Vikings analysis
J.J. McCarthy's Real Odds of Becoming a Starter
Part of an ongoing analysis of the Vikings offense (project hub).
Data through week 9 of 2026.
Bottom line.
- First seasons as rough as McCarthy’s 2025 rarely turn into starting careers. Of 16 first-year starters that bad since 2015, 2 had even one average season by year 3.
- Held to a multi-season standard, his chance of becoming a good starter is about 9%. His most likely outcome is a backup or fringe starter (65%).
- His 77 Giants dropbacks roughly doubled those odds, from a low base. His strong finish to 2025 adds nothing to the forecast.
- A new coach by itself doesn’t turn young QBs around. Goff’s and Lawrence’s breakouts looked nothing alike, and McCarthy’s Giants situation matches neither.
McCarthy’s first 3 games as a Giant came in at +0.10 adjusted EPA per dropback. His 2025 season in Minnesota was −0.23. That’s 77 dropbacks, which is a small sample, and it was enough to restart the argument about whether the Vikings gave up too early.
That argument depends on a base rate that rarely comes up: how often a first season like his turns into a starting career. So I built it, plus a forecast that updates after every game he plays.
1. Where a first season like his usually leads
McCarthy’s 2025: −0.23 opponent-adjusted EPA per dropback on 290 dropbacks, the 12th percentile of 60 first-year starters since 2015. He was 22 and the 10th overall pick.
First-year starters who had an average-starter season by year 3
First seasons 2015–23 · average starter = 300+ dropbacks at or above that year's median QB
Data table
| Row | Value |
|---|---|
| All first-year starters (19 of 60 (95% CI 21–44%)) | 32% |
| Bottom quartile in year 1 (2 of 16 (95% CI 4–36%)) | 13% |
| Bottom quartile, 1st-round picks (2 of 7 (95% CI 8–64%)) | 29% |
| Within ±0.10 of McCarthy (1 of 17 (95% CI 1–27%)) | 6% |
One in three first-year starters turns into at least an average starter. Start in the bottom quartile, where McCarthy was, and it’s about one in eight. Being a first-round pick helps (2 of 7), but look at who the two were: Jared Goff and Trevor Lawrence. Both were #1 overall picks, and both took off after a coaching change, Goff from Fisher to McVay and Lawrence from Meyer to Pederson. The first-rounders who didn’t make it: Josh Rosen, Dwayne Haskins, Bryce Young, Zach Wilson, Justin Fields.
That’s two cases. It’s tempting to read them as a recipe: bad first year, new coach, breakout. Section 6 tests that.
The base rate has a feedback loop in it. Half of the bottom-quartile QBs never got another 150-dropback season, so they never had the chance to post an average one. A bad first year costs a QB his starts, and losing the starts makes the bad first year look even more predictive. Some of that 13% reflects how teams behave.
2. Holding QBs to more than one good year
The base rate above counts a QB who has a single average season. That’s a low bar. Burrow, Mahomes and Allen are elite over several years, so for the forecast I required a QB to reach a tier in at least 2 of his years 2–4:
- Elite: top 8 in adjusted EPA twice
- Good: top 16 twice
- Average: top 24 twice
- Backup or fringe: played, but didn’t qualify twice
- Out: never got 150 dropbacks in a season again
Draft slot goes in as the actual pick on a log scale, because the drop-off within the first round is steep. Those two choices cut McCarthy’s good-or-elite odds from about 30% to about 10%.
3. The odds by tier
Sustained tier in years 2–4
Cohort: 47 first-year starters, 2015–22 · McCarthy rows are the model's forecast
McCarthy + Giants games, 80% intervals: good or elite 1%–27%, average or better 5%–57%, backup or fringe 40%–82%. Model: ordinal logistic on year-1 adjusted EPA, log draft pick, age, and his early year-2 play shrunk toward what year 1 predicted.
