your discussion on the law of large numbers is that a truly strong champion will theoretically have >50% win rate regardless of number of trials, while average champions may fluctuate high/low but eventually reach ~50%. If MF truly is above average then I would see why it further supports your point over mine. Spoiled because its not about MF + Show Spoiler + Not quite. Specifically what we would be doing would be to look at a group of pmfs which take the value of 1 for P=f(X) and 0 for 1-P where X is the vector of champions and f is an unspecified function. Looking at MF's win rate we would be looking at a function where P=f(X|MF=1) We don't know f(X|MF=1) and we more or less can't know it, though we can look at individual matchups easily enough. We note that there exists a LLN which basically states that the average of a sum of random variables converges in probability to their weighted expectations. This is saying that Miss Fortune's Win rate converges in probability to the sum of f(X|MF=1) weighted for the relative density of each possible game. That is; we look at X as a random vector and can use iterated expectations to say that E(WIN|MF=1) = MF's Win Rate This win rate can change over days as the distribution of X the random vector changes and the overall win rate will converge in a similar way with distribution changes (or changes to the function, like with a buff or nerf). But this isn't helpful to us because what we are interested in specifically is how likely MF is to win a game in the current meta(basically this just means that we constrain ourselves to relatively stable and recent periods with few buffs. This applies to every champion with a restriction such that P(Win|No info) = .5, or the sum of each champions weighted expectation will be equal to .5. In precisely the same way that f(X_1|MF=1) doesn't need to equal f(X_2|MF=1) we note that each champions expected win does not need to be equal to .5 Then we simply measure MF's win rate, though technically what we are doing is that we are gathering samples of the random events which occurred, random events with a pmf that takes the value 1 for P=f(X|MF=1) and 0 else. Technically it is correct in that the variance of our complicated random variable is larger than the variance of a simple binary variable. But once we have 60,000 samples that barely matters anymore and if a champion is consistently winning a large percentage of games then we have very good evidence that they're strong. At the very least in the current meta. That is to say that yes, a champions current win rate is basically the only good definition of their general strength, with a champions conditional win rate as a slightly better definition for the given matchup. Now, this is imperfect because it relies on the idea that the meta is exogenous from a champions win rate (which depends on the meta). However there is probably good evidence to suggest that the meta is exogenous. After all, if the meta depended on win rates we would expect that a champions counters would be played more often in response to first picking. And so either the meta has stabilized or we should see something that we don't see. The thing we should see but don't is win rates for champions moving; that only happens in specific times when champions are being figured out. If the meta is relatively stable then its endogeneity doesn't matter because it doesn't change. The things which people say can get in the way don't seem like they can make any effect on this(technically we can put them into X, but that isn't helpful) or at the worst inform us as to her strength in a different way. If MF is played only by AD mains then AD mains have to be picking her for a reason. And we believe that AD mains pick champions who are strong at their position (because their goal is to win) and so if AD mains are picking MF then they must believe that MF is strong which they would not if they could not win with her. Which pretty much the definition of champion strength PS about biostats + Show Spoiler + Biostats is not easy stats because it doesn't focus on theory. Biostats is easy stats because experimental data does not tend to have as many complications as observational data and the situations in which you need to look at do not require methods to get around those problems. | ||
|
SniperVul5
Canada166 Posts
