Dynasty 101
Home / Rookie Digest / PREDATOR Score

PREDATOR Score

Late in the 2016 offseason I started to become very intrigued by the use of metrics and numbers in dynasty fantasy football. We constantly hear about players passing the “eye test,” but what about the “number test?” Sure, we have combine measurables, college statistics, and even some newer, complex metrics. But are all of these numbers actually relevant? Surely some must be more important than others in predicting fantasy success. This is the mystery I set out to solve.

I now knew what problem I wanted to solve, but had no idea how to get there. I spent many hours searching Google, reading articles, and watching YouTube tutorial videos on statistical analysis. The method I settled on was to conduct a binary logistic regression for each of the offensive skill positions.

Methodology

I began by making a list of all players drafted (and any significant undrafted free agents) at each position over the past 10 years (15 years for QB/TE due to fewer prospects). I then compiled a list of every measurable, metric, and statistic I could think of for each player. I ended up with 20-25 different variables for each position. Each player was then given a score of “1” (successful) or “0” (unsuccessful), hence the binary nature of the regression.

I ran the regression for each position and eliminated the least statistically significant variable (using p-values). I repeated this process until I was left only with variables that were statistically significant in predicting the future success of players at the position. Thus was born the PREDictive Analysis and Testing Of Rookies, or PREDATOR Score.

Defining Success

Success is defined differently depending on the position. The details are as follows (assumed PPR scoring):

  • Quarterback – Player achieved at least one QB1 (top 12) fantasy season during his NFL career.
  • Running back – Player achieved at least one RB1 (top 12) fantasy season during his first three NFL seasons.
  • Wide receiver – Player achieved at least one WR2 (top 24) fantasy season during his first three NFL seasons.
  • Tight end – Player achieved at least one TE1 (top 12) fantasy season during his NFL career.

Results

After conducting the binary logistic regression to completion, the statistically significant variables by position are as follows (listed from most significant to least):

  • Quarterback
    • Draft position
    • Breakout age
  • Running back
    • Height
    • Draft position
    • BMI
  • Wide receiver
    • Draft position
    • Breakout age
    • College dominator
  • Tight end
    • Draft position
    • Weight
    • Height

I was very surprised by the results. For all the commotion revolving around the NFL combine, none of the drill results were found to be statistically significant at any position. Although, I think it is safe to assume that the “draft position” variable contains a lot of relevant NFL combine information. It includes everything that NFL scouts and GMs take into account when deciding which pick to use on a player. This could include combine/pro day measurements, college production, attitude, scheme fit, etc.

Predicting Success

Each position has its own cut-off for predicting success. Prospects with a score above that number are predicted to fulfill the success requirement and those with a score below that number are not. The cut-offs for each position are as follows:

  • Quarterback – 0.420
  • Running back – 0.500
  • Wide receiver – 0.450
  • Tight end – 0.430

Accuracy

The accuracy of the model at each position is as follows:

  • Quarterback – 77.3%
  • Running back – 76.6%
  • Wide receiver – 76.1%
  • Tight end – 75.7%

Using the Model

While I believe that these regression models provide valuable insight into the prospects at each position, they should not be used as the sole method of player evaluation. I use the PREDATOR score as a tiebreaker when evaluating prospects. If I have two players in the same tier, the PREDATOR score is a great way to separate them and give preference to the prospect who wins the statistical analysis component.

PHP Code Snippets Powered By : XYZScripts.com