SLIDE 1
Decision Making in American Football : Evidence from 7 Years of NFL - - PowerPoint PPT Presentation
Decision Making in American Football : Evidence from 7 Years of NFL - - PowerPoint PPT Presentation
Decision Making in American Football : Evidence from 7 Years of NFL Data Konstantinos Pelechrinis Presented by Ke Zhang ECML/PKDD Machine Learning & Data Mining for Sports Analytics Riva del Garda, Italy 2016 COACHING DECISIONS ARE
SLIDE 2
SLIDE 3
ARE COACHING DECISIONS RATIONAL?
SLIDE 4
WHAT IS A RATIONAL DECISION?
A decision that maximizes the probability of winning
SLIDE 5
WHAT IS A RATIONAL DECISION?
A decision that maximizes the probability of winning In the vast majority of the cases this is equivalent to maximizing the points scored
SLIDE 6
WHAT IS A RATIONAL DECISION?
A decision that maximizes the probability of winning In the vast majority of the cases this is equivalent to maximizing the points scored The closer to the end of the game the less good the above approximation
SLIDE 7
TWO CASE STUDIES
PAT decision 4th down decision
SLIDE 8
DATA
NFL API
- 7 years play-by-play data
- 1,792 regular season games
- 42,371 drives
- 316,630 snaps
- 9,021 touchdowns
SLIDE 9
PAT DECISION
OR
SLIDE 10
PAT DECISIONS
SLIDE 11
PAT DECISIONS
E[p]= 2*
SLIDE 12
PAT DECISIONS
E[p]= 2*
- 1*
SLIDE 13
PAT DECISIONS
E[p]= 2*
- 1*
15 14 14 15 9 24 12 13 13 16 24 21 10 17 21 11 12 14 11 12 10 14 9 16 22 6 14 5 14 22 12 18
- 0.50
- 0.25
0.00 0.25 0.50 ARI ATL BAL BUF CAR CHI CIN CLE DAL DEN DET GB HOU IND JAC KC MIA MIN NE NO NYG NYJ OAK PHI PIT SD SEA SF STL TB TEN WAS
Expected Point Gain
SLIDE 14
PAT DECISIONS
E[p]= 2*
- 1*
15 14 14 15 9 24 12 13 13 16 24 21 10 17 21 11 12 14 11 12 10 14 9 16 22 6 14 5 14 22 12 18
- 0.50
- 0.25
0.00 0.25 0.50 ARI ATL BAL BUF CAR CHI CIN CLE DAL DEN DET GB HOU IND JAC KC MIA MIN NE NO NYG NYJ OAK PHI PIT SD SEA SF STL TB TEN WAS
Expected Point Gain
SLIDE 15
4TH DOWN DECISION
SLIDE 16
4TH DOWN DECISIONS
SLIDE 17
4TH DOWN DECISIONS
- 0.4
0.0 0.4 0.8 1.2 10 20 30
Yards to go 4th down converstion success rate
SLIDE 18
4TH DOWN DECISIONS
- 0.4
0.0 0.4 0.8 1.2 10 20 30
Yards to go 4th down converstion success rate
0.0 0.4 0.8 20 30 40 50 60 70
Distance (yards) Probability of FG Success
SLIDE 19
4TH DOWN DECISIONS
- 0.4
0.0 0.4 0.8 1.2 10 20 30
Yards to go 4th down converstion success rate
0.0 0.4 0.8 20 30 40 50 60 70
Distance (yards) Probability of FG Success
Touchback line
0.00 0.25 0.50 0.75 1.00 25 50 75 100
Distance to Goal Line at the Beginning of the Drive Probability of Success
type TD FG Fail
SLIDE 20
4TH DOWN DECISIONS
SLIDE 21
4TH DOWN DECISIONS
Touchback
- 2
2 4 25 50 75 100
Distance to the Goal Line when 4th down Expected Points Gained
SLIDE 22
CONCLUSIONS & LIMITATIONS
- Coaching decisions seem to follow the status quo rather than
being rational
- Coaches of course might not want to maximize the points
score but rather minimize the variance for the points scored
- Current analysis is only focused on offensive unit decisions
- Objective: maximize the point differential
- The estimates of the fourth down conversion rate can be biased
as alluded to above
SLIDE 23
13
Questions can be directed to kpele@pitt.edu @kpelechrinis Every snap is a data point! http://fivethirtyeight.com/features/the-high-school-football-coach- who-never-punts/
SLIDE 24