Coaches who still send out the punt unit in that spot surrender 0.42 EPA compared to keeping the offense on the field. Between 2009 and 2025, attempts from exactly that down, distance and field position converted 65 % of the time; drives that succeeded produced 2.7 points on average, while failed tries left opponents only 1.9 points ahead of where a punt would have placed them. The break-even line is 48 %, so every extra yard past midfield tilts the scale further toward going for it.
Front offices now bake those figures into weekly game plans. Baltimore built a +137 EPA edge on fourth-down calls from 2018-22, the largest five-season stretch since 2000, by dialing 119 keeper plays instead of kicks. Philadelphia followed suit and cashed 78 % of its fourth-and-shorts last year, adding 34 net points-roughly half a win-to its regular-season tally. League-wide, offenses attempted 668 fourth-down conversions in 2025, up from 420 in 2011, while success rates held steady at 61 %, proving the rise stems from smarter choices, not luck.
Defenses have countered by stocking lighter, faster nickel packages: base personnel usage on fourth-and-short dropped to 19 % in 2025 from 55 % in 2015. That shift widened B-gap seams, pushing conversion odds another 3 % upward and forcing coordinators to treat every midfield fourth down as a de facto red-zone snap. Special-teams snaps, meanwhile, keep shrinking-punts from inside the opponent’s 40-yard line fell 38 % across the same span-while scoring per possession climbed from 1.79 to 2.05 points.
Quantifying Win Probability Added on 4th-and-1 from Your Own 34
Go for it. A conversion raises expected win probability from 18 % to 26 %; a punt drops it to 15 %.
Since 2018, league success rate on 4th-and-1 between the 30- and 35-yard lines is 71 %. That 0.71 × +8 % = +5.7 % baseline WPA for keeping the drive alive.
Miss and the opponent starts at the 34, worth 2.3 expected points. The resulting −7 % WPA swing still leaves the aggregate decision at −1.3 %, compared with −3 % after a punt.
Weather trims the edge. Wind above 20 mph cuts conversion odds to 63 %, shrinking WPA gain to 3.5 %. Coaches facing gusts above 25 mph should reconsider only if their offense ranks bottom-five in power-run DVOA.
Opponent red-zone efficiency matters. Against top-third units, a punt moves the ball but raises touchdown probability from 38 % to 47 %. The break-even conversion rate falls to 65 %, still below league average.
Clock situation tilts the scale. Trailing by 1-3 points with 7-10 minutes left, the go-for-it WPA jumps to +9 % because a first down burns two additional minutes. Leading by 4-7, the same failure hands the opponent a 55 % win chance, so the call dips to −1 %.
Kicker range is irrelevant here; a 52-yard attempt into the open end carries a 58 % make rate and −6 % WPA on a miss, far worse than the offensive gamble.
Bottom line: any roster that converts above 65 % in short-yardage situations adds at least half a win per season by declining the punt on this exact yard line.
Building a Conversion Probability Model with 5 Years of Next Gen Stats Tracking Data
Feed 1.2 million route snapshots into a gradient-boosted tree with 38 engineered features-release velocity, defender closing rate, field zone, score differential, temperature, altitude-and calibrate the model with isotonic regression; the resulting ROC-AUC of 0.847 on the 2026 test fold shows that a 2-yard shotgun run on 4th-and-2 between the 34- and 38-yard lines converts 57 % of the time when the back reaches 9.1 mph within 0.7 s of handoff, but only 41 % if the interior linebacker’s first step is under 0.18 s and safety depth is < 7 yd at snap.
Stack a second-layer neural net that ingests these calibrated probabilities plus micro-tracking residuals-snap-to-throw interval, WR separation at 2.5 s, OL surge displacement-to shrink the prediction error by 11 %; the blended model flags three go windows: 4th-and-1 from own 29-41 (expected gain 0.37 WP), 4th-and-2 from opponent 32-38 (0.29 WP), and 4th-and-3 from opponent 26-30 (0.22 WP). Export the inference engine as a 7 MB ONNX file, drop it into the OC’s Surface so the staff sees live heat-maps 0.8 s after the previous play ends, and update priors every mid-drive using 2026 tracking feeds to keep decision deltas inside ± 0.015 WP.
Convincing a Risk-Averse Head Coach: The 15-Minute Film Room Pitch Deck
Open with a 12-second clip: 2025 Week 11, CIN-BUF, 4:32 Q2, 4&2 at own 45. Freeze at snap: Ja'Marr Chase removes both safeties; Joe Mixon walks into 17 yards of green grass. Pause, rewind, freeze again. Overlay EPA ticker: +2.34. Point to the coach: That’s three plays shaved off the drive, 0.7 expected points added, and your punter never sees the field. Slide the next graphic-2026 league go-rate on 4&2 between 40- and 50-yard lines: 68 %. Teams that passed the sticks averaged 0.38 points per drive more than those who kicked; defense allowed 0.21 points fewer on the ensuing possession because the offense tilt field position. Hand him the laminated card: 4&2 conversion probability by personnel group-11 (64 %), 12 (71 %), 21 (77 %). Tell him you’ll practice it three times a week: one walk-through, one full-speed look, one red-zone period starting at the minus-35. Finish with the kicker: every extra conversion raises playoff probability by 1.3 % according to 10 000-season Monte Carlo; that’s the equivalent of a $2.4 M veteran minimum contract you don’t have to spend.
| Down & Distance | Own Territory | Go Rate 2026 | Success % | EPA/Play |
|---|---|---|---|---|
| 4&1 | Own 30-50 | 72 | 74 | +0.31 |
| 4&2 | Own 30-50 | 68 | 64 | +0.24 |
| 4&3 | Own 30-50 | 41 | 51 | +0.09 |
Close the door, dim the lights, queue the last slide: a bar graph showing head-coach tenure length for aggressive vs. conservative quartiles since 2016. Aggressive quartile averages 4.8 seasons; conservative 3.2. Tell him the math protects jobs, not just scoreboards. Walk out without another word; let the silence sell the call sheet.
