lichess.org

Inflation

This issue has been discussed several times, sure. It has been more or less a slow process but within the last couple of weeks some adjustments (?) have lead to a fast inflation in Blitz.

It noticed it and talked to several people, they all made a leap within some weaks.

My rating climbs up whereas my percentile decreases. Recent example: Blitz 2362 is 98.8%. Some months before much less rating equalled a higher percentile.

Watch out!

(probably feedback forum had been more appropriate)

If lichess was e.g. growing and therefore attracting an increased fraction of stronger players to play regularly on the site, then that would explain the changing percentiles - not saying inflation can't occur, but how can one measure it robustly?

It's an accepted fact that it doesn't make much sense to compare ratings from different rating pools - e.g. comparing chess . com, lichess and the official Elo system.

Arguably, the same is true for the same system at different time points. The pool of players at Lichess now is probably quite different compared to the Lichess pool a year ago. If your rating changes, it might be hard to figure out if your chess skills changed, or the characteristics of the pool of players changed.

The only straightforward interpretation of an individual rating number is to compare it to the ratings of other players in the same pool, *at the same time*.

If we think about it in this way, any talk about "inflation" becomes obsolete.

Agreed. But „hyperinflation“ within a short period? Not few made leaps of 100 points?

Is there an archive of the "weekly distribution" graphs?

lichess.org/stat/rating/distribution/blitz

If something shifted in the pool, it may show up by comparing those.

I haven't looked at it for a while, but it seems to me the Blitz distribution looks less skewed than I remember from a couple months back ( more "Gaussian" now, with more "beginner" and "average" players than "really really good" ones.) I may be totally wrong though, just an impression from memory.

there was tune up of RD parameter which caused pool medians to move towards 1500 which the intended median. Maybe prior the pool has moved away and now back? though difference shoudl not be 100 points.

But by and large no rating system is stable over time. Compositin of the pool changes so does the "meaning" if rank change. Elo type calculation will suffer alos from maximum expected resutl limitations, rating floors etc.

True, I am 1502 at a moment and it says that I am better than 47.3% blitz players. Not so long ago if you passed 1500 mark, you passed 50% as well.

Rapid does seem somewhat different. The rating of the 90th percentile has gone up a few dozen points over the last few months. I'd like to see the rating distribution over time to check my progress. Comparing with others doesn't work as they get better as well.

Don’t know where this leap comes from but at Blitz 2365 you take over a whole load of fellows: a couple of rating points but the percentile grows from 98.8 to 99.1% which is pretty much. Experienced today.

@Sarg0n well, internet chess if mostly for us weaklings.
You are just so strong stat statiscical measures start lose meaning.

Reconnecting