Passes and Patterns: Given the Boot: The Analytics of Kickoffs

Abstract

The third piece in a series on the current state of affairs regarding analytical research into football’s kicking plays, this piece looks at kickoffs. The broad conclusions to draw from the corpus are that the rule changes regarding touchbacks have had a major impact in reducing the number of kickoff returns and negatively impacting the starting field position after a kickoff; and that onside kicks are underutilized. Their success rate is almost entirely dependent on whether or not the kicking team is obliged by game situation to attempt an onside kick, and non-necessary onside kicks have a far higher success rate, high enough that their use is currently an under-exploited area of opportunity.

Introduction

Compared to punts (Clement 2018b), which are the subject of considerable new work trying to better define what makes a good punt, and field goals (Clement 2018c), which have been studied ad nauseam, kickoffs are little-studied. In principle they bear a great resemblance to punts, and the same methods for assessing the value of a punt could be transferred to kickoffs. Furthermore, since the circumstances of a kickoff vary far less than those of a punt, being nearly always from the same field position, no snap, and no variation in the distance-to-gain, kickoffs should be far simpler to evaluate with the same core metrics - flight time and flight distance, with some consideration for placement. And yet, there has been no meaningful research into this. The two areas of focus are the impact of rule changes over the last decade, with an eye to reducing the number of kickoff returns in the name of player safety, and the success rate of onside kicks and thus their recommended frequency.

Kickoffs

Touchbacks

Prior to the 2011 season, the NFL moved kickoffs from the kicking team’s 30-yard line to the 35. Prior to this touchbacks occured on 11.3% of kickoffs , but this rule change cause an immediate jump to 45.5% (Kacsmar 2014). Prior to 1994 kickoffs had also been from the 35-yard line, with a touchback rate of 25.3%. Research consistently shows an improvement in the strength and accuracy of field goal kickers, consistent with these results (Clement 2018c). As we see in Figure 1, starting field position for drives after kickoffs plunged by 5 yards wit the implementation of this rule.

Figure Field Position after a Kickoff by Year (Kacsmar 2014)

When the NCAA opted to move touchbacks after kickoffs to the 25-yard line and to allow returners to call fair catches within the 25-yard line and for the ball to then be spotted at the 25-yard line this brought the NCAA’s effective touchback rate to 56.4% (Connelly 2018). The rule changes are having their intended effect - they are reducing the number of kickoff returns in the name of player safety. The question then is how far can or should kickoff returns be reduced, or if the better option is the removal of kickoffs altogether. Proposals exist to remove kickoffs, while retaining the notion of the onside kick, proposals which may gain traction as kickoffs fall out of favour. Even without these rule changes the combination of these rule changes, the improvement in kickers’ leg strength and additional rules to limit kickoffs such as restricting the available alignment of players on the kicking team and removing the run-up to the kick have caused an ongoing trend of increasing touchback rate in the NFL as both kicking and receiving teams are incentivized to avoid kickoff return. The kicking team almost always has the ability to create a touchback, and the return team also has the ability to create a touchback. If we try to imagine a scenario that would generate a return the ball would have to be fielded at a point where the returner believes he can get past the 25-yard line. But poor kicks aside, the kicking team will only kick the ball in a position to be returned if they feel that they can tackle the returner before the 25-yard line. Imagine each kicking team has a line behind which they feel they can tackle returner before he reaches the 25-yard line, and each returning team has a line beyond which they feel they can return the ball past the 25-yard line. Only those kicks caught between these two lines will be returned. Progressive rule changes shrink this window, and improved kickers will put more hang time on kicks, further limiting the window.

Onside Kicks

Where traditional kickoffs receive short shrift and most discussion regarding them focuses on efforts to eradicate them, onside kicks are a more interesting area of research. While onside kick rate is clearly inversely correlated to win probability, shown in Figure 2, we also see in Figure 3 that onside kick success rate is inversely correlated to win probability (Burke 2009). Of course the first result is fairly self-evident, as teams with low win probabilities need to create opportunities to score and come back, and the second conclusion follows that onside kicks are far easier to defend when they can be anticipated.

Figure 2 Onside Kick Rate by Win Probability (Burke 2009)

Figure 3 Onside Kick Success by Win Probability (Burke 2009)

Where the matter becomes interesting is in the discussion of onside kick rate vs. success rate for what we might call “non-necessary onside kicks.” Seen in Figure 3, the success rate for these kicks is around 60%. While we know that this success rate declines precipitously as the onside kick becomes more predictable, the question is thus what is the optimal rate of non-necessary onside kicks and is that rate higher than the current rate? From Burke’s calculation (Burke 2009), the break-even conversion rate for an onside kick is 42.4%, far lower than 60%. Of course, any team that attempts more than the occasional non-necessary onside kick will see returning teams promptly adjusting their style to prevent this, but then we would have to consider the knock-on effects of this adjusted return scheme on the returns of ordinary “deep” kickoffs.

Onside kick success rates may be influenced by individual team and kicker skill, but even looking at seven entire seasons broken down by team does not come up with a large enough sample size for any individual team to so much as shrink the confidence interval to reasonable levels (Pelechrinis 2016). Moreover, a seven-year interval for a given team is highly unlikely to encompass a consistent set of personnel, kickers, and coaches. We do find evidence that suggests that onside kicks in the first three quarters of a game have a higher success rate (Pelechrinis 2016), seen in Figure 4. This is consistent with the idea that success rate is inversely correlated to win probability, as win probability tends toward the extremes as the game nears its end. An interesting secondary conclusion may be that teams who find themselves trailing by a significant margin should attempt onside kicks earlier, not so much driven by the necessity but by the notion that an onside kick in the third quarter where the success rate is higher may avoid the necessity of an onside kick in the fourth quarter with a much lower success rate.

