Building a Super Bowl winner is the goal of every NFL franchise, but it’s a lot easier said than done. As I discussed last week, the Titans are a team approaching a crossroads when it comes to investing in their roster. General Manager Jon Robinson will face a slew of decisions over the next 12 months, and he will need to have a clear, comprehensive plan in place that guides him as he goes.
We will find out what that plan looks like as decisions start getting made, but I tend to believe that Robinson will be influenced by analytics. When he was hired back in 2016, Buccaneers Director of Football Operations, Shelton Quarles, told Paul Kuharsky that Robinson and Bucs GM Jason Licht helped build out Tampa Bay’s analytics department upon their arrival.
“Since both him and Jason have come in, we’ve started to develop our analytics department so I know [Robinson] relies on some of that stuff,” Quarles said. “But the eyeball test -- they can play or they can’t play and sometimes it doesn’t add up from an analytics standpoint. He’s able to decipher and decode and all that stuff through the knowledge that he has. He spent 12 years with the New England organization, and they produced a few titles there, a few championships, and I think he’ll bring that winning formula to the Titans.”
It stands to reason that he will likely use analytics, at least to some degree, as he really begins to shape the future of the Titans franchise.
Football is among the toughest sports to analyze with numbers. The sample sizes are often very small, especially when it comes to wins and losses, and the stats are often cluttered with noise. However, that doesn’t mean that some valuable information can’t be derived from looking at some key metrics over multiple years.
There are two different ways to look at success in the NFL — regular season wins and Super Bowl wins. While there is no debating that Super Bowl wins are the ultimate goal, it is also the more difficult to measure what makes a Super Bowl winner due to a much smaller sample size. We have just 53 Super Bowl winners to study over the history of the league and the evolution of the league has rendered around 30-40 of those virtually irrelevant as compared to the NFL landscape in 2018. Regular season wins is a far better metric in my opinion due to the fact that just looking at the 10 most recent NFL seasons give us 320 points of data to study.
It’s no secret that turnovers are important to winning in the NFL. How many times have you listened to a broadcaster list “win the turnover battle” as one of his [Insert Sponsor Here] Keys to the Game? It happens almost every game and for good reason. Winning the turnover battle is extremely important to winning.
Using 10 seasons worth of data, I plotted each team’s end of season turnover margin against their regular season win total on the graph below and fit a trend line to it.
Even without the trend line or the regression numbers that I am about to layout, you can probably already tell there is a pretty positive correlation. Getting more turnovers than you give up is good. Really groundbreaking stuff, I know.
When you run regression analysis — a mathematical process that helps determine if two sets of data are related and how strongly they are related — you get a correlation coefficient of +0.656. A correlation coefficient is basically the measure of how two sets of data are related. They fall on a spectrum from -1.0 to +1.0 with -1.0 indicating that the two stats are perfectly inversely correlated and a +1.0 indicating that the two stats are perfectly positively correlated. A coefficient of 0 would tell us that the two numbers have nothing to do with one another. What is considered “strong correlation” really depends on what you are measuring. In some fields where predictability is high and the data is very accurate, a +0.9 or even higher may be required to meet the threshold for “strong correlation”. However, football is far from predictable so a +0.656 is about as strong a correlation as you can find in the game.
Another way to look at this result is through the coefficient of determination, which is simply the square of the coefficient of correlation. In this case the coefficient of determination is 0.431. That tells us that 43.1% of the variance in a team’s regular season wins can be explained by turnover differential. That’s a LOT for one statistic in a game full of statistics.
So the numbers agree with the obvious statement that winning the turnover battle is important, but what else can we learn from these stats? Breaking it down further, I looked at both components of turnover differential — takeaways and giveaways — to see which one was more important to winning football games.
I started with takeaways and plotted those against wins just like I did for turnover differential above.
The coefficient of correlation this time was +0.450, not as strong as the turnover differential which makes complete sense. That tells us that about 20.3% of variance in a team’s wins can be explained by the number of turnovers they create over the course of a season. That’s a relatively strong correlation.
Then I plotted giveaways against wins to see which had the bigger effect.
The giveaways resulted in a coefficient of correlation of -0.550 (negative because fewer giveaways is obviously a good thing). That’s significantly stronger than the takeaways we plotted above and tells us that roughly 30.2% of variance in team wins is determined by how many giveaways they have over the course of the season.
That result tells us that protecting the football on offense is more important than creating turnovers on defense. Spoiler Alert: That won’t be the last time offense turns out to be more important than defense.
