Advertisement

Evaluating NFL Plays: Expected Points Adjusted for Schedule

  • Konstantinos PelechrinisEmail author
  • Wayne Winston
  • Jeff Sagarin
  • Vic Cabot
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 11330)

Abstract

“Not all yards are created equal”. A 3rd and 15, where the running back gains 12 yards is clearly less valuable than a 3rd and 3 where the running back gains 4 yards, even though it will not necessarily show up in the yardage statistics. While this problem has been addressed to some extent with the introduction of expected point models, there is still another inequality omission in the creation of yards and this is the opposing defense. Gaining 6 yards on a 3rd and 5 against the top defense is not the same as gaining 6 yards on a 3rd and 5 against the worst defense. Adjusting these expected points model for opponent strength is thus crucial. In this paper, we develop an optimization framework that allows us to compute offensive and defensive ratings for each NFL team and consequently adjust the expected point values accounting for the opposition faced. Our framework allows for assigning different point values to the offensive and defensive units of the same play, which is the rational thing to do especially in a league with an uneven schedule such as the NFL. The average absolute difference between the raw and adjusted points is 0.07 points/play (p-value < 0.001), while the median discrepancy is 0.06 (p-value < 0.001). This might seem negligible, but with an average of 130 plays per game this translates to approximately 125 points/season discredited. The opponent strength adjustment that we introduce in this work is crucial for obtaining a better evaluation of personnel based on the actual competition they faced. Furthermore, our work allows to evaluate special teams’ performance, a unit that has been identified as crucial for winning but has never been properly evaluated. We firmly believe that our developed framework can make significant strides towards computing an accurate estimate of the true monetary worth of a given player.

References

  1. 1.
    Expected points (ep) and expected points added (epa) explained. http://archive.advancedfootballanalytics.com/2010/01/expected-points-ep-and-expected-points.html. Accessed 19 June 2018
  2. 2.
    Football outsiders: Methods to our madness. https://www.footballoutsiders.com/info/methods. Accessed 19 June 2018
  3. 3.
    Prospectus toolbox: Value over replacement player. https://www.baseballprospectus.com/news/article/6231/prospectus-toolbox-value-over-replacement-player/. Accessed 19 June2018
  4. 4.
    Carroll, B., Palmer, P., Thorn, J.: The Hidden Game of Football. Warner Books, New York (1989)Google Scholar
  5. 5.
    Carter, V., Machol, R.E.: Operations research on football. Oper. Res. 19(2), 541–544 (1971)CrossRefGoogle Scholar
  6. 6.
    Pelechrinis, K.: The passing skill curve in the NFL. In: The Cascadia Symposium on Statistics in Sports (2018)Google Scholar
  7. 7.
    Shapley, L.S.: Stochastic games. Proc. Natl. Acad. Sci. 39(10), 1095–1100 (1953)MathSciNetCrossRefGoogle Scholar
  8. 8.
    Winston, W., Cabot, A., Sagarin, J.: Football as an infinite horizon zero-sum stochastic game. Technical report (1983)Google Scholar
  9. 9.
    Winston, W.L.: Mathletics: How Gamblers, Managers, and Sports Enthusiasts Use Mathematics in Baseball, Basketball, and Football. Princeton University Press, Princeton (2012)Google Scholar
  10. 10.
    Yurko, R., Ventura, S., Horowitz, M.: nflwar: A reproducible method for offensive player evaluation in football. arXiv:1802.00998 (2018)

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Konstantinos Pelechrinis
    • 1
    Email author
  • Wayne Winston
    • 2
  • Jeff Sagarin
    • 3
  • Vic Cabot
    • 2
  1. 1.University of PittsburghPittsburghUSA
  2. 2.Indiana UniversityBloomingtonUSA
  3. 3.USA TodayMcLeanUSA

Personalised recommendations