SlideShare a Scribd company logo
Should a football team go for a one
or two point conversion?
A dynamic programming approach
Dr. Laura Albert
University of Wisconsin-Madison
laura@engr.wisc.edu
@lauraalbertphd
(c) 2019
The problem
• When an NFL team scores a touchdown, they score six points
• They have two choices:
– kick a point after to score 1 extra point
– try for a 2 point conversion
• When should they try a 2 point conversion? When should
they kick an extra point?
– Solution approach: dynamic programming
– Model output: a series of decisions based on the score differential and
the remaining time/remaining possessions
(c) Laura Albert 2019 2Two point conversion DP
We make several assumptions
1. There are three possible outcomes for each team’s
possession:
1. touchdown (19% of the time),
2. field goal (13% of the time),
3. no score (68% of the time).
2. The probability of success of each outcome does not depend
on the score or the time left.
3. If the game ends in a tie, each team’s probability of winning
is 0.5.
4. We know the number of remaining possessions.
5. The teams are equally matched
Note: we can easily lift this assumption
(c) Laura Albert 2019 3Two point conversion DP
Our approach
• This decision depends on two factors:
– The point differential
– The remaining number of possessions (assumed to be known)
• The state of the system depends on these two factors
– The minimal amount of information needed to make our decision
• The success rate for two point conversions is 48% and the
success rate for extra points is 96%.
(c) Laura Albert 2019 4Two point conversion DP
5
Let’s solve this with dynamic
programming (DP)
• We model a sequence of interrelated decisions for two teams.
• Each team seeks to maximize the probability they win the
game.
• DP is a general approach to problem solving, not a particular
model or algorithm.
• DP is smart enumeration. All possible outcomes are not
enumerated, but DP guarantees the optimal solution.
(c) Laura Albert 2019Two point conversion DP
6
Shortest Path Problem
INSTANCE: Directed graph 𝐺𝐺 = (𝑉𝑉, 𝐸𝐸), nonnegative
arc distances 𝑑𝑑𝑖𝑖𝑖𝑖 for each edge, a starting node 𝑠𝑠.
OBJECTIVE: Find the shortest path from node 𝑠𝑠 to
node 𝑡𝑡.
s
p
t
(c) Laura Albert 2019Two point conversion DP
7
Shortest Path Problem, cont’d.
• Observation 1: If the shortest path from s to r passes through node p, the
subpaths (𝑠𝑠, 𝑝𝑝) and (𝑝𝑝. 𝑡𝑡) are shortest paths from 𝑠𝑠 to 𝑝𝑝 and 𝑝𝑝 to 𝑡𝑡,
respectively.
– If not, the shorter subpath would allow us to construct a shorter path from 𝑠𝑠
to 𝑡𝑡, leading to a contradiction.
• Observation 2: Let 𝑉𝑉𝑗𝑗 denote the length of a shortest path from 𝑗𝑗 to 𝑡𝑡,
then
𝑉𝑉𝑗𝑗 = min
𝑖𝑖∈𝑉𝑉𝑖𝑖{𝑗𝑗}
{𝑉𝑉𝑖𝑖 + 𝑑𝑑𝑑𝑑𝑑𝑑}
In other words, the lengths of the shortest paths from every neighbor of 𝑗𝑗
to 𝑡𝑡, then we can find the length of the shortest part from 𝑗𝑗 to 𝑟𝑟.
Note that this is not necessarily an algorithm for finding the shortest path.
(c) Laura Albert 2019Two point conversion DP
8
Principle of Optimality
• Principle of Optimality: Observations 1 and 2 define the
optimal substructure of all DP models and algorithms. Each
solution is characterized by
– A sequence of decisions (more on this later),
– The optimal solution is composed of optimal solutions to
subproblems.
• In the Shortest Path problem, the DP gives the shortest paths
from all nodes to node 𝑡𝑡
– 𝑉𝑉𝑡𝑡 = 0.
(c) Laura Albert 2019Two point conversion DP
9
Shortest Path DP algorithms
• Since DP algorithms are smart enumerations, good
DP algorithms are important to find.
• Let Vj
n = the shortest path from j to r at iteration n.
• We want to find the optimal distances Vj by initially
setting Vj
0 = large number (j ≠ r)
• The idea: iteratively loop over all nodes to find the
best edge to traverse
– Because of the principle of optimality, we only improve
(lower) Vj
n.
