THE BOOK--Playing The Percentages In Baseball

Sagarin’s rankings aren’t the only game in town anymore; I use Ken Pomeroy’s if I only get to use one.  Otherwise, I weight KenPom 2x or 3x with Sagarins at 1x.

I was just thinking, actually, that it’d be cool to go back over the last 25 years of the NCAA tournament and rank teams pre-tourney using Sagarin’s and Pomeroy’s methods.  Then see which proved to predict actual results better.  25 years is probably not nearly enough to make any meaningful conclusions, but I’d really be curious.  My hunch is that Pomeroy’s are better.

I use Pomeroys..........surprisely my bracket is extremely similar to Sean.

The best “rankings” or “ratings” is the betting line.  Its almost public information, I don’t know why more people don’t use that information when doing projections like this.  Better to use a market number rather than any one individual person’s rankings.

I should have added: the betting line is only good for the first round, after that, there isn’t that much information that one can gather.  Also, when I say the betting line, I don’t mean the one that the bookmakers post at first, but rather the betting line closer to game time when bettors have put their opinions into the line by betting.

On the topic, does anyone know a good site for statistical college basketball analysis besides kenpom and basketballprospectus?

Slate had an interesting article on this.  In a nutshell, the concensus bracket may work well in small pools, but you need to gamble a bit if you are in a bigger pool.  Click on my name for the link.  I’m more of a quantitative person than the average Joe, but the denizens here can probably run circles around me, so I’d be interested in what you guys think about the article.

I skimmed over Slate’s article.  It is similar to a chapter in my book Weighing the Odds in Sports Betting (I don’t want this to go the wrong way, so if the owners of this site have an issue with my mentioning my commercial work, feel free to delete it).  My material is better IMO because I suggest taking market information (in the form of gambling lines) into account, which Slate does not.  He mentions Sagarin, but why stick with one guy when there is information from the general market?  That’s like saying the fair price in GM is what the Morgan Stanley analyst says it is and ignoring last sale in the NYSE.

Overall, its one of the complaints I’ve had with the academic and/or sabremetric fields.  For some reason, they seem to ignore betting lines even though they have to know it exists.  Instead of using betting lines, they use either one person’s power rankings to make up lines or they try to create estimated odds themselves.  Why do extra work when the market has it done for them...and the market will almost always do a better job of pinpointing fair value than most individuals?  (note, I’m not saying the betting line is always right, not at all.  But it is usually pretty close even when it is off....and more often than not, it is better than any individual projection, although it can still be beaten).

I’ve been using the method Slate mentions for the last 3 years, though I didn’t know it until I read the paper.  The biggest champion underbet on ESPN.com was BYU and it wasn’t close.  It hurt to see them go down 2.5 hours into the tourney, but I accepted that risk going in.  Bear in mind that I’m not at all sure my method is sound and that I’m a total amateur.

I used Pomeroy to construct a simple table with the probability each team would make each round.  Instead of running a simulation, I simply found the probability each team would win it’s Round 1 game and multiplied that by the probability they’d face each opponent X the probability they’d beat that opponent. 

To find the biggest underbet on ESPN.com, I used their Who Picked Whom table.  I wasn’t sure how to proceed, so I divided the probability a team would reach a round by the % of ESPN users who picked them to reach that round. So if Memphis won it all 7% of the time according to Pomeroy’s ratings and 21% of brackets had Memphis winning, then they got a 1/3.  BYU won it all ~1.7% in Pomeory and was picked by 0.0% of all ESPN.com brackets, which seemed to be a problem until I noticed that the table was still sorted by the number of brackets with each pick, and so I could assume that at least some of the 0.0%s were really just ~0.04%s that got rounded down.  For example, BYU was the top 0.0% team, so I assigned them 0.04% of ESPN.com’s brackets.  They came out well ahead of the pack at 1.7/.048=~45.  Arizona State was second at ~35. 

The method Slate mentions suggests picking the top underbet champion and filling out the rest of the bracket by pure favorites.  I decided to follow that advice this year, except I made my runner-up Arizona State.  I thought that might give me an edge over anyone using the Duke/Slate method.

The only way I can think of to test this method is to actually examine whole brackets from ESPN.com and compare the probabilities of winning the contest (taking into account tiebreakers, etc.).  I wonder how well ESPN’s new Insider feature the Bracket Optimizer fares and what it suggests.  On the whole I was happy to give my method a try.  I’m as comfortable with the idea of competing against as few other brackets as I need to as I am with any other strategy, but I have very little confidence in any particular strategy.  I just don’t have the math skills to begin to think about it.