The Chip Kelly and Bill Parcells Connection

It’s odd to say, but Chip Kelly may be influenced by Bill Parcells. On the surface, it seems a rather ridiculous statement to make. Parcells was an authoritative disciplinarian over both players and coaches. He was crass and curmudgeonly. Kelly, on the other hand, appears to be more of a collaborator, a coach with distinct ideas who is willing to delegate thoughtfully. But in the NFL there is often an ebb and flow of trends and philosophies, where a stream of thought here was originally a trickle over there. It is no different between Parcells and Kelly. A common thread attaches one to the other. For them, it all comes down to offensive plays.

According to Michael Lombardi of the NFL Network, Parcells measured offensive success by the number of rushing attempts plus pass completions. If the number was greater than or equal to 51, then Parcell’s teams most likely put themselves in the best position to win. Kelly has a similar idea; he also places a priority on offensive plays. In his white paper titled “Efficient use of Practice Time”, he wrote, “We base the success formula for offense on the total number of plays. Take those plays minus the dropped balls, offensive penalties, and negative yardage plays, and divide by the total number of plays. If the answer is 80% or better, you win the game.” Given Kelly’s desire to maximize the number of offensive plays in this fashion, perhaps there is something to Parcell’s “magic number.” Let’s examine…

Thanks to ProFootballReference.com, I compiled data from the last five years of Eagles games (n=80) and compared game outcomes when the offense reached, or did not reach, that magic number. Next, I determined if there was perhaps a better magic number. Lastly, I measured how strong the relationship is between the magic number and winning individual games, and since the magic number incorporates both rushing and passing, gauged the importance of the magic number with a balanced attack.

In the last five years (not including playoffs), there were 53 games in which the Eagles did not exceed 51 rushing attempts plus completions. Of those 53 games, the Eagles won 24 of them (45%). In the 27 games in which the Eagles offense did meet or exceed 51 rushing attempts and completions, they won 18 (67%). This is good evidence in favor of the magic number, but hardly conclusive. Parcell’s magic number may be just that… Parcell’s. So I looked at the Eagles results with various magic numbers, sliding them up and down the scale. It turns out that the Eagles’ magic number in the last five years was 55. In the 64 games they did not reach 55 rushing attempts and completions, they won 45% of the time. During the sixteen games in which they did, they won 81% of the time, which is rather impressive. Over the length of a season, that percentage projects to a 13-3 record. However, is it statistically significant?

Logistic regression can be used when we want to see the strength and significance between some independent variable (or variables) and a dependent, dichotomous outcome. In other words, we want to see how the magic number for each Eagles game in the past five years relates to wins and losses. If results are statistically significant with at least 95% confidence, the logistic regression model will produce a p-value less than .05. If results are strong, a statistic called an odds ratio, which in this case will represent the ratio of winning odds to losing odds, will be high.

According to the table above, when the Parcells magic number is measured against the last five years-worth of Eagles games, the result is not statistically significant and needs to be ignored. As the magic number increases, however, so does the odds ratio, which reaches an apex at 55. After that we begin to see diminishing returns. The odds ratio of 5.2299 at magic number 55 indicates that, with 99% confidence, the Eagles were over five times more likely to win a football game if they met or exceeded the magic number than when they didn’t.

Lastly, since the magic number involves rushing and passing, we can use logistic regression to determine the importance of a balanced attack within this model. Put another way, will teams who complete 45 passes and rush ten times have the same odds of winning as a team that completes 30 passes and rushes 25? To better control for game context, I created a statistic, “Attempts from Balance”, which is the number of attempts (rushing or passing) needed to achieve a one-to-one run/pass ratio during the first through third quarters only. The assumption here is that rushing attempts in the fourth quarter are inflated during games in which the Eagles were protecting a lead, and passing attempts (and completions) were inflated when playing from behind. The table below represents the winning probabilities resulting from this two variable logistic regression (where the model p-value is .0052 – statistically significant with 99% confidence).

The “Null” value indicates that the Eagles never played a game in which they did not meet the magic number (55) and the run/pass ratio was perfectly balanced. Itself an intriguing result. Likewise, these results seem to illustrate that a balanced attack is extremely important. As an attack is more balanced within the first three quarters, and the magic number meets or exceeds 55, the odds that the Eagles will win are extremely favorable, better than 85%.

Favorable odds are nice, but there is a small matter of execution. For Eagles fans, Chip Kelly’s offense is still a nebulous entity. There is no definitive quarterback, there are questions on the offensive line, and general concerns about desire. But given a few clues and some level of analysis, the overall picture might be getting clearer. The Eagles of Andy Reid past have reached low tide, and now the tide rises again. Kelly’s offense will be fiercely fast, the number of plays robust, and the attack balanced. And if it’s successful, it’s funny to think we could have Bill Parcells to thank.