Filed under:

# Designing the Perfect NFL Team (Part II)

In Part I, I hypothesized what the ideal professional football team would do well. In Part II, I review the results of my analyses to determine how likely (or unlikely) my hypotheses were to be correct. This post is 1800+ words, so all the TL;DR folks can skip to the "Conclusions and Closing Thoughts" section at the bottom.

Welcome to the thrilling conclusion of my study to determine what the perfect NFL team would look like. In Part I, I offered a brief overview of my goals with Crunching The Numbers and my methodology for the 2016 formula. I also posed twelve hypotheses - six on offense and six on defense - about what successful NFL teams do well to help them win consistently. Read on to see the method for my analysis and the ensuing results.

#### The Analysis

Like in past years, I used a simple Pearson correlation coefficient to determine the relationship between how teams performed in each metric that I tested and the number of wins those teams had. Since football is a constantly evolving sport, I only went back three seasons (2013-2015). I used the average of each team's performance and wins over those three seasons to assess the correlation.

But what's a "Pearson correlation coefficient"? Simply put, it is a number between -1 and 1 that quantifies whether or not a relationship exists between two sets of data (we'll call them Set A and Set B). A positive coefficient implies a positive relationship - that is, as Set A increases, Set B also increases. The easiest example is offensive scoring. The more points your team scores, the more likely they are to win. Conversely, a negative coefficient implies a negative relationship - as Set A increases, Set B decreases. The easiest example therefore is defensive scoring. The fewer points your team allows, the more likely they are to win.

You've probably heard the phrase "correlation does not imply causation," and that holds true here. Depending on your sample size (32, in this case) a correlation coefficient must cross a certain threshold in order to be considered significant. This is referred to as a "confidence level" and is expressed as a percentage. In engineering circles we like to us a confidence level of 95% so that is the standard for each metric in this study.

Previously, I've gotten into the nitty gritty of the coefficients themselves. This time around, I'm only going to display the confidence level of each metric. It gets the same point across and is much easier to digest from a layman's standpoint. If you are interested in the actual numbers, give me a shout out in the comments and I'll be more than happy to oblige. But enough geeky banter. Let's move onto the stuff that really matters.

#### The Results: Offense

Refer to Part I for a rationale on why each metric was chosen.

##### Yards Per Pass Attempt

Correlation Confidence: +99.96%
Starting off on a good foot! As was expected, yards per pass attempt had a strong positive correlation with winning: the higher a team's YPA, the more often that team won. The reasoning behind this is clear; the more bang for your buck you get every time the quarterback drops back, the more successful your offense (and team) will be.

##### Interception Percentage

Correlation Confidence: -99.99%
Another no-brainer. The less often the quarterback gifts a ball to the opponent, the more often his team wins. We all good here? Next!

##### Sack Percentage

Correlation Confidence: -96.27%
Not quite as strong as the first two, but it still meets our 95% threshold for significance. The words of the late Buddy Ryan ring true: "No quarterback ever completed a pass lying flat on his back."

##### Fumbles Per Game

Correlation Confidence: -96.64%
Discipline is the name of the game here. Take care of the rock and the scoreboard will wind up in your favor more often than not. Enough said.

##### Rushing First Downs Per Game

Correlation Confidence: 99.99%
As increasingly archaic as the running game has become, it still has its uses, and it shows here. Being able to convert first downs on the ground with a healthy lead is instrumental in closing out games, and I still think part of it comes from the ability to set up third-and-short early on in the contest. If you're wondering whether or not this correlation is caused by the fact that better teams simply get more first downs overall than bad teams (and therefore would get more rushing first downs), this is something I will investigate as I do more analysis to prepare for the upcoming season. So stay tuned!

##### Scoring

Correlation Confidence: 99.99%
Even though this technically has the same confidence level as rushing first downs per game, in terms of raw correlation coefficient value scoring had far and away the strongest relationship, and for obvious reasons. The more points you score the more you will win, plain and simple.

#### The Results: Defense

Refer to Part I for a rationale on why each metric was chosen.

##### Yards Per Pass Attempt Allowed

Correlation Confidence: -99.99%
A negative correlation is good here, since it implies that the lower yards per attempt you allow the more you will win. The almost perfect confidence level was not a shock given the heavy emphasis on the passing game in today's NFL.

##### Interception Percentage

Correlation Confidence: 63.88%
The first stumble. While a 64% confidence implies some kind of meaningful relationship, it falls far short of the 95% confidence level I'm using as my threshold. This seems counter-intuitive given the significance of turnovers, but it does mean I can't use this metric in my formula if I am to uphold my integrity. I'll discuss possible reasons for the relatively low confidence level in the conclusions section below.

