Vikings analysis
Is Kevin O'Connell a QB Whisperer?
Part of an ongoing analysis of the Vikings offense (project hub).
Data through week 10 of 2026.
Bottom line.
- Against projections built from each QB’s own track record, O’Connell’s quarterbacks have played +0.02 EPA per dropback better than expected. That’s 14th of 28 head coaches, and the interval includes zero.
- Two things do stand out. His QBs are more accurate than projected (4th of 28), and they take more sacks than projected (6th-most). The two roughly cancel.
- In a model that separates the QB, receivers, opponent and head coach, the coach explains 0.2% of the variation in teams’ passing efficiency. O’Connell’s own term is about zero.
- He changes how his QBs play, with more snaps under center and more play action, by about as much as any new head coach does. The change in play mix is worth about nothing either way.
Sam Darnold’s 2024 is the main exhibit for O’Connell as a QB whisperer. Darnold arrived with a career that projected to −0.03 adjusted EPA per dropback and played at +0.04 over 621 dropbacks. With Kirk Cousins back in Minnesota, the reputation is getting tested again.
One great season doesn’t tell you much about a coach. The fair test is every QB he’s had, each compared with what that QB was expected to do, and the same test run for every other head coach in the league.
1. Every O’Connell QB against his own projection
For every QB season since 2018 with 150+ dropbacks and an NFL track record (320 of them), I projected opponent-adjusted EPA per dropback from the QB’s previous three seasons and his age. The projection is fit league-wide and never sees the season it’s predicting. A QB coming off a bad year is expected to bounce back partway, so beating the projection means more than bouncing back.
O'Connell's QBs: projected vs actual
Opponent-adjusted EPA per dropback · projection from the QB's prior 3 seasons and age
* Partial season. Any single season's gap carries noise of roughly ±0.1. McCarthy (2025) and Brosmer have no prior NFL seasons to project from.
Data table
| Row | Projected | Actual |
|---|---|---|
| Cousins 2022 | +0.04 | +0.01 |
| Cousins 2023 | +0.02 | +0.10 |
| Mullens 2023 | −0.02 | +0.06 |
| Dobbs 2023 | −0.07 | −0.14 |
| Darnold 2024 | −0.03 | +0.04 |
| Wentz 2025 | −0.08 | +0.03 |
| Murray 2026* | +0.03 | −0.08 |
4 of 7 beat their projection. The biggest gaps, Wentz and Mullens, came on fewer than 200 dropbacks each. Darnold’s +0.07 is the most convincing because it lasted a full season. Murray this year is the biggest miss, at −0.10 so far.
2. O’Connell among 28 head coaches
How much each coach's QBs beat their projections
Head coaches with 5+ projected QB seasons, 2018–2026 · each dot is one coach, weighted by dropbacks
O'Connell, 95% intervals: EPA −0.04 to +0.08; completion % −0.4 to +4.1; sack rate −0.2 to +1.2. Each coach's intervals are wide, because each is built from a handful of QB seasons.
Data table
| Coach | EPA per dropback | Completion % over expected | Sack rate |
|---|---|---|---|
| Matt LaFleur | +0.08 (1st) | +1.9 (5th) | −0.9 (25th) |
| Sean McDermott | +0.07 (2nd) | +1.3 (7th) | −0.9 (24th) |
| Kyle Shanahan | +0.07 (3rd) | +1.0 (9th) | −0.5 (19th) |
| Mike Vrabel | +0.07 (4th) | +2.3 (1st) | +1.3 (1st) |
| Dan Campbell | +0.06 (5th) | +0.3 (14th) | +0.0 (13th) |
| Andy Reid | +0.06 (6th) | +0.4 (13th) | −0.7 (22nd) |
| Sean Payton | +0.05 (7th) | +0.4 (12th) | −0.7 (23rd) |
| John Harbaugh | +0.04 (8th) | +0.3 (15th) | +0.2 (10th) |
| Mike McDaniel | +0.04 (9th) | +1.3 (8th) | −0.1 (16th) |
| Zac Taylor | +0.04 (10th) | +1.3 (6th) | −0.6 (20th) |
