NFL play-by-play EDA

Exploratory data analysis of NFL play-by-play data using nfl-data-py

Contents

• NFL play-by-play EDA
  • Contents
  • Playcalling
  • Play results
    • Offensive skill
    • Defensive skill

Playcalling

The first goal is to develop a playcalling engine based on real NFL playcalling data. Initially the playcalling engine might be based purely on game context, such as

• Score (point differential)
• Time
• Quarter
• Yard line
• Down & distance
• Timeouts remaining

In the future, we might extend this playcalling engine to accept additional team context, such as

• Offensive playcalling style
• Passing game overall
  • Pass blocking
  • QB overall
  • WR overall
• Rushing game overall
  • Run blocking
  • RB overall

Play results

The next goal is to develop a model (or models) which can represent the outcomes of individual plays. The input to this model should be

• The play call from the playcalling engine
• Some measurement(s) of offensive skill
• Some measurement(s) of defensive skill

Offensive skill

As a starting point, we will consider the following measurements of offensive skill

• blocking: Likelihood of a TFL, sack, or QB hit against
• rushing: Yards per carry for
• passing: Passer rating for
• receiving: Incomplete passes and yards after catch for
• scrambling: Likelihood of a QB scramble and QB yards per carry
• turnovers: Likelihood of causing a turnover
• penalties: Likelihood of committing a penalty

Defensive skill

As a starting point, we will consider the following measurements of defensive skill (which mirror the offensive skill measurements closely)

• blitzing: Likelihood of a TFL, sack, or QB hit for
• rush_defense: Yards per carry against
• pass_defense: Passer rating against
• coverage: Incomplete passes and yards after catch against
• turnovers: Likelihood of recovering a turnover
• penalties: Likelihood of committing a penalty