Stat Analysis
Advanced analytics on player and team performance

Evaluating Third-and-Long Strategy with Expected Points

Guest column by Andrew Kyne, Sports Info Solutions

In a previous Football Outsiders column, we evaluated goal-to-go strategy using our Expected Points Added (EPA) model. Today, we will use a similar approach to look at offensive and defensive effectiveness in third-and-long situations. We're defining "long" as at least 8 yards to go.

Pass Routes

Offensively, the results are intuitive: intermediate and long passes are successful on average, while short passes and checkdowns are not.

Third-and-Long Target Types, 2016-2018
Best Performing Target Types
Worst Performing Target Types
Type Targets EPA Positive% Type Targets EPA Positive%
Seam 144 0.79 43% Flat 218 -0.41 20%
Deep Cross 88 0.79 52% Chip-Flat 170 -0.22 20%
Post 288 0.48 43% Screen 732 -0.13 28%
Corner 175 0.38 39% Fade 101 -0.11 21%
Dig 528 0.34 46% Chip 124 -0.09 31%
(Minimum 75 Targets)

After that group of five, the sixth-best performing route is the slant. That's somewhat interesting in that the average depth of target on those plays is only 7 yards, whereas the five shown above all had average depths of target of at least 13 yards.

Ranking among the worst target types is the fade route, similar to what we showed in the goal-to-go analysis. As a jump ball, it's a low-percentage play regardless of the situation.

Running the ball obviously is not common on third-and-long, and only quarterback scrambles have a positive EPA on average. Even draw plays -- which might be thought of as a better run type in these situations, designed to catch the defense off-guard -- produce a -0.11 EPA and 23 percent Positive%, similar to other basic zone runs.

Coverages

Defensively, Cover-1 and Cover-3 are used most often in pass coverage on third-and-long, just as they are in other situations. Though Cover-1 outperforms Cover-3 in this context, neither is a bad call.

There is significant separation, however, between coverages like Cover-2 and Cover-4. Among the main coverages, Cover-4 is the most effective third-and-long scheme. Cover-2 is the least effective.

Plays on third down with at least 8 yards to go against Cover-4 result in -0.07 EPA on average for the offense (a positive outcome for the defense). Conversely, plays against Cover-2 result in 0.10 EPA on average for the offense (a negative outcome for the defense).

Third-and-Long Coverage Types, 2016-18
Scheme Count EPA Positive%
(Offense)
Cover-4 729 -0.09 30%
Cover-1 1846 -0.07 34%
Cover-3 1556 -0.03 32%
Man Cover-2 1145 0.05 36%
Cover-2 1150 0.08 34%
(Minimum 400 Plays)

Matchups

This data can also be combined with offensive charting information to evaluate how certain target types perform against certain coverages. Among all route types on third-and-long with at least 20 targets against a given coverage, the top performers are deep crosses against Cover-3 (average of 1.19 EPA) and posts against Cover-2 (average of 1.15 EPA). Meanwhile, the worst performing target is the flat against Cover-4, which has produced -0.72 EPA on average.

Individuals

Which quarterbacks perform best and worst on third-and-long against coverages with a single-high safety (Cover-1 and Cover-3)? The leaders in 2018 were Aaron Rodgers (12 Points Earned) and Matt Ryan (10). The trailers were Derek Carr (-14) and Cam Newton (-11). Among pass-catchers, Davante Adams ranked first with 7 Points Earned in those situations, while Trent Taylor finished last with -4.

Sports Info Solutions, based just outside Allentown, Pennsylvania, provides football analytics to teams and media. Sign up for a free week-long trial of their NFL or College Football data hub today at www.SISDataHub.com.

Comments

5 comments, Last at 19 Mar 2019, 9:27am

2 Re: Evaluating Third-and-Long Strategy with Expected Points

One problem with this type of analysis is that "target type" is a poor approximation of "play design".

How many of the less productive routes were targets because the seams, deep crosses, and posts (etc) were completely denied by the coverage? Compare that to how many were in play designs with no deeper target, or no other target at all?

Still, it's a useful, thought provoking bit of analysis, taken with the necessary salt.