| Todd Bowles | +0.03 (11th) | +0.2 (16th) | −0.7 (21st) |
| Sean McVay | +0.03 (12th) | −0.4 (20th) | −1.1 (27th) |
| Nick Sirianni | +0.02 (13th) | +1.9 (3rd) | +0.2 (11th) |
| Kevin O'Connell | +0.02 (14th) | +1.9 (4th) | +0.5 (6th) |
| Pete Carroll | +0.02 (15th) | +2.2 (2nd) | +0.7 (5th) |
| Mike McCarthy | −0.00 (16th) | −0.9 (23rd) | −0.1 (17th) |
| Frank Reich | −0.01 (17th) | −1.3 (25th) | −1.6 (28th) |
| Jonathan Gannon | −0.01 (18th) | −0.5 (21st) | −0.0 (15th) |
| Mike Tomlin | −0.02 (19th) | −0.7 (22nd) | −0.0 (14th) |
| Doug Pederson | −0.03 (20th) | +0.0 (17th) | +0.3 (8th) |
| Dan Quinn | −0.04 (21st) | +0.5 (11th) | +0.4 (7th) |
| Shane Steichen | −0.04 (22nd) | −3.6 (28th) | −1.1 (26th) |
| Kevin Stefanski | −0.05 (23rd) | −1.2 (24th) | +0.8 (4th) |
| Matt Nagy | −0.05 (24th) | −0.2 (19th) | −0.3 (18th) |
| Brian Daboll | −0.06 (25th) | +0.8 (10th) | +1.2 (2nd) |
| Ron Rivera | −0.08 (26th) | −0.1 (18th) | +1.1 (3rd) |
| Bill Belichick | −0.11 (27th) | −2.7 (26th) | +0.2 (9th) |
| Robert Saleh | −0.11 (28th) | −3.2 (27th) | +0.1 (12th) |
On EPA, O’Connell sits in the middle. On accuracy, he’s near the top: his QBs complete passes at a higher rate than their own history predicts, though that interval also just reaches below zero. The sack column is the cost, and that’s the subject of the next post.
The top of the EPA list is a warning about this kind of ranking. Two of the top five coached one QB the whole time: Sean McDermott (Josh Allen) and Dan Campbell (Jared Goff). A QB who keeps improving makes his coach look like a developer, and this measure can’t tell the two apart. The coach who looks most like a QB whisperer here is Kyle Shanahan: 5 different QBs over 11 seasons, +0.07 above projection (95% CI +0.01 to +0.14).
3. Did it travel?
If O’Connell were the reason his QBs played well, their gains should fade after they leave. Two of them have played enough elsewhere to check.
After Minnesota: actual minus projected
Opponent-adjusted EPA per dropback · projection includes their Minnesota seasons
* Partial season.
Data table
| Row | Value |
|---|---|
| Cousins 2024 (ATL · 481 dropbacks) | −0.02 |
| Cousins 2025 (ATL · 283 dropbacks) | −0.08 |
| Cousins 2026* (LV · 333 dropbacks) | +0.11 |
| Darnold 2025 (SEA · 516 dropbacks) | +0.07 |
| Darnold 2026* (SEA · 237 dropbacks) | −0.13 |
Darnold kept it for a full season in Seattle: +0.07 above a projection that already counted his Minnesota year. That points to Darnold having improved, with the gains going with him. His 2026 has dropped off (−0.13 so far). Cousins was below projection both years in Atlanta and is +0.11 above it in Las Vegas this season, without O’Connell.
4. Separating the coach from everything else
The comparison above credits the coach with everything around the QB: the receivers, the line, the schedule. So I also fit a league-wide model on every dropback since 2015 with separate terms for the QB, the head coach, the receivers, the opponent and game situation.
What explains the variation in teams' passing efficiency
Share of variance in team-season EPA per dropback, 2015–2026
Fit through week 10 of 2026. "Head coach" covers the whole regime: play caller, scheme, line and protection plan.
Data table
| Row | Value |
|---|---|
| QB | 30.0% |
| Receivers | 24.0% |
| Situation | 5.0% |
| Opponents | 3.0% |
| Head coach | 0.2% |
| Unexplained | 38.0% |
The QB and the receivers explain most of what can be explained. Once they’re in the model, a head-coach term barely improves its predictions. O’Connell’s is −0.004 EPA per dropback (95% CI −0.019 to +0.012), 33rd of 42.
