Beware the Polls: Overperforming Teams Regress in NCAA TournamentPosted by Will Ezekowitz on March 14th, 2017
Preseason rankings may seem irrelevant when it comes to college basketball, but history has shown that their predictive value are surprisingly important. For example, as we have shown in this space before, the preseason poll is in fact just as predictive of a run to the Final Four as the current AP Poll. The reasoning behind this phenomenon is that the preseason polls take into account all of the changes that the mathematical models have more difficulty accounting for (i.e. coaching changes, big recruiting classes, injuries, etc.). The conventional wisdom entering the season is actually fairly sophisticated and smart.
Given this intuition, let’s examine the teams that defied that wisdom throughout the season and went from unranked in the preseason to ranked in the final AP poll before the NCAA Tournament. Had the preseason polls gotten something wrong? Or did these over-performing teams regress back to their expectations in the postseason? Using the last 10 years of data that includes seed win expectation data from fivethirtyeight.com, I investigated. Of the 97 teams who qualified for analysis, 32 (33%) of those teams outperformed their win expectations in the NCAA Tournament while 65 (67%) underperformed. As a whole, a group that was expected to win 138.8 games over the 10-year sample won just 107. A statistical T-test found this difference statistically significant. In fact, only one team — Kemba Walker’s 2011 National Championship team, Connecticut — even made the Final Four.
Moreover, in none of the tested years did the sum of the teams’ actual wins exceed their expected wins. Put another way, even though individual teams may have outperformed expectations once in a while, the group as a whole always underperformed.
So what does that mean for this year’s NCAA Tournament? Which are the teams that we shouldn’t trust?
This year’s group of SMU, Baylor, Notre Dame, Florida State, Cincinnati, Wichita State, Florida, Butler and Michigan is projected to win 11.1 games over the next three weeks. The total of nine teams is the most since 2013, when, led by #2 seed Georgetown, which infamously lost to Florida Gulf Coast, seven of that year’s nine such teams underperformed.
History has been repeating itself now for the last 10 years. Maybe a couple of these teams will beat the odds and make a run. But on the whole, this group will underperform against their seed’s expectations. You can bet on it.