I searched everywhere in FAQ but could not find any answers! • page 2/2 • Lichess Feedback • lichess.org

@StingerPuzzles Ok, I looked it up and did some tests and I'm pretty sure what I said is true, at least for Glicko 2.

In Glicko 1, RD indeed doesn't increase (or change) based on the game result but even there it still increases when not playing for a while.

In Glicko 2, the RD increases proportionally to the volatility (and decreases based on the expected outcome).

Quoting the Glicko 2 spec: "The volatility measure is high when a player has erratic performances (e.g., when the player has had exceptionally strong results after a period of stability), and the volatility measure is low when the player performs at a consistent level."

So as you say, the volatility goes up "under certain circumstances", namely when the player wins a lot in a row as I said (and of course also when losing a lot in a row).

And since the RD increases proportionally to the volatility, it can then also increase in that case.

Doing some practical tests using a Glicko 2 implementation also confirms that this can indeed happen and that the increase isn't always overshadowed by the decrease. It even seems like Lichess' system, where (I believe) each game is treated as a single rating period, leads to faster increases.

Of course, I can't say for sure what happens on Lichess since I didn't check what parameters it uses and how closely it follows the spec (e.g. I think it probably also doesn't increase the RD based on time?) but at least in principle, the RD should behave as I said with the Glicko 2 system.

@StingerPuzzles Ok, I looked it up and did some tests and I'm pretty sure what I said is true, at least for Glicko 2. In Glicko 1, RD indeed doesn't increase (or change) based on the game result but even there it still increases when not playing for a while. In Glicko 2, the RD increases proportionally to the volatility (and decreases based on the expected outcome). Quoting the Glicko 2 spec: "The volatility measure is high when a player has erratic performances (e.g., when the player has had exceptionally strong results after a period of stability), and the volatility measure is low when the player performs at a consistent level." So as you say, the volatility goes up "under certain circumstances", namely when the player wins a lot in a row as I said (and of course also when losing a lot in a row). And since the RD increases proportionally to the volatility, it can then also increase in that case. Doing some practical tests using a Glicko 2 implementation also confirms that this can indeed happen and that the increase isn't always overshadowed by the decrease. It even seems like Lichess' system, where (I believe) each game is treated as a single rating period, leads to faster increases. Of course, I can't say for sure what happens on Lichess since I didn't check what parameters it uses and how closely it follows the spec (e.g. I think it probably also doesn't increase the RD based on time?) but at least in principle, the RD should behave as I said with the Glicko 2 system.