During the NCAA tournament last year we played a fun little game. For every round, just as the teams tried to prove their worthiness on the court, I asked some sharp handicappers and analysts to put their own knowledge to the test and play ... wiseguy brackets.
Last season, Sal Selvaggio from madduxsports.com beat out Vegas legend Alan Boston when Butler covered in the final. And this season, the format will be the same. Four wiseguys from a rotating pool of eight (give or take a couple) will share their wisdom for each bracket. Then, the two with the most wins will face off in the Final Four and the final. The winner will get your undying devotion. And maybe a fleece.
Today, we're breaking down the West with Boston and the Southeast with newcomers to the blog stable, TeamRankings.com (the brains behind Insider's Bracket Predictor tool -- look to the right). I met Tom Federico, CEO and co-founder of TR during the Sports Analytics conference at MIT a couple of weeks ago. After I moderated the panel on sports gambling, Tom came up to me, introduced himself, handed me a packet describing his site and the data he used and then told me, basically, that I was often an idiot. Finally, someone is paying attention. But he did it all with a smile. Then I went and checked out his site, which was very cool, very thorough and very deep.
Earlier this week I called Tom to invite him into the wiseguy bracket. Why should I be the only one potentially embarrassing myself in the column? He graciously agreed and we started talking about his site. Turns out TeamRankings.com was started by Federico and his buddy, Mike Greenfield -- two sports-loving Stanford students who decided to use the math and engineering they were learning for good, not geek.
"We have mined about 15 years' worth of data across all the major sports," Federico says. "And we use a lot of contextual game variables in our analysis. For example, when a team that looked statistically liked Florida State played a team that looked like Texas A&M, what was the spread, how far did the team travel? We examine those types of game factors as well as team and player stats."
Ultimately, TR uses three prediction models to determine how confident it is about an against-the-spread prediction. One is the standard power rating. Two is a bucket titled "Similar Games" which incorporates the power rating as well as dozens of other stats and betting lines that featured similar teams. And three is the "Decision Tree" model, which is what I found most fascinating. "[That] came straight from the Stanford math textbooks," says Federico.
Greenfield was working as a programmer at Paypal in the early 2000s. The bosses there realized they were often getting scammed and assigned Greenfield to a team that was tasked with figuring out why. Using sophisticated programming language and mathematical models they were able to find the "signal," as Federico called it, that was common in every scam or scam attempt. TeamRankings uses similar programming techniques in its Decision Tree model. Ultimately, it identifies the elements that most accurately predict who will win and lose a game and by how much.
"It is what a guy who studies statistics would call a black box model," says Federico. He started to explain it even more and then, when he realized it was hopeless, just sent me to the website for a clearer explanation. Here it is: "The Decision Tree model is the output of a machine learning algorithm that views every college basketball game since 1999 through the lens of hundreds of input variables, ranging from contextual information like the distance traveled by squad to team statistics like effective field goal percentage."
The algorithm does what might be convenient to think of as complex, high volume, statistically significant trend analysis. It repeatedly partitions the games into smaller and smaller subsets based on the values of one or more variables. Each split is chosen so that the win probabilities of the teams in each group get further away from 50 percent and closer to 0 percent or 100 percent.
Of course, it makes total sense -- as long as it translates to wins.
Without further ado, here are the expert picks for the West and Southeast regions:
West: Alan Boston
"I think Duke will have a hard time getting super motivated for this game. It knows it can waltz through it. But Krzyzewski doesn't have teams that let up too often. Meanwhile Hampton is in a weak conference this year and pre-conference games would indicate they'll have a problem because they lost to Wake Forest, which isn't a good team. I think Duke can name the score. But I don't think they'll have motivation for it. I made it 22 and it is 22.5 right now. I am not fooling with this game. "
ATS pick: Pass
Straight-up pick: Duke
"This is a classic tourney game that, in the old days, you would drool over. It's the not-so-smart Vols team vs. the ultra-smart Michigan team. If this had been played a month ago the Vols would have been a four-point fave. The line has been adjusted. However, it's on its way back up and people are betting Tennessee because the opener of Michigan minus-1 was silly. Michigan got better as the year went on and it was no fluke. I can't imagine Tennessee being ready for the 1-3-1 and the well-timed, constantly cutting offense that Michigan has and the other stuff John Beilein will throw at them. I made it Tennessee -1, but Michigan's home run efforts this year have been spectacular. I think Michigan is going to win the game because smarts can beat athleticism in the NCAA tourney. It wouldn't surprise me if Michigan runs them out of the gym.
ATS pick: Michigan
Straight-up pick: Michigan