the CLT states that the sampling distribution of the mean is approximately normal. How do you expect your argument to be taken seriously when you don’t even get stats 101 topics right?
I don’t think you even have to go as far as that, you can simply point out that Elo is a system that ranks you based on relative performance to the other players in the pool, so unless you can prove that the average skill for all 3 races is the same then the point is pretty moot.
This has been repeated quite a lot, its in all of the links on wikipedia as well as the stuff I’ve posted, if anyone cares to actually browse through any of the links posts here then its quite clear that batz doesn’t understand the fundamentals of pairwise comparison models.
Thats incorrect, we know for a fact that the race distributions are very different in region to region, therefore it is clear that people are not randomly choosing a race and therefore we cannot expect that skill would be IID.
What’s wrong with just plain Jane win rates? Given enough sample size, it should be clear what the trend is.
In all honesty, it’s possible that Zerg is UP. But, Zerg players lost a ton of pros (used to be around 38%, now around 33%), which is why he shifted his goalposts. It used to be: “MOMMY! LOOK! ZERG HAS 38% OF PROS THEY SHOULD WIN MORE OFTEN!” But that’s not a factor any more, so he shifts the metrics he looks at. That’s why I don’t argue with him.
Could he be right this time? Sure. No one exemplifies a stopped clock more than he does, so it’s about high time he got one.
The win-rates are all over the place and if you average them over a large time period you’d see they go to 50/50. That’s because balance is a REALLY small variable compared to other variables like skill and representation. Win-rates don’t filter that out so they are a mash of the random noise caused by who shows up to what tournament or not.
Even the distributions of the tournament formats themselves affect win-rates more than balance does. Code S has one format, IEM has another, weekly cups still another.
Dude, everyone knows your conclusion before the thread even starts: Zerg players are massively more skilled than Terran players on average. Zerg is UP, Terran OP, let’s all just go home.
This analysis subtracted skill from the equation. That’s the point of the analysis. Clearly your assumption about the thread is wrong. Please read the original post before commenting in the thread.
Sometimes things are not as simple as they appear: one expect one thing and something quite unexpected hapens:
Norwegian undertakers request economic assistance from the government because the Coronavirus lockdown led to a decline in deaths: [Færre dør – krise for begravelsesbyrå]
Norway acted timely. Its Covid-19 peak of new cases was about 270 per day (7 day average) and the total deaths were some 250. The country has reopened and it now has under 10 new cases per day. The lockdown did not only prevent an out of control epidemic situation, it also stopped accidents from happening. Less stress led to fewer other incidents like heart attacks. In total the number of deaths declined!
Exactly, but that doesn’t mean you can’t compare them if you normalize.
I’m going to use the example of a normal distribution. As I discussed, if a matchup us extremely coinflippy compared to another, it will have a lower standard deviation (because player elos are more clumped up) than a matchup that is less coinflippy. This means that once you standarize the players of each matchup, when you divide by the standard deviation, the lower ELO differences from the coinflippy matchup will be divided by a lower standard deviation while the higher ELO differences for the less coinflippy matchup will be divided by the higher standard deviation. Parameters such as variance and sandard deviation take care of those coinflippy matchups generating more clumped up resuts when you normalize them, that is the power of standard distriutions. Of course this only works if the population size is large enough.
Race distributions do not necessarily correlate with skill distributions. Even if more players choose say zerg in europe because of whatever reason (and keep in mind that I want to expclude the very top players from this, because there is not enough of them to, only focusing on them, come to any conclusion regarding skill), the skill distribution of those players still will probably follow the same curve as the more common terran or zergs. Unless there was a specific factor that made say people with lower reaction times like terran because something in the brains of such people like the concept of marines or some other factor that made players that are able to have a higher “skill” potential to pick a race in particular. Bear in mind that this is speaking of rank in the particular mirror matchup of the race of eah player, not on overall race, this way factors such as different metagame changes in XvY matchups affecting winrates, or balance changes, do not come into effect. I’d be willing to bet that rank distributions for mirror matchups are extremely identical for all 3 races unless there was a hidden factor making higher skill potential players to choose a particular race. Of course, I also want to repeat the necessity of putting a filter of say 300 games played per selected player as minimum.
Edit: If somebody would be kind enough to point to me where to actually get the data of players-mirror matchup rank (for a lot of players, separated per race) I’d be quite happy to make a quik script to plot them into scatter graphs or whatever to take a peek if they are similar.
They pick their race long before their full skill is realized and the dynamics of each race are constantly being mixed up. HotS era biomine is very different from, say, current era hellion cyclone styles.
This reminds me of the great IQ debate. Everyone was saying that this might contribute to IQ, that might contribute to IQ, etc, so they did a study. They broke IQ down into as many sub-elements as they could clearly differentiate and then did studies to see which element corresponded with IQ the most. They found the ability to remember large numbers took up the majority of the correlation (~90%).
SC2 is the same way. It might reward this skill or that skill, but the bottom line is that it rewards skills which are universal to all matchups including click accuracy, apm, multitasking etc.
In probability theory, the central limit theorem (CLT) establishes that, in some situations, when independent random variables are added, their properly normalized sum tends toward a normal distribution (informally a bell curve) even if the original variables themselves are not normally distributed. The theorem is a key concept in probability theory because it implies that probabilistic and statistical methods that work for normal distributions can be applicable to many problems involving other types of distributions
