International Elo Ratings 2015/08/02
International Elo Rating
With both LCS regions on break the biggest story this week came out of Korea. KT's amazing game 2 win over the KOO Tigers solidified their position as a top 3 LCK team. SBENU also got their first match win over Jin Air.
In China, Royal Never Give Up (AKA the team that kind of used to be Team King) are, true to their name, making a last minute move for a playoff spot with an unexpected win over snake and 1-1 matches against the very strong QG and IG. The actual Team King (sort of former world finalists Starhorn Royal) became the first team eliminated from playoff contention after their draw with OMG. The LPL as a whole remains close; the last 7 weeks have had at least 6 of the 12 matches end 1-1.
Brief explanation of methodology here. A full list of teams, including the IWC regions, can be found here. The next two pages present a detailed breakdown of the NA and EU LCS regular season performances and playoff possibilities. For those interested, an excel file containing the formulas used to calculate the odds can be found here (Let me know if there's an error).
Playoff Predictions - North America
The seeding and rating of the North American playoff teams are in the table below:
As you can see there is a clear top 2 (TIP and TL) and a clear underdog (Dignitas). Here is how that translates into probabilities of different outcomes:
Team Liquid is a strong favourite to make finals and the leading contender for 1st overall. This is not just because they're the highest rated team; their path to the final is easier than either CLG's or Impulse's.
These odds are based on specific ratings which are in turn based on historical wins and losses (with more recent ones more heavily weighted). While I appreciate this method it does leave a lot on the cutting room floor. One could put much more emphasis on recent performance (without over-weighting it in a distorting way) than these ratings do and end up with the following chart:
This tracks a measure derived from elo ratings called a performance rating. More accurately it uses a weighted performance rating (a slightly different version is described here). A performance rating is a measure of how well a team did in a specific set of games. A performance rating of 1600 means that the team won as many games, against their specific opponents, as a team rated 1600 would have. This chart is a rolling average using a 6 (or 7, in the case of week 9, which includes the tiebreakers) game window to show how teams have done recently. It also shows how their performances have changed over the course of the split. Clearly C9, TDK and Impulse are much improved CLG had a few bad weeks in the middle and Dignitas and TSM have collapsed. Interestingly, going by the last 3 weeks, the 4th best performing team was C9 and the 5th best was TDK. TSM was 9th.
This sort of analysis is useful for looking at playoffs. It's also, I think, a strong argument against auto-relegation (a good alternative, which preserves auto-promotion, is suggested here). TDK did well by this measure (which does not take into account how close some of their losses were, nor how bad some of TSM's wins were). Certainly they did much better than NME. But they won't even have a chance to play for their spot until after next spring.
If you think that the ratings are a poor reflection of the team's actual skills then you might be interested in this. By editing the ratings in this file you can how, for example, TSM will do if they get much better under Reginald.
Playoff Predictions - Europe
The rankings for Europe are much more stark than NA's:
Fnatic is the overwhelming favourite, Giants are the heavy underdog and the rest are close together. These rankings, as we will see, reflect something very true about EU right now, the oldest truism in the EULCS remains true: everyone beats everyone and in the end Fnatic wins. No team in the EU besides Fnatic has won more than 5 games, no team besides CW have won fewer than 3. Compared to NA in which 3 teams have won all but one game in the last three weeks this is very even.
As you might imagine Fnatic are the heavy favourites here. The Elo ratings thinks they'd win 85% of the time. If an upset is going to happen it's most likely to come from H2K or Origen. This is not because they're the best teams, they're only slightly ahead of UoL and Roccat, it's because they don't have to play Fnatic in Semis.
Here is the same rolling average of performance ratings (6 game window again):
What's interesting here is how little difference there is between the 2nd best team over the last 3 weeks (UoL) and the worst (the Giants!). Fnatic are obviously in a league of their own.
Here at least the auto-relegated team was one of the worst performing. But in EU we once again see the potential for late split improvement, in the form of SK, and collapse, in H2K.
The file I linked before also includes the EULCS. Roccat seems like the most interesting team to me. You can use it to so see how good the currently in form Roccat would have to be to have even a 10% chance of winning (answer: Roughly 1600, or good enough to win about 2/3's of their games against Origen, H2K or the Unicorns, but only a fifth against Fnatic).