College Basketball · Analytics
Home field and home court, tested the same way across both sports - does the advantage actually hold up once you control for team strength?
Five tests, run identically for both sports: a naive baseline, a rating-adjusted regression that isolates the real effect from team-strength differences, a neutral-site sanity check on that regression, a cluster-bootstrap and permutation test for whether the adjusted effect is statistically real, and a collinearity check on the regression itself. Bars marked "significant" have a 95% confidence interval that excludes zero.
Raw score differential - average(home score − away score) across non-neutral games, no adjustment for team strength.
OLS regression of point differential on the home/away indicator and the gap between the two teams' ratings entering that game. The home/away coefficient is what's left over once team quality is already priced in - the real, adjusted home advantage.
CFB R² = 0.273 · CBB R² = 0.433
Same regression restricted to neutral-site games only, where the intercept should land close to zero if the model is behaving correctly - there's no true home team to give credit to.
CFB
17 neutral-site games · intercept +3.27 (95% CI [-5.45, 14.88])
CBB
589 neutral-site games · intercept -0.17 (95% CI [-1.07, 0.69])
A cluster bootstrap (resampling by team, 95% CI) and a permutation test on the Test 2 coefficient - whether the adjusted advantage is actually distinguishable from zero, not just a single lucky regression fit.
CFB 95% CI [-10.57, 7.94] · p = 0.9830
CBB 95% CI [2.96, 4.85] · p = 0.0000
Correlation and variance inflation factor (VIF) between the home/away indicator and the rating-difference term - rules out Test 4's result being a modeling artifact rather than a genuine finding. A VIF near 1.0 means no meaningful inflation.
CFB
corr = 0.061 · VIF = 1.004
CBB
corr = -0.024 · VIF = 1.001
One season of correlational data for each sport - not a controlled experiment. See the full write-up in Articles for the complete methodology.
