Before you point me to the search bar, I have searched and found only the threads where people are ranked in multiple game modes. However my case is the following:
Since live+19x19 is the only rank I have, why is my overall rank just a couple of points higher than that?
Because it updates faster.
Faster is perhaps not the right word. See above. Search function is useful sometimes.
TLDR (not long actually): It takes more data into consideration, i.e. reflects a different rating distribution. The numbers don’t mean the same thing.
Let’s try to explain it in detail by ignoring how rating points are calculated.
Imagine a Go player population of N=3. Let’s say you’re one of them and you’re the second best; your skill level is (N-1)/N=0.67. Now 7 new players start to play Go. Suppose their skill levels are 4…10. You do not “actually” improve in terms of reading or strategizing etc. But you’re still second best, so your skill level is now (10-1)/10=0.9.
You get bored of playing newbies and you start playing megablitz, which only the top 5 Go players in the pool play. You’re still #2 so your megablitz skill level stabilizes at (5-1)/5=0.6.
You think you deserve better, so you start playing correspondence. We all know corr players are a bit slow, which the #1 player doesn’t like - only the bottom 9 players participate. You are now #1 in that category so your skill level is displayed as (9-0)/9=1.0.
As you can see: all numbers are different because they refer to different (sub-)populations of players.
I gave the thread above a read and the summary of the comment that everyone considers to be the explanation is that each rating is compared against only the rating of that type when calculating the next score, e.g. overall is compared to overall, live-19x19 is compared to live-19x19, so if you ever play a player who has these two values different, which can be achieved by participating in the other game modes, then yours gets skewed as well.
If that summary is correct, is there any confirmation from the developers that this is the case? And if that is actually the case, then I don’t really see a point in having the overall rating in the first place, since the integrity between it and the other ratings isn’t maintained. That or it just shouldn’t be called “overall”, because it seems that there’s no difference in how “overall” is calculated comparing to say “live-19x19”, so it’s just a yet-another rating, not a summary of the ratings as the label suggests.
The overall rating (still) exists because it’s a good predictor of outcomes, ever since the most recent change it’s used in updating the other ratings, actually.
Apparently, if somewhat unsurprisingly, moar data = better.
It doesn’t really defend it. One way to check whether something like this is reasonable is to imagine that you don’t have it and then imagine that someone proposes a system like this. I can’t really see it getting implemented in my imaginary scenario. So defending it from the position that it’s already there and it’s already convoluted enough to the point of bearing results isn’t the way to go in my opinion. Additionally the points that are used in such a defence can also be used to further worsen the case by being the motivators to produce an even more convoluted solution.
I see. So you’d prefer if there was only the overall rating because it’s the best-fit and most parsimonious predictor. Yes, that’s the alternative.
Once you do that you have two options: 1) make 13x13 and 9x9 unranked or 2) keep explaining why this is the best solution, over and over again, to everyone who posts a complaint like “Why am I getting matched with this one guy who only played 9x9 until now? Our playing strength could not be further apart!”.
Not sure why you label something that you made up as being proposed by me. I imagine a system that has the current ranks minus the overall rank, and all of the matching is done according to those ranks, and then somebody just kicks the door out and says “how about we add an overall rating that isn’t a summary of all the ratings, but instead a rating of its own, and then also apply it as a margin when calculating all the other ratings?”. In that scenario I only get one question, and it isn’t “ok, when are we implementing this?”.
You wanted something that isn’t convoluted. A single rating is the least convoluted solution. However, thanks to your clarification, I now see that degree of convolution is not actually your criterion.
I don’t care either way, I just felt like being so nice as to take the time to explain to you the status quo.
I see that you don’t like the status quo. I’m fairly sure if you can come up with a demonstrably better predictor of outcomes (that’s the point of ratings), anoek will consider your suggestion.
It’s about priorities too…
Overall is the only rank that is used. The other are just informative.
All ratings are compared against the opponents overall rating,
I don’t know, why the numbers are slightly different but 4 rating points are nothing to bother to much about. My guess is some annulled games caused the numbers to diverge slightly.
I will make sure to create the appropriate thread should it become my objective.
I did have games annulled as part of the process of addressing cheaters. So if it wasn’t for the unusual circumstances would it be expected that these two ratings would always be the same?
From this example, it would seem possible that annulling games just adjusted the overall and not the other ratings?
It would make sense from the point of matchmaking, and from the point updating other users in the future if that’s all that would be used to update other users, but it would be confusing, as is the case in this instance.
These 4-6 points difference are common in everyones rating from what I’ve seen so far. The overall rating is always slightly higher then the other ratings. I would guess even with a new account without cancelled matches this would be the case. But since it will stay constantly between 3-6 higher, I’d assume that the formula that leads to the displayed numbers for specific game types is simply slightly different than the one for the overall rating, but both get their data from the same pool or something.
Because it updates faster…