Data table
| Group | Out | Backup / fringe | Average | Good | Elite |
|---|---|---|---|---|---|
| All first-year starters (n=47) | 21% | 38% | 13% | 13% | 15% |
| Bottom quartile in year 1 (n=13) | 46% | 39% | 8% | 0% | 8% |
| McCarthy, 2025 only | 21% | 66% | 8% | 3% | 2% |
| McCarthy + Giants games | 7% | 65% | 19% | 6% | 3% |
On 2025 alone, he looked like a bottom-quartile starter with a better draft slot: 5% good or elite, 13% average or better. Adding 77 Giants dropbacks at +0.10 moves him to 9% and 28%. The biggest shift is out of “Out”: 21% → 7%. Because of the feedback loop above, that matters: he’s likely to keep getting the starts most QBs in his spot lose.
Only 13 of the 47 QBs in the cohort reached a sustained good or elite tier: Mahomes, Allen, Lamar Jackson, Purdy, Watson, Tagovailoa and Goff at elite; Herbert, Burrow, Hurts, Wentz, Winston and Mariota at good.
4. Game by game
McCarthy's odds of becoming a sustained starter, game by game
Chance of reaching each tier in at least 2 of his years 2–4 · shaded = 80% interval · each point uses only games played to that date
Data table
| After | Dropbacks | Adj. EPA | Avg or better | Good or elite |
|---|---|---|---|---|
| MIN 2025 wk 11 | 167 | −0.37 | 8% (1%–22%) | 2% (0%–8%) |
| MIN 2025 wk 12 | 192 | −0.42 | 5% (0%–18%) | 1% (0%–6%) |
| MIN 2025 wk 14 | 221 | −0.33 | 11% (2%–25%) | 3% (0%–10%) |
| MIN 2025 wk 15 | 247 | −0.26 | 18% (5%–32%) | 6% (1%–14%) |
| MIN 2025 wk 16 | 266 | −0.26 | 18% (5%–32%) | 6% (1%–14%) |
| MIN 2025 wk 18 | 290 | −0.23 | 22% (8%–36%) | 8% (1%–16%) |
| NYG 2026 wk 6 | 21 | +0.32 | 29% (10%–47%) | 11% (2%–23%) |
| NYG 2026 wk 7 | 51 | +0.21 | 33% (11%–53%) | 13% (2%–27%) |
| NYG 2026 wk 9 | 77 | +0.10 | 31% (10%–49%) | 11% (2%–24%) |
Most of the climb came in December 2025, before he ever played for the Giants. From his low point after week 12 to the end of the season, his good-or-elite odds went from 1% to 8%. The Giants games added a few more points, and his third Giants game gave some back.
This chart uses a slightly different model than section 3: it fits once and feeds in his year-2 play as it accumulates, which stays stable after a single game. It reads 11% where section 3 reads 9%. Treat them as the same answer.
5. The strong finish doesn’t change the forecast
McCarthy’s rookie season had two halves: −0.37 adjusted EPA per dropback before week 10, −0.03 after. That looked like a QB figuring it out. Across the cohort, a jump like that from the first half to the second doesn’t predict year 2 at all. Adding it to the model made the forecasts worse.
His closest matches on rookie level, second-half jump, age and draft slot:
| QB | Pick | Year 1 | Adj. EPA | 2nd-half jump | Sustained tier, years 2–4 |
|---|---|---|---|---|---|
| D. Haskins | 15 | 2019 | −0.24 | +0.30 | Backup / fringe |
| J. Fields | 11 | 2021 | −0.15 | +0.11 | Backup / fringe |
| Z. Wilson | 2 | 2021 | −0.22 | +0.09 | Backup / fringe |
| D. Mills | 67 | 2021 | −0.16 | +0.14 | Backup / fringe |
| D. Kizer | 52 | 2017 | −0.19 | +0.08 | Out |
| J. Allen | 7 | 2018 | −0.09 | +0.17 | Elite |
| L. Jackson | 32 | 2018 | −0.09 | +0.12 | Elite |
| D. Jones | 6 | 2019 | −0.10 | +0.07 | Average |
Two of eight became elite: Josh Allen and Lamar Jackson. Five never stuck. Haskins is the closest match overall.
