THE POWER RANK

The Power Rank best at predicting the 2015 NCAA tournament

By 1 Comment

Screen Shot 2015-04-13 at 12.26.06 PMYou have a lot of choices for predicting the NCAA tournament.

My probability for each team to advance through each round appear in this interactive visual. However, these other quants perform the same calculation.

Which tournament win probabilities were the most accurate in 2015?

Reuben Fisher-Baum of FiveThirtyEight took a quantitative approach to determining which tournament predictions were most accurate. He looked at how a predicted probability deviated from the actual result in the tournament.

For example, my numbers gave a 70.7% chance for Duke to advance to the Elite 8. Since Duke did advance to that round, we assign that result the number 1.0. The square of the deviation, or (1.0 – 0.707) squared, gives a Brier score for that prediction. For an event that happened, a larger probability is better, so a lower Brier score implies a better prediction.

From FiveThirtyEight’s analysis, The Power Rank had the most accurate 2015 tournament predictions.

fivethirtyeight_2015_tourney

Duke’s run through the tournament helped out my method rise to the top. Of all the models, my numbers had the highest probability for Duke to advance to the Sweet 16 and every round after that.

FiveThirtyEight gave this very kind conclusion.

Given that Duke is traditionally undervalued in bracket pools, fans who built their brackets on The Power Rank’s numbers likely had a pretty good tournament.

Indeed, some people did have such success.

Ryan Peters bought my book on how to win your pool and knew the importance of picking a value champion, or a team with a high win probability overlooked by others in his pool. He chose Duke and sent this tweet on the eve of the championship game.

I had Duke ranked third in my team rankings heading into the tournament. They got credit for playing a tough ACC schedule which included Virginia, Notre Dame (3 times), North Carolina (twice) and Louisville, all top 20 teams in The Power Rank. These strength of schedule adjustments moved Duke higher compared to other rankings.

The Power Rank came out on top this year, but don’t expect these results every year against stiff competition. One tournament represents a small sample size of 67 games. Had Kentucky won the tournament like my numbers expected (38% win probability), the results could have been different.

John Calipari is a better tournament coach than Tom Izzo

By 4 Comments

calipariJohn Calipari evokes many different emotions in sports fans.

If you’re a Kentucky fan, you probably love Calipari. In six years as coach, he has won a national championship, and his 2015 team might win another with an undefeated record.

If you’re not a Kentucky fan, Calipari represents all that’s wrong with college basketball. His teams at Massachusetts and Memphis had to vacate wins during Final Four years because of NCAA rules violations. While the NCAA never found Calipari guilty of anything, it seems unlikely he knew nothing about the infractions.

Moreover, Calipari pisses off his colleagues. At a press conference, he got former Temple coach John Chaney so mad that Chaney attacked him, yelling “I’ll kill you.” Yes, this really happened; check out the video.

Numbers reveal a third side Calipari: he’s an amazing tournament coach. This article looks at how teams perform in the tournament compared with a regular season baseline. With a high degree of statistical certainty, Calipari’s teams play better in March than the regular season.

Calipari’s ability to get more out of his teams during the tournament is neither a typical part of his narrative nor the story in which I was originally interested. Michigan State coach Tom Izzo usually gets praised for his excellent coaching in March. I didn’t believe this conventional wisdom, so I dug into the numbers.

Comparing tournament performance with the regular season

To test Michigan State’s play in the tournament, I compared their margin of victory in the post season with expectations from the regular season.

For a regular season baseline, I used my college basketball team rankings at The Power Rank. Developed from my Ph.D. research in statistical physics, this algorithm takes margin of victory and adjusts for strength of schedule.

From 2002 through 2014, the higher ranked team in my pre-tournament rankings won 71.3% of games. In addition, the rankings provide a predicted margin of victory in each game. We’ll use this prediction as a baseline for tournament performance since, unlike the point spread in Vegas, it makes no preference for Michigan State or any other team in March.

For all tournament games from 2012 through 2015, I looked at Michigan State’s actual margin of victory compared with The Power Rank’s prediction. For example, in 2015, Michigan State beat Virginia by 6 points and exceeded the baseline prediction by 12.1 points. In 43 games tournament games, Michigan State has exceeded their expectation from The Power Rank by an average of 2.07 points.

Two points might not seem like a lot, but it’s a huge jump in performance. If the betting markets favor a college basketball team by 2 points, this teams wins the game 58.4% of the time, much more than the 50% for a game with a zero point spread.

Are these results statistically significant?

However, we can’t just assume that Michigan State performs better in the tournament based on this 2.07 points. There’s randomness in this estimate. We don’t know whether Michigan State performed at the same level as the regular season and got lucky by two points a game. Or Michigan State could be 4 points better than the regular season and got unlucky in the tournament.

Statistics gives us tools to account for the randomness in this estimate. A t-test, a method first developed at the Guinness Brewery, provides a probability that this estimate of 2.07 points is better than zero. This test, using this nifty calculator, gives a 92.6% confidence that Michigan State performs better in the tournament. (For those who want to check my work, the standard deviation of sample mean over 43 games was 1.40 points.)

