http://www.espn.com/mens-college-ba...nalytics-experts-consider-creating-new-metric
Interesting. I don't think too many people gripe over the current process (except the teams that just didn't make it in). What do you think of this Professor?
As a statistician who's also taught sports courses, I agree the weak link is the selection process. Even if a good team is selected but unfairly seeded in the teens, they are likely one and done (or if they win, the seeders have unfairly penalized a top team and wiped out their great 30-game season!). So here's my take on the issues raised and my solutions:
Margin of victory:
Solution:
should be considered but only use the logarithm (they'll likely not do because they are too dumb). Why? The RPI and Sagarin, by ignoring MOV, treat blowouts the same as squeakers and "moral" victories the same as big losses. On the other hand, MOV creates perverse incentives: forcing coaches to pour it on instead of putting in the reserves and developing younger players, risks injuries to starters after a game has been decided.
Ranking and Strength of Schedule: SOS and Rankings are used horribly. Look at the RPI ranking score or Sagarin SOS score (not the rankings themselves). Compare the difference in that score among #1, #11, #21, #31, etc. Notice how for the rankings, scores get closer as they bunch together:
.7045, .6596, .6376, .6111
As a result there's as big a difference between #1 and #11 (about .045) as between #11 and #28, or between #28 and #65, or between #65 and #130! Rankings magnify tiny differences that do not matter at all. Notice how the score is proportional to the log of the rank. Solution:
Either ignore rankings and use the score or take the log of the rank.
If that's not bad enough, combining individual game outcomes into a single index score commits another major source of "noise" in the rankings. Computer rankings award identical weights (i.e., importance) to each game regardless of abnormal conditions (due to travel, injury, biased refs, etc.) apt to create atypical "outliers" in the data. In statistics, they rightfully refuse to relying on the arithmetic mean which are sensitive to outliers. Thus,
the proper solution is to use "trimmed means" that drop the high and low extremes (like they do in Olympic scoring of gymnastics).
