If you take a quick scan of the percentiles below, it looks like the stats models are performing pretty well this year. But the Elite Eight results will start to create separation among the 15 models. The biggest bracket killer was the Indiana loss. Five of the 15 models liked the Hoosiers to cut down the nets. There was one Gonzaga and one Kansas bracket as well. So that leaves eight brackets with their champion still intact (Louisville, Duke or Florida). Here are the results:
- M11 Contrarian – 98.5 (4 F4, Duke)
- M5 Pulse Check – 94.9% (2 F4, Louisville)
- M12 Nate Silver – 90.2% (2 F4, Louisville)
- M13 Coach+Six – 90.2% (2 F4, Indiana)
- M1 Kenpom – 86.7% (2 F4, Florida)
- M8 Seed Match – 86.7% (2 F4, Indiana)
- M2 F4/Champ – 82.8% (2 F4, Indiana)
- M10 ESPN BPI – 82.8% (2 F4, Louisville)
- M14 Costanza – 82.8% (2 F4, Duke)
- M6 Factor PASE – 73.6% (1 F4, Kansas)
- M7 Baseline – 73.6% (1 F4, Gonzaga)
- M15 Keeper – 73.6% (3 F4, Indiana)
- M4 Efficiency+Coach – 67.0% (2 F4, Florida)
- M3 Outcome – 49.6% (3 F4, Louisville)
- M9 Upset – 29.1% (1 F4, Indiana)
Surprisingly, the Contrarian model leads the pack–and it’s the only model that can still get all four Final Four teams right. If you’ll recall, I made a conscious effort here to avoid one seeds, so I had Duke, OSU, Florida and Syracuse advancing. Those are probably the odds-on favorites for these Elite Eight match-ups. Two other models, Keeper and Outcome Matching, have three possible Final Four teams…but Outcome Matching holds an edge because it retains its champion (Louisville).
In the battle of famous metrics-based models and no-brainer seed picking, Nate Silver holds a slight edge over Ken Pomeroy and ESPN BPI. Bragging rights all come down to the champion. Nate and ESPN like Louisville; Ken likes Florida. The Baseline higher-seed/higher-margin strategy favored Gonzaga. Ouch! So the “YourMom” approach fell flat this year.
Here’s another thing to consider: if Ohio State, Wichita State, Michigan, Marquette or Syracuse win the championship, none of the models will likely remain in the 90+ percentile level, with the exception of the Contrarian.