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.
Competition-Adjusted College Production vs NFL Playing Time
Study, 2026-10-29, run as pre-registered in PLAN_COMPETITION.md
(a5ac964). Drafted QBs, RBs, WRs and TEs, classes 2014-2023.
Verdict: no evidence on either question. - Adjusting final-season production for opponent defensive strength doesn't predict NFL playing time better than raw production (better at 1 of 4 positions; pooled change in R squared -0.001, 90% interval -0.003 to +0.002). - Neither raw nor adjusted production adds to draft position (pooled +0.007, interval -0.007 to +0.022). Draft position carries nearly all of it.
Why the adjustment does nothing
Opponents matter game to game: a running back gains 0.9 more scrimmage yards per game for each SP+ point worse the defence is; a receiver 0.2 receiving yards; a QB 0.15 adjusted yards per attempt. But drafted players' schedules are alike: the spread (SD) of average opponent rating is only 3-4 points. So the adjustment moves a running back by about 3 yards a game when players differ by about 40. Raw and adjusted values correlate 0.96 (QB) to 1.00 (WR).
Results (out-of-sample R squared, leave-one-draft-class-out)
Outcome: log(1 + NFL offensive snaps in first three seasons).
| Model | QB (83) | RB (157) | WR (241) | TE (97) |
|---|---|---|---|---|
| Draft pick only | 0.401 | 0.281 | 0.225 | 0.195 |
| Pick + raw | 0.385 | 0.316 | 0.218 | 0.212 |
| Pick + adjusted | 0.382 | 0.314 | 0.218 | 0.214 |
| Pick + raw + Power 5 | 0.369 | 0.312 | 0.201 | 0.222 |
| Raw alone | 0.034 | 0.143 | 0.002 | -0.023 |
| Adjusted alone | 0.036 | 0.146 | 0.003 | -0.023 |
- Running backs are the one position where production seems to add something beyond the pick (+0.035). That is a per-position look the plan did not make a claim on, so it's a lead, not a finding.
- A crude Power 5 flag helped only for tight ends and hurt elsewhere.
Deviations from the plan
- The box scores carry no positions, so each slope was estimated on every player-game of that kind (all passers, all ball carriers, all receivers). WRs and TEs therefore share one receiving slope.
- 788 drafted players: 588 linked to a roster, 578 with box scores in their final season (the rest mostly FCS or unmatched names).
Caveats
- Snaps partly measure draft capital: teams play high picks. The pick-only model is the bar for that reason, but it also makes it hard for anything else to show up.
- Drafted players only. Scouts have already filtered on competition, which narrows schedule differences; undrafted and small-school players, where competition level differs most, aren't in the sample (FCS players have no box scores here).
- Final college season only, per-game volume measures. Shares of team production and efficiency per route aren't tested.
Files
fetch_box.py (needs CFBD_API_KEY; ~180 calls), competition.py,
competition_results.json, data/box.parquet, data/sp.parquet,
data/competition_players.parquet.