The biggest problem with pro elo threads is they assume that the logical consistency of ELO proves that it is a reliable predictor of behavior. The problem is that the first is completely analytic (non-testable), the other can be only be empirically tested and potentially falsified.
A simple test of the latter (reliability: http://en.wikipedia.org/wiki/Reliabi...ychometrics%29), is to ask what the mean win:loss ratio is for the entire player base. Then test if this average is significantly different from 50:50 a coin flip.
Using statistics, I'm extremely confident that this average is significantly lower that a 50:50 coin flip.
My argument,
1) 135,000 players are listed on the official latter (1200 elo+).
2) Riot reported at the end of season 1 that they have 11million monthly players
2a) Accordingly I think it's safe to say that maybe 1,500,000 players actively play ranked in a given month,
3) This implies that approximately 9% of the player base is at or above 1200 ELO.
Axiom) The only reliable way (95% of the time) to get below 1200 is to lose more matches than you win
4) Therefor, the vast majority of players (~90%) have a w:l ratio lower than .500.
5) If ELO reliably measures performance in LoL, then the average win:loss ratio must average out to a coin flip
5a) this is a corollary of the law of large numbers.
QED:
the ELO system does not reliably measure performance.
PWN'd by scientific method, statistics, and logic.
1) 135,000 players are listed on the official latter (1200 elo+).
2) Riot reported at the end of season 1 that they have 11million monthly players
2a) Accordingly I think it's safe to say that maybe 1,500,000 players actively play ranked in a given month,
3) This implies that approximately 9% of the player base is at or above 1200 ELO.
At the end of season 1 all of our players weren't only on the NA server. You would have to use data from all our servers.
[PROVEN] ELO doesn't reliably measure performance.
A simple test of the latter (reliability: http://en.wikipedia.org/wiki/Reliabi...ychometrics%29), is to ask what the mean win:loss ratio is for the entire player base. Then test if this average is significantly different from 50:50 a coin flip.
Using statistics, I'm extremely confident that this average is significantly lower that a 50:50 coin flip.
My argument,
1) 135,000 players are listed on the official latter (1200 elo+).
2) Riot reported at the end of season 1 that they have 11million monthly players
2a) Accordingly I think it's safe to say that maybe 1,500,000 players actively play ranked in a given month,
3) This implies that approximately 9% of the player base is at or above 1200 ELO.
Axiom) The only reliable way (95% of the time) to get below 1200 is to lose more matches than you win
4) Therefor, the vast majority of players (~90%) have a w:l ratio lower than .500.
5) If ELO reliably measures performance in LoL, then the average win:loss ratio must average out to a coin flip
5a) this is a corollary of the law of large numbers.
QED:
the ELO system does not reliably measure performance.
PWN'd by scientific method, statistics, and logic.
1) 135,000 players are listed on the official latter (1200 elo+).
2) Riot reported at the end of season 1 that they have 11million monthly players
2a) Accordingly I think it's safe to say that maybe 1,500,000 players actively play ranked in a given month,
3) This implies that approximately 9% of the player base is at or above 1200 ELO.