Roberson Rating System

[Peter Hornsby]

http://aigames.net/Research/Elo-Roberso ... ystem.html

Hi all,

A friend of mine from Burlington USA has written the following paper and would be grateful for any feedback, just comment below on your thoughts

[Chris Goodall]

The K component of Elo is a total fudge - I'm with him that far. Let's do these beautiful calculations, and then let's multiply it by whatever number makes it look right. A bit like the heat index formula. https://xkcd.com/2026/

Where he loses me, is here:

The primary problem is the fundamental concept around its creation - "the more you play the more consistent you become". Another way of saying this is the more you practice the less likely you are to improve. Also, your ability to improve has nothing to do with your current level of play. This flies directly in the face of points 2, 3, 4 and 5. Of extreme importance is that it defies the age old concept of "practice makes perfect".

Do not trust age-old concepts. That's not science. Perfect practice makes perfect. Imperfect practice merely makes practiced. His theory runs into the indisputable fact that chess players don't all become World Champion, regardless of how many games they play. They hit a point of not just diminishing returns, but zero returns. They are practicing playing bad moves. Glickman is absolutely right that the more you do that, the better you get at it.

I suggest that any server using the Glicko system replace it with another system that more properly adheres to the behavior charactersitics that model human competitve sport performance...

No, no, a thousand times no. He is committing the cardinal sin of data analysis, that of overfitting. Suppose I create a chess-playing automaton and instruct it to play at a strength of exactly 2000. If I let it loose on the Elo-Roberson System, or any system that has been optimised for human behaviour, that system will take longer to identify the automaton's true strength of 2000 than a system that is, shall we say, hardware-agnostic.

Now replace the 2000 automaton with a human who has attained a stable rating of 2000 on chess.com and lichess, found a mouse and a Wi-Fi hub that work well, and then creates a new account on EloRobersonChess.net. What's the difference?

A system designed to outperform Elo for a group of Intel employees who are all equally new to playing OTB blitz chess with a clock in a cafeteria, will underperform Elo for players who don't share those characteristics, which is almost everyone. In particular, the online version of the system would have no idea what point in your career trajectory you've already reached.

[Phil Neatherway]

Let's do these beautiful calculations, and then let's multiply it by whatever number makes it look right.

I used to work for an insurance company, and that sounds exactly like the approach the actuaries used in the annual valuation of the business.

[JustinHorton]

Ha ha surely not, that would be wrong

[John Clarke]

Let's do these beautiful calculations, and then let's multiply it by whatever number makes it look right.

I used to work for an insurance company, and that sounds exactly like the approach the actuaries used in the annual valuation of the business.

Ah well, even Einstein wasn't above that sort of caper, with his "cosmological constant".