Turnovers are a tricky stat though when it comes to team building. How do you influence the numbers of giveaways and takeaways? The first and most obvious way is to find a quarterback who doesn’t throw interceptions. It probably won’t surprise you to find out that the Patriots — who have won an astounding 123 of a possible 160 regular season games over this time period — also averaged the least interceptions thrown by FAR from 2008 through 2017. New England has averaged just 8.7 interceptions thrown per season over the last 10 years. The 2nd best team at avoiding picks, the Green Bay Packers, have averaged 10.6. The Pats have thrown over 20% fewer interceptions than the next best team over that time frame! The league average at 14.7 picks is nearly double the Patriots rate.
How do you prevent losing fumbles though? That’s a little tougher, partially because recovering a fumble is almost always a 50-50 proposition. Sure, you may have some seasons with bad fumble luck where the ball just doesn’t bounce your way, but over a large enough sample size there is virtually zero skill involved in recovering fumbles. So the only way to truly control your fumbles is to prevent them. Once again, we come back to the quarterback. Out of 644 total fumbled footballs in the NFL in 2017, 275 of them — or 42.7% — fell out of the hands of a QB, easily the highest percentage of any position.
It should come as no surprise then, that the Patriots have also lost the fewest fumbles over the last 10 years, averaging just 6.9 fumbles that result in a turnover per season. There are plenty of Patriots conspiracy theories surrounding their suspiciously low fumble rate, but a big part of that number has to do with Tom Brady’s pocket presence and ability to protect the football. From a team building standpoint you can also teach and reinforce fumble avoidance through practice and snap distribution. Some non-QBs do fumble more than others, but those numbers tend to be very small. Likely just a difference of 1 or 2 more fumbles per season for the most turnover prone back compared to an average back. While 1 or 2 fumbles are certainly important, that gap is never as large as the same gap for quarterbacks. In 2017, Jameis Winston led all NFL players with 15 fumbles in just 13 games while only 6 other quarterbacks even managed to hit double figures.
Creating turnovers is an even more difficult challenge for a GM. Sure, a strip-sack artist pass rusher or ball-hawking defensive back help, but just having one or two of those players isn’t enough. For example, the Broncos ranked 6th from the bottom at creating turnovers in the past 10 years despite having two of the league’s most prolific fumble causers in DeMarcus Ware and Von Miller alongside Aqib Talib — an elite turnover creator in his own right — for part of this time frame.
Creating takeaways in football is an extremely high variance endeavor from year to year. Plotting each team’s 2016 takeaway numbers versus their 2017 numbers produced a correlation coefficient of +0.003 — which essentially suggests that there was no carryover in the ability to create turnovers from one year to the next. That falls in line with other studies that suggest turnovers are more than 50% caused by random luck.
That adds up when you think about how turnovers often occur. We know that recovering a fumble is basically a 50-50 proposition so each time a ball hits the ground, luck comes in to play. Further, how many interceptions are created by a deflected ball that just happens to land where a defender is standing instead of safely crashing to the ground in the open field?
Because of the amount of luck involved, there is no surefire way to rank among the tops in the league in turnovers year after year. However, the best way to influence turnovers is to generate pressure on the quarterback. In 2017, quarterbacks under pressure threw interceptions on 2.6% of dropbacks versus just 2.0% of dropbacks without pressure according to Pro Football Focus. That seems like a small number, but that’s a 30% increase in the likelihood of creating a turnover that’s before you factor in fumbles recovered by the defense as a result of strip sacks.
Offense vs. Defense
Another angle to think about from a team building standpoint is whether to invest heavily on offense or defense. To get a gauge on this I decided to use Football Outsiders’ Defense-adjusted Value Over Average — or DVOA — as my measuring stick for successful units. Stats like total yardage or yards per play are fine, but they lack much of the context that DVOA adds. The classic example being that a 3-yard rush on 3rd and 2 is far more valuable to a team’s success than a 3-yard rush on 3rd and 20. Total yardage treats both of these results the same, but DVOA does not. For more information about what DVOA is, you can get it from the source here.
Using the same method I used above for the turnover discussion, I plotted Offensive DVOA versus Regular Season Wins for each team going back to 2008.
Running regression analysis for this data resulted in the largest correlation coefficient among all the variables I compared in this study at +0.70. That tells us that having an excellent offense is extremely closely tied with winning football games.
That’s an expected result, but how does that compare to the defensive side of the ball?
You can see visually that the grouping of data points isn’t quite as tightly bunched around the trend line as they were for the offensive side. The correlation coefficient reflects that as well with Defensive DVOA netting just a +0.49 compared to the +0.70 for the offense. That’s still a strong correlation, but nowhere near the offense. When you look at the coefficient of determination numbers it shows that offensive performance is roughly twice as important when it comes to influencing the number of wins a team collects during the regular season.
It should follow that investing heavily on the offensive side of the ball makes far more sense than spending big on defense.