The late-season version of McCarthy did look different in one place. From shotgun, his throws past the first read were close to a league-average QB’s. From under center, in O’Connell’s play-action game, they weren’t.
McCarthy, throws past the first read, weeks 10–18 of 2025
Opponent- and situation-adjusted EPA per dropback · 95% interval (games resampled)
Both intervals are wide and cross zero. His overall improvement over the season was ordinary for a rookie, so this looks like where he was comfortable. It doesn't show the formation sped up his development.
Data table
| Row | Estimate | 95% CI |
|---|---|---|
| Shotgun | −0.03 | −0.27 to +0.40 |
| Under center | −0.45 | −1.07 to +0.08 |
6. Is there a “Goff and Lawrence play”?
The story people tell about Goff and Lawrence is a bad first year, a new coach, a friendlier offense, then a breakout. The Giants move looks like the same story for McCarthy. I tested both halves.
A new coach alone doesn’t make young QBs better. Among 37 first-year starters who kept playing in year 2, those with a new head coach improved by −0.02 EPA per dropback relative to those without one, once you account for how bad their first year was (95% CI −0.11 to +0.07). In the bottom half of year 1, 2 of 15 reached an average season with a new coach and 2 of 15 without. A new coach did change who kept the job: 53% of those QBs kept starting with a new coach, 27% without.
Goff and Lawrence got different things.
Time to throw, year 1 → year 2
NGS average seconds from snap to throw · lower = quicker offense
Data table
| Row | Year 1 | Year 2 |
|---|---|---|
| Goff | 2.54s | 2.93s |
| Lawrence | 2.87s | 2.59s |
| McCarthy | 3.02s | 2.88s |
Receiver quality, year 1 → year 2
League percentile of his receivers, judged on throws from other QBs · higher = better cast
McCarthy's 2026 numbers rest on 77 dropbacks.
Data table
| Row | Year 1 | Year 2 |
|---|---|---|
| Goff | 5 | 48 |
| Lawrence | 36 | 12 |
| McCarthy | 93 | 48 |
Goff’s offense got slower under McVay, and his receivers went from the 5th percentile to the 48th. Lawrence’s offense got much quicker under Pederson, with receivers a little worse than the year before. What they had in common was draft slot and patience: both were #1 picks, and both got a full second season (510 and 639 dropbacks).
McCarthy got a quicker offense and the starts. He also lost one of the league’s best supporting casts: his receivers went from the 93rd percentile with Jefferson and Addison to the 48th with the Giants. That’s the biggest drop of any QB in the cohort who changed coaches.
So what
- Moving on was defensible by the base rates. One in eight bottom-quartile starters had even one average season by year 3. For a team trying to win now, that’s a reasonable call. It was still a bet. Had the number been 40%, it would have been premature.
- The Giants gave McCarthy the one thing Goff and Lawrence shared: the starts. A new coach on its own hasn’t helped QBs like him, and he’s throwing to a much weaker group than he had in Minnesota. If he plays well with this cast, that’s a stronger signal than the same numbers would have been with Jefferson.
- What to watch: his odds after each Giants start. The chart above updates after every game, and each start puts more weight on his Giants play and less on his rookie year.
Takeaway: a rough first season like McCarthy’s is a strong negative signal. By the base rates, a team that moves on from one usually isn’t giving up a future starter. The two exceptions since 2015 were #1 overall picks whose teams gave them a full second season.
What this doesn’t tell us
- 47 QBs. The intervals are wide: backup or fringe runs from 40% to 82%.
- The base rates partly measure opportunity (section 1). Some of the QBs who lost their starts might have improved with them.
- A three-season window clips late bloomers. Sam Darnold became a top QB in year 7.
- Tiers are adjusted EPA rank only. No contracts, no reputation, and no front-office reasoning. Why Minnesota decided what it did isn’t in the data.
Method: every NFL dropback 2015–26, with opponent effects removed. First-year starter = first season with 150+ dropbacks, rookie year 2015 or later. Full analysis: Q24, Q22, Q26. Time to throw: Next Gen Stats. Data: nflverse and FTN charting (CC-BY-SA 4.0).