Expected Rating Feature

Sort:
jntrcs

I've been thinking of a feature that would be really cool to see and fairly easy to implement. As part of the post game analysis, I want the analysis to provide statements like "During this game you played like someone rated XXX (e.g. 920) and your opponent played as XXX (e.g. 830)"

This would be extremely interesting to me, because when I go on 8 game losing streaks I am really interested in whether I am making way more mistakes than is typical for me, or whether I'm just getting match ups where my opponent is significantly outplaying their current rating. It would also be interesting to see games where you and your opponent both significantly overperformed or underperformed your ratings!

As a statistician, I really don't think this would be a difficult metric to create. There's be more complex approaches, but I would start with a multiple linear regression where the response is the person's rating going into the game and the predictors are the number of best, excellent, good, mistake, and blunder moves (all divided by the total number of moves in the game so we are modeling the rate, so games of different lengths are comparable). Throw in some splines to address non-linear relationships and we've got a model to predict your rating from a single game. This prediction would be really interesting to look at after each game!

I hope Chess.com runs with this. If they do, I definitely think they should give me a free lifetime membership for suggesting it happy.png

notmtwain
jntrcs wrote:

I've been thinking of a feature that would be really cool to see and fairly easy to implement. As part of the post game analysis, I want the analysis to provide statements like "During this game you played like someone rated XXX (e.g. 920) and your opponent played as XXX (e.g. 830)"

This would be extremely interesting to me, because when I go on 8 game losing streaks I am really interested in whether I am making way more mistakes than is typical for me, or whether I'm just getting match ups where my opponent is significantly outplaying their current rating. It would also be interesting to see games where you and your opponent both significantly overperformed or underperformed your ratings!

As a statistician, I really don't think this would be a difficult metric to create. There's be more complex approaches, but I would start with a multiple linear regression where the response is the person's rating going into the game and the predictors are the number of best, excellent, good, mistake, and blunder moves (all divided by the total number of moves in the game so we are modeling the rate, so games of different lengths are comparable). Throw in some splines to address non-linear relationships and we've got a model to predict your rating from a single game. This prediction would be really interesting to look at after each game!

I hope Chess.com runs with this. If they do, I definitely think they should give me a free lifetime membership for suggesting it

 

That's sort of what the new enhanced analysis does.   The accuracy score gives you an assessment of how you played.

You can't really say you played like a 920 because it only takes one move to throw away a game.

How many of those reports have you run?

Here is one from your most recent game.

jntrcs

Right, I think it would be just one more step to map the accuracy to an expected rating. Obviously it's going to be highly variable (especially depending on whether you resign after a blunder or continue onward), but even in the cases where one blunder totally ruined your game that number would be a useful indicator of how well you played the rest of the game. Accuracy contains the information I want, but reporting it in terms of ratings is much more interpretable since we're familiar with it. I have no idea what my average accuracy is, but I know what my long run average rating is.

jntrcs

Also, one of the reasons I don't know what my "normal" accuracy is is because it's not available to us non-paying plebeians.