##### Sack Percentage

Correlation Confidence: 10.56%
And now for the shocker. As much as we all love sacks, they don't really seem to matter as much to winning as of late. Again, the conclusions section below will have some speculation on the surprising relationship (or lack thereof) between sacks and winning.

##### Fumble Recovery Percentage

Correlation Confidence: -52.56%
As I admitted in Part I, this was my biggest reach and I paid the price for it. Not only does recovering fumbles not correlate with winning, this indicates that teams who were better at recovering forced fumbles actually did worse overall than teams who struggled in this area. Let's just chalk this one up to a soundly disproved stretch and move on.

##### Third Down Conversion Percentage

Correlation Confidence: 99.99%
Back on track! In Part I I said that one of the defense's primary objectives was to get off the field, and that holds up here. We don't have to go any further into this, right?

##### Scoring

Correlation Confidence: -99.99%
Like offensive scoring, defensive scoring easily held the strongest relationship, and it wasn't even close. And it's not hard to see why: the fewer points you allow, the greater your chances of winning. Duh.

#### Conclusions and Closing Thoughts

If you're keeping score at home, all of my offensive metrics (yards per pass attempt, interception percentage, sack percentage, fumbles per game, rushing first downs per game, and scoring) easily surpassed the 95% confidence threshold. I was not as successful on defense, where only yards per pass attempt allowed, third down conversion percentage, and scoring showed a confidence level greater than 95%. Interception percentage was a middling 64%, sack percentage interestingly only finished at 10.6%, and fumble recovery percentage was a disaster at 52.6%... in the wrong direction. Clearly I'm going to need to do more work in this area before I can put together my formula.

What can all of this teach us? First and foremost, offense seems to pass the "gut check." That is, the statistics that seem to be intuitively important to winning, like sacks and interceptions, bear out the expected result when put to the test. On the other hand, the corresponding defensive statistics which we seem to place an equal emphasis on don't show the same relationship. What gives?

In my opinion, the key to this is third down conversion percentage. A raw metric like that does not discriminate against how that number is achieved. It is only concerned with the results. Whether you stop the opponent on third down by stuffing the box on a run, deflecting a pass in the secondary, or dropping the quarterback to the turf is irrelevant. The only thing that matters is that the conversion was prevented. This focus on the end result becomes the key difference between offense and defense. An efficient offense requires clean, consistent execution and discipline. Ball security and sound decision-making are highlighted on every play, because every play has the potential to become a mistake that the defense can capitalize on. The defense is much more concerned with the "bigger picture." You have the same desired end - prevent the opponent from scoring. But with highly sophisticated offenses in today's NFL, each opponent will attempt to score in a unique way. This requires the defense to adjust its strategy week in and week out, mitigating the impact that "process" metrics like interceptions and sacks have on the overall result.

DEFENSES FIND SUCCESS BY FOCUSING MORE ON "THE BIG PICTURE."

For example, let's look at sacks. On first glance, the weak connection between sacks and wins might suggest that generating pressure doesn't matter in modern football. But that would only hold true if the only possible outcome of pressure is a sack. Of course, there are many things that pressure can cause - incomplete passes, interceptions, fumbles, tackles-for-loss (if its a run play), just to name a few. And all of those outcomes, along with sacks, help achieve the objective of preventing a score. Focusing on sacks is missing the forest for the trees, and while we may not want to hear that with Jim Schwartz in town, keep in mind that in his quotes he never singled out sacks. He said he wanted "to attack and be aggressive" and "create negative plays." He's looking at the big picture with his scheme, which is the mindset that you want for your defense.

With all that in mind, where is the way forward with the formula? The most logical approach is to follow the evidence, and the evidence right now suggests that I should distance myself from "process-driven" metrics and focus more on statistics that highlight results. In the coming weeks, I will look over the other statistics available to me from my source, www.teamrankings.com, and pick three new ones to replace the three that failed to meet my criterion of 95% confidence. Look for the end results of this study to be revealed when I present the new formula for Crunching The Numbers in August, and best of luck surviving the rest of the offseason... we're all in this together.

---

(Note: I feel compelled to address yards per pass attempt on defense since I have a feeling that it will be pointed out in the comments as a "process-driven" metric with a strong connection to winning. And I won't deny that, although I contend there is flexibility in how a low YPA is achieved, just as there is flexibility in how a low third down conversion percentage is achieved. Overall, there are usually exceptions to every rule, and with the huge role that quarterbacks play in the game YPA seems as good of a candidate for an exception as any.)