By Michael Bond '25
Introduction
When the game is on the line, few players face as much pressure and public scrutiny as kickers. Some of the most memorable games have ended with similarly legendary field goal attempts, from Justin Tucker’s record-breaking 66-yard make to Cody Parkey’s “Double Doink” miss. Certain players may handle the pressure better than others, and the clutch factor could be an important consideration when front offices look for their next kicker. This paper aims to evaluate kicker performance at the end of close games in order to determine whether performance significantly changes in clutch time and discover which players have historically stepped up their game when it matters most.
Data and Methods
NFL play-by-play data was acquired using the nflfastR package. For this project, field goal attempts from the 2000 through 2022 seasons were included. In order to more fairly assess kicker performance, blocked kicks were removed from the dataset. Next, field goal attempts were flagged as occurring in “clutch time” if they met the following two conditions:
1. The attempt occurred within the final two minutes of the game.
2. The game was within six points.
This resulted in a dataset of 22,856 kicks, 1,016 of which were in clutch time. Next, a logistic regression was run based on the distance of the field goal attempt. This assigns a probability of an attempt succeeding, which can then be used to establish an expected number of successes for a given subset of attempts.
Results and Discussion
Model of Field Goal Success
The logistic regression resulted in the following equation, where p represents the probability that the attempt would be successful and d represents the distance of the attempt.
The model resulting from the regression can be seen more clearly in the
following graph:
Finally, it is worth testing the accuracy of the model. The model is considered to have correctly predicted the result of a field goal attempt if it gave a probability < 0.5 and the kick was missed, or if it gave a probability ≥ 0.5 and the kick was successful. Under this definition, the model correctly predicted the result of approximately 85% of attempts.
Attempt Success During Clutch Time
In order to judge the effect of clutch time on kickers, we need to establish the expected number of successes for their kicks. To do this, let A1, A2, . . . , An be Bernoulli random variables representing n independent field goal attempts, each with probability of success pi. Then, using the linearity of expectation:
That is, the expected number of made kicks among a set of n attempts is the sum of the probabilities that each kick is made. Applying this formula, to the clutch time and non-clutch time attempts, we have the following table:
As seen in the table, kickers during clutch time performed worse than expected, missing 16.72 more kicks than expected. In order to determine if this difference is statistically significant, the error (observed successes - expected successes) was calculated for ten thousand random samples of 1016 kicks. Roughly 12% of these samples had errors with absolute values larger than 16.72. This
is not sufficiently low to reject the null hypothesis that kickers perform just as well during clutch time.
Top Performing Kickers
Next, we can group attempts by kicker, and use the same logic to determine how players performed compared to expectations. Among kickers with at least 50 attempts over the years included in this project (2000-2022), Ravens legend Justin Tucker comes out on top in terms of both total successes over expectation (SOE) and successes over expectation per attempt. Below are the top 10 players
in terms of total SOE:
We can also look at players exclusively in terms of their clutch time performance. Among those with at least 10 clutch time attempts, Justin Tucker once again comes out on top in terms of total SOE, but Matt Prater has the higher SOE per attempt. Below are the top 10 in terms of clutch time total SOE:
Limits of the Model
The model created by the logistic regression is certainly imperfect. First, it fails to account for outside factors, namely weather. Furthermore, it produces seemingly high probabilities of making long attempts, likely due to the data used to train the model. Long attempts are inherently difficult, so coaches will not choose to attempt them unless they know they have a talented kicker. This likely pushes up the ratio of successful attempts, compared to if kickers of all skill levels attempted the same kicks.
Conclusion
This project utilized a logistic regression model to evaluate kicker performance during the final moments of tight games. Sufficient evidence was not found to conclude that kickers perform differently during these moments, although future research and model improvement could shed more light on the subject.
Finally, the model was used to rank kickers in terms of successful attempts over expectation, both in general and during clutch time. Justin Tucker dominated as expected, although Matt Prater had the highest number of successes over expectation per clutch time attempt.