Figure 4 Onside kick Success Rate by Quarter (Pelechrinis 2016)

In fact, the underutilization of non-necessary onside kicks is well established. Defining from parsed play-by-play data can give us a good estimate of what kickoffs are onside attempts according to the length of the kick, and whether they are necessary or not by game context, but this effort was followed by watching every kickoff from the 2009 NFL season to determine by eye which kickoffs where in fact onside attempts (Urschel and Zhuang 2011), and to decide if the kick was necessary or non-necessary (expected or unexpected in the parlance of the original researchers. Thereafter a similar methodology to those above was implemented: Determine the average EP value of an ordinary kickoff, determine the EP of a successful and unsuccessful onside attempt, and determine the conversion rate of non-necessary onside attempts. The use of EP in these calculations over field position is obligatory. Because either team may end up with possession on the play, and because the kicking team recovering possession is not just infrequent but central to the discussion at hand, we cannot consider use field position as a proxy for value because the two sets of possible field position outcomes exist in two subsets and cannot be equated.

Notably, Urschel and Zhuang (2011) use only “raw” Expected Points (EP), which is quadratically correlated to field position, instead of ordinary EP, which is roughly linear with respect to field position (Clement 2018a). They also consider scoring plays in terms of the actual points scored, the set {-8, -7, -6, -3, -2, 0, 2, 3, 6, 7, 8}, rather than make the typical adjustments for the subsequent change in possession following a scoring play (Clement 2018a). Adjusting the scores in this context would likely only make slight changes to an already belaboured exercise, as there exist already myriad EP models that could have been substituted to simplify the analysis. The use of raw EP is more questionable, since it ignores downstream effects of field position and significantly undervalues field position near midfield relative to regular EP models. Given that they are looking at non-necessary onside kicks, which occur far more often earlier in the game it is important to consider how these decisions can have a more lasting impact. While they perform a linear regression on their EP model as divided into 10-yard bins, the data is clearly non-linear, as we see from Figure 5 derived here from their results in Table 1. Unsurprisingly, their R2 value was very low, a 3.4%. Criticisms aside, Urschel and Zhuang do come to the same conclusion as other works, that surprise onside kicks are underutilized, though the methodological choices make it impossible to estimate the extent of the lost value.

Figure 5 Raw EP for 1st & 10 by Yardline (Urschel and Zhuang 2011)

Field position bin (yards)	Number of cases	Mean Points Scored
0-9	32	0.88
10-19	362	1.11
20-29	1433	1.35
30-39	464	1.59
40-49	177	1.78
50-59	36	2.03
60-69	40	2.33
70-79	16	4.38
80-99	14	4.71
100	18	7.00

Table 1 Mapping field position of the receiving team (after a regular kickoff) to expected points scored during the initial post-kickoff possession. (Urschel and Zhuang 2011)

Taking a more analytical look at the problem with a dynamic programming method, Sackrowitz (2004) developed a table of when to attempt an onside kick based on the estimated number of possessions remaining and the score differential. The result is very much an approximation, mostly based on average probabilities of scoring plays after receiving a kickoff, and neglects the impact of field position caused by an unsuccessful onside kick attempt, but it does suggest that in some cases a team that is leading late in the game should attempt an onside kick in order to give itself a chance to secure possession and end the game rather than kick for field position and play defense.

Regarding various proposals to eliminate kickoffs and replace them with a 4th down attempt in one’s own end, the exact distance and field position of the 4th down attempt should be determined with an eye to maintaining the current balance. Converting the 4th down attempt should be of similar probability to current attempts to convert an onside kick, and a successful conversion should result in similar field position. From Figure 6, the 4th down attempt would be approximately 4th & 15 from the -35 yard line (Pelechrinis 2018).

Figure 6 P(1D) of 4th Down by Distance vs Onside Kick Success Rate (Pelechrinis 2018)

Conclusion

While kickoffs have been the means of putting the ball in play in football since its inception, there is a consensus that kickoffs are more injurious than scrimmage plays (Yard and Comstock 2009; Houck et al. 2016). There have been significant rule changes to the kickoff, including moving the placement of the kickoff, the placement of a touchback after a kickoff, and allowing for a fair catch to be treated as a touchback. Each of these have served to reduce the number of kickoff returns. As a result, the kickoff risks becoming a defunct play, a vestige of an earlier era. The kickoff remains in large part because of the need to preserve the onside kick, as comebacks would become difficult to impossible late in games if a trailing team could not steal a possession. Many of the rule changes have had the knock-on consequence of making onside kicks more difficult to convert.

The more philosophical question is what equivalent onside kick success rate is desirable? Are more onside kicks a desirable thing for football a a whole? For example, if onside kicks became much easier to recover, to the point where non-necessary onside kicks became relatively common, would we see a reduction in the number of kickoff returns? Kicking teams would have the option to attempt an onside kick, with relatively lesser impacts, or to kick deep for a touchback. As returning teams would adjust their alignment to better defend against onside kicks, they would be less well-positioned to return the occasional deep kick that does not become a touchback, and would be inclined to take advantage of the fair catch rules.

Kickoff returns are in decline, both by design and by the improved performance of kickersThe value of predictive analytics may be growing in general in football, but the variables involved in modern kickoffs are shrinking rapidly.