Home court advantage in college basketball
By George Boyle · Updated Nov 9, 2026 · 6 min read
Our projections use one home-court number for every true home game. Looking at where that number goes wrong turned up a real pattern, and a version built to fix it that still didn't make the cut.
- Bias, Nov–Dec (current model)
- −1.57 pts
- home teams underrated early
- Bias, Jan 1 on (current model)
- +1.33 pts
- home teams overrated later
- Conference home edge (fitted)
- +2.2 pts/100 poss
- vs. +4.7 non-conference
- Crowd effect (fitted)
- +1.3 pts/100 poss
- per standard deviation of attendance
A bias that flips with the calendar
The home-court number in our projections, worth 2.877 points per 100 possessions to the team that is truly at home, is a season-long average.
Splitting the unseen 2025-26 season by month shows it fits neither half well. In November and December the model underrates home teams by about a point and a half. From January on it overrates them by about the same amount.
The same pattern shows up in the three seasons the model was built on. That made it worth investigating, not one season of noise.
The idea: it's about the crowd
The working theory: home-court advantage is less about being at home and more about the building being full.
November and December carry the non-conference schedule, including "buy games" where a high-major team hosts a heavy underdog in a half-empty arena. Conference play, later in the season, reliably fills seats.
Splitting the building seasons by game type, not by month, shows where the miss sits. In November and December non-conference games the home side is underrated by nearly two points. In conference home games the lean is about one point the other way, and it is the same in both halves of the season.
| Window | True home, conference | True home, non-conference | Neutral site |
|---|---|---|---|
| Nov–Dec | +1.00 pts (n 646) | −1.93 pts (n 4,773) | −0.87 pts (n 1,178) |
| Jan 1 on | +0.98 pts (n 9,357) | −0.22 pts (n 181) | −0.90 pts (n 1,063) |
Six ways to model home court
Six versions of the home-court term were compared. They run from the single number in use today, to separate numbers for conference games, non-conference games and neutral sites, to versions that add crowd size.
Crowd size is the standardized log of a team's average attendance in its home games so far that season. A team with no home games yet falls back to last season, then to the league median.
Every version was scored on the three building seasons, before 2025-26 was looked at, using the same accuracy score the current model was chosen by: the average of spread error and total error.
| Version | Accuracy score (lower = better) | Nov–Dec bias | Jan+ bias |
|---|---|---|---|
| Current model (one number) | 22.2533 | −1.45 pts | +0.77 pts |
| By game type (conference / non-conference / neutral) | 22.2740 | +0.07 pts | +0.05 pts |
| By calendar month | 22.2721 | −0.11 pts | −0.07 pts |
| Crowd size only | 22.2458 | −1.45 pts | +0.72 pts |
| Game type + crowd size | 22.2515 | +0.07 pts | +0.02 pts |
| Game type + a non-conference "buy game" stretch | 22.2711 | +0.65 pts | +0.06 pts |
No version that removed the bias improved the accuracy score by the margin needed to replace the current model, a bar set before any version was run. A bias of about one point barely moves an error measure on games that miss by eleven points on average. Accuracy and bias are different tests, and a rule built for one doesn't catch problems in the other.
A second rule, and one look at the unseen season
A second rule was then written, aimed at the bias itself: stay within three-quarters of a point in both halves of the season, then take the most accurate version that does.
That picked the game-type-plus-crowd version: conference home worth 2.2 points per 100 possessions, non-conference 4.7, a designated home team at a neutral site 0.9, plus a crowd term worth about 1.3 points per 100 possessions per standard deviation of attendance.
That version, and only that one, then got a single look at the 2025-26 season.
| Model | Spread error | Total error | Nov–Dec bias | Jan+ bias |
|---|---|---|---|---|
| Current model | 9.322 pts | 13.757 pts | −1.57 pts | +1.33 pts |
| Game type + crowd | 9.362 pts | 13.757 pts (no better) | −0.15 pts | +0.47 pts |
Verdict: the current model stayed. The game-type-plus-crowd version fixed the bias, with both halves of the season inside the target. But its spread error on the unseen season was slightly worse and its total error was no better, and the rule required the replacement to be no worse on both.
What held up anyway
The crowd effect is real, and it did not go away when the version that used it lost. Home teams gain about 1.3 points per 100 possessions for every standard deviation of pregame attendance, a stable number in every version that included it.
Conference home games really are worth less than non-conference ones in the fitted numbers: roughly half as much.
Both findings describe the sport. Neither was strong enough, on its own, to earn a place in the live projection.
Method
how this was measured
Every version kept all of the current model's other constants and changed only how the home-court term is computed.
Coefficients were estimated the way the current home-court number was: an unregularized term in each building season's full-season ridge fit, averaged over 2022-23 through 2024-25.
The crowd term uses only games played before the one being projected, never that game's own attendance.
The selection rules and their limits were written down before any version was run. The second rule was written after seeing the first rule's results on the building seasons, but before the unseen season was looked at. That exception was approved in advance and is disclosed here.
What this does not claim
This does not say the current model has no home-court bias. The split above shows exactly where it does.
It is a record of one disciplined attempt to fix it that came up short of its own bar, and of why a visibly smaller bias is not, by itself, a reason to change a live projection.
If the current model is still live after 1,000 graded 2026-27 games, its bias by part of season gets reported again, whatever it says.
See it live
Written by George Boyle, who builds The Sport Stack. Corrections and questions: hello@thesportstack.io. Who runs this.