Making Sense of the Madness: Forget What You’ve Just Seen

A detailed view of a March Madness branded basketball. Patrick Smith/Getty Images/AFP.

Predicting college basketball games during March Madness requires an avalanche of data. With the sheer number of teams involved, gamblers need as much information as they can possibly get their hands on, including the most up-to-date data available on each team. Or do they?

Incredibly, the ability to predict upsets and winners in March could come down to the same skill that makes for a good three-point marksman or football cornerback—a short memory.

Who’s hot? Who’s not? Who cares? Over time, one of the best predictors of March success has been some of the oldest information available on the season that’s coming to an end—the preseason poll.

Older Is Better?

On Saturday, March Madness got off to an early start with a historic number of upsets. Seven of the top 10 ranked teams suffered losses, including everyone in the top 5. It was the most top 10 upsets in a single day in the history of the AP poll.

No. 1 Gonzaga lost by 10 to St. Mary’s, No. 2 Arizona by 16 to Colorado, No. 5 Kansas by 10 to Baylor. No. 3 Auburn (against Tennessee), No. 4 Purdue (Michigan State), and No. 6 Kentucky (Arkansas) all had closer games but still went down to upsets.

What happened to everyone, and, more relevantly, could any of it have been foreseen?

If we turn back the clock to mid-October and look at how the AP poll expected the basketball season to shake out, we notice a lot of familiar names. Baylor was No. 8. Arkansas No. 16, and Tennessee No. 18.

Michigan State was at the top of the “others receiving votes” category, just outside No. 25, and Colorado also received votes in the opening poll.

The teams that were springing upsets at the end of the regular season were all expected to be contenders heading into the year.

Does It Work?

Studies have shown that the preseason poll is at least as good a predictor of March results as other, more current analytic tools. Preseason poll positioning was actually more predictive of upsets than RPI, the model the selection committee used to rank NCAA teams until a few seasons ago.

At every seed line in the tournament, teams that did better in the preseason poll win more games. No. 1 seeds with higher preseason rankings win an average of 1.5 more games in the tourney than top seeds not expected to have as good a year. Two seeds win 0.86 more games, six through eight seeds, get twice as many wins if they did better in the preseason poll.

Why Does It Work?

There are a couple of reasons why the preseason poll may prove to be so useful. One is that it clears out all the midseason drama. People tend to overreact to the most recent thing that happened, and one upset loss—or even a cold spell for a week or more—just doesn’t prove to be all that predictive.

Are Gonzaga, Arizona, or Auburn suddenly more vulnerable in the tournament because they lost on Saturday? Probably not. Years ago, the selection committee stopped looking at teams’ records in their last 10 games as a seeding tool, because it just wasn’t all that predictive of tournament success. Looking at the preseason poll wipes out that recency bias.

Another reason could be that it measures things that don’t appear in a box score—by necessity. Voters in the preseason poll don’t have much information available to them. The most recent poll was seven months earlier, and rosters have been overhauled since then. There are no recent game results, no injury information.

Instead, decisions are made based on the expected talent level on a team’s roster. Do they have a bunch of good players? Do they have veterans? A strong point guard? A big man?

Tournament season is another time when information is severely limited. All through conference play, teams are playing rivals, often for a second time in just over a month. Opponents are well-scouted, tendencies committed to memory, and it’s a battle of Xs and Os.

In March, teams often don’t know their next opponent until 36 hours before tip. It limits the amount of scouting and prep that can be done. Instead, coaches have to just let their guys play, and results often depend on a team’s talent level, experience, guard play, and inside presence.

Worth a Look

Clearly, gamblers should utilize all the information available to them, but the preseason poll is a tool that they should not forget when gambling. It can make the difference between picking an upset and being upset.