Predicting NCAA Tournament Upsets: The Underdogs

By John Ezekowitz

A few weeks ago, when I started analyzing the effects of tempo on NCAA Tournament upsets, I got a crazy idea: what if I could find a model that would predict factors of successful underdogs? I’ve always tried especially hard to successfully predict first and second round NCAA Tournament upsets. Those rare occurrences that stick in our sports memories are even more sweet when you can brag about predicting them beforehand.

Now, armed with a better understanding of the practice of statistics and aided by the ever invaluable Ken Pomeroy’s website, I created a database of the 144 potential upset games in the last six years of March Madness and attempted to make a predictive model. The (very interesting) results after the jump.I limited myself to the last six years because those are the years for which we have tempo-free statistical data. For the uninitiated, tempo-free stats are those that do not depend on the pace of the game, as counting stats such as points, rebounds, and turnovers do. For instance, instead of tracking offensive rebounds, tempo-free statisticians track offensive rebound percentage–what percentage of a team’s shots does it rebound.

Using the statistical technique of logistic regression, and coding a successful upset as 1 and a loss as 0, I was able to test the predictive strength of a wide range of variables, both defensive and offensive. Most of the stats are explained here.

Two stats were statistically significant for predicting upsets: turnover percentage and the ratio of free throws allowed to field goals allowed. The two stats had identical P-values of 0.013, which are well inside the predictive threshold. A decrease in turnovers by one percent increases the chances of an upset by 26%. Likewise, a decrease in opponent free throw rate increases the chances of an upset by 10%.

What do these stats mean? Successful underdogs have experienced guard play that handles the ball well. They also are disciplined on defense, avoiding shooting fouls. Of the best 10 underdogs in terms of taking care of the ball, 7 pulled off upsets.

So which teams have the requisite traits to pull off upsets this year? Interestingly, Houston, who just got in by winning the CUSA auto-bid, turns the ball over at a historically low rate of 12.6 percent.  Siena has the best opponent free throw rate in the country, and only turn the ball over 17.4 percent of the time. Utah St., woefully underseeded as a 12, also has the typical traits of the successful underdog.

Potential 2nd-round upsetters in the 7-10 seed range include St. Mary’s, and BYU, who may have drawn the weakest 10 seed in Florida and the 2nd weakest 2 seed in Kansas State. Join me tomorrow as I identify the factors of favorite vulnerability.

This entry was posted in NCAA Basketball and tagged , , , , , . Bookmark the permalink.

9 Responses to Predicting NCAA Tournament Upsets: The Underdogs

  1. Brian says:

    Can you post the teams that don’t have the ability to pull of an upset?

  2. Lou says:


    I’m immediately wondering how much of this is causation/correlation and how many of these upsets were by teams of equivalent strength despite their seeds? For example, this season Houston over Maryland would be a much more substantial upset than say Siena over Purdue.

    I’m sure you’ll cover this in part 2.

  3. Jacob says:

    Any chance you can put the dataset up somewhere to share? I’d love to play around with it.

  4. Pingback: Badgers in the Final Four?? « Courtside Analyst

  5. Dave K. says:

    Totally agree with Lou. Ideally the logistic regression model should include both teams.

  6. Pingback: Predicting NCAA Tournament Upsets: Favorite Vulnerability « Harvard Sports Analysis Collective

  7. Pingback: Bar Bracketology

  8. Nice post. It will be interesting to see how predictive analytics tools will be used in the future for predicting the outcomes of sporting events and other types of events.

  9. Pingback: Best NCAA Tournament Bracket ‘Underdog’ Upsets | Texas Thunder Radio

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s