Writers of Pro Football Prospectus 2008

Most Recent FO Features


» 2016 Slot vs. Wide: Defense

Our final part in looking at slot vs. wide targets focuses entirely on defenses. Which first-round rookie corners had tough times in 2016? Who was the weak link in Denver's No-Fly Zone? Also: the Falcons have a secondary to watch out for in 2017.

22 Apr 2004

Guest Column: Turnovers and the Unpredictability of Defense

Guest Column by Jim Armstrong

During the football season, we often hear the media talk about the turnover battle. Winning the turnover battle is always one of the keys to winning football games. But what does this really mean? How important are turnovers and how much control do teams have over them? Rather than looking at turnovers on a game-by-game basis, it's perhaps more useful to step back and take a broader look at teams' general turnover tendencies over the course of one or more seasons. In this article, I will explore how closely a team's tendency to give up or force turnovers correlates with winning relative to its ability to move the ball and score points, how offense and defense differ in this respect, and how predictable a team's turnover tendencies are from one year to the next.

To compare turnover rates to other measurable aspects of offense and defense, I looked at drive charts from all NFL regular season games from 1998-2003, a six-year period. I computed yards per drive, points per drive, and turnovers per drive for each team's offense and defense for each of the six years. Then I computed the correlation coefficient of these values with respect to that team's number of regular season wins. The table below shows the average yearly correlation of each of these rate stats.

  Offense Defense
Yards/Drive .553 -.415
Points/Drive .692 -.654
Turnovers/Drive -.393 .531

We see that, in terms of yards and points, offensive performance correlates slightly better with winning than defensive performance does. But a team's turnover rate per drive seems to be more important on defense than on offense. That suggests that it's particularly important to have a defense that can "force" turnovers. But to what extent can a defense force turnovers? Isn't a big part of forcing turnovers just being in the right place at the right time, or random luck?

To help answer this, I looked at year-to-year correlations of the drive rate stats over the same six year period. There were 155 pairs of consecutive team-seasons during this period. Recall that the Cleveland Browns re-entered the NFL in 1999 and the Houston Texans entered the NFL in 2002. Using this data, I computed the correlation coefficient for each of the drive rate stats mentioned above. The idea behind year-to-year correlations is that if a particular measure closely represents actual ability, then that stat will hold relatively consistent from one year to the next. However, if there is a lot of variation from one year to the next, that suggests that performance measured by the statistic is largely due to luck or other factors. Thus, the lower the correlation, the less predictable it is, and the more it can be attributed to luck.

However, we have to be careful when making these conclusions at the team level because, especially in today's NFL, teams change players and coaches every year, and hence a team's general ability does change from one year to the next, probably more so than that of an individual player. So it's important not to view these correlations only in absolute terms, but to view them relative to the other yardage and points-based correlations.

  Offense Defense
Yards/Drive .484 .360
Points/Drive .455 .280
Turnovers/Drive .296 .069

Here we see that offensive performance is much more predictable from year to year than defensive performance by any measure. And by far the least predictable of these drive rate stats is the defensive turnover per drive rate. This suggests that defenses have relatively little persistent ability to force turnovers. Perhaps if we were to dig a little deeper, we might find a few teams or players who do have such persistent ability to some extent, but in general it's not at all an ability that is pervasive throughout the NFL. If a coach can teach such an ability, it's probably so rare that most teams can't do it to a significant degree year after year.

But there is some possible bias in these turnover rates. Both interceptions and fumbles are lumped into the turnover category, but interceptions can only occur during a pass play. And how often a team passes is somewhat dependent upon the game situation--a poor team which falls behind quickly has to resort to passing more often in an attempt to catch up, thus increasing its risk of a turnover. To account for this, I decided to break down turnovers further by looking at fumbles lost per play and interceptions per pass play. The numbers used to compute these are more widely available from NFL.com and various annual publications. I computed these stats for each team over the same six year period of 1998-2003.

First, the correlation of these rate stats with winning.

