Extracting Rating Deviation and Volatility from Lichess Game Data

I am trying to compare the Glicko-2 and Elo rating models using Lichess game data for a research project. To do this, I plan to create a large Lichess game dataset, use the two different rating models to calculate their respective expected outcomes for each game, and then compare it to the actual game outcome to see how well the model performed. However, I am having difficulty calculating the expected game outcome with the Glicko-2 model because, as far as I know, there is no data that gives player rating deviation (RD) and volatility in the Lichess API (Glicko-2 is dependent on RD and volatility). Is there a way to access such data or a workaround to ensure I can calculate Glicko's expected outcome without directly using player RD and volatility? Perhaps there's something in the existing chess data that already accounts for these two variables?

@efoldemort said in #1:
> Is there a way to access such data or a workaround to ensure I can calculate Glicko's expected outcome without directly using player RD and volatility?
If you want to compare the different rating systems and you're going to simulate Elo ratings, I think it would be better if you simulated Glicko ratings as well
www.glicko.net/glicko.html
This should allow you to do this relatively easily

I'm not sure I understand. Lichess uses Glicko (technically Glicko-2) ratings, so I would need to
1. Calculate expected game outcomes with the Glicko model (which is difficult if RD and volatility isn't widely available)
2. Convert Glicko ratings into Elo, which would allow me to calculate the expected game outcome of the Elo rating models

I'm mostly focused on step 1 as of now, but I am also keeping step 2 in the back of my mind.

Not sure I understand, is it that you are working with games and it's cumbersome that you need to find the RD and volatility data from players, not games?

@efoldemort said in #3:
> Convert Glicko ratings into Elo, which would allow me to calculate the expected game outcome of the Elo rating models
Wdym "convert"?

@GoWithTheFLU said in #4:
> Not sure I understand, is it that you are working with games and it's cumbersome that you need to find the RD and volatility data from players, not games?

Yes. I want to calculate the expected outcome of each game using the Glicko rating system, but it's difficult to do so because each player's RD and volatility in each game isn't available in the Lichess API, as far as I know.

@aentrenus said in #5:
> Wdym "convert"?

"Convert" might not be the right term, but I'm not sure if there's necessarily an appropriate one. I want to compare the efficacy of Elo and Glicko models in terms of predicting game outcomes, but because Lichess doesn't use Elo, I'm trying to use translate/transform the Glicko rating model to the Elo model.

@efoldemort said in #7:
> but because Lichess doesn't use Elo, I'm trying to use translate/transform the Glicko rating model to the Elo model.
This will not give you meaningful results. What you should do (in my opinion) is get a large batch of games and literally simulate both systems yourself. This way, you will have all the information you want.

@efoldemort said in #6:
> Yes. I want to calculate the expected outcome of each game using the Glicko rating system, but it's difficult to do so because each player's RD and volatility in each game isn't available in the Lichess API, as far as I know

To be honest I haven't used the api, so I was just guessing. You can see RD of a player for each time control or chess variant in lichess.org/@/username/perf/rapid for instance. So that data point should be there. Maybe volatility not shown, but my guess was that those data points are stored within the player data and not within a certain game data. Just a guess.

You raise a good point that player RD is stored in the player data; however, this only gives the player's latest RD rather than historical game RD, which is what I'm interested in.