2016-2017 NCAAWD2 Model Performance
Season
Every ranked model predicted the same 3208 games (2016-11-26 to 2017-03-24, 17 weekly windows) out of sample: each week's games were predicted by models fit only on the season's earlier games. 28 of 3236 games are left out because a model had no rating for a team yet.
Model Performance Leaderboard
Ranked by full-season AUC. The final week is the season's last 7 played days (7 games, 2017-03-18 to 2017-03-24), predicted from every earlier game: a small sample that shows the latest form.
| # | Model | AUC | Acc | Brier | LogLoss | AUC final wk | Acc final wk |
|---|---|---|---|---|---|---|---|
| 1 | Points Off/Def Recency Points Off/Def Recency Off/def points regression with exponential recency weights. More โ | 0.831 | 74.4% | 0.171 | 0.513 | 1.000 | 71.4% |
| 2 | Points Off/Def Points Off/Def Raw points regression with separate offensive and defensive team parameters. More โ | 0.831 | 74.3% | 0.171 | 0.513 | 1.000 | 71.4% |
| 3 | Margin Recency Margin Recency Margin regression with exponential recency weights on newer games. More โ | 0.830 | 74.3% | 0.171 | 0.514 | 1.000 | 71.4% |
| 4 | Margin Margin Linear team-strength model fit on point differential instead of binary wins. More โ | 0.829 | 73.9% | 0.171 | 0.514 | 1.000 | 71.4% |
| 5 | Core Ensemble Core Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and points off/def. More โ | 0.827 | 74.1% | 0.171 | 0.511 | 1.000 | 71.4% |
| 6 | Recency Ensemble Recency Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and recency points off/def. More โ | 0.827 | 74.1% | 0.171 | 0.511 | 1.000 | 71.4% |
| 7 | Adjusted Efficiency Adjusted Efficiency Opponent-adjusted efficiency model with separate offensive and defensive components. More โ | 0.821 | 73.5% | 0.176 | 0.523 | 1.000 | 71.4% |
| 8 | Log Adjusted Log Adjusted Log-scale adjusted efficiency model that downweights blowout leverage. More โ | 0.821 | 73.5% | 0.176 | 0.524 | 1.000 | 71.4% |
| 9 | Adjusted Context Blend Adjusted Context Blend Chronologically trained component blend with rest and signed venue context. More โ | 0.818 | 73.3% | 0.183 | 0.554 | 1.000 | 71.4% |
| 10 | Avg Margin Baseline Avg Margin Baseline Predict from simple average scoring margin in the training window. More โ | 0.812 | 72.8% | 0.178 | 0.529 | 1.000 | 57.1% |
| 11 | Dynamic Bradley-Terry Dynamic Bradley-Terry Time-evolving paired-comparison model with latent team strength drift. More โ | 0.805 | 72.0% | 0.181 | 0.538 | 0.800 | 71.4% |
| 12 | Bradley-Terry Bradley-Terry Static logistic paired-comparison model with one team strength parameter. More โ | 0.802 | 71.6% | 0.182 | 0.540 | 1.000 | 71.4% |
| 13 | Bradley-Terry Recency Bradley-Terry Recency Static Bradley-Terry with exponential recency weights on newer games. More โ | 0.802 | 71.5% | 0.182 | 0.540 | 0.800 | 71.4% |
| 14 | Pythagorean Pythagorean Pythagorean win expectation from raw points scored and allowed. More โ | 0.800 | 72.7% | 0.187 | 0.555 | 1.000 | 71.4% |
| 15 | Elo Elo Streaming paired-comparison rating with recency baked into sequential updates. More โ | 0.770 | 69.0% | 0.196 | 0.573 | 0.800 | 71.4% |
| 16 | Home Team Baseline Home Team Baseline Always favor the home team with a fixed prior. More โ | 0.569 | 56.9% | 0.246 | 0.686 | 0.800 | 71.4% |
Methodology
Out-of-sample windows
- The season is split into 7-day windows ending on its last played day
- Each window is predicted by models refit on all earlier games of the season
- The first window starts after 21 days of games
- Every model is scored on the same games; a model that cannot predict most of them is listed as not compared
Elo / Bradley-Terry
- Elo: the published league Elo (preseason priors and carryover where configured), frozen at the window start
- Bradley-Terry: static logistic fit on the earlier games
- Recency variants weight games with a 120-day half-life in NCAA leagues (keeping the non-conference games that link conferences) and a 30-day half-life in pro leagues
Pythagorean and margin models
- Raw, adjusted and log-adjusted Pythagorean from scores; efficiency from box-score possessions
- Margin: ridge regression on point margin with a learned home advantage
- Ensembles average the log-odds of five families
Baselines
- Home Team: the home team wins 60% (50% at neutral sites)
- Avg Margin: the higher average margin so far wins
- Models should beat these to add value