There has been a lot of discussion on the topic of Starcraft 2 rankings recently, and I thought we might take this time to sit down and have a chat about them. At their core, rankings seek to address the eternal questions of “Who am I?” and “Am I a man?” and more frequently, “How can I prove that I’m a better person than my friend Bob?”
Throughout the ages, man sought to prove his dominance over his fellow man. This was instinctual ever since we first left the trees for African plains. However, early man lacked the comforts of Battle.net 2.0. Instead, he was forced to resort to cave paintings, local competitions and iCCup. In these days, ranking was not something that happened between you and your 99 friends, but rather a letter grade which was bestowed upon you like a tardy student at the end of 2nd year English.
iCCup rankings were based upon the earlier Elo ranking system designed by Arpad Elo for comparing chess players. An accomplished physicist, Elo knew that performance could not be measured accurately, but could be inferred, much like a black hole can be inferred by its effect on surrounding stars. By treating individual matches as samples, a win, loss or draw would suggest the relative skill levels between the players. An accomplished chess player himself, Elo designed his system so that the average score would approximate 1500 with a difference of 200 points indicating a .75% chance of win or draw.
The application of Elo across various competitive sports is well documented, and the iCCup’s choice to use this as their rating metric can be seen as “standard play.” To further help users understand their rating, iCCup designed a set of rankings based on various ranges. For greater accessibility, these rankings were designated along the lines of grade school report cards from D- to A+, with each difference in grade (C to C+) indicating Elo’s 200 point difference in skill. It’s worth noting that given that the average level of play in iCCup is on par with a mid-level Diamond player in SC2, the vast majority of players found this system largely inaccessible.
It’s plain to the average person that Blizzard considered this system when designing their own games (the WoW Arena famously uses a hidden Elo ranking to match players). Initially, Blizzard made this system public but quickly learned the lesson of iCCup where new players found the system inaccessible. In an effort to fix this, Blizzard made the system less transparent. A player’s Elo rank was hidden, and instead, they were shown a point value. This point value had a number of friendly properties:
- Players started at 0 and could not drop below this number
- Players received points based on their hidden ranking
- Hidden rankings were used across different teams (preventing smurfing)
The combination was arguably a success over the previous system. In the creation of the Starcraft 2 system, Blizzard also added the following:
- Users were placed in separate leagues (this was only partly based on their Elo)
- Leagues contained a smaller amount of the player base as players went up them
- Match making was based on Elo with a predictive component to quickly adjust for differences in skill
- Players were assigned to a division in their league (giving them only a rank compared to 99 other players)
- Rank in a division was based on visible points
- Bonus points were added as visible points
For players coming over from iCCup, they quickly found themselves in the top league. This is likely an accurate measurement of the average, C rank, iCCup player as compared to the general populous. For new players, they found themselves in other leagues with similarly ranked players, with most players existing in the bottom 2 leagues. Movement between leagues is governed by the following rules:
- To move to a new league, you must have a 50% expected win chance against members of that league
- You must have at least 7 losses (this is to prevent statistical anomalies)
- Your hidden rank must be above, or below, a certain number (dependent on the target league)
Inside a given league, players were given points to differentiate rank 1 from rank 100. These points are given based on the difference in ranking between the players and additional bonus points. All players accrue bonus points based on the time that has passed since the beginning of the season. They are then awarded these points as bonuses to their win. Since all players receive bonus points at a rate of 1 every 2 hours, this rating inflation is evened out for players with an empty pool. Finally, this visible score is reset at the end of every season.
Despite the effectiveness of this system at encouraging new players, an Elo system has natural flaws. As new members enter into the system, there is a natural rating inflation. Additionally, players who do not play enough are no longer accurately ranked. Blizzard hides these flaws by hiding the system, but they still exist.
As I sit here in my luxurious arm chair, sipping my expensive brandy next to a roaring fire, I want you to know something. I am not better than you because I have an expensive after dinner jacket or because of my Scrooge McDuck style money bin or my harems of beautiful women. No, I’m better than you because I’m diamond rank and you’re not.