Is there any provision to “re-initialize” a rating that is sufficiently stale as to be useless?
Again looking at Asher Josephson, they now have a regular rating of 1429 from a tournament that ended yesterday, but a quick rating of 1179 that is over a year old. If they didn’t have that quick rating, their quick rating would be initialized to something much closer to 1400, which I suspect is much more accurate than the old quick rating.
There are currently no provisions for this, deciding when a stale rating would be sufficiently improved by reinitializing it is not a trivial task.
I suspect there would be complaints if it was applied to regular OTB ratings, too. (There has always been extreme reluctance on the part of players to have any other ratings information affect their OTB regular rating.)
It’s not a trivial task, but there does appear to be a fair amount of relevant work already done in the rating system:
Possibly a comparison of the Adjusted Weight on the other rating system and the hypothetical Adjusted Weight on the existing rating could be used, with a minimum age of the stale rating for re-initialization.
There’s a bit of a fairness issue here for opponents - playing someone that is severely underrated is never fun, and it seems sub-optimal to have to slowly improve each rating system separately.
If we had to go through the initialization process for every player with ratings in other ratings systems for every event to see if a reset would be helpful, that would be a significant increase in the computational resources needed for rating.
My understanding is that the RC is currently looking at the staleness issue, but I have not yet seen their proposal yet.
I’m sure that could be reduced significantly by some reasonable optimizations, such as only checking monthly to update stale ratings before publication of the next supplement. The number of players with a) stale ratings in a rating system and b) activity in a different rating system in the prior month would not be expected to be that large.
The first step in the ratings process for a section is looking up the previous rating (if any) for that player in that ratings system. If the data returned included the date of the player’s most recent event, that would be the first part of a staleness check. The absence of a previous event in that ratings system means that player’s rating needs to be initialized via other procedures. Whether the current data structure and lookup process returns that information is beyond my knowledge of the rewrite of the ratings programming.
A stale rating may mean the player has been playing in other ratings systems (possibly not US Chess rated events) or it may mean that player hasn’t been playing chess at all. And the age of the player can factor into that, someone who played a handful of events at the age of 10 but is now 24 may be a better player without having played a single game of chess in the interim, just because that person is older and has more developed mental faculties.
What the RC is looking into appears to be staleness in the absence of other information, such as play in other US Chess ratings systems. Utilizing that other information would likely trigger different measures to deal with staleness, but that’s all possible.
But it will take time to formulate and test the various methods to deal with staleness and make sure they are both mathematically valid and politically acceptable.
