CFB Matchups

NFL Matchups
Loading…

EPA

Expected points added per play, from ESPN play-by-play via cfbfastR. Garbage time is left out, and every team is adjusted for the offenses and defenses it has faced, so a soft schedule doesn't inflate a rating. New to EPA? EPA Explained walks through every part of this page.

Head to head

Longer bar, better rank among FBS teams; the edge goes to the side ranking at least 26 spots better. Adjusted EPA / play strips out what each opponent was expected to allow or gain. Plays with a lopsided score are left out (a lead over 43 points in the 1st quarter, 37 in the 2nd, 27 in the 3rd or 21 in the 4th), along with kneels and plays wiped out by a penalty. Early-down pass is EPA on 1st- and 2nd-down passes. For a defense, lower is better.

EPA line vs. the market

The EPA line is what the ratings alone say about each game. A big gap is not a bet: more often the market knows something the ratings can't see, such as a quarterback injury, the weather or a team resting starters. Treat it as a game to look into.

EPA ratings, every FBS team

Net is offense minus defense, in adjusted EPA per play. The shift columns show how many spots a unit moves once schedule and garbage time are accounted for. The flags list results likely to even out. Select a column heading to sort.

How the numbers are built

  1. Take every pass and run play this season from cfbfastR play-by-play, which carries EPA and success for each play. FCS opponents are pooled into one team.
  2. Drop garbage time (a lead over 43, 37, 27 or 21 points in quarters 1 to 4), kneels and plays wiped out by a penalty.
  3. Fit one regression across all plays so each team gets an offense rating and a defense rating, with opponent strength and home field separated out. Pass and run get their own fits.
  4. Project each game: each offense's rating plus the other defense's, times expected plays, plus 2.5 points for home field (none at a neutral site).
  5. Pull the projected totals toward the league average. The spreads are left as they are: the regression already shrinks the ratings, and pulling them further made them worse.

Not included: quarterback injuries, weather, rest and travel, recruiting and last season's results.

Does it predict anything yet?