lichess.org
Donate

My opinion of the new rating system.

Also, if I had to guess, I'd say the main reason the ratings dropped is that all the points from players who at any point cheated got dropped from the pool.

Many of the players caught cheating didn't cheat all games, or even most games, so there were a lot of rating points floating around that had been gained from beating people who decided at some point or another to cheat.

So, if I had at some point gained 40 points by having a good run against someone who got frustrated and ended up cheating in a couple games, those 40 points are gone. That ends up affecting a lot of people, because then all my subsequent opponents are playing me with 40 fewer rating points, so they lose more by losing to me and gain less from winning.

I don't have any hard evidence of this, but that seems the best explanation so far :)
I would expect the opposite: cheaters steal points from other players by cheating. Removing them will recover the lost points, so the average rating should rise.
@ChessWhiz: You missed the crux of the whole thing. As I said in that post:

"Many of the players caught cheating didn't cheat all games, or even most games, so there were a lot of rating points floating around that had been gained from beating people who decided at some point or another to cheat."

The point is that many of players who got caught cheating lost a lot of games before they decided to cheat. Very few of them just got an account and cheated every game.
Arg, that's what I get for pre-solving the captcha...I accidentally post before I'm finished :)

Anyway, you are correct to note that removing cheaters doesn't necessarily mean that the average rating will go down (or up, for that matter).

It would all depend on whether the cheaters as an aggregate lost more rating points or gained more. If they in total gained rating points, then you would expect average ratings to rise. If they in total lost rating points, you would expect average ratings to fall.

Even that's not quite right, because the biggest effects on other people's ratings would happen when a person who ended up cheating at some point in their games played a very provisional player.

If players in their first few games played, and beat, someone who ended up cheating, that could easily cause a swing in their rating of a couple hundred points.

All that to say, whether removing games played by people eventually caught cheating lowers or raises people's ratings on average depends on several factors, and would be very hard to predict.

In this case, it seems to me the most plausible explanation for the drop. Of course, I'm one of those silly humans, so I could be really, really wrong :)
Ah, now I understand, thanks.

One thought: a provisional player who loses a lot of rating points in a single game can also regain them in a single game, so the "weight" of that rating is less influential.
Makes sense to me.

About the standard rating matter; as you said,

"The Standard rating is just a separate rating that changes whenever you play a game that isn't 960 or KotH, and the rating change is calculated by comparing your standard rating to your opponent's standard rating."

I'm not quite sure that's right.

My last opponent standard rating was 1900+ and I gained only 1 point on mine by beating him in a blitz game. His blitz rating was surprisingly quite a lot lower than mine on the other hand (there was a bug after the update; you could be paired up with an opponent whose rating was outside your seeking limits, if his standard rating matched your time control seeking... not sure if that's clear).

So I'd say standard rating is a balance between all your ratings except variants.
All games involving cheaters were counted as if they never counted.

The new ratings are a weighted average. So if you've played 4,000 bullet games at 2000 level and 4,000 blitz games at 1000 level, you average out at 1500. If you'd played 2,000 blitz games at 1000 level, you'd be more like 1750. So, if you're a top end player, specialising in bullet, and want a high bullet score, you're better off:

i) only playing bullet and ignoring blitz/classic
ii) ignoring the pools as they also drag your average rating down.

This is why I think this form of rating is flawed, because it seemingly penalises those who diversify, but ignores those who only play one type.

A better solution would, perhaps, to be done with the idea of a standard rating entirely, and have the three ratings completely stand alone without any average or aggregate. When you advertise a game, the rating of that time limit should be displayed. In this way, you have three ratings, independent, rather than one average/aggregate, as it provides misleading ratings either way.
@Chesswhiz: No problem, I have a problem typing a lot of material, and I know I don't always say things in the clearest way :)

Well, the trick is that your rating change in the provisional stage (a bit of a misnomer, actually, since Glicko doesn't have a separate "provisional" formula. "Provisional" players just have a higher RD) slows down quite a bit as you play games, so the first games are more important.

So, if you're a 2000 player playing for the first time here, and you win your first 3 games against a 1900, and then say, have a bad run and perform at 1900 over the next 10 games, your rating will drop quite a bit if it turns out that since you played, that 1900 decided to cheat, and we recalculate your rating without those 3 games.

Hopefully what I'm saying makes sense and isn't utter rubbish :)
It does make sense.

And cynosure's solution does too.

A standard rating as a weighted average on 3 time controls doesn't say much to me.

That doesn't matter that much though :)
Yeah, I didn't think the weighted average was actually how Standard worked, but it looks like I was wrong about that.

One of these days I'll actually have to get conversant enough in Scala to actually read the code instead of just guessing based on context :)

This topic has been archived and can no longer be replied to.