Comparison of go rank with chess rating

[uPWarrior]

When I read the first post I automatically assumed percentile. Top 50% of the players (casual, tournament, however you want to define it) or any other percentile you wish to consider. The fact that the rating distribution of the go players is significantly different to that of the chess players will make the analysis hard, but not impossible by just looking at raw data.

[msgreg]

I was interested in a generic comparison. A general rule of thumb type of thing.

This doesn't work. Until you specify what you're comparing, you can't even get a rule of thumb, because the right rule of thumb will vary depending on whether you want rarity, chance of beating the best player in the world, number of skill differences from a beginner, etc.

Clearly you are correct in that there are differences depending on which factors are considered. But how do we know there isn't a "rule of thumb" if we don't know how inaccurate such a rule of thumb would be? It seems like the guesses in this thread could be getting close to a rule of thumb.

Interesting mathematical discussion based on probability of winning.
http://en.wikipedia.org/wiki/Go_ranks_and_ratings

Includes USCF vs go histograms based on "percentage of players" distribution
http://senseis.xmp.net/?RatingHistogramComparisons

The "old" go rank comparison page
http://senseis.xmp.net/?RankWorldwideComparison

The "new" go rank comparison page, uses ranges
http://senseis.xmp.net/?Sh%2Frankcomparison

(You could calculate the ELO difference in chess ratings between one go rank by looking at the percentage chance of a win in an even game on average. I wouldn't be surprised if it were close to 100 ELO points for one go rank to get the same win percentage in chess. I believe the "spread" for go ranks is bigger than chess, so 85 approximates this spread for chess.)

Do we know it's a linear relationship (i.e. a constant conversion of 85 or 100 ELO points per go rank)? It looks like on the RatingHistogramComparison's page (in particular, the blue graph) that it is non linear when referring to percentage of players at a level. The USCF line is very linear while the AGA line (for example) is not. If they were both linear but not overlapping, it would indicate a different scaling between the left axis and right axis. But I think that since one is linear and the other is not indicates that a static scale (X points = Y ranks) is not technically accurate. I think that I read on the graph that such a static scale would be in error about 4 ranks in AGA (just a guess based on loosely eyeballing the graph).

Totally eyeballing the blue graph, it looks like about
2800-400 USCF (2400 Elo points)
23k - 10d AGA (32 ranks)

Declaring each of the above scales to be equivalent, it comes out to 75 Elo points per AGA rank. And if you used this rule of thumb, you might miss the equivalent rank by about 4 stones/300 Elo points depending on where along the scale you were observing, particularly in the 7d or 22k area.

The wikipedia link indicates that the EGF ratings are based on the Elo rating system, and both are based on chance of winning a specific game against an opponent.

Wikipedia
The EGF rating system attempts to establish rough correspondence between ratings and kyu/dan ranks. This is done by taking the Elo rating system and varying some of the components of its formula to achieve a close match to the table to the right. The probability (SE) that the player with the lower rating, player A, wins against a higher rated player B is given by the formula

Another interesting graph is the "Estimated probability of winning against players at specified rating distance".

Wikipedia
Note how different the expectations of each system are regarding even games between players of unequal strength. If a player can win 90% of even games against a 2 kyu player, the AGA believes he or she is 1.33 ranks higher, the EGF believes (s)he is 2.42 ranks higher, and the IGS believes (s)he is 2.80 ranks higher. The lack of agreement stems from a tradition of playing handicap games between players of different ranks, so there is a lack of data regarding non-handicap games between mismatched opponents.

[hyperpape]

I think we're only converging because people are assuming that the highest go ratings are equivalent to the highest chess ratings. For some conversions, that will make sense, for others it will not (a pure ELO scale with non-negative scores will have much higher top go ratings than others).

You could have worse assumptions for Chess vs. Go, but it would become a ridiculous assumption for a number of other games.

[badukJr]

I am not sure why there is this obsession with comparing chess with go. They both are played on a board with pieces and require thinking, but the structure of each game is totally different.

This is like trying to figure out how many points a pro futsal player would score playing basketball if his ability could somehow transfer to a pro basketball league just because both of these games uses a ball in a court and require athletic skill.

[speedchase]

I am not sure why there is this obsession with comparing chess with go. They both are played on a board with pieces and require thinking, but the structure of each game is totally different.

chance to win converts.
Either way, I don't think they compare linearly, because go has a steeper learning curve (see ogs rating system).

[cdybeijing]

I am not sure why there is this obsession with comparing chess with go. They both are played on a board with pieces and require thinking, but the structure of each game is totally different.

This is like trying to figure out how many points a pro futsal player would score playing basketball if his ability could somehow transfer to a pro basketball league just because both of these games uses a ball in a court and require athletic skill.

I think it's simple: people who play or have played both want to know which game they are better at, relatively speaking.

[icehorse]

I think this is a fun AND potentially useful discussion.

I mostly play on IGS and I do lots of experimenting, but if I had to pick my "true" IGS rank I suspect it would be 2d+. (I've achieved an official 3d rank several times.)

I used to play chess, and when I retired from tourneys I was a B player.

When I talk to people about Go and tell them that I'd probably be an AGA 4d, I tell them that my "guess" is that that's about equivalent to a weak chess master, but that's speculation on my part. But it's a useful comparison for establishing context in conversations.

