Is it time we revisited topic once again? Can anyone make those cool comparison tables?
The data is in OGS now for people who have filled in their rank from other servers. Can anyone access this to make a comparison?
Yes, anyone can access it. There’re settings to hide things but I’m sure couple of thousands people don’t hide it.
So how does one access this data?
It’s pretty easy to download it through api. The hard part is the analysis. I saw what previous surveers did, what confidence intervals and so on, it’s way too complicated.
Well then, maybe better to keep it simple. After all, ranks are only an approximate measure of strength anyway. I’d just assume a linear relationship between ranks on two servers, plot one against the other, apply a simple straight line fit and that will do. I’d say it’s more about calibrating one server’s ranks against another’s, e.g. OGS 7 kyu ~ DGS 6 kyu (true of me right now). Just something simple like this: Rank - worldwide comparison at Sensei's Library No need for confidence intervals - that’s just overcomplicating it.
So, how does one access that data in the API?
I presume one has to do something a tiny bit complicated.
For example while you might get people that play very regularly on all servers they listed, you might also get people who haven’t played on some of the servers they’ve listed in quite a while, or who play much less frequently on one than the others. It could be the case that some ranks lag behind others.
Either one tries to restrict to which users are included for the mappings for each server (remove the outliers), which seems impossible to figure out without more information or one has to do something to adjust mapping, or just leave it as it is and if it works it works and if it doesn’t it doesn’t
I guess scatterplotting OGS vs KGS ranks (OGS vs IGS…) etc and doing a straight line fit would work ok provided most people sampled play ranked on both servers frequently. Otherwise I just expect some nonsense like everyone is the same (eg mean) rank (imagine it being completely uncorrelated) Then one has to know how to remove outliers again to get a better plot.
Maybe just getting out an R^2 value (Coefficient of determination - Wikipedia) for each rank comparison would be useful though as an indicator of how reasonable the fit is, but one could probably eyeball it from the line through the scatterplot.