Chapter 28 — Key Takeaways: Analytics and Offensive Decision-Making

Key Takeaways

  • Analytics replaces memory with counted evidence — and admits where evidence runs out. Marisol's job: "I'm not here to make the decision. I'm here to make sure you know what you're deciding."
  • Expected points (EP) value a down, distance, and spot; EPA is the change on one snap. Third-down results, turnovers, and explosives swing EPA far more than ordinary early-down gains.
  • Success rate (40/60/100) measures staying on schedule; explosive rate (12+ run, 20+ pass) measures skipping it. You need both engines. Past six yards to go, third-down conversion falls off a cliff.
  • Selection effects distort run/pass comparisons. Compare like with like (same down, distance band, box count, score). Code RPOs as their own family.
  • Grade the family, not just the parent. A run's value can live in the play-action and RPOs that look like it — but prove it at the defender level (whose feet change?), not with averages alone.
  • Fourth down is a break-even problem. Go when your conversion chance beats the break-even; on fourth-and-short across most of the field, it's far below what teams convert.
  • The model assumes a league-average play. Making that true is the coordinator's job. The head coach's question — "What's the call if we go?" — tests it.
  • Decide before third down. "Four-down" turns third-and-short into one of two downs and changes the third-down call.
  • Name it, then price it. An override must name something specific the model can't see and translate it into a probability. Most named objections change the call, not the decision.
  • Two-point decisions are driven by the score, because on average a two-point try and a kick are worth about the same.
  • Small samples lie; perfect streaks may be traps. Counts, not percentages; question marks on streaks; the opponent reads the data too.

Action Items (this week)

  1. Add a "This doesn't know…" column to every stat on your weekly report.
  2. Grade last game's snaps by success rate by hand. Find your third-down distance distribution.
  3. Find your comfort call: the play you call most after a negative play. Compare its success rate to the same situations without the bad play before it.
  4. Build or adopt a fourth-down chart (zones × distances), adjusted for your kicker, and write the call that makes each GO box true.
  5. Put the two-point table on the laminated card and a two-point play on your call sheet every week.
  6. Pick one family (a featured run plus its play-action and RPOs) and chart the conflict defender's first step for the next two games.

Common Mistakes (and the fix)

Mistake Fix
Grading a fourth down by its result Monday review in order: decision, call, execution — then result
Treating EPA as a player grade Attach EPA to the call and situation first; assign blame with film
"Establish the run" by volume Earn play-action by picture and by the specific defender's habit; chart his feet
The comfort call after a bad play Make it half of a Kill call so the defense has to earn it
Trusting a big percentage on a small sample Counts first; under about 20 snaps it's a lean, not a fact
Aggregated tendencies ("15 of 26") Split by the condition that matters for your team (e.g., QB experience)
Going for it without a trusted call Build the short-yardage package that makes the model's assumption true
Practicing only the defense's tendency Rep the exception that would beat your call on the snap that matters

Decision Framework: When Should the Number Override the Gut?

Step Question If yes If no
1 Is the sample big enough (counts, ~20+ snaps; a "tell" is ~70% on 10+)? Go to step 2 Treat it as a lean; don't let it override anything
2 Is this the situation the model assumes (average play, kicker, defense; healthy players)? Follow the number unless step 3 finds something Go to step 3
3 Can you name what it can't see — and price it as a change in probability? Re-run the comparison; usually change the call, sometimes the decision The number wins
4 Is the objection about the play — or about the blame? — If it's blame, the number wins

Fourth-down quick procedure: (1) Before third down, know the chart box → say "Four-down" if it's a go. (2) Find the break-even: value of converting × p + value of failing × (1 − p) = value of kicking. (3) Name your call and its likely failure. (4) Head coach decides; write it down; grade it Monday before looking at the result.