Avg Margin Baseline
Predict from simple average scoring margin in the training window.
| Status | implemented | Short Name | AvgMarg |
|---|---|---|---|
| Rating Style | probability | Where Shown | model performance |
Model Card
Training-only average margin baseline. Another cheap sanity check for whether a richer team model adds signal beyond raw per-game margin.
Validation here is out of sample: each week of a season is predicted by models refit on the season's earlier games, and every model is scored on the same games. Weekly fits use shared v2 matchup states, including joint-fit ensembles, symmetric Dynamic BT forecasts and the Kitchen Sink trained only on earlier games. Priors are last season's ratings and the stored preseason priors; when those priors were written is not recorded. Current scope: Ranked Games. Only games where at least one team had an AP Poll rank on game day. All models are re-evaluated on that same subset, and AP Poll joins the comparison set here.
Selected League / Season
Change league with the League selector at the top of the page.
| Full-season AUC | 0.825 (#15 of 17) ยท 878 games |
|---|---|
| Full-season log loss | 0.526 |
| Final-week AUC | 0.583 (#17) ยท 7 games |
Held-out game predictions
Weekly out-of-sample games in the comparable full-season cohort. Each point is a game: the predicted probability that team 1 wins against team 1's actual margin, so a surprise sits high-left or low-right. Plot: 400 of 878 games (up to 25 largest errors and an even sample of the rest); calibration uses all 878 games. Select a point to inspect its game and prediction; a verified game link appears when available. Browse every game by surprise.
Calibration by predicted probability
| Prediction range | Games | Mean predicted win probability | Observed win rate |
|---|---|---|---|
| 0โ10% | 32 | 5.2% | 0.0% |
| 10โ20% | 49 | 15.8% | 16.3% |
| 20โ30% | 106 | 25.4% | 15.1% |
| 30โ40% | 130 | 35.4% | 34.6% |
| 40โ50% | 126 | 44.9% | 42.1% |
| 50โ60% | 120 | 54.8% | 61.7% |
| 60โ70% | 120 | 64.7% | 73.3% |
| 70โ80% | 107 | 74.5% | 86.9% |
| 80โ90% | 67 | 84.5% | 95.5% |
| 90โ100% | 21 | 93.6% | 95.2% |
Largest errors
Peers in NCAAM 2025-2026
Every model on the same games, ordered by full-season AUC.
| Model | AUC | Rank | LogLoss | AUC final wk | Final wk rank |
|---|---|---|---|---|---|
| Points Off/Def | 0.867 | #1 | 0.478 | 1.000 | #1 |
| Points Off/Def Recency | 0.867 | #2 | 0.479 | 1.000 | #1 |
| Margin | 0.866 | #3 | 0.474 | 1.000 | #1 |
| Margin Recency | 0.865 | #4 | 0.474 | 1.000 | #1 |
| Adjusted Context Blend | 0.864 | #5 | 0.461 | 1.000 | #1 |
| Core Ensemble | 0.856 | #6 | 0.477 | 1.000 | #1 |
| Recency Ensemble | 0.856 | #7 | 0.477 | 1.000 | #1 |
| Adjusted Efficiency | 0.854 | #8 | 0.472 | 1.000 | #1 |
| Log Adjusted | 0.854 | #9 | 0.473 | 1.000 | #1 |
| Elo | 0.852 | #10 | 0.475 | 0.833 | #13 |
| Dynamic Bradley-Terry | 0.844 | #11 | 0.501 | 0.833 | #13 |
| Bradley-Terry Recency | 0.843 | #12 | 0.493 | 0.917 | #11 |
| Bradley-Terry | 0.843 | #13 | 0.493 | 0.917 | #11 |
| Pythagorean | 0.838 | #14 | 0.495 | 0.667 | #15 |
| Avg Margin Baseline | 0.825 | #15 | 0.526 | 0.583 | #17 |
| AP Poll | 0.811 | #16 | - | 0.958 | #10 |
| Home Team Baseline | 0.657 | #17 | 0.650 | 0.667 | #15 |
League ร Season Summary
Avg Margin Baseline in every evaluated league-season; rank among the models on the same games. The final week is the season's last seven days of games: in the pro leagues that is the last playoff series, a handful of games, so its AUC can read 0.000 or 1.000. Games '-' means the model was not evaluated over that full season.
