Introducing – NFL “CORP” 1.0

NFL – Championship Odds over Replacement Player (“CORP“): an estimate of the percent odds a player provides his team to win a Super Bowl

The origin of the term “CORP” was attributed to Backpicks‘s metric of the same name to measure the odds a player provides a random team to win a championship. Due to the lack of widespread value metrics in American football, I created an NFL equivalent to provide a framework for quantitative player valuation.

Rationale

CORP, as stated earlier, is an estimate of a player’s championship equity. It’s an attempt to quantify a player’s contributions toward the most important goal of an NFL team: to win a Super Bowl.

  • Expected Points Added

To quantify CORP, the formula required a base value to estimate a player’s impact on a rate basis. RBSDM supplies an EPA/Play statistic to estimate the average point value of a play in which a player was involved. EPA/Play, along with the Plays tallies provided by the site, created an estimate of a player’s total point contributions in a given season. EPA totals were used to calculate a player’s expected value in a sixteen-game season as well as scale for “pace” (normalizing a team’s offensive plays per game). 

  • Game-Scaling Value

As mentioned earlier, a step in the CORP calculation results in an “expected” EPA per sixteen games for players. Expected EPA/G (xEPA) could act as a functioning element in CORP, although there’s additional room for team context. The aforementioned “pace” adjustment relies on the principle that more play opportunities typically garner stronger xEPA results. For the purpose of eliminating skewing based on pace, xEPA normalized results to a 65-plays per game (for teams) basis. The percentage of plays a player was involved in was also included to draw outlier seasons closer to the mean.

  • Calculating Title Odds

The concluding step in the CORP calculation involved converting game-scaled xEPA to championship equity. Sports-Reference‘s team evaluation metric, Simple Rating System (opponent-adjusted point differential), was regressed onto ten seasons of win percentage to create the formula to approximate the odds a player provides his team with winning a game against an average team. The equal opponent difficulty for players contextualizes the raw data (eliminate biases based on opponent difficulty), as EPA measures value relative to expected values, similar to SRS. The regression formula creates the CORP score.

Context

CORP is not a perfect metric, and as a result, it doesn’t pinpoint certain aspects of a player. It is not a measurement of talent; it’s a measurement of a player’s contributions in an entire season. It’s extremely dependent on health; a talented player isn’t helping his team if he isn’t playing. Additionally, regular season and postseason games are weighed equally. As indicated by the “1.0” element of “CORP 1.0,” a follow-up model may place more stock into Playoff games to account for settings. At the moment, all games are treated the same. 

Due to the inherent noisiness of contemporary metrics, they’re often presented in three-to-five-year intervals to allow a level of play to stabilize. CORP is no exception; multi-year CORP scores weigh three seasons worth of data, with a given season weighed twice as high as the one preceding it. Multi-year CORP scores, displayed on Cryptbeam, are only calculated for the current season. Single-year results will (eventually) date back as far as the 1999 season. Scores are only calculated for quarterbacks now due to data accessibility, although all offensive positions will (eventually) be measured. To view the Cryptbeam database for CORP, click here.

Conclusion

As is with every custom metric on Cryptbeam, the end result is not intended to be a definitive player ranking. CORP is not perfect, and is planned to act as a framework to model player value and promote more analysis.


3 responses to “Introducing – NFL “CORP” 1.0”

  1. Austin Setzler Avatar
    Austin Setzler

    A few questions and a suggestion:
    1. Does the winning percentage start at zero or from a certain replacement level?
    2. Does this have a theoretical maximum percentage?
    3a. Does it take team wins into account?
    3b. Would a player on a zero win team be given a value fo 0%?
    4. It would be cool if you made a Win-Loss% W/ Avg. Team metric for the NFL.

    1. 1. A replacement-level player is worth a 0% title increment throughout a season, although a player with a very poor season who played in a majority of his team’s snaps (think Josh Rosen in 2018) could fall past the 0% mark into the negatives.
      2. Yes, the “theoretical” maximum is 100%. Once a player passes the threshold for an expected perfect win percentage, the odds won’t increase.
      3a. It does not.
      3b. Not necessarily.
      4. Could you expand on that a bit more? Thanks for your comment

      1. Austin Setzler Avatar
        Austin Setzler

        You know how, in baseball, there is a waaWL% metric that uses WAR (I believe) to estimate the winning percentage of an average team, but it has the player in question on it? In the NFL, where value is very positionally skewed, it would be more difficult to make but very interesting to look at. So this statistic would estimate the player’s winning percentage with a perfectly average team.
        I am not sure how this would be accomplished, but I would love it if it was possible!


Leave a Reply to chromeder Cancel reply

Your email address will not be published. Required fields are marked *