Glicko RD

Hi All,

I understand that with an rating of 875 and a Glicko ratings reliability (RD) of 40 that I have an ideal real strength somewhere between 835 and 915 with a 95% confidence level.  What I haven't been able to understand is why some descriptions of RD on this site claim that a higher RD indicates a better player.

It would seem intuitively that someone with a lower RD would be more consistent.  The strength of previous opponents and the number of games, among other things, is factored into the RD.  Thus, the more games one plays and the lower their RD, the more consistent they should be.

Is this a correct assessment?

I am sorry if I misunderstood your question which is likely for the answer is in there. The higher the GRD is the less accurate the rating is. It means that a player with high GRD may be much stronger/better than their rating shows. Since the top rated players do not play that many games their GRD is usually quite high. 

I don't know if I'm dense (probably) or if your question is, but it took me awhile to figure out what you were getting at. Now that I "get it" I have no idea if your assessment is correct or not, but maybe this will help you

http://www.glicko.net/glicko.html

If not, I suppose you could try contacting Dr Glickman directly.

Thanks guys...I understand what you are saying Boletus but you are only looking at the right hand tail of the bell shaped curve.  What about the people two iterations out on the left, don't they cancel out the very good players?

I guess if a person has a relatively low score with a high RD probably not good (all things being equal more or less) but a high score probably shows relative strength.  

Or maybe not.  I suppose it is like every other statistical average.  One must examine each data point to understand the underpinnings of what that data point represents in context of the whole sample.

"What I haven't been able to understand is why some descriptions of RD on this site claim that a higher RD indicates a better player."

I think that's either an error in their wording or your interpretation, but it's not correct.

From one Scott to another, I think it was a bald statement that those with a higher RD rate a better, but I suppose each player has to be reviewed to determine what their RD rate really means.

Jamie,

It basically means that your relative strength as a player is ideally between 796 and 904.  That's a pretty good spread so it means that you can bascially compete at a level over 900 with a 95% confidence level, at least based on an ideal formulation of a lot of factors combined to determine your RD.

It doesn't mean you will win all games over 900 but that you are competitive and have shown in the past that you can beat players at about a 900 level.

RD is ONLY a measure of how accurate your rating is based on frequency of play. higher RDs mean nothing more than lower accuracy :)

Think of RD almost as a margin of error in the rating. You are correct to say that the lower the RD, the more accurate the rating. As you play more games in a short span of time, you can be more confident that the rating accurately reflects your performance.

But even a perfectly accurate rating is only a measure of past performance and is not a perfect predictor of future results. If you have an RD of 0 and you play someone rated 200 pts higher than you will still have a 25% chance of beating them.

I think an easy way to think about it Jamie is that each game you play,  there is a very high probability that you will play at a rate somewhere between 796 and 904.

My point was that if I am playing games against oppnents that have a somewhat lower or higher score and my RD is low, it points to the fact that I am playing at a consistent level of my ideal strength.

RD means ratings deviation.  You can think of it as a mathematical formula that determines the probable range of your lowest score to your highest score on every game you play.  Some days we play better than others days so think of the range of 796 to 904 as the strength of your game each game you play.   

As SFan alluded to, if you are playing a much higher or lower player than yourself, if you lose it won't affect your RD ranking very much.  If you play someone who is close to your own score than if you win or lose it will have a much greater effect on your RD ranking.  

RD can't measure unrated players.  You need a minimal number of games played for the formula to be able to make a prediction of your relative playing strength.

Forget the RD and hit the chess books instead, @smb.  Your rating is in the tank.

zborg you know nothing about me, when I started to play chess, how often I play, etc.  You are an obnoxious person.  This is a constructive forum and should be used to edify and educate.  If I knew you personally I would show you how men work out disagreements if one of them is not a gentleman.

Even if players had the same rating but different Rating Deviations - the one with the higher RD could be way stronger OR way weaker, so a "claim that a higher RD indicates a better player" doesn't seem to have any basis.

According to Dr. Glickman, unrated players start at 350, so that should be the maximum. Also, Dr. Glickman recommends that it never be allowed to drop below 30.

Blow a little harder, please.  It becomes you.

You've played about 2000 games, to little avail.  It's clearly time to start a study program. You'll be a better "gentleman" for it.

Beats me I didn't read it but I thought it might be more comprehensive and/or helpful than chess.coms description of the Glicko rating system - or that it might provide an answer for smb1959

PS: I tried reading it, but the math equations were way over my head.

That's a great answer Timothy.  This has been a good conversation thread.  Do you know why Glickman thought it wasn't wise to fall below 30?

Perhaps having a score under 30 indicates that one is not playing opponents having a higher score to be challenging, or a person is not playing enough,  or person may be playing too many people with a lower RD?

She recommends that the formulas not allow the RD to go below 30 or a similar threshold value. If you let the RD fall below 30, she says, players who are improving may not see any change in their ratings.