Filed under:

# Game of Throws: Does Rushing Matter Anymore?

Madness has overtaken this city and grasped in its claws my children. But now we must drive it back under the rocks whence it came.

— Pete Carroll just prior to his 2018 1st round running back selection.

You may not be aware of it, but there is a war raging. On the hallowed batlleground of twitter there is conflict waged over the usefulness of the run game in the modern NFL. Was Indy savvy by waiting until the 4th round to add to their RB stable. Will Barkley reverse the fortunes of the Giants? These are the life questions that set finger to keyboard, pitting dork against jock.

On one side, a new breed of stat-jockeys produce a never ending series of charts and detailed analysis showing that “rushing doesn’t matter”. On the other side, stand the NFL stalwarts defending their eye test, convinced that the run game is the key to NFL victories.

To well and truly tell this tale, I have to go back to the beginning. In 1988, the hidden game of football was published . . . no, not far enough . . .

In the early 1900’s, Karl Pearson advanced statistical theory . . . let me start again . . .

It was Pascal and Fermat . . . wait . . . The ancient Greeks understood . . .

You know what, let’s just look at some graphs. The argument for or against rushing takes many forms, but I am going to focus on just two facets:

1. Does rushing have a significant impact on wins?
2. Does a good rush game help the passing game.

#### RELATIONSHIP TO PASSING

The data warriors correctly explain that if a good rush game assists the pass game, then their respective metrics should have a moderate to strong correlation to each other. In other words, if one performs well, then so should the other. Often the analysis looks something like this:

This measures passing using Adjusted Net Yards per Attempt (ANY/A) compared against both yards per carry (YPC) and run volume; all commonly used stats in these kinds of analyses. Each dot represents a team’s season totals, but since teams with late leads run a lot and teams that are behind don’t, the 4th quarter was excluded from all games to get a cleaner signal.

The results are that both YPC and run volume have very low correlations to ANY/A, suggesting that neither rushing efficiency nor volume assist the passing game. In fact, the R-squared values listed can be interpreted as the percent of passing performance that is explained by rushing performance. They are both basically 0.

Data geeks cry havoc!

#### RELATIONSHIP TO WINS

The problem I have with the above analysis is that it relies on YPC and run volume which are crap stats. To illustrate this and also show the relationship of various stats to wins, I plotted the correlations of selected metrics to season total point differentials (which is actually a better predictor of wins than past win rates).

For passing, I used ANY/A, Yards per Attempt (YPA) and Expected Points Added per Attempt (EPA/A). For rushing, the list was comprised of YPC, Run volume, EPA per Carry (EPA/C) and 2 stats based on Success Rate (SR)(1).

In a previous article, I wrote about SR as a robust metric because instead of yards, it measures successes (TDs, First Downs, chunk % gains of ytg). Adding to that methodology, I have developed a new stat that I am calling weighted Success Rate (wSR)(2).

For these measures, I used all game data not just the first 3 quarters.

It is clear that not all stats are created equal.

Notice how passing stats (green) universally dominate the run stats (blue). They all have very high correlations to point margins (dark shades) and also strong predictive capability (light tints).

This is the dagger that analytic heroes thrust into the digital chests of football Luddites who think that running is equally as important as passing. It isn’t.

While running stats are inferior to passing across the board, some are more inferior than others. Based on these numbers, you can understand why I shake my head every time I see an analysis whose conclusions are based entirely on YPC. It’s just about the worst stat you can use.

On the other hand weighted Rushing Success Rate (wRSR) at least somewhat approaches the level of the passing stats and is the best of the bunch (did I mention that I invented it).

So armed with a decent stat, I again tried to see if running is correlated to passing at all. Here is how the Colts numbers rank by year for wRSR and ANY/A.

I’m no statistician, but that looks correlated to me . . . not to mention sporadically horrible. 2011 is excused but 2015 and 2017, see me after class.

Let’s expand this to all teams. To make things even more sciencey, I put the metrics on a common scale by converting them to z-scores. If you don’t know what that means just think of it as numbers relative to the average. Postive numbers are better than average and negative are worse.

Mmmm, Skittles.

So, does rushing matter?

First of all, when using wRSR the correlation to ANY/A jumps all the way to 0.51. The R-squared value suggests that 26% of the season total passing game is explained by the rushing game, which is significant and waaaaaay more than YPC (take that stat-nerds).

Furthermore, calculating the win rate for each quadrant of the graph results in the following:

### Win Rate

Quadrants <avg Rush >avg Rush All Rush
Quadrants <avg Rush >avg Rush All Rush
>avg Pass 54.20% 64.10% 61.10%
<avg Pass 35.30% 46.50% 39.00%
All Pass 41.00% 58.40% 49.90%

The right half of the graph are teams that rush well and they win about 58% of their games. The teams that don’t rush well win 41% of the time. That difference is about 2.8 games per team per year.

So yeah, it matters. It matters a whole lot.

But don’t feel bad for the diagnostic soldiers out there. If you do the same math for passing, the differences are even more dramatic (61.1% - 39.0% * 16 games = 3.54 games). And if you compare teams that are only good at one phase, then passing has the clear edge there too (54.2% - 46.5% * 16 = 1.23 games).

Utilizing a more traditional method, the results of a regression with wRSR and ANY/A against point differentials shows that passing is far more impactful (has a much higher coefficient).

### Regression Results

Measure Value
Measure Value
R-Square 61.30%
wRSR_z Coeff 25.5
ANY/A_z Coeff 63.4

#### CONCLUSION

In this fight, I think both sides are a little right and a little wrong.

Clearly rushing matters. It explains wins, it is predictive of future wins and it describes a good portion of the passing game. However, passing sits on the throne. It is far more related to past/future wins and when teams are only good at one phase, they are more successful when that skill is passing.

But I don’t think this is the end of wisdom. All of the data above are cumulative measures from the last 9 years and in that time, the NFL has been changing. QBs are becoming much more efficient at passing while the run game has actually stagnated a bit. The result? A rush to pass correlation that is falling.

So while the cumulative results are fairly strong, the trends of last few years add uncertainty.

#### FOOTNOTES

1) Success Rate utilized the Football outsiders methodology, which defines success as a TD, First Down, 40% of yards to go on first down, 60% of ytg on second downs. %’s are adjusted in 4th qtr to 50%/65% for teams behind by more than one score and 30%/50% for teams with any lead.

2) Weighted Rushing Success Rate defines success as a TD. First Down, 45% of ytg on first down or 60% of ytg on second down. Successes are given weights of TD: 2.0, First down: 0.9, 1st %ytg: 0.65, 2nd %ytg: 0.55. 4th quarter adjustments apply with <11 minutes = (11 - minutes remaining) * (# of scores point differential) * x where x = 0.025 for teams with a lead and x = 0.0125 for trailing teams.

Thanks to Pro Football Reference, Armchair Analysis, Football Outsiders and the nflScrapR project for being excellent data sources.