(c) Laura Albert 2019Two point conversion DP
10
Shortest Path DP algorithm 1
Bellman’s Algorithm
1. Let Vr
0=0 and Vj
0 = M, j ≠ r (M is a large number) and let
n=1.
2. For all j ∈ V{r}
Vj
n=mini{dji+Vi
n-1}
3. If Vj
n < Vj
n-1 for any i, let n=n+1 and return to step 2. Else
stop.
Note: only consider nodes where an edge from i to j exists.
Problems: In step 2, most of the time we have to consider
edges with distance M. A better algorithm only
considers nodes with paths less than M
(c) Laura Albert 2019Two point conversion DP
11
Characteristics of DP Problems
1. Problem divided into stages with a policy decision required at
each stage.
Shortest path = which node to visit next
2. Each stage has a number of states associated with it.
3. The effect of each policy decision in each state (at each stage)
transforms the current state into a state in the next stage
(deterministic or stochastic).
4. The solution procedure is designed to find the optimal policy.
Prescription for finding the best decision at each stage for each
state (state-action pairs)
(c) Laura Albert 2019Two point conversion DP
12
Characteristics of DP Problems
5. Given current state, an optimal policy for the remaining stages is
independent of the policy in the previous stages.
This is the principle of optimality for dynamic programming
6. The solution procedure begins by finding the optimal policy for the
last stage.
This is usually trivial, such as 𝑉𝑉𝑡𝑡 = 0
7. A recursive relationship identifies the optimal policy for stage 𝑛𝑛,
given an optimal policy for stage 𝑛𝑛 + 1
Called the value functions or optimality equations or Bellman
equations or Hamilton-Jacobi-Bellman equations
(c) Laura Albert 2019Two point conversion DP
Dynamic Programs as shortest or
longest paths
• Almost any DP problem can be formulated as a shortest or longest
path (by maximizing) in a graph with directed arcs (digraph)
• Nodes of the digraph correspond to states in an incomplete
solution
• Arcs of the digraph correspond to decisions. They link each state to
a subsequent state to which the decision leads.
Implication for our football problem:
• The nodes represent the triplet: (1) who has the possession, (2) the
current score differential, (3) how many possessions are left.
• We solve a longest path problem to maximize win probability (not
distance)
13
(c) Laura Albert 2019Two point conversion DP
Dynamic programming
Let
• 𝑉𝑉𝑛𝑛
𝐴𝐴
(𝑝𝑝) = the probability that team A wins the game if they are
𝑝𝑝 points ahead, they have just gotten the ball, and there are 𝑛𝑛
remaining possessions.
• 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝) = the probability that team A wins the game if they are
𝑝𝑝 points ahead, their opponent team B has just gotten the
ball, and there are 𝑛𝑛 remaining possessions.
• 𝑛𝑛 = 0 means the game is over.
(c) Laura Albert 2019 14Two point conversion DP
Boundary conditions
• 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 1 for 𝑝𝑝 > 0
• 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 0 for 𝑝𝑝 < 0
• 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 𝑉𝑉0
𝐴𝐴
𝑝𝑝 = 0.5 for 𝑝𝑝 = 0 [a tie]
• Goal: maximizing 𝑉𝑉𝑛𝑛
𝐴𝐴(𝑝𝑝) for some 𝑛𝑛 and 𝑝𝑝
– Find by conditioning on the possession outcome and
applying the law of probability
(c) Laura Albert 2019 15Two point conversion DP
Law of total probability
• Let 𝑋𝑋 denote a random variable.
– 𝑋𝑋 is a binary variable denoting the outcome of the game
(win/loss)
• We can find the probability of 𝑋𝑋 given some
additional information 𝑌𝑌
– 𝑌𝑌 is a random variable denoting the outcome of the
current possession (touchdown, field goal, or no score).
[ ] ( | ) ( )
y Y
P X P X y P y
∈
= ∑
(c) Laura Albert 2019 16Two point conversion DP
Finding an expected value
• Let 𝐼𝐼 be an indicator variable if the team wins:
– 𝐼𝐼 = 1 if they win and 𝐼𝐼 = 0 if they lose.
• The expected value of 𝐼𝐼 is the probability that the team wins
𝐸𝐸[𝐼𝐼] = 𝑃𝑃(𝐼𝐼 > 0)
• 𝑃𝑃(𝐼𝐼 > 0) = 𝑉𝑉𝑛𝑛
𝐴𝐴
(𝑝𝑝)
– This can be found by computing the expected value of an
indicator variable 𝐸𝐸[𝐼𝐼]
– This is the solution to a longest path problem!
(c) Laura Albert 2019 17Two point conversion DP
Law of total probability
• 𝑃𝑃𝑇𝑇𝑇𝑇 = probability of a touchdown
• 𝑃𝑃𝐹𝐹𝐹𝐹 = probability of a field goal
• 𝑃𝑃𝑁𝑁𝑁𝑁 = probability of no score
• 𝑃𝑃𝐴𝐴𝐴𝐴 = probability of scoring a point after a touchdown
• 𝑃𝑃2𝑃𝑃𝑃𝑃 = probability of a two point conversion after a touchdown
Three outcomes of the current drive: TD, FG, or NS
( 𝑃𝑃𝑇𝑇𝑇𝑇 + 𝑃𝑃𝑃𝑃𝑃𝑃 + 𝑃𝑃𝑃𝑃𝑃𝑃 = 1)
𝑉𝑉𝑛𝑛+1
𝐴𝐴
𝑝𝑝 =
𝑃𝑃𝑁𝑁𝑁𝑁 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝)  No score
+ 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝 + 3)  Field Goal
+ 𝑃𝑃𝑃𝑃𝑃𝑃 max{𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝 + 7) + (1 − 𝑃𝑃𝐴𝐴𝐴𝐴) 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝 + 6) , TD with AT
𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝 + 8) + (1 − 𝑃𝑃2𝑃𝑃𝑃𝑃) 𝑉𝑉𝑛𝑛
𝐵𝐵
(𝑝𝑝 + 6) }  TD with 2PT
Note: define 𝑉𝑉𝑛𝑛+1
𝐵𝐵
𝑝𝑝 in an analogous way
(c) Laura Albert 2019 18Two point conversion DP
Series of equations
Team A:
𝑉𝑉𝑛𝑛+1
𝐴𝐴
𝑝𝑝 =
𝑃𝑃𝑇𝑇𝑇𝑇 max{ 𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 7 + 1 − 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 6 , 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵(𝑝𝑝 +
(c) Laura Albert 2019 19Two point conversion DP
Team A’s win probability
Note: the points P are before the touchdown is scored
V^A_n(p) Possessions remaining n
P 0 1 2 3 4 5 6 7 8 9 10
-16 0 0.000 0.000 0.003 0.002 0.014 0.010 0.030 0.022 0.047 0.036
-15 0 0.000 0.000 0.006 0.004 0.020 0.014 0.038 0.029 0.058 0.044
-14 0 0.000 0.000 0.015 0.010 0.035 0.025 0.056 0.042 0.077 0.059
-13 0 0.000 0.000 0.024 0.016 0.050 0.036 0.075 0.056 0.097 0.075
-12 0 0.000 0.000 0.026 0.018 0.057 0.042 0.086 0.066 0.112 0.088
-11 0 0.000 0.000 0.035 0.026 0.074 0.056 0.108 0.084 0.136 0.108
-10 0 0.000 0.000 0.045 0.034 0.092 0.070 0.131 0.102 0.161 0.129
-9 0 0.000 0.000 0.062 0.046 0.119 0.090 0.162 0.126 0.193 0.154
-8 0 0.046 0.031 0.108 0.079 0.159 0.122 0.198 0.155 0.226 0.182
-7 0 0.091 0.062 0.154 0.113 0.199 0.153 0.234 0.186 0.260 0.212
-6 0 0.186 0.127 0.252 0.183 0.283 0.216 0.303 0.240 0.318 0.258
-5 0 0.190 0.135 0.271 0.202 0.311 0.242 0.334 0.268 0.350 0.288
-4 0 0.190 0.141 0.282 0.218 0.332 0.265 0.362 0.296 0.382 0.319
-3 0 0.255 0.198 0.362 0.288 0.409 0.332 0.430 0.356 0.441 0.371
-2 0 0.320 0.243 0.427 0.336 0.465 0.378 0.480 0.398 0.485 0.410
-1 0 0.320 0.251 0.440 0.357 0.489 0.405 0.509 0.430 0.518 0.444
0 0.5 0.660 0.500 0.622 0.500 0.601 0.500 0.588 0.500 0.579 0.500
1 1 1.000 0.749 0.805 0.643 0.714 0.594 0.667 0.569 0.639 0.555
2 1 1.000 0.757 0.817 0.662 0.736 0.621 0.695 0.600 0.670 0.588
3 1 1.000 0.803 0.849 0.713 0.774 0.668 0.733 0.644 0.708 0.628
4 1 1.000 0.859 0.896 0.782 0.832 0.734 0.789 0.703 0.758 0.681
Team A’s last
possession
Team A’s second to
last possession(c) Laura Albert 2019 20Two point conversion DP
Boundary
conditions
How do we determine whether to go
for two points?
Test the condition in the max statement to see what was
selected:
For Team A who gets the ball with 𝑛𝑛 + 1 possessions to go and is
up by 𝑝𝑝 points 𝑉𝑉𝑛𝑛+1
𝐴𝐴
𝑝𝑝
𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 7 + 1 − 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 6 <
𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 8 + 1 − 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛
𝐵𝐵
𝑝𝑝 + 6
Then go for two points is better than going for one point.
A similar approach works for Team B by solving 𝑉𝑉𝑛𝑛+1
𝐵𝐵
𝑝𝑝
(c) Laura Albert 2019 21Two point conversion DP
When should team A go for 2 if they score a
touchdown
Note: the points P are before the touchdown is scored
Possessions remaining n
P 0 2 2 3 4 5 6 7 8 9 10
-10 1 1 1 1 2 2 2 2 2 2 1
-9 1 1 1 1 1 1 2 2 2 2 1
-8 1 1 1 1 2 2 2 2 2 2 1
-7 1 1 1 1 2 2 2 2 2 2 1
-6 1 1 1 1 1 1 1 1 1 1 1
-5 1 1 1 1 2 2 2 2 2 2 1
-4 1 1 1 1 2 2 2 2 2 2 1
-3 1 1 1 1 2 2 2 2 2 2 1
-2 1 1 2 2 2 2 2 2 2 2 1
-1 1 1 1 1 2 2 2 2 2 2 1
0 1 1 2 2 2 2 2 2 2 2 1
1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1 1 2 1
3 1 1 2 2 2 2 2 2 2 2 1
4 1 1 1 1 1 1 1 1 1 1 1
5 1 1 1 1 1 1 1 1 1 1 1
6 2 2 2 2 2 2 2 2 2 2 2
7 1 1 1 1 2 1 2 1 2 1 1
8 1 1 1 1 1 1 1 1 1 1 1
9 1 2 2 2 2 2 2 2 2 2 1
10 1 1 2 2 2 2 2 2 2 2 1
Team A’s last
possession
Team A’s second to
last possession(c) Laura Albert 2019 22Two point conversion DP
When should team A go for 2 if they score
a touchdown
Two point conversion DP (c) Laura Albert 2019 23
Team A’s last
possession
Team A’s
second to
last
possession
When should Team A go for 2?
On their last possession (from 𝑉𝑉1
𝐴𝐴
𝑝𝑝 ):
When 𝑝𝑝 = −8
i.e., when Team A is down 8 and then scores a touchdown
Team A would be tied or down 2 after this possession.
On their second to last possession (from 𝑉𝑉3
𝐴𝐴
𝑝𝑝 ):
When 𝑝𝑝 = −16, −14, −11, −8, −5, −4, −2, −1
(Or 𝑝𝑝 = −10, −8, −3, −2, 1, 2, 4, 5 including the 6 points from
the touchdown)
Note: the points 𝑝𝑝 are before the touchdown is scored
(c) Laura Albert 2019 24Two point conversion DP
When should team A go for 2 if they score
a touchdown
Two point conversion DP (c) Laura Albert 2019 25
Go for it
when down 2
Go for it
when down 8
Go for it
when up
1, 2, 4, 5
Go for 1
here,
otherwise
the other
team could
win with a FG
When should Team A go for 2?
When we set 𝑃𝑃𝐴𝐴𝐴𝐴 = 0.95 (from 0.96) the results change slightly:
• Team A should go for 2 if they score a touchdown when down 15 before
the touchdown
• Note: the win probability does not change very much
On their second to last possession (from 𝑉𝑉3
𝐴𝐴
𝑝𝑝 ):
When 𝑝𝑝 = −16, −𝟏𝟏𝟏𝟏, −14, −11, −8, −5, −4, −2, −1
Note: the points 𝑝𝑝 are before the touchdown is scored
How much does going for 2 improve the win probability?
(c) Laura Albert 2019 26Two point conversion DP
How much does going for 2 improve
the win probability?
Answer: not by much.
Rationale: if you are down by 15, you have to score two touchdowns while
keeping your opponent from scoring.
• If down by 8 (before the 6 TD points) with 1 or 3 possessions to go, the
strategy improves the win probability by 4.6% or 4.5%.
• If down by 1, 5, or 11 (before the 6 TD points), the strategy improves the win
probability by 0.6 – 1.8% depending on the remaining number of possessions
• If down by 14, 15, or 16 (before the 6 TD points), the strategy improves the
win probability by <1%
In some scenarios, the win probability gets worse (!)
• Your opponent is also finding the best strategy
The win probability can be further improved by improving other decisions
• E.g., Going for it on fourth down, better play calling
Two point conversion DP (c) Laura Albert 2019 27
Why would a team go for 2 when they
are down by 8?
Let’s say a team is down by 14 and scores a touchdown, putting
them down by 8. Let’s say there will be 4 possessions left after
this possession.
• Traditional decision: go for 1 point.
If they succeed (with probability 0.96), they will have a 11.3% probability
of winning.
If they do not succeed, they will have a 7.9% probability of winning
Overall: 11.2% win probability
• Going for a two point conversion
If they succeed (with probability 0.48), they will have a 18.3% probability
of winning.
If they do not succeed, they will have a 7.9% probability of winning
Overall: 12.9% win probability
Two point conversion DP (c) Laura Albert 2019 28

More Related Content

PDF
Should a football team run or pass? A linear programming approach to game theory
PPTX
Game theory ppt
PPTX
Nash equilibrium and applications
PPT
Beyond Nash Equilibrium - Correlated Equilibrium and Evolutionary Equilibrium
PPTX
Oligopoly and Game Theory
PPTX
Min-Max algorithm
PPTX
Stat451 - Life Distribution
DOC
mingdraft2.doc
Should a football team run or pass? A linear programming approach to game theory
Game theory ppt
Nash equilibrium and applications
Beyond Nash Equilibrium - Correlated Equilibrium and Evolutionary Equilibrium
Oligopoly and Game Theory
Min-Max algorithm
Stat451 - Life Distribution
mingdraft2.doc

What's hot (20)

PDF
Quantitative Methods for Lawyers - Class #15 - R Boot Camp - Part 2 - Profess...
PPT
Machine Learning Chapter 11 2
PDF
Rademacher Averages: Theory and Practice
PDF
A Game Theoretic Analysis of Intrusion Detection in Access Control Systems - ...
PDF
Reinforcement Learning
PPTX
What Is Dynamic Programming? | Dynamic Programming Explained | Programming Fo...
PDF
CAGT-IST Student Presentations
PPTX
Newgame (2)
PDF
Study_Pricing_Digital_Call_Options
PDF
DOCX
A brief introduction to game theory prisoners dilemma and nash equilibrum
PDF
Bioalgo 2012-04-hmm
PPT
Lecture11
PPTX
Game throy
PDF
Machine learning Lecture 2
PDF
Black Scholes
PDF
Daa chapter 3
PDF
Dias Da Cruz Steve - Seminar
PPT
AI Lecture 5 (game playing)
Quantitative Methods for Lawyers - Class #15 - R Boot Camp - Part 2 - Profess...
Machine Learning Chapter 11 2
Rademacher Averages: Theory and Practice
A Game Theoretic Analysis of Intrusion Detection in Access Control Systems - ...
Reinforcement Learning
What Is Dynamic Programming? | Dynamic Programming Explained | Programming Fo...
CAGT-IST Student Presentations
Newgame (2)
Study_Pricing_Digital_Call_Options
A brief introduction to game theory prisoners dilemma and nash equilibrum
Bioalgo 2012-04-hmm
Lecture11
Game throy
Machine learning Lecture 2
Black Scholes
Daa chapter 3
Dias Da Cruz Steve - Seminar
AI Lecture 5 (game playing)
Ad

Similar to Should a football team go for a one or two point conversion? A dynamic programming approach (20)

PPTX
World Series Problem
DOCX
Unit 7 dynamic programming
PPT
dynamic-programming unit 3 power point presentation
PDF
Dynamic programming
PPTX
Introduction to dynamic programming
PPT
Dynamic programming 2
PDF
Dynamic programming
PPTX
dinosourrrrrrrrrrrrrrrrrrrrrr formula .pptx
PPTX
Dynamic programming
PPTX
unit-4-dynamic programming
PDF
Volleyball analytics: Modeling volleyball using Markov chains
PPT
Dynamic pgmming
PPTX
Floyd warshall algo {dynamic approach}
PDF
Dynamic Programming Algorithm CSI-504.pdf
PDF
Time Machine session @ ICME 2012 - DTW's New Youth
PDF
Unit 4- Dynamic Programming.pdf
PPTX
introduction to Operation Research
PPT
4900514.ppt
PPTX
Graphical Solution in LPP along with graphs and formula
PPTX
Computer science in Dynamic programming .pptx
World Series Problem
Unit 7 dynamic programming
dynamic-programming unit 3 power point presentation
Dynamic programming
Introduction to dynamic programming
Dynamic programming 2
Dynamic programming
dinosourrrrrrrrrrrrrrrrrrrrrr formula .pptx
Dynamic programming
unit-4-dynamic programming
Volleyball analytics: Modeling volleyball using Markov chains
Dynamic pgmming
Floyd warshall algo {dynamic approach}
Dynamic Programming Algorithm CSI-504.pdf
Time Machine session @ ICME 2012 - DTW's New Youth
Unit 4- Dynamic Programming.pdf
introduction to Operation Research
4900514.ppt
Graphical Solution in LPP along with graphs and formula
Computer science in Dynamic programming .pptx
Ad

More from Laura Albert (20)

PDF
Tackling hard problems: On the evolution of operations research
PDF
Optimization with impact: my journey in public sector operations research
PDF
On designing public sector systems in emergency medical services, disaster re...
PDF
2018 INFORMS Government & Analytics Summit Overview
PDF
Designing emergency medical service systems to enhance community resilience
PDF
Modeling Service networks
PDF
Translating Engineering and Operations Analyses into Effective Homeland Secur...
PDF
Delivering emergency medical services:Research, theory, and application
PDF
Advanced analytics for supporting public policy, bracketology, and beyond!
PDF
Bracketology talk at the Crossroads of ideas
PDF
Wicked problems in operations research
PDF
Spring new educators orientation
PPTX
engineering systems: critical infrastructure and logistics
PDF
Operations Research for Homeland Security and Beyond!
PDF
Discrete Optimization Models for Homeland Security and Disaster Management
PDF
2015 Fuzzy Vance Lecture in Mathematics at Oberlin College: Locating and disp...
PDF
Integer programming for locating ambulances
PDF
Screening Commercial Aviation Passengers in the Aftermath of September 11, 2001
PDF
Delivering emergency medical services: research, application, and outreach
PDF
Women in engineering luncheon presentation at CASE 2013 (IEEE conference on a...
Tackling hard problems: On the evolution of operations research
Optimization with impact: my journey in public sector operations research
On designing public sector systems in emergency medical services, disaster re...
2018 INFORMS Government & Analytics Summit Overview
Designing emergency medical service systems to enhance community resilience
Modeling Service networks
Translating Engineering and Operations Analyses into Effective Homeland Secur...
Delivering emergency medical services:Research, theory, and application
Advanced analytics for supporting public policy, bracketology, and beyond!
Bracketology talk at the Crossroads of ideas
Wicked problems in operations research
Spring new educators orientation
engineering systems: critical infrastructure and logistics
Operations Research for Homeland Security and Beyond!
Discrete Optimization Models for Homeland Security and Disaster Management
2015 Fuzzy Vance Lecture in Mathematics at Oberlin College: Locating and disp...
Integer programming for locating ambulances
Screening Commercial Aviation Passengers in the Aftermath of September 11, 2001
Delivering emergency medical services: research, application, and outreach
Women in engineering luncheon presentation at CASE 2013 (IEEE conference on a...

Recently uploaded (20)

PPTX
PPH.pptx obstetrics and gynecology in nursing
PPTX
Lesson notes of climatology university.
PPTX
PPT- ENG7_QUARTER1_LESSON1_WEEK1. IMAGERY -DESCRIPTIONS pptx.pptx
PDF
Black Hat USA 2025 - Micro ICS Summit - ICS/OT Threat Landscape
PDF
The Lost Whites of Pakistan by Jahanzaib Mughal.pdf
PPTX
master seminar digital applications in india
PDF
Insiders guide to clinical Medicine.pdf
PDF
102 student loan defaulters named and shamed – Is someone you know on the list?
PDF
FourierSeries-QuestionsWithAnswers(Part-A).pdf
PDF
Supply Chain Operations Speaking Notes -ICLT Program
PDF
2.FourierTransform-ShortQuestionswithAnswers.pdf
PDF
O7-L3 Supply Chain Operations - ICLT Program
PDF
VCE English Exam - Section C Student Revision Booklet
PDF
Basic Mud Logging Guide for educational purpose
PDF
Computing-Curriculum for Schools in Ghana
PDF
BÀI TẬP BỔ TRỢ 4 KỸ NĂNG TIẾNG ANH 9 GLOBAL SUCCESS - CẢ NĂM - BÁM SÁT FORM Đ...
PDF
Sports Quiz easy sports quiz sports quiz
PDF
RMMM.pdf make it easy to upload and study
PPTX
IMMUNITY IMMUNITY refers to protection against infection, and the immune syst...
PPTX
Pharma ospi slides which help in ospi learning
PPH.pptx obstetrics and gynecology in nursing
Lesson notes of climatology university.
PPT- ENG7_QUARTER1_LESSON1_WEEK1. IMAGERY -DESCRIPTIONS pptx.pptx
Black Hat USA 2025 - Micro ICS Summit - ICS/OT Threat Landscape
The Lost Whites of Pakistan by Jahanzaib Mughal.pdf
master seminar digital applications in india
Insiders guide to clinical Medicine.pdf
102 student loan defaulters named and shamed – Is someone you know on the list?
FourierSeries-QuestionsWithAnswers(Part-A).pdf
Supply Chain Operations Speaking Notes -ICLT Program
2.FourierTransform-ShortQuestionswithAnswers.pdf
O7-L3 Supply Chain Operations - ICLT Program
VCE English Exam - Section C Student Revision Booklet
Basic Mud Logging Guide for educational purpose
Computing-Curriculum for Schools in Ghana
BÀI TẬP BỔ TRỢ 4 KỸ NĂNG TIẾNG ANH 9 GLOBAL SUCCESS - CẢ NĂM - BÁM SÁT FORM Đ...
Sports Quiz easy sports quiz sports quiz
RMMM.pdf make it easy to upload and study
IMMUNITY IMMUNITY refers to protection against infection, and the immune syst...
Pharma ospi slides which help in ospi learning

Should a football team go for a one or two point conversion? A dynamic programming approach

  • 1. Should a football team go for a one or two point conversion? A dynamic programming approach Dr. Laura Albert University of Wisconsin-Madison laura@engr.wisc.edu @lauraalbertphd (c) 2019
  • 2. The problem • When an NFL team scores a touchdown, they score six points • They have two choices: – kick a point after to score 1 extra point – try for a 2 point conversion • When should they try a 2 point conversion? When should they kick an extra point? – Solution approach: dynamic programming – Model output: a series of decisions based on the score differential and the remaining time/remaining possessions (c) Laura Albert 2019 2Two point conversion DP
  • 3. We make several assumptions 1. There are three possible outcomes for each team’s possession: 1. touchdown (19% of the time), 2. field goal (13% of the time), 3. no score (68% of the time). 2. The probability of success of each outcome does not depend on the score or the time left. 3. If the game ends in a tie, each team’s probability of winning is 0.5. 4. We know the number of remaining possessions. 5. The teams are equally matched Note: we can easily lift this assumption (c) Laura Albert 2019 3Two point conversion DP
  • 4. Our approach • This decision depends on two factors: – The point differential – The remaining number of possessions (assumed to be known) • The state of the system depends on these two factors – The minimal amount of information needed to make our decision • The success rate for two point conversions is 48% and the success rate for extra points is 96%. (c) Laura Albert 2019 4Two point conversion DP
