PDA

View Full Version : Rating Differentials

Garvinator
27-06-2006, 04:52 PM
Does anyone know of a site that states how much one person should score against another over a series of games- ie players with 300 point difference play a 10 game match, what should the overall score be based just on rating.

I am sure there has to be a site somewhere and I admit I havent searched for one :oops:

Bill Gletsos
27-06-2006, 04:58 PM
Does anyone know of a site that states how much one person should score against another over a series of games- ie players with 300 point difference play a 10 game match, what should the overall score be based just on rating.

I am sure there has to be a site somewhere and I admit I havent searched for one :oops:Check the FIDE site at http://www.fide.com/official/handbook.asp?level=B0210

That shows the expected % based on the normal curve.

You can however use the superior logistic curve formula which Elo recommended in his book and is used by the USCF, Glicko and others:

P = 1/(1 + 10 ^ [D/400]) where D = the rating difference.

Garvinator
27-06-2006, 05:24 PM
Check the FIDE site at http://www.fide.com/official/handbook.asp?level=B0210

That shows the expected % based on the normal curve.

You can however use the superior logistic curve formula which Elo recommended in his book and is used by the USCF, Glicko and others:

P = 1/(1 + 10 ^ [D/400]) where D = the rating difference.
Thank you Bill, the reason I asked was that I wanted to see if AO was willing to state how many points start he could give an opponent. Since he thinks of himself as a GM, it should be plenty :P

I just wanted to see if his thoughts would match reality :uhoh: :wall:

qpawn
27-06-2006, 06:38 PM
Won't that site give only the ELO results?

Don't we use a funny glick thingy now?

Kevin Bonham
27-06-2006, 06:39 PM
Won't that site give only the ELO results?

Don't we use a funny glick thingy now?

The scaling of Glicko and ELO is the same so the same differentials should apply.

Bill Gletsos
27-06-2006, 07:10 PM
Won't that site give only the ELO results?I pointed out the FIDE site uses the normal distribution.
The USCF, Glicko amongst others use the logistic function.

Don't we use a funny glick thingy now?Garvin was interested in expected percentage. The difference between normal and logistic is minimal up to around a difference of 350 points, however even at 600 points there is only a 1.4% difference in expected percentage.

Igor_Goldenberg
28-06-2006, 10:06 AM
It would be an intresting experiment to see how accurate ELO formula in estimating the result. E.g a regression of result on three variables (ELO prediction for the game, colour and smallest rating in the pair) will show whether:

1. Whether ELO formula is a good estimator
2. How is it affected by the colour.
3. How is it affected by the level of players.

I strongly suspect that results will be different for high and lower rated players.

ACF rating officer must have a vast database of results.

Bill, do you want to do this exercise? Or to send data to me so I can run some statistical analysis?

Glenno
12-06-2007, 06:05 PM
It would be an intresting experiment to see how accurate ELO formula in estimating the result. E.g a regression of result on three variables (ELO prediction for the game, colour and smallest rating in the pair) will show whether:

1. Whether ELO formula is a good estimator
2. How is it affected by the colour.
3. How is it affected by the level of players.

I strongly suspect that results will be different for high and lower rated players.

ACF rating officer must have a vast database of results.

Bill, do you want to do this exercise? Or to send data to me so I can run some statistical analysis?

The website http://www.chesscenter.com/twic/event/sonas/sonasrat.html includes a graph that displays White's percentage score in relation to the difference in the ratings.

pax
18-06-2007, 04:31 PM
It would be an intresting experiment to see how accurate ELO formula in estimating the result. E.g a regression of result on three variables (ELO prediction for the game, colour and smallest rating in the pair) will show whether:

1. Whether ELO formula is a good estimator
2. How is it affected by the colour.
3. How is it affected by the level of players.

I strongly suspect that results will be different for high and lower rated players.

ACF rating officer must have a vast database of results.

Bill, do you want to do this exercise? Or to send data to me so I can run some statistical analysis?

I believe Bill did exactly this sort of modelling when Glicko was originally introduced, and concluded that Glicko is a far better predictor than ELO.

Igor_Goldenberg
19-06-2007, 09:23 AM
I believe Bill did exactly this sort of modelling when Glicko was originally introduced, and concluded that Glicko is a far better predictor than ELO.
As far as I understand, Glicko uses the same formula for expected result P=1/(1+pow10((D1-D2)/400)). That the one I'd like to test.