Applied Bracketeering: So, who saw that final four coming?

By: Richard W. Sharp

The question

Welcome back sportsball fans. As we head into the home stretch, how goes our experiment in bracketeering? Who’s hitting it out of the park and who’s throwing ducks (Oregon, obviously)?

Is it time to move the goalposts? Maybe.


How’s it shaping up?

  • RPI has moved up to the middle of the pack, but there’s little room left for growth. Other approaches are better placed to win.
  • True noise seems to be pulling ahead of mascot noise.
  • The Final Four was tricky this year, but approaches that picked 2/4 teams are well positioned to take it all. Some have potential to almost double their current score over the last three games.

Time to update our pre-tourney predictions to foreshadow the forthcoming Final Four. As always, we’re using a simple scoring system, where each 1st round game is worth 1 point, each 2nd round game 2, 3rd round game 4, etc. With this system, remember, there are a total of 32 possible points each round.

Brackets below are ranked in ascending order of current score (not potential).


Stand-alone brackets

Here is an updated leaderboard of the stand-alone brackets (vs. statistics over many brackets produced by different randomization strategies). Upsets are any game in which the team with the better seed loses. 

  1. Obama’s bracket
    • Points: 59
    • Final Four (1/4)
      • Duke (2) vs. Arizona (2)
      • Kansas (1) vs. North Carolina (1)
    • Championship: North Carolina (1) defeats Duke (2)
    • Potential Points: 59 + 16 (UNC over Oregon) + 32 (UNC wins title) = 107
  2. Sagarin with mascot weighting #14, (bracket.ranker_SAG.mascot_14.csv – this is the one that really counts. We entered it in an actual pool)
    • Points: 63
    • Final Four (1/4)
      • Villanova (1) vs. Gonzaga (1)
      • Iowa St. (5) vs. Kentucky (2)
    • Championship: Villanova (1) defeats Kentucky (2)
    • Potential Points: 63
  3. RPI
    • Points: 66
    • Final Four (1/4)
      • Villanova (1) vs. Arizona (2)
      • Kansas (1) vs. North Carolina (1)
    • Championship: Villanova (1) defeats Kansas (1)
    • Potential Points: 66
  4. Harraton’s bracket (without revealing sources and methods, I will reveal that this source is close enough to the author to lord this over him and attributes its success in the early rounds to an unrepentant Big 10 bias)
    • Points: 70
    • Final Four (2/4)
      • Villanova (1) vs. Gonzaga (1)
      • Oregon (3) vs. Butler (4)
    • Championship: Gonzaga (1) defeats Butler (4)
    • Potential Points: 70 + 16 (Gonzaga over S Carolina) + 32 (Gonzaga wins title) = 118
  5. Seeds (although RPI is officially used “to aid in the selecting and seeding” process, apparently cooler heads prevailed). The AP Poll was used as the tiebreaker for the final games.
    • Points: 78
    • Final Four (2/4) 
      • Villanova (1) vs. Gonzaga (1)
      • Kansas (1) vs. North Carolina (1)
    • Championship: Villanova (1) defeats Kansas (1)
    • Potential Points: 78

Randomized bracket sets

Once again, the brackets we posted on GIT were created by adding either random noise or mascot-weighted noise to some standard model. For each set of brackets listed below, we report the performance of the brackets at or above the 90th percentile in each group. Note that these final fours are not the same as the pre-tourney predictions since we are focusing on the current leaders only. If we play random brackets from these strategies year after year, we should expect to generate one this good in every ten entries.

  1. Mascot-weighted Sagarin
    • 90th pct. Correct: 42
    • 90th pct. Points: 69
    • Most common outcomes at 90th pct.
      • Final Four (1/4)
        • Villanova (1) vs. Gonzaga (1)
        • Kansas (1) vs. Kentucky (2)
      • Championship
        • Villanova (1) defeats Kentucky (2)
    • Potential points: 69
  2. RPI with standard deviation 5 random noise
    • 90th pct. Correct: 41
    • 90th pct. Points: 74
    • Most common outcomes at 90th pct.
      • Final Four (2/4)
        • Villanova (1) vs. Arizona (2)
        • Oregon (3) vs. North Carolina (1)
      • Championship
        • Villanova (1) defeats North Carolina (1)
    • Potential points: 74 + 16 (UNC over Oregon) = 90
  3. Sagarin with standard deviation 5 noise
    • 90th pct. Correct: 42
    • 90th pct. Points: 76
    • Most common outcomes at 90th pct.
      • Final Four (2/4) 
        • Villanova (1) vs. Gonzaga (1)
        • Kansas (1) vs. North Carolina (1)
      • Championship
        • Gonzaga (1) defeats North Carolina (1)
    • Potential points: 74 + 16 (UNC over Oregon) + 16 (Gonzaga over S Carolina)+ 32 (Gonzaga wins title) = 138
  4. Model composite with standard deviation 5 random noise
    • 90th pct. Correct: 42
    • 90th pct. Points: 77
    • Most common outcomes at 90th pct.
      • Final Four (2/4)
        • Villanova (1) vs. Gonzaga (1)
        • Kansas (1) vs. North Carolina (1)
      • Championship
        • Gonzaga (1) defeats North Carolina (1)
    • Potential points: 77 + 16 (UNC over Oregon) + 16 (Gonzaga over S Carolina) + 32 (Gonzaga wins title) = 141
  5. Mascot-weighted composite
    • 90th pct. Correct: 44
    • 90th pct. Points: 79
    • Most common outcomes at 90th pct.
      • Final Four (1/4)
        • Villanova (1) vs. Arizona (2)
        • Oregon (3) vs. UCLA (3)
      • Championship
        • UCLA (3) defeats Villanova (1)
    • Potential points: 79

Coming up

It’s time to play for all the marbles. After the championship, we’ll take a look at what worked and how we might adjust for next year.

About The Author

Richard is a Seattle area data scientist who builds predictive models and the services that deliver them. He earned a PhD in Applied and Computational Math from Princeton University, and left academia for the dark side of science (industry) in 2010, following his wife to the land of flannel. Fan of coffee, beer, backpacking and puns. Enjoys a day on the lake fishing, and, better, cooking up the catch for a crowd.

No Comments on "Applied Bracketeering: So, who saw that final four coming?"

Leave a Comment