Monday, April 06, 2015

When the best team doesn't win (UK-Wisconsin edition)

On June 10, 1992 the Portland Trail Blazers beat a favored Chicago Bulls team led by Michael Jordan and Scottie Pippen 93-88. It was the Bulls' 22nd, and final, loss of the season. Did that game prove that the Trail Blazers were better than the Bulls? No. It was Game 4 of the NBA Finals, and Chicago would win the next two to close out the seven-game series in six games.

On April 4, 2015, Wisconsin beat a favored Kentucky 71-64 in the NCAA Final Four. It was UK's first, and last, loss in 39 games. Did that game prove that the Badgers were better than the Wildcats? No. But the NCAA does a one-and-done tournament, so you don't get more evidence.

The problem with most sports "journalists," pundits, coaches, and fans is that they don't apply statistics to their analysis and make some real errors of logic. Two of the better-known ranking systems out there (Pomeroy, Sagarin) use ELO models to judge teams against each other based on games, points, and possessions while using opponents' (and opponents' opponents) measures of the same. Even with the head-to-head win, Wisconsin still shows up rated lower than Kentucky in terms of the full body of evidence, the entire season:



Wisconsin executed various strategies that Kentucky (who allegedly watched no film of Wisconsin) could likely counter in the second game (and vice-versa). Nate Silver points out a crucial flaw in Kentucky's strategy at the end of the game-- the better team should maximize number of possessions, rather than minimize, as Kentucky apparently did (I was asleep, but had an email from someone echoing Silver's point exactly at the time). Silver's model, which uses a composite of other ELO models (not all ELO, but I'll use it as generic term) like the one above, gave Kentucky a 69% chance of beating Wisconsin.

Ask yourself this: If there's a 31% chance of rain, do you take your umbrella? Do you get mad if it rains and call the weatherman "wrong" for giving it such a low probability? (I saw a college baseball coach do that once on Twitter. Oh, the irony as a .310 hitter is "good" by professional standards). 

If Silver's model is reasonably accurate (and it was similar to other projections) then if you played the game 1,000 times, Wisconsin would likely win more than 300 games (with some variance). But the odds of them winning more than 500 times would be pretty slim.

A team estimated at a 40% chance of winning the tournament before the first game tips is unheard of in the modern era. Having a 69% chance of beating another #1 seed is ridiculously high. If the models are reasonable, then Kentucky is still a "historic" team, still maybe the best ever, even if sports writers will write them out of history. (1991 UNLV, 1999 Duke were also very good teams).  Kentucky will still likely be #1 in most models on Tuesday, after the official champion is crowned.

This post is a little crude, but I feel better having ranted a bit.

No comments: