Glicko-2 anomalies in historical data?

Based on real game results, both OTB, on OGS, and on DGS, I know I have improved drastically from April 2016 when I started and could barely play (a 20k player was a difficult match for me), until now, when I can hold my own against a 10-12k OGS player.

Under the old rating system, this improvement was reflected in the graphs, both on OGS and DGS. With the new Glickko-2 system, my rating is essentially flat all way back over the last 18 months.

Is this expected?


I’ve been thinking about cases such as this. I think it’s to do with the fact that all our past games were reassessed with the new system. Overall this should be accurate for most players but there will be outliers. Say from the day you started you only ever played against players with rank 16k. Some you beat and some you lost to but more importantly some games were easier than others. Under the new system, you played the same players at the same time but they may have had a rank of 20k or 10k instead of 16k. Does that make sense?

1 Like

Not exactly.

I almost always play correspondence. Most notably, my 19x19 correspondence rating has declined 300-500 points in 18 months, while I’ve been beating stronger players and almost never losing to weaker players.

Every other rating system, shows that I improved at least 7 stones during that same time (and OTB play confirms the same). Down 3-5 versus up 7 (a difference of 10-12 stones!) seems like a huge discrepancy that is hard to explain away.

1 Like

My past data is really weird - it shows I hit something like 3k around a year or so ago (which I would say is a bit of a ridiculous idea - I wouldn’t say I have ever been stronger than 5-6k), but am now rated around 9k (which is probably too low). Other people who I play against have similar graphs. I think the error bars around the graphs are way too narrow. I may have been playing a bit more lax in the last few months (but I don’t really think so). And if anything I have probably improved a bit since 2016 - certainly not dropped 6 stones. Still, the graph does look cool :slight_smile:


Perhaps I wasn’t clear. The point I was trying to make was that while most people have more accurate rankings, statistically a few people were paired with weird combinations that have them a false high or a false low.
I think playing new games under the new system will quickly return you to where you are supposed to be (though this is all speculation)


Yep. I’m not too bothered about it. I just hope it has the capability of allowing for the possibility that the whole player pool is increasing in strength over time rather than assuming that that is static, otherwise I think we may have a slow effect of rank deflation due to people registering weaker than they are… I think it should be much better for relative strengths though.

Historical data are more than bit wierd, lowest point on my glicko2 graph is 12.6k, which is quite funny concidering I came to OGS just after I learned that goal is to surround territory and how to capture stones :smiley: determined to solve 9x9 board :smiley: :smiley: :smiley:

Yeah - have a look at this one too

The glicko graph has hardly any points on it. Compare to the old graph, which has many many.

It looks like this player played a lot of games on single days. New graph appears to group by days rather than by games. Old graph adds a point for every game. The old graphs were weird because the x axis was not aligned by time. It was aligned by number of games.

I think the new graph, aligned by time, is a better representation of player strength.


Yeah while glicko has been tweaked to reflect your per game performance it actually processes batches, every 15 games and monthly.


I prefer the opposite (isn’t that always just the way with software!?).

Time is irrelevant, number of games is much more relevant: am I improving each time I play or not? :slight_smile:



1 Like