There’s a few points of error for your math, MicroRNA. I’ll try my best to explain, but I’m afraid I can’t give you a full step-by-step of the formula.
We started with a list of records generated, by region, for each class. These spanned a variety of criteria. I can’t speak to the specifics, as I didn’t generate the original list myself. Only select teams at Blizzard have direct access to certain data, and I’m afraid those are not my secrets to share.
From the provided data, we adjusted the estimated actual Greater Rift level performance by either subtracting or adding to it accordingly based off of either lowering their Paragon level to 5000 or raising it (in the case of Seasonal players). Finally, we averaged each final entry across all regions to get a global number.
With your math, (1) you’re taking an aggregated snapshot far later in the Era than when we took ours (our data was from roughly early December), and (2) you only have a small slice of overall player data. You’re also (3) excluding players who aren’t within your chosen range, as well as (4) from other regions, rather than accounting for them by adjusting their performance numbers to fit that range. This is important to account for many factors, including a wide variety of skill, build differentiation, and other balance considerations like augment levels.
Those levels of averaging are incredibly important, because otherwise your data is skewed to a particular type of player. Our goal is to make sure our balance changes appeal and benefit broadly, and not to any one particular camp.
Data can be very easily manipulated to prove one thing or another by changing only one or two factors. This is part of the reason that raw Leaderboard data is often very misleading and shouldn’t be taken in a vacuum. It’s easy to say one class is over-performing and supply information by excluding any number of details to prove it. Any data analyst will tell you that taking as much data as possible is the best way to get the closest absolute “truth,” and the Leaderboards really don’t provide enough on their own to give the full picture since they only ever represent the highest level of play. Imposing too many filters will always produce a biased result, and so must be used judiciously.
The table I provided in that blog isn’t the only one we looked at, or will look at. It’s meant as an illustrative example, not a representation of cold, hard truth.
