Are Fumble Recoveries Actually Pure Luck?

This post is inspired by an interaction on X, where some fans mentioned that they didn't think our NFL Defensive Player Ratings were valuable because they included fumble recoveries, which they argued were pure luck and not correlated with player ability. 

Effectively, they argued that fumble recoveries were purely random regarding which player recovers the fumble. So I dug into the research, and most of the same sources reiterated that fumble recoveries as a team, not as individuals, were random. In other words, the ratio of fumbles the offense has to the amount they recover is almost entirely determined by randomness.  None of this research was recent, with some dating from 2007. In addition, the study did not attempt to uncover if better defensive players were statistically more likely to recover fumbles. So I looked!  

I pulled player defensive player data from the 2018-2024 (recently completed) regular season (2018 was randomly chosen based on data availability, but I think 6 years of the most recent data should suffice), filtered down to essentially DT, DE, and LB (the same positions in our original NFL Defensive Player Rating), summed up stats by player, and explored correlations between FRs and other major defensive stats. 

TLDR: We test the thesis that players who get higher stats, like Tackles, Sacks, etc., and are therefore regarded as higher skill/impact players, also recover more fumbles. If this is true, we would expect to see a correlation between FRs and other stats. If FRs are purely random and player skill has no relation to the amount of FRs a player gets, there should be no correlation between FR and other stats.  Lo and behold, there were several medium-strength correlations. 

For correlations, here's a rough breakdown of the categories: 

0-0.4 = no correlation/weak

 0.4-0.6 = weak/medium

0.6-0.8 = medium/strong

 0.8 and higher = strong/very strong

Note that these are not definitive; they are only correlations (which don't necessarily imply causation), but I think the overall story makes sense and deserves further exploration from a better data scientist than yours truly.  The following graphic looks intimidating, but it's simple, and I'll break it down.  RED = Strong Correlation, WHITE = No Correlation

This chart shows the correlations of every metric with every other metric in the database, where all stats are represented on both axes.  On the x-axis, I've placed FRs first (furthest left) so we can go vertically up that column and explore every correlation.   There's a perfect diagonal becuase that's where each stat intersects with itself on the y-axis, where naturally, there is a perfect 1.00 correlation.  For example, If you move one column over,  you can see G (Games Played) has no color indicator when it intersects with itself on the y axis, but has a stong (0.8) correlation with Games Started, which makes intuitive sense.  Other intuitive relationships exist, such as weak relationships between INTs and Sacks. Because pass rushers are less likely to get INTs than LBs, who are in coverage more often, their relationship is weak (0.22). You can explore the correlation of any stat by going through the matrix in this fashion.  

Getting back to FR, you can see several correlations fall into the  0.6-0.8 = medium/strong category. Such as: 

• Games Played - 0.62

• Games Started - 0.67

• Pass Deflections - 0.64

• Forced Fumbles - 0.63

• Combo Tackles - 0.66

• Solo Tackles - 0.67

• Tackle Assists - 0.63

• Tackles for Loss - 0.66 

There were other weaker correlations like Sacks at 0.57, and QB hits at 0.55, but anything below 0.6 is not strong enough to really consider.  What was also interesting is the lack of correlation with INTs and INTtds.  

TDLR: Overall, these data suggest that players who make lots of tackles (and are consequently around the ball carrier often), are adept at forcing fumbles, and who get into the backfield often on running plays are more likely to recover fumbles than their counterparts.  

Since TJ Watt was our #1 ranked player last year,  we decided to look at some tape. Granted, some plays like this one certainly show that some FRs are lucky, where, as the Tweet mentioned, the opposite end knocks the ball loose to Watt's benefit. Still, there are many more plays where one could argue that Watt's pressure, proximity to the ball, and ability to force fumbles contribute to his total FRs throughout his career.  In particular, Watt's league-leading ability to Force Fumbles (32 since 2018) certainly contributes to the number of FRs he can achieve since he can either gather the fumble or is at least close, and the tape seems to back that up.  

We pulled the top 10 FR getters since 2018, and you can see that the correlations aren't perfect. This certainly isn't a who's who of the NFL, but the list does have some big-name defensive players.

In conclusion, FRs certainly have a significant element of luck to them.  They are in a different category of stats, from sacks, tackles, and others, which more directly display a player's actual ability and impact on the game. Still, this analysis has shown that above-average players are more likely to achieve FRs, indicating they should receive some type of credit for recording an FR.  The way the SSAT system works is that we essentially devalued FRs relative to other stats because there is so much luck involved.  We tried to place emphasis on more central stats like Sacks, Tackles, and TFLs. Thus, we stand by our decision to include FRs as just one small component in evaluating a player's performance for the season. 

That being said, 2024 SSAT NFL Defensive Player Ratings, coming soon!