EPTI assume a player's SC2 rank includes their league, point total, and possibly their win-loss ratio. How would I normalize the SC2 rank so that I can rank players on a defined scale (say from 0-1000 points).
Background
This normalization function would be used by the Collegiate StarLeague to solve a problem we encountered last season: We grouped teams into divisions geographically which ended in both weaker and stronger divisions - with the stronger having 3-4 "strong" teams fighting for 2-3 playoff spots.
This season, we would like to augment division selection with a "power rank" for each team. Since players register with their b.net ID, we have SC2 rank info available for almost each player(*). Since a team needs a minimum of 5 players to compete, I wanted to compute this "power rank" as an average point total of a team's top 5 players.
League Distribution
There are some 1850+ players with:
1063 diamond players
363 platinum players
167 gold players
86 silver players
68 bronze players
145 unknown players (*)
Preliminary idea
I was going to use the distribution above to generate scales for each league. For example:
The final scale would be from 0-1000 points.
If there were 50% diamond players, their scores would be scaled from 500-100 points.
20% Platinum players would scale from 300-499 points.
15% gold players would scale from 150-299 points.
... etc ...
However, after reading Excalibur_Z's posts here and here this would not be a fair approach as high ratio platinum player could be > low ratio diamond players.
I would appreciate any insight and help, as statistical analysis is not one of my strengths
----
(*) - We pull rank info from SC2Ranks.com so we don't have every player's info as SC2Ranks relies on scrapping b.net profiles that it needs to first discover. Furthermore, players can enter false character names and codes, as we have no way of validating b.net profiles without asking for a full URL to a player's profile.
