RD2L EST-SUN Playoffs: Ranking the Teams, Part I
Overcoming the Initial Rankings which have Less Diversity than USC Athletic Admissions
As RD2L playoffs start, there will be the inevitable comparisons, punditry, and discussions to determine who’s going to make a deep playoff run while who is set to crater in the first round. The problem you may quickly find is that all of the teams from a record perspective appear quite similar.
Needless to say, lots of teams are bunched together. However, not all these teams are created equal. Some teams got their wins by beating talented squads, and others scraped by with byes, beating up garbage teams and sneaking in at the last minute.
Ranking Methodology: Rank-Order Scoring
In order to rank teams, we will be using a rank-ordering based method. A team’s total score is the sum of the reversed rank-orders of all the teams they’ve beat multiplied by the number of times they’ve beaten them. For example, beating the top team twice would earn one a score of 70 (35 x 2) for the series. This metric has the benefit of being easy to compute, easy to see what’s happening and giving some differentiation to all the teams.
However, there is an issue. After scoring all the teams the order can change. In order to combat this, the team’s rank-order scores are changed after the algorithm is run, new rank-order scores are assigned to each team and the ranks are re-calculated. This is run until the algorithm converges (until no teams change places after re-running).
This allows us to see what happens each time we run the algorithm, how the teams change places, and better understand the implication of the algorithm run.
Final Team Ranks
In the interest of not burying the lead, I’ll put the final ranks at the start of the article and leave the runs and computations for those interested in how I arrived at this result. I’ve also showed the results of all the runs
- Torn (+1), 258 points
2. Insane Corndog (-1), 254 points
3. Econ, 235 points
4. Logical (+2), 231 points
5. Iceyeti (-1), 221 points
6. Chup (+2), 213 points
7. Znyper, 200 points
8. Squid (+1), 186 points
9. Tiramisu (+2), 152 points
10. Hullcity, 149 points
11. Pepega (+1), 144 points
12. Shmeezy (+3), 139 points
13. Mao (-8), 137 points
14. Droziki??? (+8) 124 points
15. Acidrain (-2), 120 points
16. Smacktrick, 119 points
17. Warcraft (+2), 116 points
18. Foosquash (-4), 106 points
19. United States Daniel (-1), 102 points
20. Unregister, 85 points
21. Trav2s, 86 points
22. Mu (-5), 60 points :-(
What follows is a summary of each run, how the ranks change, and what results seem to stand out. This can show how we reached these final rankings, why the Silicon gods are so confident in Droziki (and not so confident in Mao) and how an algorithm can quickly converge.
Run 1: First Run Rankings and Scores
The first run produced some strong separation across the various teams. Teams that earned their wins beating up squads at the bottom were heavily penalized, while teams putting up quality wins were rewarded. Torn and Logical were both rewarded for a helping of quality wins over strong teams, boosting them over Corndog.
On the other end of the spectrum, Droziki was heavily awarded for wins over Shmeezy and Unregister near the start of the season.
The sixth bar (in purple)shows a striking picture of how Logical’s team has a significantly higher score in the new rankings.
In the second run, Logical’s score falls back in line with some of his opponents and he drops 3. There is essentially a seesaw effect as the lower ranking of Corndog hurts logical, which in turn lets Corndog rise again. Generally speaking the scores themselves are becoming less volatile with fewer large swings, although the computers really seem to have a thing for Droziki.
The computers wisely decide fuck Smactrick.
Run 4: Final Run
It may seem strange for only one team to drop, but this is the result of Mao and Shmeezy being tied on the previous run. Re-running the algorithm resolves the tie. Re-running again doesn’t change the order of any teams.
You’ll notice very little changes between the graphs in runs 3 and 4.