New to these reports? Start here
- Dotted-underlined words have a plain-English definition — hover or tap them. Every term is also on the glossary page.
- "Null" means we found nothing, not that something broke. Most reports here are negative results, on purpose — knowing an idea doesn't work is the point.
- Two questions get asked separately. First, is the effect real? Second, is it already priced into the betting odds? An effect can be completely real and still useless to bet on.
- A "calibration" row is a self-check. It runs the same method on something already known to be true. If that fails, the whole report is unreliable — so it's reported alongside the findings.
- If a confidence interval includes zero, the real effect might be nothing at all, so no claim gets made.
New play-callers and closing totals (2026-11-01)
Primary result: nullnull resultA test that found nothing. "Null" is the starting assumption that there is no real effect; a "null result" means the data gave us no reason to abandon that assumption. It does not mean the data was missing or the test failed to run.. In weeks 1–4, offenses with a new head coach or new play-caller score what
the closing lineclosing lineThe final odds right before a game starts. It reflects everything the betting market knows, which makes it the hardest benchmark to beat. implies. Tests were fixed in analyze.py before results were seen. 2006–2025
regular season, free data only: nflverse closing lines and head coaches, plus Wikipedia offensive
coordinators (build_list.py → data/staff.csv).
The list
- Head coach (primary signal): nflverse schedules, complete. 139 of 640 team-seasons had a week-1 head coach who coached none of the team's games the year before.
- Offensive coordinator: Wikipedia team-season pages, known for 88% of team-seasons.
- Sources: the infobox
off_coach(who started the season), else the end-of-season staff box. - Spot-checked 25 random rows: right except mid-season replacements when only the staff box has an OC (e.g. Rams 2015 shows Rob Boras, who took over mid-year).
- No free source says who called plays. Head coaches listed as their own OC (McVay, Shanahan: 22 team-seasons) count as play-callers. Reid, McCarthy and Payton aren't marked, so their OC changes count as play-caller changes. Both kinds of noise push results toward zero.
Points scored vs implied team total (flagged minus other teams)
| Signal | Weeks 1–4 | 2006–15 | 2016–25 | Weeks 5–8 | Weeks 9–18 |
|---|---|---|---|---|---|
| New head coach | −0.13 ± 0.42 | +0.01 | −0.28 | −1.28 ± 0.46 | −0.17 ± 0.32 |
| New play-caller | −0.01 ± 0.38 | −0.04 | +0.04 | −0.45 ± 0.41 | −0.18 ± 0.28 |
| New OC only (same HC) | +0.30 ± 0.50 | +0.20 | +0.45 | +0.21 ± 0.53 | +0.01 ± 0.37 |
The market prices new play-callers correctly in the first month. The idea that "the market anchors on last season" doesn't hold.
A side pattern, not the tested one: in weeks 5–8, new-head-coach offenses scored 1.3 points under their implied total (z −2.8, −1.16 and −1.38 by half). It was a secondary window, one of about 27 cells. It fades by weeks 9–18, and only team totals could bet it. Those have no free history, so it can only be tested forward, from captured team-total lines. Not pre-registered.
Points allowed by new-head-coach teams: no effect in weeks 1–4 (+0.10 ± 0.48).
Game totals, weeks 1–4 (over % at the close)
| Teams with a new head coach | Games | Over | 2006–15 | 2016–25 |
|---|---|---|---|---|
| 0 | 750 | 51.5% | 53.3% | 49.7% |
| 1 | 444 | 48.4% | 50.9% | 46.0% |
| 2 | 49 | 55.1% | 50.0% | 59.3% |
Nothing beats 52.4%.
python3 backtests/nfl_playcaller_totals/build_list.py # ~15 Wikipedia API calls
python3 backtests/nfl_playcaller_totals/analyze.py