Football Analytics with Python & R

Chapter 3. Simple Linear Regression: Rushing Yards Over Expected

Football is a contextual sport. Consider whether a pass is completed. This depends on multiple factors: Was the quarterback under pressure (making it harder to complete)? Was the defense expecting a pass (which would make it harder to complete)? What was the depth of the pass (completion percentage goes down with the depth of the target)?

What turns people off to football analytics are conclusions that they feel lack a contextual understanding of the game. “Raw numbers” can be misleading. Sam Bradford once set the NFL record for completion percentage in a season as a member of the Minnesota Vikings in 2016. This was impressive, since he joined the team early in the season as a part of a trade and had to acclimate quickly to a new environment. While that was impressive, it did not necessarily mean he was the best quarterback in the NFL that year, or even the most accurate one. For one, he averaged just 6.6 yards average depth of target (aDOT) that year, which was 37th in the NFL according to PFF. That left his yards per pass attempt at a relatively average 7.0, tied for just 20th in football. Chapter 4 provides more context for that number and shows you how to adjust it yourself.

Luckily, given the great work of the people supporting nflfastR, you can provide your own context for metrics by applying the statistical tool known as regression. Through regression, you can normalize, or control for, variables (or features ...