From 2016 to 2026: Charting Every Aggressive 4th Down Call and Its EPA Outcome
Filter every 2016-2026 regular-season play for scrimmage downs with 1, 2, 3, or 4 yards-to-go inside the opponent’s 40; keep only the attempts that chose go-for-it. 2,847 such snaps produced +0.29 EPA per try, a 57 % success rate, and a 3.4-point swing in expected victory probability for the offense.
- 2016 baseline: 214 attempts, +0.19 EPA, 52 % conversion.
- 2017 spike: 267 attempts, +0.32 EPA, 59 % conversion.
- 2018 dip: 248 attempts, +0.21 EPA, 54 % conversion.
- 2019 surge: 301 attempts, +0.35 EPA, 61 % conversion.
- 2020 covid year: 356 attempts, +0.31 EPA, 58 % conversion.
- 2021 peak: 398 attempts, +0.38 EPA, 63 % conversion.
- 2025 plateau: 401 attempts, +0.28 EPA, 57 % conversion.
- 2026 record: 415 attempts, +0.33 EPA, 60 % conversion.
Inside the 5-yard line the EPA jumps to +0.71; outside the 35 it collapses to -0.12. Distance matters more than field position once past midfield.
- RPO slants and QB sneaks from 12 personnel generated +0.47 EPA on 412 tries.
- Shotgun outside zone reads averaged -0.03 EPA on 218 tries.
- Empty-dropback passes versus man coverage returned +0.52 EPA but only +0.09 against zone.
Coaches who went for it at least 25 times in a season saw their offense score 0.9 more points per game the following year even after removing those extra possessions, hinting at a feedback loop of practice emphasis.
Seven clubs-the Ravens, Bills, Browns, Cardinals, Eagles, Chargers, and Dolphins-accounted for 38 % of all tries yet only 22 % of the league’s total offensive snaps, posting a collective +0.41 EPA. The bottom five-Falcons, Panthers, Texans, Titans, and Cowboys-managed -0.07 EPA and left 47 win probability points on the table in 2026 alone.
Takeaway for 2026 play-callers: any 4th-and-2 or shorter between your own 45 and the opponent’s 30 is a default green-light if your offense converts 55 % of 3rd-and-2s in game scripts faster than 2.3 plays per minute; below that threshold treat it as a yellow-light and lean on pre-snap motion to confirm man coverage before keeping the offense on the field.
Real-Time Decision Boards: How Tablets Feed Live Optimal Go-for-It Thresholds
Green Bay’s surface shows 0.02 win probability lost if the punt unit jogs out on 4th-and-2 from the opp 38; the tablet blinks +4.7% go and Matt LaFleur keeps the offense on the field.
Each sideline slate pulls four micro-models: pre-snap defense alignment, 2020-23 tracking data for personnel on the grass, weather-corrected EPA for the distance, and clock delta. Inputs refresh every 3.2 s, delivered through secure 5 GHz hotspots bolted to the replay tower. Result: the break-even line moves from 48% to 51% after the opponent inserts a heavy-nickel look, then to 44% once wind gusts spike to 17 mph.
Coaches do not see probability ribbons; they see a traffic bar: red for kick, amber for conditional, green for go. If the bar sits within ±2% of the knife-edge, the play-caller hears a short vibrate in the headset. Baltimore installed the system in 2019; its punt rate on midfield 4th-and-1 dropped from 68% to 31% within eight weeks.
league rule caps in-game data to 8 Mbps, so graphics are prerendered as 4-kilobyte SVG layers. Each tablet stores a 1.3-million-row cache of opponent-specific tendencies; only three columns update live: win-prob delta, score delta, and time. Encryption uses 256-bit rotating keys swapped every drive, keeping average lag under 0.4 s.
Accuracy check: 1,947 fourth-down calls in 2026 followed the board; 1,741 (89.4%) aligned with post-game optimal, up from 62% in 2018. The exceptions came mainly when the model misread defensive tackle substitutions, a bug fixed by adding RFID tags to interior linemen helmets this spring.
Next frontier: merging quarterback RFID and passer-tracking chips to adjust break-even for fatigue. Early beta in Arizona lowered the go-threshold by 1.8% when Kyler Murray’s accumulated drop-backs topped 28, adding roughly 0.15 wins over a season. Expect every club to run that module by September.
FAQ:
Which single piece of data swung the most coaches from kicking to going on 4th-and-1 inside the opponent’s 40?
The 0.9 expected points added that models show for going instead of kicking. Once that number leaked into every staff Slack channel, coordinators who still sent the field-goal unit out were effectively lighting a tenth of a point on fire. No coach wants to walk into the Monday review meeting having voluntarily spotted the other side nearly a point, so the 0.9 EPA became the line in the sand: cross it and you keep the offense on the field; stay south of it and you’re still in the old punt-or-kick world.
How do teams reconcile the models with the fact that one failed 4th-down try can swing win probability by 20 % and get a coach fired?
They bake the downside into a season-risk column rather than treating each 4th-down as an isolated bet. Carolina, for example, runs 10 000 season sims every Tuesday night; if the model says the staff’s seat is lukewarm even after a worst-case streak of three straight failures, they green-light the aggressive calls. If the same streak drops their playoff odds below 15 %, they throttle back and take the points. The math stays the same—go is still plus-EV—but the organization’s risk budget for that year is spent, so the play sheet gets conservative until the standings stabilize.