The model also has an “unexplained” term for each season: what’s left after the QB, receivers and opponents. Darnold’s 2024 left +0.081 unexplained, the largest positive year of O’Connell’s tenure. That could be coaching, a career year, or luck. The model can’t tell which, and the same coach left −0.080 unexplained the next year. Across his 5 seasons it averages −0.014.
The one place O’Connell’s regime does register is sacks: +0.47 percentage points of sack rate (95% CI +0.20 to +0.74), the 2nd-most of 42 head coaches. Unlike his efficiency effect, its interval excludes zero.
5. What he does change
O’Connell’s QBs play differently in Minnesota than they did elsewhere. They line up under center more and run more play action.
Under center: elsewhere → Minnesota
Share of dropbacks, 2022–26 · the QB's own games under other coaches vs under O'Connell
Data table
| Row | Elsewhere | Under O'Connell |
|---|---|---|
| Cousins | 18% | 30% |
| Darnold | 31% | 29% |
| Wentz | 10% | 20% |
| Dobbs | 15% | 26% |
| Murray | 11% | 24% |
Play action: elsewhere → Minnesota
Share of dropbacks, 2022–26 (FTN charting)
Data table
| Row | Elsewhere | Under O'Connell |
|---|---|---|
| Cousins | 20% | 28% |
| Darnold | 25% | 27% |
| Wentz | 20% | 24% |
| Dobbs | 19% | 25% |
| Murray | 21% | 35% |
Across eight style markers, his QBs moved 0.65 of the way from their own style toward the style of his other QBs (95% CI 0.44 to 0.90). The league median for a QB who changes head coaches is 0.66, and O’Connell ranks 6th of 18 coaches. Darnold moved least. Most of his games elsewhere are in Seattle, in an offense that already looks a lot like O’Connell’s. What a QB keeps is how far downfield he throws and how often he throws into tight coverage.
Does the reshaping cost anything? I split each QB’s gap between Minnesota and elsewhere into two parts: the change in the mix of concepts (screens, play action, straight dropbacks, under center or shotgun), and how well he played within each concept.
The play-mix part of each QB's gap
EPA per dropback from the change in concept mix alone · 95% interval (games resampled)
Total gap = adjusted EPA per dropback under O'Connell minus elsewhere. The rest of each gap comes from how well the QB played within the same concepts.
Data table
| Row | Estimate | 95% CI |
|---|---|---|
| Cousins | −0.00 | −0.02 to +0.02 |
| Darnold | −0.01 | −0.03 to +0.01 |
| Wentz | −0.01 | −0.04 to +0.04 |
| Dobbs | −0.03 | −0.07 to +0.02 |
| Murray | +0.04 | −0.01 to +0.07 |
The mix part is within ±0.04 for every QB, and every interval includes zero. Whatever is better or worse in Minnesota happens inside the same plays.
So what
- The reputation rests mostly on one season. Darnold’s 2024 was real, and it went with him to Seattle, which points to Darnold improving.
- What O’Connell measurably adds is accuracy, and what he measurably costs is sacks. His QBs complete more passes than their history predicts and get sacked more than their history predicts. Net, they play about as well as they were expected to.
- What to watch: Cousins. He’s +0.11 above projection in Las Vegas this year. If the sack pattern holds, expect his sack rate to rise in Minnesota and some of that gain to go with it. The prediction is logged before his first start.
Takeaway: the fair way to grade a coach known for developing QBs is to compare every QB he’s had with what that QB was expected to do, then run the same test on every other coach. On that test, O’Connell’s QBs have played about as expected: more accurate, sacked more often. Darnold’s 2024 is the best of the seven seasons.
What this doesn’t tell us
- Seven QB seasons. Four are under 200 dropbacks, and each season’s gap has noise of about ±0.1. The coach rankings are close enough that a few seasons could reorder the middle of the list.
- The projection is modest. It explains about a fifth of the year-to-year variation in QB play, which is typical for this kind of forecast. That’s why it’s compared against every other coach instead of read alone.
- “Coach” means the whole regime: play calling, scheme, the line and the protection plan. Whether the sacks come from the line or the design is the next post.
- QBs without a track record are left out. McCarthy’s 2025 and Brosmer have nothing to project from.
Method: every NFL dropback 2015–26, with opponent effects removed. Projections are cross-fitted by QB, so no QB is projected by a model that saw him. Full analysis: Q2, Q3, Q13. Data: nflverse and FTN charting (CC-BY-SA 4.0); time to throw from Next Gen Stats.