RD for problems.

[drahacikfm]

I'm wondering about the RD (rating deviation) for the problems, as opposed to the RD for players.  The idea for players is that the RD goes up when you don't play for a while.  That's because during that time off, the player could have studied a lot and gotten better.  Or the player might have done no chess at all, and gotten worse.  In either case, the player's rating is more uncertain after not playing for a while, so the RD goes up.

But for a problem, it really doesn't matter if the problem is "active" or not... it's exactly the same problem with exactly the same difficulty whether it was attempted 50 times in one day, or 50 times in one year.  So it seems the RD for problems should depend only on the total number of attempts, and should not depend at all on any kind of time factors.

Maybe that's how it's done already?

[richard]

Currently RD for problems is calculated in a similar way to RD for players.  Up until a week ago it was calculated exactly the same way. However I've recently modified problem RD calculations to slow down the rate at which it goes up as it appeared that with the current number of users the problems still had a little too much volatility due to average RD remaining relatively high.

I think the idea of problems volatility being tied to the number of times they have been done is an interesting one.  Intuitively it makes sense, however I think time based volatility still has a couple of advantages.  Firstly if someone (or group of someones) made a wide scale cheating attempt that brought a large number of problems down in rating under a "problems done" RD approach it would be harder for the ratings to recover. Also if I change aspects of the problem solving (such as the recent hiding the ratings when tactic is live) in ways that impacts how hard the problems are, the time based approach allows the problems more of a chance to mutate to a more appropriate rating, a "problems done" approach would lock the problem into it's rating more and more over time. I could look at resetting RD when these types of situations arise but I like the self-healing nature of time based RD.

So in general I think your observation is correct, however the assumption that problems don't change in difficulty over time probably doesn't hold in a couple of important corner cases (cheating and changes in tactic solving conditions).

Regards,
Richard.