  Offense Defense
Interceptions/Attempt -.400 .358
Fumbles Lost/Play -.326 .377

Here we see that interceptions seem to be a bit more important on offense, while recovering fumbles is a bit more important to the defense. But the differences are fairly small, and the correlations are generally lower than the drive-based correlations. Now here are the year-to-year correlations:

  Offense Defense
Interceptions/Attempt .300 .278
Fumbles Lost/Play .185 .073

Here we start to see a bigger difference between interceptions and fumbles, and it suggests that interception tendencies are definitely more persistent than fumble tendencies. This is particularly true on defense, perhaps because interceptions, to some extent, result from good pass coverage skills and a good pass rush, both of which are also key components of a good overall defense. But the tendency to force (and recover) fumbles on defense appears to be mostly random from year to year.

So what conclusions can we make from all this?

1. Defensive performance is significantly less predictable from one year to the next than offensive performance. Exactly why this is true is still a mystery. Perhaps it has something to do with the scarcity of talent among the players who play offense compared to those who play defense. Maybe injuries are a contributing factor, possibly because an offense can more easily exploit a weak link in the defense than the other way around. Or maybe there are some economic reasons that dictate inequalities and perhaps inefficiencies in player salaries. More research is definitely needed to shed some more light on this.

2. A defense's tendency to force turnovers is fairly important to the team's success, but it seems to be even more unpredictable. In general, a team's ability to force fumbles seems to be almost entirely luck. There is a little bit more persistence in a team's ability to force interceptions, though it isn't clear how much of this ability is just a residual effect of general defensive ability. Again, this only pertains to the team level. Whether there are some individual players with a special ability to force turnovers significantly above average rates would be an interesting subject for further study.

In any event, the implication of these numbers is that a general manager should be very cautious about attempting to build a perennial championship contender around a dominating defensive unit unless he has a deeper understanding of why defensive performance is so relatively unpredictable.

Editor's Notes from Aaron: Thanks to Jim for doing some really great analysis here.  Two more implications for these numbers that I find interesting:

1) Remember our article on the St. Louis defense from early March?  At the end, I wrote, "Note that compared to other teams that had a lot of takeaways in 2003, the Rams had a high percentage of fumble recoveries."  In fact, the Rams had 22 fumble recoveries in 2003; no other team had more than 17 fumble recoveries.  The Rams may be near the top of the league in interceptions again next year, but they won't recover 22 opposition fumbles again, which means the defensive outcome will suffer even if their actual performance doesn't get any worse.  That's good news for Seattle and bad news for the Rams -- as is the fact that the rate of interceptions thrown by the offense is somewhat consistent from year to year.

2) The fact that defensive turnovers are so unpredictable from year to year also explains why team defenses in fantasy football are so inconsistent from year to year -- and should never be drafted high.

Jim Armstrong is a Boston-area software developer by day and a Packers fan by birthright.  Plus, he was one of Football Outsiders' earliest readers.  (Thanks for your guiding advice in those first couple of months, Jim!)  If you are interested in writing a guest column, something that takes a new angle on the NFL, please email us your idea at info @ footballoutsiders.com

Posted by: admin on 22 Apr 2004



by John (not verified) :: Thu, 03/11/2010 - 4:39pm

Great stuff but I do have one question.

How is it that when correlating yrds, pts, and TOs, with winning for both the offense and defense, the strengths of the correlations for each category are not reflections of each other? How can this be?

For instance, if a defense "forced" a turnover and is .531 correlated with winning, how can that not be the same for the offense if they did not turnover the ball?

Shouldn't it reflect the value of a TO? If forcing a turnover correlates strongly with winning, in theory, so should not turning over the ball, with the same correlation strength.

Thanks so much!

P.S How does one get into this fun job anyhow?

by rania (not verified) :: Thu, 05/05/2011 - 12:34pm

It is not what they want ....you know every body has his own views and desires...
Halloween costumes 2011

by rochester john (not verified) :: Thu, 01/05/2012 - 12:25pm

Loved this article- my first read of FO but not my last!! I think what you said about the makeup of teams changing year to year has the most to do with poor year-year correlations in all categories.

As far as interceptions forced by defenses being more predictable, teams that have a good pass rush will tend to keep that intact year after year (eg- today's Giants), and that really does affect a QB's ability to throw in control.

Lastly, the discrepancy between the win correlations of yards allowed versus points allowed gives hope to the Pats & Packers fans that have witnessed historically high yards allowed by their defenses but not high points allowed.