Chuck Kennedy wrote:...Eventually the ratings in those isolated areas will integrate with the big mixing pot but then only because more players from that area travel and have their ratings "adjusted" in that arena. When Europe and Japan were starting to ramp up their international play against U.S. players, there was a feeling some of them were overrated and took those ratings back to Europe and "artificially" boosted European and Japanese ratings. There may have been a little of that at the time. But the intermixing of players at big events like USDGC and Worlds over time seems to have adjusted International ratings among the top DG countries accordingly.
Definitely, these melting pot events are great for coupling, and should help a lot. But I still think there are some residual issues, owing to players with a higher rating doing most of the coupling between different regions...
The ratings system is essentially a linear map between scores in a round and the ratings of players. To describe a line, you need 2 parameters: a baseline, and a slope. The novelty of your approach is that you set these 2 parameters for every rated round to best match the propagators, and then apply the ratings to all the players who played that round. It's a great system, and works great especially in terms of being self-referential and self-correcting in a well-coupled population.
Now, when considering the coupling between different regions with players who travel further and thus provide more overlap, you have to consider how well they carry over each the 2 linear parameters, in terms of their coupling of ratings between the regions. If the distribution in ratings of the coupling set of players is small, then you are likely to get a very good coupling for the baseline (but only around the mean ratings level of the coupling players), but you're unlikely to obtain a good coupling of the slope. To obtain good coupling of the slope, you need a broader distribution of ratings among the coupling set of players.
For example, imagine if the only 6 players who played shared events between 2 distinct populations had ratings of 1000, 1001, 998, 999, 1000, and 997. This distribution of ratings does not provide enough spread-leverage to constrain a robust mapping of the slope. However, players in the 2 regions with ratings near 1000 would be very well-coupled, and you could probably compare them to one another with few worries. But these players do little to couple players at the 800, 900, or 1050 levels, which could be very far off when comparing the one region to another.