Suggestion to Improve Rating Stability

As I understand it, chess.com uses the Glicko formula for chess ratings. One of the flaws (IMO) with this formula is the built in assumption that strength falls off as the player becomes less active. Activity however is relative. I may choose to play unrated chess and keep active and hence the assumption made by the formula is unreliable. The K factor increases as time passes between rated games.

I didn't play chess about one year,maybe more and my rating is now 200-300 points less then before.Rating is good,people don't play unrated games too much.Why would you play unrated?

This formula doesn't affect ratings as much as people like to make it seem.
Your rating will always catch up to your playing strength, so even if you stop playing here for years, once you come back, your rating will come back to where it was after a few games.
 

I had a glitch when I was typing this original posting on my Android. I thought the message was lost and I never came back to finishing my thought regarding the suggestion I had in mind.  I misspoke about the K Factor.  That was inaccurate.  I meant to reference the RD factor of the Glicko instead.

Suggestion:  Given that unrated play isn't true "inactivity" I was going to propose the idea that the rating formula should be applied to all unrated games.  The only difference is you throw away the rating change but keep the RD factor current. 

In this way, the RD factor does not increase over time and when the player does resume rated play, his actual rating stays relatively stable based on his perfomance.

That's not how it works.

It makes the assumption that if you haven't played a rated game for a while that your rating may be less accurate than someone who has continued to play rated games. Less accurate, not less strong.

A fair assumption I'd say.

Under your idea someone could win 20 unrated games in a row through studying a lot (or vice versa) and be treated as if their rating is still an accurate and up to date reflection.

in that scenario - if your study improved your play, and you win games, your rating still increases.  Just not by 30-50 point increments but by 8-10 (roughly speaking). 

Yeah so it takes longer to get to an accurate rating, and more players have to play a player with an inaccurate rating.

So, not an improvement.

Ratings are a measure of performance.  Not strength.  Fast acceleration with a high RD factor is misleading.

A measure of performance used to match people of similar strength. Semantics. Anyway your idea isn't an improvement and that's why it won't be implemented.

That's a matter of opinion.  It would be an improvement on how unrated games are played. If you don't play unrated games you would not be affected.

You could always design your own rating system where unrated training games have an effect on your rating coefficients. It wont take off though.

I play a lot of unrated games and I think it's ridiculous to see 50-60 point swings in my few rated games that I play.  Fundamentally, my strength hasn't improved or declined significantly.

You can "game the system" by allowing your rating to go "stale" with unrated games. Then when you do play a rated game, select a weak opponent to assure a better than average chance of winning.  In winning, you gain 50 points in a single rated game. 

By keeping the RD current, as in my suggestion, you can prevent people from gaming the system and improve the overall accuracy of the ratings for those people who play rated games regularly.

Glicko's forte over Elo is just that, the RD. It's a very well programmed rating system, arguably the best out there, and it's only logical that someone with a high RD will have greater rating swings (and likewise affect the opponents rating less) per game.

If someone were to, like you outline, play unrated games and only once in a while play rated against opponents a little weaker, and when they're on a winning streak... Yes, their rating would be skewed, but it's a matter of little consequence since this cannot affect the rating pool as a whole. Despite having a lesser average playing strength than their rating suggests, their performance in rated games would be consistent with the measured rating. Which is the point.

The system is designed so that rating changes reflect your probability of winning.

So you can't on average 'game it' by playing weak opponents.

If you are 200 points higher than your opponent you should win 3 out of 4 games. If you can do better you're not gaming the system, you deserve to be rated higher.

Hmm, I am trying to figure out the effect of all those unrated games and study... Weren't you saying that your strategy would allow players like you to game the system? What do you think your real rating should be?

Mike, 

It sounds like you are upset with the rating system because you have seen your rating change. The Glicko system accurately measures your  performance, which fluctuates. I'm sure there are ways to tweak it so that it could be even more accurate, but artificially stabilizing ratings makes the ratings less accurate. 

Are you really arguing in favor of less accuracy in favor of more stability?

Mike

There's more to the RD increase than performance slipping with inactivity.  You play unrated games.  These are not factored into your rating.  Thus, the uncertainty in your rating increases because your recent performance hasn't been measured.  Your present performance probably differs from your rating, and the only way to capture that is the RD.

Maybe not the most accurate, but...

The best, Jerry. The best!

@notmtwain - I'm not really making any effort to game my rating here on chess.com so your graph isn't relevant.  To answer your question I should be rated between 1600-1650.  Those few rated games you see on my chart were accidents.  The Android app can be sneaky and sometimes I accidently play rated games that I didn't intend to. 

FYI - I did try gaming the rating on FICS (which also uses Glicko) and managed to bump it to about 1750 before I lost interest.  My USCF is floored at 1700 and is currenly 1902 (1729 Quick). My FIDE rating is 2054. I have not played active tournaments since January 2013.  I'll leave you to decide if my current chess.com rating is accurate or not.

@SmyslovFan - I'm not "upset" with Gliko.  As a rating formula it has it's strengths and weaknesses. I just believe Gliko can be manipulated by altering the number and frequency of rated games played. 

IMO - Gliko actually discourages rated tournament play by those people who wish to manipulate their rating.