Guild Wars 2

I’m not sure how exactly the season 1 groupings were created, and I was wondering if my suggestions below had been considered on how to statistically approach the question.

I had initially thought that a time series forecast approach based on server scores might be good when doing inter-server comparisons, but I don’t think it would let you bring in covariates (e.g. counts of people playing WvW, maybe averaged somehow e.g. rolling 12-hour) in sufficient detail to make this approach work well.

How about a linear mixed effect model, where you can assign both fixed and random effects, which uses a repeated measures design to approach the problem? This is a type of regression, where you could have (for example) overall season 1 score as the dependent variable and as many covariates as is sensible. Because it’s a repeated measures design, you could include covariates like points per period (e.g. 24 hours), interquartile range of player counts for the period, server population without having the issue of dependence between observations cropping up. You could use the studentised residuals to identify:
- the servers that achieved below expectation
- the servers that achieved at expectation
- the servers that over expectation
for points and use that information in helping to form the groupings for season 2. The other advantage of this type of model is that it is easily explainable with respect to the effects of the covariates/factors.