| League | Season | AUC | Rank | LogLoss | Games | AUC final wk |
|---|---|---|---|---|---|---|
| NCAAW | 2025-2026 | 0.888 | #15/17 | 0.421 | 709 | 0.500 (5 g) |
| NCAAW | 2024-2025 | 0.918 | #15/17 | 0.429 | 65 | - |
| NCAAW | 2023-2024 | 0.968 | #14/17 | 0.290 | 64 | - |
| NCAAW | 2022-2023 | 0.936 | #14/17 | 0.344 | 89 | 0.500 (3 g) |
| NCAAW | 2021-2022 | 0.897 | #15/17 | 0.432 | 76 | - |
| NCAAW | 2020-2021 | 0.843 | #14/17 | 0.503 | 63 | 1.000 (2 g) |
| NCAAW | 2019-2020 | 0.973 | #14/17 | 0.234 | 45 | 1.000 (6 g) |
| NCAAW | 2018-2019 | 0.857 | #15/17 | 0.469 | 34 | - |
| NCAAW | 2017-2018 | 0.935 | #15/17 | 0.398 | 64 | - |
| NCAAW | 2016-2017 | 0.899 | #15/17 | 0.417 | 56 | 1.000 (3 g) |
| NCAAW | 2015-2016 | 0.819 | #16/17 | 0.534 | 53 | - |
| NCAAW | 2014-2015 | 0.970 | #13/17 | 0.281 | 40 | - |
| NCAAW | 2013-2014 | 0.855 | #13/17 | 0.534 | 44 | - |
| NCAAW | 2012-2013 | 0.835 | #15/17 | 0.519 | 77 | - |
| NCAAW | 2011-2012 | 0.876 | #11/17 | 0.436 | 70 | - |
| NCAAW | 2010-2011 | 0.791 | #5/17 | 0.507 | 38 | - |
| NCAAW | 2009-2010 | 0.893 | #14/17 | 0.415 | 46 | - |
| NCAAW | 2008-2009 | 0.951 | #10/17 | 0.358 | 68 | - |
| NCAAW | 2007-2008 | 0.815 | #11/17 | 0.506 | 30 | - |
| NCAAW | 2006-2007 | 0.836 | #15/17 | 0.447 | 27 | - |
| NCAAW | 2005-2006 | 0.967 | #6/17 | 0.272 | 29 | - |
| NCAAW | 2004-2005 | 0.901 | #10/17 | 0.370 | 27 | - |
| NCAAM | 2025-2026 | 0.825 | #15/17 | 0.526 | 878 | 0.583 (7 g) |
| NCAAM | 2024-2025 | 0.801 | #15/17 | 0.556 | 972 | 0.500 (5 g) |
| NCAAM | 2023-2024 | 0.776 | #10/12 | 0.570 | 1015 | 0.500 (5 g) |
| NCAAM | 2022-2023 | 0.825 | #7/17 | 0.427 | 22 | - |
| NCAAM | 2018-2019 | 0.922 | #13/17 | 0.443 | 62 | - |
| NCAAM | 2017-2018 | 0.000 | #17/17 | 0.889 | 6 | - |
| NCAAM | 2014-2015 | 0.750 | #13/17 | 0.630 | 25 | - |
| NCAAM | 2011-2012 | 0.795 | #15/17 | 0.466 | 23 | - |
| NCAAM | 2010-2011 | 0.745 | #16/17 | 0.582 | 53 | - |
| NCAAM | 2006-2007 | - | - | 0.771 | 2 | - |
| NCAAM | 2005-2006 | 0.733 | #11/17 | 0.597 | 19 | - |
| NCAAM | 2004-2005 | 0.819 | #9/17 | 0.530 | 18 | - |