Conference Bias in the NCAA Tournament

By Kurt Bullard

With the brackets now set and Harvard cementing its status as an Ivy League dynasty with four straight NCAA Tournament bids, over 40 million people will begin to fill out brackets for the chance to win bragging rights and, for those to whom that is not enough, possibly a cash prize.

March Madness is by far my favorite week to geek out and do copious research to try to gain an advantage over my friends and family. One question that I’ve always had is whether the selection committee tends to favor or malign a conference in choosing the seeding. My colleague Austin Tymins looked at a similar issue in college football rankings earlier in the year and found that the SEC tended to be ranked higher than what they should have been. Anecdotally, I had always thought that the Pac 10 – now the Pac 12 – was undervalued because of some sort of east coast bias. Rather than speculate like I had in the past when I was a kid, I thought I’d take a quantitative look to see if conferences consistently outperform or underperform their respective seeds.

I only looked at the first through six seeds because of a weird spike at the seventh seed that cannot be explained just by UConn’s run to the championship last year. That having been established, the following table shows the average number of wins per tournament by seeding:

Using seeding data from KenPom since 2002, I looked at how many wins each seed has averaged in the past thirteen years.  Then, I found out how many each conference should have over that period based on every team’s expected performance. Then, I calculated how many wins that the conference actually accumulated to see the difference between actual and expected wins by a conference. Theoretically, there should be no difference between the two if there isn’t bias.

I then regressed predicted wins by conference wins and then added a dummy variable for all of the power conferences and the Atlantic Ten, which has consistently been sending teams to the dance as of late and therefore will have a large enough sample size to run analysis.

If the dummy variable for a conference was significant, then the conference either overperformed or underperformed their seedings whether it was negative or positive, respectively.

Using a significance level of 5%, no conference is significantly maligned or favored by the selection committee. The Atlantic 10 has most negative residual, meaning that the conference has been over-seeded in the past thirteen years, but not to a significant extent. But, using a slightly bigger alpha level, the Big Ten is the closest to being significant. The residual is positive, meaning that the selection committee has undervalued the conference. The SEC is not too far behind, but it would be too much of a stretch to say that it has been undervalued significant.

So, don’t get swayed by the 7 seed of Michigan State and the 9 seed of Purdue. There’s a chance they’re better than their seeds suggest.