Overall rank vs. speed ranks discrepancies

General Go Discussion
1

First of all, this is not a demand to change anything, I want to write down my thoughts about the rank discrepancies, because I have noticed people saying patently false things (e.g. “your overall rank simply converges faster to your real new strength”).

For many players, the overall rank is different from the three other ranks in a puzzling way, for example mine is 14k in all three speeds and 13k overall. To understand why this is so, I will look at an extreme example where we assume that the skill of the players does not improve during the example.

Player Black is a 10k correspondence player who is bad at live games because he really needs his cigarette pause between moves, therefore his live rank is 19k, but he plays them so rarely that his overall rank is 10k.

Player White is a 10k live player who practically never plays correspondence games (you may assume that she lacks the patience and she forgets her own strategy after a couple of days and her correspondence rank is 19k). Accordingly, her overall rank is 10k.

Now what happens if Black and White play live games against each other and White plays no other games? Their correspondence ranks cannot change. Their live ranks cannot change, either, if they reflect their actual live game strength. So what happens to their overall ranks?

Black sees his overall rank plummet, but he can compensate by playing lots of correspondence games between the live games to hold it at 10k. Now White’s overall rank eventually has to reflect the fact that she is 9 stones better than her opponent at 10k overall rank. This means that her overall rank will converge quickly to 1k, although this does not reflect her strength, it just reflects the fact that she is 9 stones better at live games than someone who is 10k at correspondence games.

Now, how does this contrived example occur on this site in practice?

  1. I rarely played blitz games, so my blitz rank lagged behind my other ranks. I recently decided to “play it up to my current strength”. In doing so, I pushed down my overall rank a stone while improving my blitz rank a stone, because the blitz rank had an ascending zig-zag line while the overall rank followed with an descending zig-zag line.

  2. There is a sizable community of correspondence players who do not have much experience at life games. Every time they play against a mainly live player they push his or her overall rank way beyond their actual capability.

Why would it not help to delete the overall rank from the profile?

Because deleting it just hides the fact that the other ranks are still an amalgam of different skills - different board sizes, different handicap sizes, different times of the day.

It is much more interesting to have this discrepancy out in the open and understand that the other ranks also have this kind of problem.

However, we can take heart in the fact that people do not usually have a 9 stone difference between their live ranks and correspondence ranks, so we can expect ranks to be inflated no more than one or two stones by this phenomenon.

6 Likes
2

I sort of understand your explanation…

But my brain can’t translate your explanation to my situation where I’m above 11k in the individual rankings but my overall rank is 12k albeit very close to 11k.

But my correspondence is nearly ‘half way’ to 10k and for some live games that I played I am ‘half way’ to 10k.

Is it because I’ve had a run of loses…?

3

Okay so basically your overall rank affects the rating points more than anything else. So a 10 over all playing a 10 is going to be an even rating gain or loss. What happens sometimes is you’ll be on the cusp of gaining or losing ranks in multiple game types and gain or lose enough rating points to raise or lower your overall above or below the other game ratings.

4

No, it’s calculated completely separately from the speed based ranks and is only affected by your opponents overall rating/ranking

2 Likes
5

However, we can take heart in the fact that people do not usually have a
9 stone difference between their live ranks and correspondence ranks,
so we can expect ranks to be inflated no more than one or two stones by
this phenomenon.

I have actually played people who’s overall ranks were 15+ stones above their correspondence ranks. They got to SDK by playing live games, but didn’t play correspondence games so their correspondence rank remained at 25K. The result? My correspondence rank dropping way more than it should when I lose to someone 9 stones stronger than me. This is why I think overall rank is a much better indicator.

Maybe it would be a good idea to make each specific rank provisional until a certain number of that speed of games are played, have the specific rank start out equal to the overall rank at the time of the first game of that speed, and reduce the K-factor of a provisional’s opponent?

