Packing the Stats: Rushing to Conclusions Follow-Up

Packing the StatsIn response to the low yards per attempt by Alex Green last weekend, we had some good discussion in the comments regarding my statistical research on how the rushing game affects the success rate of NFL teams in the past ten years. The data seemed to show that the number of attempts had a higher correlation with winning than average yards per attempt.

Some people agreed, bringing up the Packers’ success during the two halves of the Seahawks and Colts games when they committed to running the ball more. Others argued that this was more of a secondary outcome, in which winning causes increased attempts and not the other way around.

We even had some suggestions for further statistical research, such as parsing out big runs and even comparing correlations in the passing game.

Before continuing my future research, I wanted to do a quick aggregate of both total attempts per game and total yards per game. These two categories showed a higher correlation to winning than yards per attempt, so what if we looked at them together?

Below is a table that charts win percentage in relationship to both attempts and yardage. Take a look:

Rushing Statistics and Success Rate, 2002-2011

Rushing Statistics and Success Rate, 2002-2011

First and foremost, this was an outcome I did not expect at all. Logic told me that we’d see more of a “diagonal” relationship with win percentages. In other words, the higher the yards and attempts combined, the better the percentage.

But that didn’t happen at all. Instead, we see a reinforcement of the idea that total attempts matters more than anything else. If you look vertically across the chart, you’ll notice that the success rate is more consistent by column rather than by row or diagonally.

Honestly, one of the things that really stunned me was the 97.14% success rate where teams had only 60-79 yards and 30-34 attempts in a game. That’s at worst a 1.76 YPC and at best a 2.63 YPC.

I’d also like to note that most of these percentages are based on at least 100 games that match the criteria. So we’re not really looking at small sample sizes across the board.

I’m interested in hearing your comments. What do you make of this, and how does it apply to our ongoing conversation?