As a complement to the previous post, I have some creep killing stats. GPM (and Net Worth, but Net Worth is a pain for me to calculate given the API information so I ignore it) is the better metric for evaluating a team performance, but isn’t so great for providing a benchmark for individual farming. Creep Kills per Minute is also a better way of evaluating how teams distribute their limited farming opportunities between different heroes.
For all players in the sample, the CK/min stats are as follows:
When we restrict it to just the top farmer on each team:
So we see a 25-30% increase between the Normal and Very high brackets.
It’s difficult to test for and I don’t have any definitive results yet, but from what I have seen my suspicions are that the typical farm rate of a successful carry tends to increase as the game goes on. This effect is most pronounced in the Normal bracket, but still exists in a lesser degree in the higher brackets. Given this, I would say that 4 ck/min is a solid benchmark to shoot for by the 15 minute mark in a relative free farm situation for someone who wants to be able to play a hard carry at an acceptable level in the upper brackets.
Changing gears, I also have some charts that display the distribution of the ck/min stat for every hero in every bracket in my sample. In lieu of trying to get WordPress to play nice with them, I’ll just provide the external links.
Very High Skill Bracket: http://jsfiddle.net/8vFdN/89/embedded/result/
High Skill Bracket: http://jsfiddle.net/FcRgh/8/embedded/result/
Normal Skill Bracket: http://jsfiddle.net/9HDRz/13/embedded/result/
All Three Skill Brackets, kinda unwieldy, contains all 270 entries grouped by hero: http://jsfiddle.net/QtTrP/16/embedded/result/
Just select the heroes you want to compare in the legend. Supports will tend to have very high peaks early in the graph. Carries will have lower peaks as they tend to be spread out over a much wider range of farm performances.
For example, Crystal Maiden, Ancient Apparition, and Wisp (among others) all have very similar extreme early peaks, marking them as heroes that tend to be played as hard supports or a 5 slot. Tidehunter, Omniknight, and Sand King have support oriented distributions, but they’re slightly more shifted towards farm, indicating a greater propensity to play a 4 or even sometimes a 3 in team compositions.