1 Like
6

It seems to me the past decisions to:

  1. Grant ratings on 9x9 and 13x13 games
  2. Separate the ranks/ratings for Blitz, Timed and Correspondence
  3. Keep the old single rating separately from the three in #2
  4. Call the old single rating 'overall’
    have complicated the matter, and made understanding of issues/
    problems difficult.

It may be time for a coherent, high-level policy to be established/published, and all above four items to be re-evaluated based on this.

Everyone understands 9x9 Blitz and 19x19 Correspondence could have different characteristics. But the issues are multi-dimensional:

  1. Board size
  2. Game speed
  3. Tournament/Ladder/Other
  4. Multiple / Provisional accounts
  5. Sandbaggers/Airbaggers
  6. Human/Bots
    and the Elo system, originally developed for Chess, does not cope well with such disparities without careful adaptations.

note: Sandbaggers are those who claims/maintains lower rank/rating than his/her true strength for the joy of destroying opponents in games. Airbaggers are those who tries to obtain higher rank/rating than his/her true strength taking advantage of loopholes in rating systems. ‘Airbagger’ term is quite new, might be unique to OGS, and may be a creation of an OGS user, Rikhon.

2 Likes
7

Sidenote:[quote=“Tokumoto, post:6, topic:3419”]
Airbaggers
[/quote]

I thought the term was “Heliumbagger” <shrug>

1 Like
8

I thought “Hindenburger” (http://en.wikipedia.org/wiki/Hindenburg_disaster) might be a better naming, but for clarity, it can’t beat airbagger :stuck_out_tongue_closed_eyes:

2 Likes
9

you are not completely wrong. your explanation is good and well considered.
BUT: in real life experience i can say that there are enough players with 5 or even 10 ranks difference from live to correspondence or blitz… i understand that OGS is not willing to change all this ranking mess. at least that’s what i get out of all the discussions and chat rooms. But is it that difficult to have an option when limiting rank in a game search to chose whether i want a e.g. 15k overall player or a 15k insertgamemodeofyourchoice player to challenge me.

i saw a 15k blitz game. entered and saw in the ingame chat that he was a 5K… is this supposed to be good user experience?

2 Likes
10

Maybe overall rank really should be a weighted sum of the other ranks.

11

What if someone never plays one or two of the modes?

12

What if a majority never plays one or two of the modes? Couldn’t we then just get rid of this confusing experimental feature?

13

Then the weight for that mode is zero and it doesn’t affect the overall rank

14

I wonder what Wulfenia thinks of this idea. ??

15

What if someone starts playing all three modes, but then stops playing one of them?

16

I have to agree, that the difference between ranks is causing issues. I always feel like a sandbagger, when I issue a challenge for a mode that I don’t play too often and it would then show me as 19K. Then the game begins and suddenly… surprise, 14K! I had many people just quitting right away after I send a message in chat showing my ‘real’ rank.

It also leads to the weird situation, that if you lose to someone with an even higher rank discrepancy than yours that instead of closing in, the difference between the ranks is increasing.
It is difficult enough that if you try to explain someone your strength you have to add the site, since ‘14K on OGS’ and ‘14K on KGS’ can be quite a difference, on OGS you have to differentiate even further, ‘14K overall but xxK on live games’?

1 Like
17

That’s one issue. The second one comes when you want to restrict the rank range in a certain game offer. The restriction doesn’t count for the certain game type (blitz, live or corr) but for your opponents overall ranking.

2 Likes
18

Gradually the other two modes will take on more weight.

20

Typically when you inform a moderator of a >3 stone rating difference they will adjust it. I always thought this should be automatic/forced in some way, e.g.

  • no rating can be more than 3 stones lower than your overall rating
  • you have a button to set all your game mode ratings equal to your overall rating
3 Likes
21

Agreed. Accounting for rating would probably mitigate most of the problems. Whole History Rating seems to be a interesting approach that produces accurate results.

IMO, if the ratings remain separate, they should at least be crossweighed, so as to increase the other ratings when there is a large jump in one rating. Even if the performance levels at different time control naturally vary, they are still very correlated.

1 Like