Skip to content

Latest commit

 

History

History
121 lines (88 loc) · 8.34 KB

bracket-notes.md

File metadata and controls

121 lines (88 loc) · 8.34 KB

Bracketology calculations, reminders and helpers

A collection of links and notes to demystify the NCAA basketball tournament selection committee decisions for at-large teams. These ideas help inform the algorithm for the Nebrasketball NCAA Tournament Odds Machine. Email bvankat [ at ] gmail with additions, corrections and feedback.

Calculations Notes

If you're Team A with metrics X, Y, Z — Will you get an at-large bid?

  • Question: "Who should get a bid?" Answer: Use results-based metrics - KPI and SOR.
  • Question: "What seed will they get? Who wins head-to-head?" Answer: Use predictive metrics.
  • Approx. 500 at-large bids from 2008-2022. Only five at-large teams with NET over 70: All 10 or 11 seeds. 20 at large bids with NET/RPI above 60
  • NET rank in the 60s/70s = Not good enough. Only two at large bids above 60 in 2021 or 2022. MSU had NET 70 in 2021. Miami was a 10-seed in 2022, NET 62. (#)
  • "The committee’s approach can be assessed — at least to some extent — using a model that projects NCAA Tournament seeds by averaging each team’s resume and quality metrics and weighting them evenly." (#)
  • No Power Six conference 24-win team has missed tournament. Only three 23-win teams have missed out: Florida (2009), Mississippi State, Virginia Tech (2010), Arizona (2012)
  • Gotta have two Quad 1 wins - Only four teams have gotten at-large bids with 1 Q1 victory (#)
  • A bunch of NET bests/worsts that haven't been analyzed here

Worst teams to get in

  • Lowest-ranked at-large team in each metric
    • 2019 — NET: 73, SOR: 55, KPI: 62, BPI: 72, KenPom: 78, Sagarin: 68 (#)
    • 2020 — No tournament
    • 2021 — NET: 72, SOR: 66, KPI: 64, BPI: 97, KenPom: 74, Sagarin: 70 (#)
    • 2022 — NET: 77, SOR: 52, KPI: 64, BPI: 74, KenPom: 74, Sagarin: 64 (#)
    • 2023 — NET: 67, SOR: 54, KPI: 66, BPI: 72, KenPom: 77, Sagarin: 84 (#)
  • Since 2018 when NET started: Worst SOR/KPI average to get at-large bid was 57.5 (2022 Rutgers)
  • Lowest NET ranking to get bid: 77 in 2022 (Rutgers), Lowest RPI: 74 in 1999
  • 2022: Last 4 In: Wyoming (NET 50), Rutgers (77), Indiana (38), Notre Dame (53)
  • Worst SOR to get in was 66 (2021 Utah State)

Best Teams Left Out

Compiling some data on teams that didn't make the tournament

  • No team with 6+ Q1 wins and < 15 losses has been left out (#)

By Year

  • 2021 — Kenpom: Highest-ranked teams left out: 33 Duke, 35 Penn State, 38 Memphis, 43 Arizona, 46 Indiana (#)
  • 2022 — First Four Out, according to NCAA: Dayton (NET 58), Oklahoma (NET 40), SMU (NET 44 or 45?), Texas A&M (NET 43)
  • 2023 — First Four Out, according to NCAA: Oklahoma State, Rutgers, North Carolina Clemson

By Metric

  • RPI — 49 - Missouri (2014) highest in 68-team tournament era (#)

Team Sheet basics

NCAA Selection committee has six metrics on the team sheets used to compare teams in the runup to Selection Sunday

NET

  • NCAA's proprietary combination of metrics. Seems to lean predictive rather than resume-based.
  • Started in 2018, replacing the RPI. Rankings (but not precise formula) made public in 2021. (#)
  • Release date: Late November or early December (12/5/22 and 12/4/23), but recreated and available early on other sites
  • LINK

ESPN's BPI

  • Predictive
  • Release date: Early November, a couple days before season's first games
  • LINK

ESPN's SOR

  • Results-based, resume metric
  • Release date: same as BPI above
  • Preaseason SOR for team is 1. So this metric takes a couple weeks to develop some reliability.
  • LINK

KPI

  • Results-based, resume metric
  • Release date: Early December (12/5/22)
  • LINK

Kenpom

  • Predictive
  • Doesn't measure accomplishment. Aims to predict results going forward. (#)
  • Records back to 2002
  • LINK

Bart Torvik

  • Predictive
  • How's it different from Kenpom? T-Rank has a wider spread between the top and bottom teams, partly because Kenpom caps margin of victory. T-Rank excludes garbage time by no longer adjusting the game efficiency scores when the outcome is assured. T-Rank has a slight recency bias and slightly discounts blowouts in extreme mismatches. (Explainer)
  • Added to team sheets for 2024-25 season to replace Sagarin ratings. (Sagarin retired after 2023 season)
  • LINK

Wins Above Bubble

  • Results-based, resume metric
  • "How much better is this team than an average bubble team?"
  • Example: Team A wins a game at the best team in the country. The expected wins (i.e., win probability) in that matchup for an average bubble team on the road might be 20%. So take that actual 1.0 wins, subtract 0.20 expected wins, and get +0.80 Wins Above Bubble. Add up the number for all games to calculate WAB.
  • Added to team sheets for 2024-25 season
  • LINK

Quadrants

"Using the quadrant system, the quality of wins and losses will be organized based on game location and the opponent's NET ranking. The number of Quadrant 1 wins and Quadrant 3/4 losses will be incredibly important when it comes time for NCAA tournament selection and seeding." (#)

  • Quadrant 1: Home 1-30, Neutral 1-50, Away 1-75
  • Quadrant 2: Home 31-75, Neutral 51-100, Away 76-135
  • Quadrant 3: Home 76-160, Neutral 101-200, Away 135-240
  • Quadrant 4: Home 161-353, Neutral 201-353, Away 241-353

"The NET is designed to define the quadrants, not to choose or seed teams. It's not a tiebreaker or anything like that. Teams are not compared by NET or other computer rankings." (#)

"Generally speaking, if that team went 2-1 (in Quad 1 games) compared to a team that went 5-10, that’s going to be looked at more favorably by the committee than just looking at the total number of Quad 1 wins." (#)

Sources