clock menu more-arrow no yes

Filed under:

Post-Madness Bracket Comparison and Thoughts

Every year, millions of Americans fill out March Madness brackets, and, every year, millions of Americans wonder to themselves, "What's the best way of doing this?" While picking the school with the tougher mascot or the higher average graduate salary are certainly valid strategies, there are at least a few resources out there that attempt to provide order to an inherently random and imperfect science. This year, along with my own expert picks, I filled out a couple brackets based completely on both available computer rankings and random selection methods. Here's how they did, along with some conclusions from this year. (To preserve my dignity, I am not including the performance of my expert picks)

System ESPN Total Points ESPN Percentile
Ken Pomeroy Rankings 1320 92.4
LRMC Rankings 1310 91.8
AP Poll 1220 84.0
Chalk 1200 81.8
Coin Flip when Reasonable* 920 56.7
Weighted on Seed History* 890 54.1
RPI 820 46.9
Coin Flip 180 0.4

*Random number generator was used for all selections. Obviously, one's results will vary when repeated. For "Coin Flip when Reasonable," I gave each team a 50/50 chance...unless the seeds were separated by more than 10, in which case I advanced the higher seed. For "Weighted on Seed History," I assigned each team a probability of advancement based on the historical percentages between the two seeds (repeated for each matchup in all rounds), and used a random number generator.

What can we learn? A few lessons from this year's Madness are:

1. Not much.

As with any process involving randomness, we can't read much into the outcome of a single trial. This is true for a few reasons. Mainly, a team may be "correctly" ranked higher than another, but the better team may not always win; regardless of what happened this month, anyone that believes Lehigh is better than Duke or Norfolk State over Missouri is not a super-genius. For the same reason, it is not reasonable to read too closely into any single-year results.

2. RPI is terrible

One thing I feel comfortable concluding is that RPI is not a useful tool in comparing teams against each other. Every year, there are a couple teams that RPI significantly overrates or underrates; because of the committee's emphasis on RPI in seeding, it's a good way to seek out a couple upset picks. This tournament, some of these teams were "hidden," as underrated Memphis and St. Louis played eachother, then had to take on Michigan State. While its miserable score mostly stems from the fact that it had Syracuse winning it all, stay away from using it in your picks. As we saw this season, you are better off picking ignoring teams and just choosing seed numbers.

3. The coin flip thing didn't work out either

A darn shame. Though if my UNC-Asheville, my eventual champion, had slid past Syracuse, things could've gotten interesting!

4. Use KenPom and LRMC for guidance...

If you are looking for undervalued or overvalued teams, look no further. Both significantly outperformed any human method of evaluation. Even when KenPom advances a favorite, you can learn when a matchup appears closer than believed, making it a prime upset choice (for example, his computer gave #12 USF a 49% chance of upsetting #5 Temple).

5. ...but if you want to win your pool, you'll have to get creative

Nothing new here. If you choose a chalk bracket, you'll be a good bet to finish in the upper echelon of your office pool...but you won't win. A winning bracket is one that uses the available information on teams, while also switching some things up, going with the all-knowing "gut," and hoping the ball bounces your way.