CBB Studies
The research behind Matchups and Ratings, written up in full: what was tested, what held up, and what didn't.
A possessions-times-efficiency model, fit on three seasons and tested blind on a fourth against the market close.
Conference home edges are smaller than non-conference ones and crowds matter — but the fix that captured both didn't improve the projection out of sample, so the simpler model stayed.
Shot volume from three is a real skill and shot accuracy from three barely is; blocks and steals are genuinely defense-driven; the slower team leans the game toward its own pace.
Returning production predicts next season's rating change, clearly — but the effect isn't pinned down precisely enough yet to safely correct a preseason number.
CBB Studies collects the research write-ups behind the Matchups and Ratings pages: how the tempo-free projection model was fit and tested against a season it never saw, what actually controls each of the four factors when two teams meet, whether home-court advantage is really about the crowd, and whether roster turnover in the offseason says anything about next season's rating. Every article states its data, its method and what the result does not claim — including the tests that came up short.
What these numbers mean: model calibration.