On June 17 2014 15:53 Goumindong wrote: Spoiled because its not about MF + Show Spoiler + Not quite. Specifically what we would be doing would be to look at a group of pmfs which take the value of 1 for P=f(X) and 0 for 1-P where X is the vector of champions and f is an unspecified function. Looking at MF's win rate we would be looking at a function where P=f(X|MF=1) We don't know f(X|MF=1) and we more or less can't know it, though we can look at individual matchups easily enough. We note that there exists a LLN which basically states that the average of a sum of random variables converges in probability to their weighted expectations. This is saying that Miss Fortune's Win rate converges in probability to the sum of f(X|MF=1) weighted for the relative density of each possible game. That is; we look at X as a random vector and can use iterated expectations to say that E(WIN|MF=1) = MF's Win Rate This win rate can change over days as the distribution of X the random vector changes and the overall win rate will converge in a similar way with distribution changes (or changes to the function, like with a buff or nerf). But this isn't helpful to us because what we are interested in specifically is how likely MF is to win a game in the current meta(basically this just means that we constrain ourselves to relatively stable and recent periods with few buffs. This applies to every champion with a restriction such that P(Win|No info) = .5, or the sum of each champions weighted expectation will be equal to .5. In precisely the same way that f(X_1|MF=1) doesn't need to equal f(X_2|MF=1) we note that each champions expected win does not need to be equal to .5 Then we simply measure MF's win rate, though technically what we are doing is that we are gathering samples of the random events which occurred, random events with a pmf that takes the value 1 for P=f(X|MF=1) and 0 else. Technically it is correct in that the variance of our complicated random variable is larger than the variance of a simple binary variable. But once we have 60,000 samples that barely matters anymore and if a champion is consistently winning a large percentage of games then we have very good evidence that they're strong. At the very least in the current meta. That is to say that yes, a champions current win rate is basically the only good definition of their general strength, with a champions conditional win rate as a slightly better definition for the given matchup. Now, this is imperfect because it relies on the idea that the meta is exogenous from a champions win rate (which depends on the meta). However there is probably good evidence to suggest that the meta is exogenous. After all, if the meta depended on win rates we would expect that a champions counters would be played more often in response to first picking. And so either the meta has stabilized or we should see something that we don't see. The thing we should see but don't is win rates for champions moving; that only happens in specific times when champions are being figured out. If the meta is relatively stable then its endogeneity doesn't matter because it doesn't change. The things which people say can get in the way don't seem like they can make any effect on this(technically we can put them into X, but that isn't helpful) or at the worst inform us as to her strength in a different way. If MF is played only by AD mains then AD mains have to be picking her for a reason. And we believe that AD mains pick champions who are strong at their position (because their goal is to win) and so if AD mains are picking MF then they must believe that MF is strong which they would not if they could not win with her. Which pretty much the definition of champion strength PS about biostats + Show Spoiler + Biostats is not easy stats because it doesn't focus on theory. Biostats is easy stats because experimental data does not tend to have as many complications as observational data and the situations in which you need to look at do not require methods to get around those problems. Once again thank you for the response. I have to admit I need to think through the explanation you gave a bit more but it was insightful in that it shows me I still have a long way to go to understand the statistics better. With respect to biostats + Show Spoiler + I'd have to somewhat disagree with your reasoning, since biological data (particularly human trials) are arguably just as complicated, albeit for different reasons. In a non-inferiority trial comparing a new drug to a set standard, we have to adjust methodologies to cater towards our population, while making sure that they adhere to the regimen (setting aside other issues of contamination and/or adjustments for human diversity). While I acknowledge that I don't have the same level of expertise with respect to discussing the theories, we require significant methodological manipulations. The data is difficult to interpret in that sense and our methodology focuses more on the subjects as opposed to the calculations. Edit: I am curious if there are more reference materials that I can get regarding our discussion. Always want to pick up on more knowledge. | ||
|
Goumindong
United States3529 Posts
