Saturday, March 10, 2012

Deriving a Heuristic for Predicting NCAA Tournament Results

It's not an althorithm,

Since the inception of the 64-team tournament in 1985, each seed-pairing has played a total of 108 first-round games, with the following results:
  1. The #1 seed is 108–0 against the #16 seed (100%).
  2. The #2 seed is 104–4 against the #15 seed (96.30%).
  3. The #3 seed is 92–16 against the #14 seed (85.19%).
  4. The #4 seed is 85–23 against the #13 seed (78.70%).
  5. The #5 seed is 72–36 against the #12 seed (66.67%).
  6. The #6 seed is 72–36 against the #11 seed (66.67%).
  7. The #7 seed is 65–43 against the #10 seed (60.19%).
  8. The #8 seed is 51–57 against the #9 seed (47.22%).
Therefore, when you fill out your NCAA Tournament bracket, always choose the higher-seeded teams in the first round, with possibly the #8 seed. 

Later on, go with a higher-seeded team.  It's not perfect; after all, how many tournaments ended with the Final Four lacking and first-seeded teams?


    1. So a #16 seed is mainly in it for the bus ride.

    2. I checked, a number 16 seed team never beat a number 1 seed team.

      Duckbutt, using the additive rule for probability (32 events/32 first round games), your method would be right for an average of 20.4 games out of 32. It's effect is strongest for the top four seeds: going with the top four seeds yields 14.3 out of 16 correct picks.

    3. How did you do on the first day?

    4. Sounds a little seedy to me ...

    5. Well, my system is not foolproof. Two #2 seeded teams (Mizzou and Duke) choked yesterday.

    6. Thanks for the observations, Mike Elvis, Bilbo, Svejk.

      My final four is Ky., Mich St. NC and Ohio St.