Evaluating how teams actually spend versus their success on the field further proves this idea to be true. I took positional spending data for each team over the last five years as listed by OverTheCap.com and normalized it versus the salary cap for each season. I then ran a regression analysis for offensive spending versus wins and defensive spending versus wins. Again, the offense wins out. Offensive spending resulted in a correlation coefficient of +0.27 versus just a +0.16 for the defensive spending. Neither correlation number is particularly strong — which tells us that spending does not always yield success among other things — but it does show that spending on offense tends to be a better habit than buying expensive defenders.
Part of the reason for this has to do with the nature of offense versus defense. Offense is far more reliant on the execution of a detailed plan, while defense is about reaction. That allows offensive players at most positions to play at a high level far further into their careers than most defenders. Quarterbacks are playing well in to their late 30’s with ease now. Offensive linemen excelling well in to their early 30’s is not uncommon either. Take a look at the aging curves below from really good study by So Called Fantasy Experts.
As you can see, defensive positions almost universally begin a steep decline around age 29. This must be accounted for in roster building because most players come to the end of their rookie contract around age 26 or 27. Investing heavily in a second contract for a defender generally results in paying big money for a declining player towards the end of that deal.
From an analytical perspective, it makes far greater sense to invest in long term assets on offense than it does on the defensive side of the ball.
Passing Attack vs. Ground Game
If you know the modern NFL, you can probably guess which way this evaluation is going to lean, but you might be surprised exactly how big of a lean it is.
Using the same methodology as before, I plotted Passing DVOA versus wins for every offense since 2008 and ran regression analysis against it.
The result was pretty amazing. The correlation coefficient between Passing DVOA and wins was a whopping +0.70, just a fraction below the correlation for the entire offense. Running the same study for Rushing DVOA yielded very different results.
Rushing DVOA correlated to wins with just a +0.38 coefficient. When you extrapolate out the coefficient of determination — or the amount of variable A that can be explained by variable B — you find that rushing offense has less than a third of the effect on wins that passing offense does.
It’s not surprising that passing is more important than rushing, but these numbers suggest that having a successful rushing attack is virtually irrelevant if you don’t have a successful passing game to go with it.
Looking at the opposite side of the ball the trend holds there as well, but it’s not quite as drastic. Defending the pass is clearly more important than defending the run, with a correlation coefficient of +0.49 compared to +0.30 for run stopping.
One frequent argument in defense of the importance of a running game is that a successful rushing attack will open up opportunities in the play action passing game. However, Ben Baldwin wrote an excellent breakdown for Football Outsiders showing statistically that a strong rushing game has no effect on the success of play action passing.
Obviously, no one would suggest abandoning the run altogether, but when it comes to an overall team building strategy it makes far more sense to focus resources on players that most improve the passing attack.
Finding and keeping a franchise quarterback is still the single most important thing an NFL team can do and that should come as no surprise. The Titans think they’ve got one in Marcus Mariota, though its certainly fair to want to see more before they fully commit to a long term, bank-breaking deal. However, if Mariota bounces back with a good season in 2018, you pay him and don’t look back.
The team should also be more than willing to pay Taylor Lewan and keep him in Tennessee long term. As an elite offensive player who plays a position with a favorable aging curve, it’s very likely that they will get good value out of that contract even if they make him the highest paid offensive lineman in the league. The same question will eventually come for Jack Conklin as well, but that is still a few years down the road for now.
While the numbers show that investing in defense does not yield the same return as spending on offense, Kevin Byard also seems like a wise investment for the team. Safeties have a favorable aging curve compared to other defensive positions and his ability to create turnovers is extremely valuable to a team.
The spots where the team may hesitate to spend big money would seemingly come at positions that primarily impact the run game or run defense. A running back who doesn’t contribute much in the passing game, for instance, is typically not a wise long term investment. Look at the shape of the aging curve for running backs in the charts shown above. Running back is the line that starts the highest on the left and finishes the lowest on the right. It is a position where players take little time to get up to speed, but tend to fall off relatively dramatically as they reach their late 20’s. NFL teams know that and its reflected in their reluctance to pay for expensive second contracts at that position. It’s generally better asset management to draft a young back, maximize his performance under a cheap rookie contract, and then let him walk so you can draft another one before the decline hits. Sure, you may be missing out on a year or two of his peak, but you can spend that money to greater effect elsewhere while getting nearly the same level of play from a younger back.
That analysis could come in to play with regards to a possible second contract for Derrick Henry. We expect to get more of a taste for what Henry can do in a full time role in 2018, and hopefully he can expand his effectiveness in the passing game under a new coaching staff. If the Titans pay top dollar for Mariota, Lewan, and Byard, it may be tough to justify a big contract extension for Henry, even if he runs well over the next two seasons.
The most important lesson that I learned from this exercise was that the NFL is 100% a passing league. Nearly every valuable metric points directly back at the quarterback and a team’s ability to throw the football successfully. So it should follow that the best way to build a sustainable winner is to invest heavily at positions that most effect that aspect of the game.