Math Takes On The Baseball Playoffs
By Hank Campbell | October 12th 2013 10:44 AM | Print | E-mail | Track Comments

I'm the founder of Science 2.0® and co-author of "Science Left Behind".

A wise man once said Darwin had the greatest idea anyone...

View Hank's Profile
NJIT math professor Bruce Bukiet wrote an article here on his Markov process predictions for the baseball playoffs. That wasn't something new, he is in his 13th season of doing just that, often to maddening success.

How did he do this time?

The Pirates didn't advance, the Cardinals are now facing the Dodgers, but otherwise he nailed it, with the Boston Red Sox and the Detroit Tigers getting ready to square off for the pennant. The math doesn't always work; last year his numbers said Detroit would win the World Series. Nope, Giants again, my gut beat reason and sanity.

But having 3 out of 4 teams in a 7-game series helps in accuracy. The Markov process he uses creates a probability that a team with its hitters, bench, starting pitcher, lineup and relievers will, factoring in scoring any number of runs and home field advantage, win a game. Obviously it is self-correcting for each game and each series; if a team is wiped out, it starts over with the new teams.  Hey, Nate Silver at the New York Times was shockingly wrong two weeks before the 2012 presidential election but he and everyone else who did a projection the day of the election got all 50 states right. Even bettors in Europe called the electoral results correctly by aggregating their bets, that is the power of things like Bayes.

We've written about baseball projections and Bukiet a lot in the past and he uses a Markov process that is a little more elegant for this type of projection because in a Bayesian analysis the densities aren't analytically tractable: you can only integrate them - basically, the only thing that matters in Bukiet's method is the present position and the parameters, not how many 9th-inning doubles a player hit in the month of August. You have seen the drunken walk problem somewhere, in physics you learned about Brownian motion - if a drunk guy can only take one step left or right and the problem is figuring out where he can end up - that is a Markov chain.

So what do the numbers say about the National League Championship Series?  Well, I watched the 13-inning game last night and the computations were done before that game but what the numbers said was that the Red Sox would be facing the Dodgers in the World Series.  Bukiet had the Dodgers going to the World Series 68 times out of 100.

A Cardinals win against the Dodgers number two starter could upset Markov a lot. Credit: Jeff Roberson/Associated Press. Link: Washington Post.

Of course, there are only 7 games, not 100. It just dropped to a lot less if we were really starting from scratch because the Dodgers lost last night - and one game in the books is better than probabilities (as is Carlos Beltran). 13-inning games where both starting pitchers do well are real anomalies and Kershaw is likely going to win his game again tonight.

A win for the Cardinals instead of the Dodgers does not doom Los Angeles, no mathematical process was realistically projecting that the Dodgers or the Cardinals would sweep the other team in 4 games, though Bukiet did project that the Dodgers had a 12% chance of that while the Cardinals only a 3% chance - that is the power of two strong starting pitchers in a 7-game event.

In the American League, the Red Sox are 57% likely to move on against the Tigers, despite the Tigers having two great starting pitchers - that is the power of hitting making the difference.

Despite the setback, I am still predicting the Dodgers in 6. Play ball!