I was wrong. The numbers suggests a high likelihood that Tom Izzo’s teams perform better in the tournament. Conventional wisdom wins this time.

The greatness of John Calipari

With the code to perform this test for Tom Izzo, I decided to repeat the test for the other Final Four coaches in 2015.

Duke coach Mike Krzyzewski has a reputation for getting the most of his players in March. However, since 2002, they have performed 0.41 points worse than The Power Rank’s expectation from the regular season.

Wisconsin coach Bo Ryan did slightly better than Krzyzewski in the tournament. However, his teams still performed 0.21 points worse than expected over 37 games.

John Calipari was a different story. During his years at Kentucky (2010 to present), his teams have performed 3.86 points better than their regular season expectation. Even with the smaller sample size than the other coaches (25 games), we can be 98.1% sure Kentucky has played better in the tournament.

Calipari also coached at Memphis before taking the Kentucky job. From 2002 through 2009, his Memphis teams exceeded their regular season expectation by 1.38 points in 20 tournament games.

It’s probably best to combine the tournament performances of Calipari’s Kentucky and Memphis teams, which gives a 2.76 point improvement over 45 games. That implies a 96.8% confidence that his teams play better in the tournament. In addition, Calipari’s tournament improvement is 35% larger than Tom Izzo’s improvement.

The following list summarizes the difference in tournament performance from the regular season since 2002.

  • John Calipari: + 2.76 points per game.
  • Tom Izzo: +2.04 points per game.
  • Bo Ryan: -0.21 points per game.
  • Mike Krzyzewski: -0.41 points per game.

For the 2015 Final Four, the main story should be John Calipari’s greatness as a tournament coach.

Check out the latest NCAA tournament predictions for 2015

By 1 Comment

Screen Shot 2015-03-09 at 12.24.43 PMWhile everyone else is trying to predict which teams make the tourney, I like to know which teams might win.

I know bracketology is fun for fans. I don’t pay attention at all. I just take Joe Lunardi’s projection and calculate win probabilities for each team. To see the interactive bracket with win probabilities, click here.

However, I messed with his field of 68.

Wisconsin keeps landing as the 2 seed in Kentucky’s region. The Badgers are one of 5 teams clustered within a point of each other behind Kentucky in my team rankings. With this field, Kentucky has a 31.9% to win the tournament.

Kansas is not one of those 5 teams ranked behind Kentucky. The Jayhawks are 9th in my rankings and got clobbered by Kentucky earlier this season. However, Kansas is projected as a 2 seed.

The interactive bracket shows results with Kansas as the 2 seed in Kentucky’s region. Kentucky’s win probability for the tournament increases from 31.9% to 35.5%.

We shall see what the actual committee gives us on Sunday.

What is Kentucky’s win probability for the 2015 tourney?

By Leave a Comment

Screen Shot 2015-02-20 at 8.43.07 AMYou know Kentucky can ball. They’re threatening to become the first undefeated college basketball team since Indiana in 1975-1976.

Coach John Calipari has coaxed this team into playing lights out defense. They’re not just first in my defense rankings by points per possession adjusted for strength of schedule. Kentucky is more than 3 points per 100 possessions better than second ranked Arizona.

It’s hard to poke holes at this team. They don’t shoot well from the outside except for reserve Devin Booker? Karl Anthony Towns isn’t quite as athletic as his competition for the first pick of the NBA draft, Jahlil Okafor of Duke?

I wasn’t always a believer in this Kentucky team. After they thrashed Kansas early this season, some of my friends had them as a Final Four lock. I disagreed, thinking that almost no team has better than a 50% chance to make the Final Four. Now I’m not so sure.

Let’s put some numbers behind Kentucky’s chance to make the Final Four and win the NCAA tourney.

Tourney win probabilities

I took Joe Lunardi’s projected bracket from Thursday, February 19th and used my college basketball rankings to calculate tourney win probabilities for each team.

I’ve made some modifications from last season. The rankings provide a margin of victory between any two teams, and this spread implies a win probability. This year, I’m taking a data driven approach to translating the spread into win probability.

This change has led to a higher win probability for the favorites. For example, Arizona would have had a 17.9% chance to win last year’s tourney, higher than the 9.5% I showed last season.

Check out the interactive visual for tourney

This interactive visual shows the probability for each team to advance to each round. Hover over a team to view its chance to advance, or hover over a circle to see the odds that each team wins that game.

Kentucky has a 34.1% chance to win the tourney. And this estimate is probably low, as this bracket puts a strong Wisconsin team as the two seed in Kentucky’s region. If a Villanova or Kansas were that two seed, Kentucky’s win probability would be even higher.

Since the 2002, only North Carolina in 2007 had a higher win probability before the tourney. This team, which featured Ty Lawson and Tyler Hansborough, had a 37.8% to win the tourney but lost in the Elite 8 to Georgetown. Florida won that tourney for their second straight title.

The visual also shows Kentucky has a 61.4% chance to make the Final Four.

How to win your tourney pool

I’ve been digging into data from past seasons as part of my research into optimal bracket strategies. To learn when my short ebook on how to win your pool becomes available, sign up for my free email newsletter.

Enter your email and click on “Sign up now.”