[Tom]

Originally Posted by DuckSauce

No it's not. The player who only has a few games isn't secured the same win ratio if he/she continues on, whereas the player with more games has played more, and has kept that ratio. Someone with more games, of the same win ratio is much more deserving than someone who could potentially meet his ratio (and for arugment sake, rise higher).

I am probably going to get nowhere with this, but here goes...

The problem with small sample sizes

I agree that in the example of a 60-40 player versus a 6-4 player, there is truth to what DuckSauce is saying. Results derived from a sample size of 10 could easily be an artefact, whereas a sample size of 100 allows one to be far more confident that the result is not due to sampling error (ie the more games they play at the same win%, the more confident we can be the player didn't simply fluke the results).

Carried across MMR

But I think we are across purposes in this "debate". I am not asking the system to assign me a rating based on a small handful of games. My previous season's MMR was carried across to this season, and I have played more than enough games for the system to be confident in my MMR. While it is necessary to play further games to "test" whether or not I am still "deserving" of my rating, it is not necessary to play mass games for the following reason:

Diminishing effect on confidence levels of increasing sample size

As anyone with a basic knowledge of statistics will know, once a certain sample size is reached, there is a diminishing effect on the confidence level in the results. For example, whilst a sample size of 30 might give you a confidence level of 95% +/- whatever (ie, there is a 1 in 20 chance the result is a fluke and a 19 in 20 chance the result is accurate), doubling the sample size may only increase the confidence level in the result by a couple of percentage points.

Obviously, it is necessary for a player to play more games as their relative skill may change over time. But this doesn't require many games. If the player only plays 3 or so games per week, but continues to win against high MMR players at a similar rate, we can reasonably infer the player still deserves their rating.

Level of decay should reflect number of games necessary to be confident in player's assigned rating

If the decay system is designed to test the continued accuracy of the rating it has assigned to a player (which in my view is the only reason to have a decay system), then it should be desiged to decay at a level that encourages a player to play the amount of games necessary for the system to be confident that player's rating is accurate, but not at a greater level. Anything more than that is cruel and unusual punishment.

It seems to me the present system of decay requires an unnecessary number of games to "test" whether a player still deserves their rating, and is therefore too high.

Quick Comments

nirvAnA:

love your posts tom