Had I thrown the other 3 and 4 in (Indo and India) I'd have had them all. doh.

BTW -- my critcism was based on my misunderstanding of the term "occupied". My point about touching starting points was entirely obviated once I was told it meant picks.

Mexico = Central America (9th)

Any requests for what type of data/stats I should look in to next?

I'd love to see a ladder with autopicked spots based on a running analysis of the highest win % spots. Randomly select the starting spots in pairs where the win % is within 1.5 or maybe 2%. Everyone's ability to infer where their opponent was starting out would be exactly even instead of dependent upon the interplay between first pick luck and pick order...

How bad luck on the first couple turns translates into wins and losses is something that I would greatly like to see.

Thanks for the work you've done!

I would like to know who wins more often, the player who attacks on first contact or the player that defends

Average income after turn 1, 2, 3, 5, 10.

|>"I'd love to see a ladder with autopicked spots based on a running analysis of the highest win % spots. Randomly select the starting spots in pairs where the win % is within 1.5 or maybe 2%. Everyone's ability to infer where their opponent was starting out would be exactly even instead of dependent upon the interplay between first pick luck and pick order..."

Sounds like you are like me -- I am not a big fan of manual distribution. But what I would prefer to see, rather than what you suggest, is a more balanced map and random distribution (with wastelanded bonuses not included in the random distribution).

I know, that would be great in a perfect world, but I think that any map that isn't purely symmetric will end up being imbalanced in one way or another, so doing autopicks based off of %win per starting spot statistics is a way to provide a starting balance to an asymmetric map.

No map will be perfectly balanced, but I don't think your method assures a good balance, either. The win % numbers are averaged across all games, meaning your opponent could be anywhere. There could easily be two bonuses with even win % numbers across all games, but one happens to be stronger when matched up against a particular bonus with similar stats.

Idk.. win% can also be relative to the other locations.. IE W.Rus is the highest %, but would be coutnered by the opponent having a good start spot in Caucasus..

means say you get 1 high 1 med 1 low win percent spots.. you just had your high win spot ruined by his low win spot, and then he knows you aren't in locations with a higher win ratio such as indo-ant-scand-E.china-India while all he could remove your possible locations from would be low percentage spots, such as W-E US, Canda, E.Russia-Europe-S.Africa etc.

you could provide that a persons 3 spots would add up to a certain win%.. and both players have roughly the same.. in which case the same scenario woudl case similar problems.. Caucasus is 39% win rat, while W.Rus is 57%, that means that you just canceled out one of his bonus, with conceivably 18% win rate difference in your favor, just by luck of the draw

Right, that's pretty much exactly what I was trying to say, but you explained it in a lot more detail. Thanks for that.

I was typing it before your's showed up, lol

Stop reading after the comment where you posted the results, so maybe someone else pointed out the glaring error, but one thing sticks out that was not mentioned up until the point. That is that different skilled players have tendencies to pick different spots. For example, lots of newbs love picking either india east china west china as their top 3 or scadninavia west russia and greenland while vets do otherwise. this, among several other discrepancies (such as different bonuses being more or less likely to have a wasteland, and also affects adjacent bonuses viability) makes the data too unreliable in my opinion.

The data is 100% reliable if the XML has been handled right, it's the conclusions we draw from it that are hit and miss. IMO though, any data is better than no data, and as the sample size increases, it will become more useful.

Well if you read the thread, you will see that the wasteland issue has been discussed (and does not really affect things). As for the first issue you mention, my gut feeling is that it will not have as much of an effect as you seem to think. I suppose it would be easy enough for me to run the numbers for only high rated players/low rated players/etc. I may look at that in the future, but it's not at the top of my priorities list right now.

Basically, correcting for the player is possible by using a generalized linear mixed model. However, this would be a huge model and the errors would be quite large.

With proper software, it's worth a shot, but as the data are rather limited and the differences between the different territories are often not significant, these results would be even less interpretable than crafty's original analysis.

I vote that Crafty and MW come up with a suitable rating system for Wl and help Fizz implement it.

Here here.