Tuesday, March 29, 2011

2011 Community Playing Time Forecast

Once again, Tom Tango is asking everyone - including you - to predict the playing time for everyone on your favorite MLB team. So if you're willing to make some educated guesses about how much everyone on your favorite team, head over to

http://www.tangotiger.net/survey/

Thursday, March 17, 2011

Sabermetric Bracketology 2011

I've been applying sabermetric (or, more appropriately, APBRmetric) principles to my NCAA bracket selections for the past three years. In the contest that uses upset points, I was the big winner the first two years, then fell to a disappointing fourth place last year. So, I've been looking to change my methodology for this year.

Fortunately, master prognosticator Nate Silver came up with his own bracket forecast this year, which I was able to modify for my upset contest's scoring system.

To recap the upset scoring system, a team gets a point per round for winning (one point for winning in the round of 64, two for the round of 32, up to six for winning the championship game), plus bonus points for defeating a lower ranked seed. Bonus points are the difference between the two seeds. For example, if a 12 beats a 5 in the first round, they get one point for the win, plus 12 - 5 = 7 bonus points, for a total of 8.

I then took the potential points each team gets for the win and multiplied by the probability of winning to get the team's expected value for each game. In the first round, this was easy - it's just the probability of winning times the difference in the seeds. For example, Notre Dame has a 91% chance of beating Akron. So ND's expected value is .91*1 = .91, while Akron's is .09*(1+15-2) = 1.26.

In later rounds it's a bit more complicated, since the number of points a team gets for winning depends on who they play. If Marquette wins their first round game, they might play Syracuse, a lower seed (which would earn them bonus points with a second round win), or Indiana State, a higher seed. So, the calculation adds a layer.

In the Marquette example, they have a 90% chance of playing Syracuse in the second round (since that's the Orange's probability of winning the first round game) and a 10% chance of facing the Sycamores. Marquette has a 16% chance of advancing past the second round, and would earn 2 points for beating ISU and 10 for beating Syracuse. So the calculation becomes .16 * (.90*10 + .10 * 2) = 1.47.

I propagated this formula for every team for every round, until I got a total expected value for all six rounds for every team in the tournament. From there, I started pairing off the teams into games. I determined the winner by choosing the team who had the greater expected value over the remainder of the tournament, figuring that that would help maximize my value. I'm not sure if this is the best method - some sort of Monte Carlo simulation would probably be the ideal - but it was at least something I could take for a test run.

Some interesting observations along the way:
  • Silver's bracket has Notre Dame as the "worst" two seed, with just a 1.8% chance of winning it all. Of course, that may be more because it likes Purdue so much, and not because it dislikes Notre Dame.

  • Upset points considered, Clemson is a heavy favorite over West Virginia. Of course, before I could write Clemson into the second round on my bracket, I had to make sure they first won their play-in game.

  • As with my past methods, the odds still favor putting all the #1 seeds in the Final Four. Since I'll be playing against a bunch of Ohio State homers in my upset contest, I'm hedging my bets by having Duke beat them in the semis and beating Kansas in the finals.

  • I also used Silver's bracket to fill out a bracket that had a more straightforward scoring system. But I also took some liberties, putting Notre Dame in the finals, where they'll lose to SDSU. Apparently I'm not as big of a homer as Luke Harangody, who has the Irish winning it all in the "celebrity" bracket he did for Fox Sports Ohio.