Rating Declining

[Jacob Marley]

I'm glad the ratings system is now up.  My main area of concern is the ratings decline due to inactivity.  In general, I agree with having this be part of the ratings system, my concerns are related to the actual mechanics.  My situation is that I don't play nearly as much as some others due to family/work demands.  Certainly I don't play every day, and I can expect stretches of inactivity due to family life/vacations etc where I could go several weeks with no activity.   This happened to me with MF as well and I often went back to zero rating because of it.

My proposal is two-fold.  First, there should be a grace period of maybe a week where ratings do not decline due to inactivity.  Secondly, there should be a floor (maybe level 20?) that you cannot fall below due to inactivity, only loses.  That way, if I'm gone for several months, I don't have to go back to the start every time I'm away for a while.

[JW]

I recall that in previous statements the developers have said that the ratings decline will be very small and much smaller than on the previous Making Fun / Goko platform. So I do not think this will be a concern.

[wyxxyr]

This was far and away the biggest reason I stopped playing on the other site. It was worse than having no ratings system at all, really.

*Makes ratings perpetually inaccurate as long as players' activity levels vary. (and thus uneven -- matches are more about active player vs. active, inactive vs. inactive than about skill).
*Difficult to want to resume playing after a break.
*Generally frustrating to have a measure of skill tied to a variable unrelated to skill.

The way to do this is to not change a rating, but simply remove inactive players from being displayed on leaderboards.

[markus]

If I understand correctly, the rating will only decline over time because of the standard deviation (phi) going up, but the expected value (mu) stays constant. I think it is fine to be conservative for the leaderboard and use the current rating, such that people with a high variance can't end up at the top.

But what should matter for matching purposes is the best guess for the skill, which is mu instead of mu-2*phi. That way, even after some inactivity you'd get matched with similar people as before.

[markus]

I did some calculations and the rating decline due to higher phi is really not something to worry about too much: at the moment it will take around 2 months of not playing to lose 1 level (you can see that also from people not playing yesterday having a rating of 19.99 today).

After playing more and having a lower phi, that decline will be a bit faster, but it will still probably take you 3 weeks to lose 1 level.

What really matters more at the moment is that everyone who plays significantly reduces phi and therefore increases the level - winning ratio is less important. That's why I really think that matching should care more about mu than about mu-2*phi. Today, I wouldn't be matched with someone who hasn't played yesterday evening, but I could (hypothetically) get matched with people who lost 12/18 games.