[tapir]

I am not sure why there is this obsession with comparing chess with go. They both are played on a board with pieces and require thinking, but the structure of each game is totally different.

chance to win converts.
Either way, I don't think they compare linearly, because go has a steeper learning curve (see ogs rating system).

Chess players are the wrong people to convert. Try backgammon players... black and white stones of equal value and already playing a fun game.

P.S. OGS rating is seriously *ucked up. (Longer thinking time, rarely full handicap and rated 9x9 games mainly.) Many established European dan players play deep in the kyu ranks there.

[skydyr]

I am not sure why there is this obsession with comparing chess with go. They both are played on a board with pieces and require thinking, but the structure of each game is totally different.

chance to win converts.
Either way, I don't think they compare linearly, because go has a steeper learning curve (see ogs rating system).

Chess players are the wrong people to convert. Try backgammon players... black and white stones of equal value and already playing a fun game.

P.S. OGS rating is seriously *ucked up. (Longer thinking time, rarely full handicap and rated 9x9 games mainly.) Many established European dan players play deep in the kyu ranks there.

As far as I can tell, all of the turn-based servers place a user several ranks weaker (sometimes 5 or more) for a given strength compared to real-time servers and go organizations.

[xed_over]

As far as I can tell, all of the turn-based servers place a user several ranks weaker (sometimes 5 or more) for a given strength compared to real-time servers and go organizations.

but that's not a fault of the server, as much as it is of the time limits.

with weeks and months (sometimes even years) to finish a game, the server can't really keep up with your real life progression.

[skydyr]

I don't mean to suggest that it's a problem... the server's ranks are internally consistent. It can just be surprising to someone new to them expecting that they are about X kyu or dan.

[Krama]

A rough linear estimate of http://www.goratings.org/ and chess elo rating (FIDE) is like this.

We start at 2000 points and every 200 points in goratings is approximate to 100 points of chess elo rating.

Top GMs are around 2800 points (+800) and on goratings top pros are around 3600 (+1600)

Also a quote from wiki.

Players with an Elo rating greater than 2700, such as Viswanathan Anand, Garry Kasparov, Magnus Carlsen, and Vladimir Kramnik are sometimes informally referred to as "Super-GMs". There were 47 players in the May 2014 FIDE ratings list with a rating of at least 2700

Now if +700 chess rating is equal to +1400 goratings.org we get to around 3400 which comes to ~30 top players which is not a bad estimate.

It is impossible to tell for real which elo corresponds to a go rating but this rough estimate is just fine by me.

[Uberdude]

P.S. OGS rating is seriously *ucked up. (Longer thinking time, rarely full handicap and rated 9x9 games mainly.) Many established European dan players play deep in the kyu ranks there.

As far as I can tell, all of the turn-based servers place a user several ranks weaker (sometimes 5 or more) for a given strength compared to real-time servers and go organizations.

Hi there! I was 8 dan on OGS when I was 3 and then 4 dan in EGF/BGA (and similar on KGS). But yes I was unusual in this regard and many European dan players had lower ranks there.

Regarding the comparison of Go and Chess ratings, there was a nice chart on Sensei's library comparing the ratings by percentage of the playing population at each level but I can't seem to find it now. As I recall Chess ratings (maybe USCF, are they are bit stronger than FIDE Elo?) were a bit stronger than the same number in EGF GoR. So I'm an EGF 2368 (4 dan) currently (lost quite a bit lately!) which would correspond with about 2250 in (some) Chess rating.

[dfan]

i there! I was 8 dan on OGS when I was 3

Now that's precocious!

I think the chart you mean is http://senseis.xmp.net/?RatingHistogramComparisons, although the graph created from it seems to have gone missing.

USCF ratings are weaker than FIDE ratings, in that a 2000 USCF player is weaker than a 2000 FIDE player.

My 2000 USCF rating "feels like" 1 dan to me, and my 4k AGA rating "feels like" 1600, but that's not based on anything quantitative.

[uPWarrior]

I decided to gather some data so that we can answer this question in the way that, for me, makes the most sense.

I started by downloading the FIDE list of all players standard ratings (243252 players, found at http://ratings.fide.com/download.phtml) and the EGF list of all players who have participated in any tournament (38928 players, found at http://www.europeangodatabase.eu/EGD/EG ... system.php).

The question now is: someone who is better than 50% of the chess players, what EGF rank should they reach in order to be better than 50% of the go players as well?

The next figure shows the FIDE vs EGF same-percentile isoline. It shows the rating that players need to have in both systems in order to achieve the same relative position in the respective populations.

go_vs_chess_ratings.png (6.76 KiB) Viewed 17570 times

According to this comparison, an EGF 10k (rating 1100) is equivalent to a FIDE 1800 player, a EGF 1d (rating 2100) is equivalent to a FIDE 2170 (expert), a FIDE master (rating 2300) corresponds to a EGF 4d and a FIDE grandmaster (2500) corresponds to an EGF rating of 2800 (EGF 7d is at 2700).

The EGF 10k = FIDE 1800 might be the most surprising comparison, the rest seems acceptable to me. We might be skewed to believe that 10k is easy to reach, or maybe this says something about the data sources.

(I had to remove everyone with a EGF rating of 100, as 5500 players clustered at this exact rating)