Author’s note: This article serves as a replication exercise - in Winston’s 2009 book “Mathletics”, he conducts a similar exercise concerning the 2006 Indianapolis Colts. Following Winston’s methodology, I’ll ask and answer a few questions about the 2019 Horned Frog offense.
By missing a bowl in 2019, the 5-7 TCU Horned Frogs induced more than a little frustration in Fort Worth. In this article, I’ll use state and value analysis to ask and answer a few interesting questions about the Horned Frogs’ offense in an attempt to highlight what went right and diagnose what went wrong.
- On any down and yards to go situation, is running more or less effective than passing? For example, on first and 10, is running more effective than passing?
- Are runs more effective than passes overall?
- Was Sewo Olonilua a more effective rusher than Darius Anderson?
Using the magic of CollegeFootballData.com and #cfbscrapR (now on Twitter!), I pulled complete play-by-play data for TCU’s entire 2019 season. For each play, I have the following information:
- Down, distance, and yards from opponent’s end zone before the play
- Whether the ball was a run or a pass
- The ball carrier on a rushing play
- The passer and the intended receiver
- The final result of the play, down, and yards to go for a first down and a new yard line after the play is completed.
- The expected points before and after the play.
For example, here is the entry for Max Duggan’s third-quarter touchdown to Jalen Reagor against Texas Tech:
1 and 10 from the 55: Max Duggan Pass complete to Jalen Reagor 55 yards for a TD. EP_before: 2.537, EP_after: 7, EPA: +4.463.
I wish we had data like the direction of the rush, the depth of target, but alas we are limited by data availability. (My offseason project is to create a depth-of-target database for TCU, but more on that later.)
For each state of TCU’s offense - down, distance, and yard line - we can use expected points to determine the value-added by each play, and those values will provide answers to our questions.
1. On any down and yards to go situation, is running more or less effective than passing? For example, on first and 10, is running more effective than passing?
The Horned Frogs rushed 205 times on first and ten and averaged -.237 expected points per run. They passed 142 times on first and ten and averaged -.316 points per pass. For the TCU Horned Frogs in 2019, running and passing both hurt their value and state, but running hurt their state by a little bit less. This is mostly influenced by turnovers - TCU had 3 interceptions on first and ten, 9 sacks, and one fumble recovery. If we were to ascribe some randomness to turnovers and only examine the plays where TCU successfully executed a run or a pass, then TCU’s rushing attack was worth -.210 expected points per play, and their passing attack was worth -.184 points per play.
Some readers may take exception to the fact that I eliminated more negative value from passes than rushes with those turnovers, but if we are trying to accurately depict reality and understand efficiency, we have to eliminate some noise; mostly, that comes out in eliminating turnovers. Here, by taking out four plays, we improve our understanding of TCU’s offense: the rushing attack, often thought of as TCU’s strength, when things went well, was still less effective than TCU’s passing attack on first and ten.
In fact, the only scenario where rushing was more effective than passing was second and less than five, and third and less than 8, although admittedly many of those third-down rushes were Max Duggan scrambles, which came on designed passes. In 2019, even despite the passing struggles, TCU’s passing attack was more efficient than its rushing attack on first and ten.
2. Are runs more or less effective than passes overall?
On all plays in 2019, TCU averaged .1296 expected points per rush and -0.163 expected points per pass. Were we again to filter out turnovers, which involve no small amount of randomness and disproportionately affect passes, that relative difference does not disappear: rushes are worth .15 EPA on average and passes -0.05. That difference is driven almost entirely by third and fourth downs, where rushes are worth 1.114 and 1.97 EPA, respectively.
While overall rushing was more effective than passing in terms of average EPA, we can see how the situational use affects those numbers: TCU faced an average distance of 4.1 yards on the 66 third-down rushes, and an average of 8.4 yards on the 126 third-down passes.
This is one of the few encouraging facts about TCU’s 2019 offense: despite early-down struggles, they selectively employed the rush on third down, averaging 1.114 expected points and 5.9 yards per third-down rush, rushing 34% of the time on third downs.
The flip side of this, of course, comes through looking at early downs. TCU ran on 58% of first and second downs, averaging -0.031 expected points per rush as opposed to -0.178 expected points per pass. The passing game was not good, but we have to impose some nuance in our argument: TCU’s rushing offense was high-functioning, as designed. This is the rushing offense that the current coaching staff wants to employ, and it was still negative in value on early downs. The passing game obviously needs some work, but we heard that all season. The rushing game was supposed to be TCU’s strength, and it still put them in a worse position on early downs. TL;DR: Rushing a lot on second and third and short is great! Rushing a lot on first and second and long is not!
Speaking of the rushing game...
3. Was Sewo Olonilua a more effective rusher than Darius Anderson?
TCU featured a two-headed monster in the backfield, and both had their moments this season. Sewo Olonilua’s running plays averaged 0.179 expected points per play, and Darius Anderson’s running plays averaged -.08 expected points per play, so Olonilua actually created more value on a per play basis. Anderson took 20 more carries than Olonilua, and he actually averaged more yards per carry: 5.55 ypc compared to Olonilua’s 4.38.
This disparity informs us that while Olonilua was overall more valuable, Anderson’s usage put him at an inherent disadvantage. DA carried the ball on early downs 94.5% of the time, whereas Sewo only carried the ball on early downs 77% of the time. Anderson took 58.5% of the team’s 140 of the team’s 239 first and second down runs , whereas Olonilua took the ball on 78.3% of TCU’s third-down runs. So, while Olonilua technically has a higher EPA per rush, that is mostly an effect of usage- by being the “big back”, Olonilua was put in positions to rush in advantageous situations (third and short) more often than Anderson, reflected by the fact that Olonilua’s average distance on a running play was 6.7 yards and Anderson’s was 8.4. The different levels of running back effectiveness are less about talent differential and more about varying levels of usage.