Is Glicko biased?


#1

does the new scoring system weigh heavier on lost gams rather than won games, also does it slant more on games that time out vs. scoring?


#2

You’re referring to the rating system (Glicko2), not the scoring system (probably Tromp-Taylor).

Also, the answer is No.


#3

then how is it then when losing to a higher ranked opponent, you can lose 1 to 2 spots immediately, but winning against the same, rank does’t budge?


#4

It would be VERY interesting to see an example of losing a single game to a higher ranked opponent and have your rank drop by one or two kyu. Can you show us in your game lists where this happened?

Also - you should be careful in charging a system with some kind of slant or bias. There would only be bias if

  1. you lost one or two stones by losing a game, and then when you win against the same opponent your rank doesn’t change

AND

  1. you gain one or two stones by winning a game, and then when you lose against the same opponent, you lose those one or two stones again immediately.

THAT would be a bias.

I bet that what actually happens when rank thresholds are crossed is that you move across the barrier with the first game, and then if you get the opposite result in the next game, nothing happens - and that this goes both ways for transitions up a rank, AND for transitions down a rank. That would be a fair - not a biased - system.


#5

not charging a bias, asking clarification on how ranking is determined when you play against stronger vs. weaker opponents, also i misspoke when saying losing against a stronger opponent, I should say weaker


#6

Well, i’m not so much worried about “charging” as much as you seem to be assuming there is a bias because of some behavior you’ve seen.

My point in my response was to note that you haven’t communicated to us enough information to claim that there might be a bias or a difference in how rank is determined. You noted that you saw something happen - you lost stones when you lost a game, but didn’t gain them right back when you won a similar match. You claimed that this indicated a possible bias (it’s your subject line so I have nothing to go on in determining your assumptions). I noted that unless you saw the behavior you saw, AND noted that when the opposite was true (a win, with a gain in stones, and then a loss, followed immediately by a loss in stones) that there is no reason to question the system as biased.

So - your note to me just now doesn’t make any sense since your first question didn’t ask about the difference between playing against stronger vs weaker opponents.

Perhaps you should ask your question again with a different wording.


#7

It’s important to note that it is extremely hard to reverse engineer what is going on with Glicko by watching your own rank.

There are just so many factors at play - including the fact that the rank impact of your most recent bunch (15?) of games is recalculated each time, because your opponents rank will also have changed, and it takes that into account.


#8

Sorry for the short answer. Somewhere on the forum are lengthy explanations how the rating system works.

The rating system is not biased in one direction. If it would be the case, OGS would have a big problem with rank deflation.
Rating changes are calculated using

  • your rating, deviation and volatility at the begin of a rating period (volatility is not shown on OGS)
  • the current rating and deviations of all of your opponents in that rating period
  • the info if you won/lost the games.

A rating period consitsts of up to 15 games.

The rating change is antisymmetric in rating difference (losing against a 100 points weaker opponent = - winning against a 100 points stronger opponent)

Neither reason for win/lose nor the score have any impact on rating changes

There is more about the rating system


And the math of glicko
http://glicko.net/glicko/glicko2.pdf


#9

@DrQuantum OP chose a rather nondescript title so I changed it to a question that summarizes his post and uses the relevant term. I also re-categorized it but that’s probably less important.

@anoek Can we have a discourse macro like

!glicko

or something that automatically pastes flovo’s above post? That would make it so much more efficient to answer the same question that keeps popping up at least once a month. :blush:


#10

So then if i play a stronger opponent and lose, and then that same person plays again(not necessarily me) and rank goes up, I can lose even more ground because the scoring would be based on what his is now and not what is was at the time we played?


#11

I think it’s the other way round. You are expected to lose against stronger opponents so you any adjustment made from the initial loss would become even smaller if that opponent’s rank increases subsequently. Where you would lose more would be if that opponent went on to lose so as to become weaker than you. I think in that case when your rank is next calculated the system would consider your original loss against a stronger opponent as now being a loss against a weaker opponent and so your rank would be decreased. If that decrease is greater than other effects from the set of games used to calculate your rank, you could go down even after a win. I hope I have that right!


#12

-If you loose against a weaker opponent, or win against a stronger opponent, that will affect your rating more than loosing to a stronger opponent or beating a weaker opponent.
-If you win or loose against someone, and they then go on a winning streak, your rating will go up, since they were actually stronger than their rating indicated when you played them.
-If you win or loose against someone and they go on a losing streak, the opposite will happen, since they were actually weaker than their rating indicated.
-There is no bias towards lost vs. won games.
-Glicko doesn’t inherently bias towards time out vs. scoring, but OGS has a rule where if someone times out multiple correspondence games in a row, those games don’t affect rank past the first one. This rule has led to massive, unproductive arguments.