Finding the Optimal Pass/Run Ratio
Guest column by Lau Sze Yui
Every football fan probably knows that passing in football is generally more efficient than rushing. But how much passing is really enough?
To answer this, we first need some prior knowledge of how pass/run balance works in NFL. For example, Sean Clement found that previous run attempts have no effect on a team's running success rate; hence, theoretically speaking, no matter how many runs a team has, its efficiency is still the same (and lower than that of passing). But one of the reasons why teams run the ball is to set up play-action, which utilizes the threat of running and is more efficient than normal passes. Therefore, the more precise question to ask is: is the value of play-action good enough for teams to mix runs into their play calling?
1. Play-action success doesn't depend on past success running the ball, nor on the number of previous runs.
2. Inversely, rush success doesn't depend on either past play-action success or the number of play-action passes.
3. Play-action is more effective in a run-heavy situation.
But still, no teams would call 100% pass plays in real competition to see if passing is still better than rushing. To tackle this problem, I've built a classifier to predict chance of passing on a given play based on offensive formation, score margin, time remaining, down, etc., and observe how expected points added (EPA) of play-action passes, non-play-action passes, and rushing plays change for different pass probabilities. As someone said, "In football, everything interesting is probably selection bias," so there are few things to note before drawing any conclusion.
- Teams that have a worse passing game (and to a lesser extent running game) would fall behind more and thus run more plays in high-pass probability scenarios. Therefore the classifier has to be controlled for season offensive and defensive DVOA as well.
- Although passing is more efficient than rushing most of the time, there are still cases where rushing is better -- i.e., short-yardage situations on third and fourth down. To limit the scope of our research, it's better to consider a fixed down-and-distance situation, and first-and-10 would be most convenient here.
- The final two minutes of each half have been discarded.
With those qualifiers in mind, here is the result of EPA on first-and-10 assuming an average offense and defense for both passing and rushing game (EPA data courtesy of NFLscrapR; DVOA data courtesy of Football Outsiders):
(Ed. Note: The original version of this graph was slightly off; this is a new, more accurate graphic.)
The graph shows that rushing and non-play-action pass efficiency is mostly constant for any pass probability, while play-action pass efficiency decreases when pass probability is high. As a result, it's likely better off to pass most of the time on first down, but let's examine play-action some more.
Different teams obviously have different tendencies to call play-action passes instead of normal passes. For example, here are the play-action rates of the 2017 Steelers and 2018 Rams, two teams that called play-action passes at drastically different rates:
By using the EPA of different play types and their frequencies, we can calculate how the projected EPA values change across pass probability by summing the statistical expected value. Here, two teams -- one with low play-action usage (corresponding to 17 PIT) and one with high play-action usage (corresponding to 18 LAR) -- are plotted to demonstrate how play-action percentages affect projected EPA value for a team, assuming an average offense and defense:
The result shows that while higher play-action rate will lead to higher EPA, since it's less effective when pass probability is low, overall EPA is still at its peak when teams pass 100% of the time.
But does that mean teams should blindly pass 100% of the time on first down? The answer is simple: "it depends." Consider two teams in 2018 as a case study.
Patrick Mahomes was crowned as 2018 MVP with 50 touchdown passes, and his passing numbers were of course elite. In 2018, the Chiefs had a 62.6% unadjusted VOA on non-play-action passes but "just" 49.6% on play-action passes, while having a 12.2% unadjusted rushing VOA. As a result, we would expect their overall offense efficiency would be the highest when passing at a high volume:
Conversely, Jared Goff is obviously not as talented a quarterback compared to Mahomes, but thanks to Sean McVay's scheme and strong defense his team was able to get to the Super Bowl in 2018. The 2018 Rams had a 42.5% unadjusted VOA on play-action passes but only 27.4% on non-play-action passes. Here is the projected EPA versus pass probability:
Unlike what we saw with the Chiefs, the EPA for the Rams peaks when pass probability is close to 50%.
When designing run/pass balance, player ability should be heavily considered as well. For a rule of thumb, when non-play-action pass efficiency is higher than the average of play-action passes and rushing plays, teams should look to pass at much as possible on first downs. Otherwise, if the non-play-action plays are less efficient than that average, teams should look to play-action half of the time in order to "establish the play-action."
There are a few other factors teams need to consider as well:
1. Dropbacks increase a quarterback's chance of injury more than handing the ball off, and teams may not want to risk too many in the regular season.
2. There is no magic way to determine a quarterback's quality besides on-field snaps; therefore, at least in the first few weeks of a season, teams need to try different things to determine the best strategy.
3. Player emotion is also an important thing to consider. For example, some players, such as Ben Roethlisberger, don't feel as comfortable executing a play-action pass from under center; it's the major reason why the Pittsburgh Steelers have a very low play-action percentage with him as the quarterback.
To conclude, there are two type of optimal run/pass ratios in NFL (except for the case where your quarterback is so bad that rushing will be more efficient): if your quarterback is good at play-action but not that good on non-play-action passes, teams should run about 50% of the time in order to "establish the play-action." Otherwise, teams should theoretically pass 100% of the time.
Although it seems that it's the takeaway of the article, I am not suggesting that passing 100% of the time is a viable strategy. Instead, I'm suggesting that the correct passing equilibrium is still far away from current run/pass ratios.
Lau Sze Yui studies sports analytics in his spare time and can be found on Twitter @903124S.