Team-, venue-, and night-specific variants do not improve opening-gap prediction beyond the global baseline.
The Ledger · Travel and venue
Team, Venue, and Night Home Field
Do team-specific, venue-specific, or night-game effects improve on the global home-field advantage already in the OnVarsity Line?
No separate adjustmentGlobal home-field advantage outperformed every contextual variant.Updated August 30, 2026
Decision
Keep global HFA in the official model. Do not publish a numeric team-, venue-, or night-specific adjustment without substantially more evidence.
- Historical seasons
- 2017–19, 2021–25
- Primary sample
- 1,685 primary games
Start here
What the evidence says
The granular home-field variants do not explain additional sampled movement.
Contextual home-field estimates also fail to improve final-margin accuracy.
Atmosphere can matter in one game without supporting a stable team- or venue-specific number across seasons.
2026 prospective implementation
Team, venue, and night-game states can be identified for 2026 from existing schedule and venue data; no incremental point adjustment is approved.
Current inputs
- Home team
- Venue
- Kickoff time
- Global HFA baseline
- Additional source need
- No new source is required for basic detection.
- Refresh cadence
- Recompute after venue or kickoff changes.
- Coverage and limitations
- Any matchup appearance should explain the null incremental result rather than imply a venue-specific edge.
Study results
| Test | Games | Result | Validation |
|---|---|---|---|
| Opening-market RMSE | 1,682 | +0.072 team-HFA RMSE | Contextual model worse |
| Final-recorded RMSE | 1,685 | +0.062 team-HFA RMSE | Contextual model worse |
| Market-movement RMSE | 1,682 | +0.001 team-HFA RMSE | No useful movement gain |
| Actual-margin RMSE | 1,685 | 15.391 global vs 15.400 team HFA | Contextual model worse |
| Night sensitivity | 644 night games | +0.065 to +0.068 market RMSE | Worse across all cutoffs |
| Older external check | 25 openers | 3.106 zero vs 3.134 team RMSE | Frozen model worse |
What we tested
Every team, venue, and night variant worsened pooled held-out prediction for opening-market, final-recorded-market, and result targets. Partial pooling pulled the most tempting venue-night estimates strongly toward zero.
- Home team and venue identity
- Venue-local kickoff time with night thresholds from 17:00 to 20:00
- Ridge-partially-pooled team and venue effects
- Market and final-margin residuals after global HFA
Leakage and validation safeguards
- Complete-season outer holdouts and inner tuning holdouts
- Multiple rating-baseline configurations
- Older-season frozen external check
- Sparse team-night estimates shrunk rather than reported as raw averages
What we learned
- Raw venue and night splits can look compelling while failing out of sample.
- The existing global HFA is the better holistic foundation.
- A longer sample or richer attendance and travel mechanisms could reopen the question.
What this does not prove
- Rare venue-night combinations remain difficult to estimate precisely.
- A null incremental adjustment does not mean atmosphere never matters in a particular game.
How uses this finding
No contextual HFA points are added. A salient matchup may disclose that team, venue, and night effects were tested and did not improve beyond global HFA.
Historical data record
Revalidated August 30 under market-line contract 2026-08-30.1. The primary panel includes 1,685 final-recorded cases, 1,682 provider-paired opener cases, 133 home teams, 134 venues, and 644 night games. Older final/result checks remain retained separately.
Public game-level table: in preparation