Great article. Thanks for the investigation & analysis.
The 1400 rating isn't a "floor" it's actually more like a "cliff". If you drop under 1400 you actually lose your rating & have to requalify for a new rating again.
Then when players lose their rating, while they are unrated, their opponents don't gain (or lose) rating in games played against the unrated player.
I've noticed more people in the 1400's have been losing their rating. It seems to be more than what occurred when it was a 1000 minimum. Especially older players playing in tournaments with lots of juniors.
A "floor" would mean players who drop under 1400 are raised back up to 1400 as a minimum rating, rather than becoming unrated. It would also also mean future games with their opponents would be rated for both players, as people wouldn't lose their rating.
@sahope said in #41:
> Great article. Thanks for the investigation & analysis.
>
> The 1400 rating isn't a "floor" it's actually more like a "cliff". If you drop under 1400 you actually lose your rating & have to requalify for a new rating again.
>
> Then when players lose their rating, while they are unrated, their opponents don't gain (or lose) rating in games played against the unrated player.
>
> I've noticed more people in the 1400's have been losing their rating. It seems to be more than what occurred when it was a 1000 minimum. Especially older players playing in tournaments with lots of juniors.
>
> A "floor" would mean players who drop under 1400 are raised back up to 1400 as a minimum rating, rather than becoming unrated. It would also also mean future games with their opponents would be rated for both players, as people wouldn't lose their rating.
This is an artifact of the "fake 1800" opponent draws that bake into the initial rating calculation. It has moderate inflationary pressure on the system. Allowing players who would otherwise be too weak to qualify for an initial FIDE rating = their points get cannibalized by the rest of the pool, while they fall off the cliff.
I am releasing a full book with in-depth analysis in the next few days! Will probably write a preview blog post beforehand.
There are a couple of incorrect assumptions baked into Elo which should be reviewed.
1. The points exchange between two players with equal K factors is equal.
Why?
Don't we expect to learn from at least some games? Instead, at the moment, active players facea deflation tax whenever they play an lower rated opponent.
2. The Sonas adjustment fixed a problem.
The deflation problem is clearly continuing, and the mathematical logic that causes it has not been addressed. The problem should be expected to get worse long term.
3. Elo is assumed to be a measure of strength but it is actually a measure of results in rated events.
Worse, the number of FIDE-unrated games for many players is huge. I looked at the lichess accounts of two 20-something CMs J know and both seem to be playing 4k games per year.
If we assume that a "play-game-review-game cycle" will normally drive improvement, these games increase the likelihood that the gap between results and strength grows.
For problem 1, I like the idea of a points bonus at the end of the year based on numbers of games played and the size of any deflation effect in their cohort. Psychologically, it will give players a boost at the end of the year.