Specializations

This part concerns itself with specialization power. Rather than using the model created by Subcreation, we employ our own models in order to estimate the specialization power. Two models exist:

• The first model is based on raider.io leaderboards. We sample in the leaderboards and use that data with a Bayesian model based on a Gamma distribution.
• The second model is based on the same dungeon run data as Subcreation. The model used here is a Bayesian model which controls for dungeon difficulty and uses a nesting strategy. We first model players. Then we nest the players within specializations. This allows us to estimate player power by looking at the player population who plays the given specialization.

Of the two models, the leaderboards model have the best fit. On the other hand, estimation of power via the dungeons has more data pulled from the top players, and could be a better description of what is happening at the tip. However, the fit isn’t as good, and the data is more susceptible to noise.