3 edits
Originally posted by AThousandYoungThis doesn't make sense to me. The Elo system doesn't say the draw expectancy is zero; it just doesn't specify what that expectancy is.
So, if there are problems with the system, it's in the Win Expectancy and implied Loss Expectancy figures. Either the fact that the Draw Expectancy is always zero throws off the calculations or the Win Expectancy calculation is simply not accurate based on the statistics. .
It attempts to predict your expected score against various rated opposition, not your expected wins and losses.
For example, an expected score of 0.76 points from (let's say) a one-off game includes drawing possibilities.
That is
W + D/2 = 0.76
L = unknown
not
W = 0.76
L = 0.24
D = 0
where
W = probability of winning
L = probability of losing
D probability of drawing
Originally posted by AThousandYoungIf both players would perfectly fit the statistic behind the rating system, then the difference in playing strength should in the long run not influence the ratings. However, there are several factors making the real statistical behaviour imperfect.
I don't know actually. I'm just going by what someone said.
I'm talking about the RHP system. Is it the same as the ELO system?
Let's try a hypothetical.
Suppose we take a 1600 named A and have him play, over and over, people with exactly 200 rating points less.
The Win Expectancy for A will be according to the formula 76%. If this ...[text shortened]... rs would be the ratings of both players at the beginning of the game and the result of the game.
The most important one is the fact that here on RHP, most players do not have the rating that reflects best their playing strength. One obvious reason for that is that we all start at 1200. The provisional system is an attempt to improve on this. But it doesn't change one fundamental factor: the average of the ratings is still close to 1200 (we give and take equal amounts), and there is nothing that justifies this average. Perhaps that starting level should also have inflated along with the average of the 'established' players. A related problem is that there is no absolute truth, because the RHP-ratings cannot be compared meaningfully with any other.
Anyway, I believe that the 'true RHP average' rating should be above 1200. The population with rating above 1200 are probably under-rated, and so are the differences in ratings. As a result, the win expectancy is also underestimated for the stronger player. Which makes me conclude that the statement is probably true: playing lower rated players has a positive impact on the rating.
Another attempt to correct this could be a 'league' system, where rated games are played mainly withing a small(er) range of ratings, eliminating the above mentioned effect.
Originally posted by THUDandBLUNDERAs I said, I am talking about the way RHP calculates ratings, not ELO ratings - unless they are the same.
This doesn't make sense to me. The Elo system doesn't say the draw expectancy is zero; it just doesn't specify what that expectancy is.
It attempts to predict your expected score against various rated opposition, not your expected wins and losses.
For example, an expected score of 0.76 points from (let's say) a one-off game includes dr ...[text shortened]... = 0
where
W = probability of winning
L = probability of losing
D probability of drawing
Here's where I got the formula:
http://www.redhotpawn.com/help/index.php?help=faq
In the formula is a value named the "Win Expectancy". If a person wins the fraction of games predicted by "Win Expectancy", and loses the rest (giving a "Lose Expectancy" of LE = 1-WE), then the person's rating will not change. An implication of this model is that the person will never draw. If the person draws, then this couldn't be part of "Win Expectancy" because a draw is not a win, so it would be taken from the "Not Win Expectancy" and the rating would increase over time for all players.