Preview #3: The secret to 100 yard NFL rushers

To listen to this episode, check out Apple Podcasts or click on the right pointing triangle. A written version of the episode is below.

You need to run the ball to win the game. Talk to any football fan yet to engage with the analytics revolution, and you will hear this.

In the NFL, nothing could be further from the truth. The analytics community has continually shown the insignificance of running the ball to winning football games.

In working on a project last year, I wanted to add to this body of work. In particular, I was thinking about 100 yard rushers.

Teams don’t need to run the ball to win. They run the ball when they’re up late in games. I wanted to show this in the data on 100 yard rushers.

However, funny things can happen when you actually dig into the data to test a hypothesis. Let me tell you what I found out about 100 yard rushers and what it means for the NFL in 2019.

Play calling depends on game situation

First, let’s look at the data behind why 100 yard rushers should get their yards late in the game.

In a nutshell, teams don’t need to run to win the game. Teams run because they’re winning late in the game. Running the ball keeps the clock moving and prevents the opponent from scoring points.

In 2018, NFL teams ran the ball on 41% of all plays. However, this play selection strongly depends on the game situation.

Teams up by seven or more points in the fourth quarter ran the ball on 65% of plays, significantly more than the 41% NFL average on all plays.

In contrast, teams down by seven or more points in the fourth quarter ran the ball on 21% of plays. NFL teams understand they need to throw the ball to get back in the game.

Hence, 100 yard rushers must get a significant chunk of their yardage late in games when their team is ahead.

100 yard rushers in the NFL

To test this hypothesis, I downloaded data on every 100 yard rushing performance in the NFL since 2009 from The Football Database.

In this dataset, the team with the 100 yard rusher won 72.6% of the games. So far, the data seems to confirm the relationship between big rushing performances and winning.

In addition, 100 rushers on winning teams tended to have more carries. 35% of the 100 yard rushers on winning teams had over 25 carries, while only 18% on losing teams had 25 or more carries.

The 100 yard rushers on winning teams tend to get more carries. Now let’s dig into how these carriers are distributed over the game.

How 100 yard rushers get their yards

To investigate the distribution of rushing yards over the game, I dug into the play by play data. My data set included the 2016 through 2018 seasons.

Let’s look at the distribution of carries and yards for 100 yard rushers on a winning team.

  • 53.6% of carries came after halftime
  • 53.1% of yards came after halftime

100 yard rushers do get more of their yards late in the game. However, 53% doesn’t support my hypothesis that 100 yard rushers get their yards because of opportunities late in games.

So how do players get a 100 yard rushing games? Big plays.

For 100 yard rushers on winning teams, 40% of the yards came from their top two carries.

In case you missed that or couldn’t read the italics, 40% of the yards for a 100 yard rusher on a winning team came from the top two carries.

What if the players is not on the winning team? Then he really needs big plays to make 100 yards. Remember, 100 yard rushers on losing teams get fewer carries.

On non-winning teams, 100 yard rushers had 47% of their yards on their top two carries.

A small disclaimer

The play by play data was from ESPN, and there were small discrepancies with the data on The Football Database. If you define a discrepancy as a 100 yard rusher being off by two or more caries or 10 or more yards, I found 10 discrepancies.

There were a total of 304 100-yard rushing games. 10 games is a very small fraction of the total games, and I’m presenting this data because I don’t think a perfect dataset is going to change these results.

How to predict explosive plays

If almost half of the yards for 100 yard rushers come from the top 2 carries, it would help to predict these explosive plays. However, the data suggests this is difficult.

In 2017, Bill Connolly published a study on SB Nation about explosive plays. He began with a definition for success rate. A play is a success if the offense gets more than 50% of the necessary yards on 1st down, 70% on second down and 100% on 3rd and 4th down.

To measure explosiveness, he only considered successful plays. On these successful plays, he considered points per play, a metric which increases with every additional yard on the play.

By looking at successful plays, you eliminate the plays that don’t go anywhere. However, this keeps a big enough sample of plays to determine the explosiveness for an offense.

For college football data, Connelly found that success rate was sticky. This means that an offense with a high success rate early in the season tends to have a high success rate later in the season. Success rate is predictive.

Connelly found the opposite for explosiveness. Teams that were explosive in the first part of the season saw severe regression to the mean for the remainder of the season. Team that break off a lot of big runs early in the season do not have the same explosiveness later in the season.

This study doesn’t imply the lack of skill in explosive plays, or that somehow more athletic players aren’t more likely to break off a big run. There’s a reason that college football coaches spend so much time recruiting the best athletes.

However, the Connelly study does imply that randomness plays an enormous role in big plays.

An example of regression

As an example, think of Stanford’s Bryce Love. In 2017, he ran for over 2000 yards and averaged over eight yards per carry. How did he get these gaudy numbers? Big play after big play.

Love decided to come back to Stanford in 2018. Unfortunately, regression hit his performance. He only mustered 4.5 yards per carry, as he and his offensive line dealt with injuries all season. Love ended up tearing his ACL in Big Game against Cal.

Bryce Love is a talented football player. Despite his tough season and his injury, he got drafted in the fourth round of the 2019 NFL draft. Despite his talent, you had to expect regression in 2018 because of the randomness in explosive plays.

The NFL in 2019

Connelly did his study for college football, but I suspect that the same results hold in the NFL. In particular, the same results will hold for running backs. Randomness plays a big role in predicting when and if a player breaks off a big run.

While you can’t predict explosive plays based on previous explosive plays, Connelly’s study does suggest looking at success rate. The more often an offense can succeed in getting the necessary yards, the more opportunities it gets to break off a big play.

Last year, I started looking at success rate on the NFL run plays. And because I can’t help myself as a math guy, I adjusted these numbers for a strength of schedule. Two teams jump out at me.

The first team was the Green Bay Packers. They ranked second in rushing success rate adjusted for schedule. It’s not exactly what you expect for an offense with Aaron Rogers, but they ran the ball way better than they threw the ball last year.

In addition, the Packers bolstered their offensive line this off season by signing free agent guard Billy Turner. This means that a guy like Aaron Jones, who’s expected to get a lot of the carries for Green Bay, might have a big season.

On the flip side, the New York Jets were terrible at running the ball last year. They ranked last in adjusted success rate. In addition, they probably did not get better in the off season. They lost starting center Spencer Long to free agency, and they were much better with Long in the lineup last year than without.

It’s unlikely the Jets get much better at running the ball in 2019. Le’Veon Bell might need some extra rest before he starts the season with the New York Jets.

Speak Your Mind


This site uses Akismet to reduce spam. Learn how your comment data is processed.