Frankly i would say that the majority of the underlying theory and understanding you won't actually have to know as a biostatistician. You will either pick it up as you go (running into small problems that need small fixes until you've acquired a breadth of knowledge that informs the whole) or you will simply run the statistical tests and methods that others come up with in the same way that I cannot build the car i drive but it does get me places. If you're looking to know more, it would be good to know how familiar with linear algebra, difference equations, and calculus? If you're very familiar then the various wikipedia entries aren't bad. Though i would not begin to understand how to go through the various articles in order to make sense of them from a theoretical standpoint and come out with a coherent framework. Franky the best thing to do would be to take the set of undergraduate probability and mathematical statistics courses that your university statistics department offers. Its very unlikely that they will bung anything up. The problem with this, is that getting from undergraduate probability and mathematical statistics to graduate level theory takes a lot of time. If the mathematical statistics course does not require calculus its not the course for you. If the course on linear regression theory does not require linear algebra its not the course for you. | ||
|
Anakko
France1934 Posts
| ||
|
ticklishmusic
United States15977 Posts
That's my hunch anyways, could be miserably wrong. | ||
|
ZERG_RUSSIAN
10417 Posts
| ||
|
Bam Lee
2336 Posts
| ||
|
Nemireck
Canada1875 Posts
On June 23 2014 11:42 Bam Lee wrote: Is there any lane MF struggles with? Her high damage and grievous wounds make her quite an obnoxious lane, so i am wondering if there is any good way to shut her down? Caitlyn is a skill match-up. Tristana performs well against her, so do Quinn and Varus. I also remember having a tough time against Draven players back in the day, but they're so rare now I'm not sure I've come across a good one in months. | ||
|
ZERG_RUSSIAN
10417 Posts
Trist has a strong level 6 all-in which can interrupt your ult plus a massive gapcloser for gank support. Draven should get totally dunked unless you screw up huge by trying to trade with him before you have at least 2 ranks in E. In general against Draven you want to all-in a lot because his harass is much stronger than his commit. As always, support dependent, though, but a bit of jungle pressure + a well placed Make It Rain should make the lane quite easy as long as you don't let him get the item advantage. A good Lucian is hard to deal with because he can cleanse your slow and avoid most of the damage from Make It Rain. As long as you don't eat a lot of Qs aimed at minions from him you should be fine, but play conservatively and aim for teamfights in Lucian matchups because Make It Rain into Bullet Time just totally owns ~DA CULLING~ in a team skirmish. | ||
|
Alaric
France45622 Posts
Also The Culling has a shorter cooldown than Bullet Time, keep that in mind.  | ||
|
Nemireck
Canada1875 Posts
On June 23 2014 19:43 ZERG_RUSSIAN wrote: Quinn with either Thresh/Morgana/Leona is probably the hardest matchup because the blind disables 100% of your auto-q-auto Trist has a strong level 6 all-in which can interrupt your ult plus a massive gapcloser for gank support. Draven should get totally dunked unless you screw up huge by trying to trade with him before you have at least 2 ranks in E. In general against Draven you want to all-in a lot because his harass is much stronger than his commit. As always, support dependent, though, but a bit of jungle pressure + a well placed Make It Rain should make the lane quite easy as long as you don't let him get the item advantage. A good Lucian is hard to deal with because he can cleanse your slow and avoid most of the damage from Make It Rain. As long as you don't eat a lot of Qs aimed at minions from him you should be fine, but play conservatively and aim for teamfights in Lucian matchups because Make It Rain into Bullet Time just totally owns ~DA CULLING~ in a team skirmish. Against Lucian I just play the old reliable Q max, W second build. E is almost pointless because he can just dash out of it. I find going up against Lucian to be pretty safe as MF, Double-Up trades well with Lucian Q so harass is a wash, and the lower CD on Double-Up means by level 6 you should be able to get 2 off in an all-in (getting my second Q in an all in is what I miss most when I try your E-max). His E is just a bitch because he can escape your Make it Rain and Bullet Time so easily. | ||
|
Nemireck
Canada1875 Posts
On June 23 2014 20:39 Alaric wrote: What's the point of 2 ranks in Make it Rain? The very high damage + the slow that, if well-timed, can force Draven to miss an axe? Also The Culling has a shorter cooldown than Bullet Time, keep that in mind. He plays an E-max Miss Fortune against immobile ADCs (Twitch, Varus, Draven, Ashe, et al). I've been testing it but I still miss having my second Q available to me in all-ins, I'm not convinced it's any good yet. | ||
| ||