  • 5. 5 Let’s solve this with dynamic programming (DP) • We model a sequence of interrelated decisions for two teams. • Each team seeks to maximize the probability they win the game. • DP is a general approach to problem solving, not a particular model or algorithm. • DP is smart enumeration. All possible outcomes are not enumerated, but DP guarantees the optimal solution. (c) Laura Albert 2019Two point conversion DP
  • 6. 6 Shortest Path Problem INSTANCE: Directed graph 𝐺𝐺 = (𝑉𝑉, 𝐸𝐸), nonnegative arc distances 𝑑𝑑𝑖𝑖𝑖𝑖 for each edge, a starting node 𝑠𝑠. OBJECTIVE: Find the shortest path from node 𝑠𝑠 to node 𝑡𝑡. s p t (c) Laura Albert 2019Two point conversion DP
  • 7. 7 Shortest Path Problem, cont’d. • Observation 1: If the shortest path from s to r passes through node p, the subpaths (𝑠𝑠, 𝑝𝑝) and (𝑝𝑝. 𝑡𝑡) are shortest paths from 𝑠𝑠 to 𝑝𝑝 and 𝑝𝑝 to 𝑡𝑡, respectively. – If not, the shorter subpath would allow us to construct a shorter path from 𝑠𝑠 to 𝑡𝑡, leading to a contradiction. • Observation 2: Let 𝑉𝑉𝑗𝑗 denote the length of a shortest path from 𝑗𝑗 to 𝑡𝑡, then 𝑉𝑉𝑗𝑗 = min 𝑖𝑖∈𝑉𝑉𝑖𝑖{𝑗𝑗} {𝑉𝑉𝑖𝑖 + 𝑑𝑑𝑑𝑑𝑑𝑑} In other words, the lengths of the shortest paths from every neighbor of 𝑗𝑗 to 𝑡𝑡, then we can find the length of the shortest part from 𝑗𝑗 to 𝑟𝑟. Note that this is not necessarily an algorithm for finding the shortest path. (c) Laura Albert 2019Two point conversion DP
  • 8. 8 Principle of Optimality • Principle of Optimality: Observations 1 and 2 define the optimal substructure of all DP models and algorithms. Each solution is characterized by – A sequence of decisions (more on this later), – The optimal solution is composed of optimal solutions to subproblems. • In the Shortest Path problem, the DP gives the shortest paths from all nodes to node 𝑡𝑡 – 𝑉𝑉𝑡𝑡 = 0. (c) Laura Albert 2019Two point conversion DP
  • 9. 9 Shortest Path DP algorithms • Since DP algorithms are smart enumerations, good DP algorithms are important to find. • Let Vj n = the shortest path from j to r at iteration n. • We want to find the optimal distances Vj by initially setting Vj 0 = large number (j ≠ r) • The idea: iteratively loop over all nodes to find the best edge to traverse – Because of the principle of optimality, we only improve (lower) Vj n. (c) Laura Albert 2019Two point conversion DP
  • 10. 10 Shortest Path DP algorithm 1 Bellman’s Algorithm 1. Let Vr 0=0 and Vj 0 = M, j ≠ r (M is a large number) and let n=1. 2. For all j ∈ V{r} Vj n=mini{dji+Vi n-1} 3. If Vj n < Vj n-1 for any i, let n=n+1 and return to step 2. Else stop. Note: only consider nodes where an edge from i to j exists. Problems: In step 2, most of the time we have to consider edges with distance M. A better algorithm only considers nodes with paths less than M (c) Laura Albert 2019Two point conversion DP
  • 11. 11 Characteristics of DP Problems 1. Problem divided into stages with a policy decision required at each stage. Shortest path = which node to visit next 2. Each stage has a number of states associated with it. 3. The effect of each policy decision in each state (at each stage) transforms the current state into a state in the next stage (deterministic or stochastic). 4. The solution procedure is designed to find the optimal policy. Prescription for finding the best decision at each stage for each state (state-action pairs) (c) Laura Albert 2019Two point conversion DP
  • 12. 12 Characteristics of DP Problems 5. Given current state, an optimal policy for the remaining stages is independent of the policy in the previous stages. This is the principle of optimality for dynamic programming 6. The solution procedure begins by finding the optimal policy for the last stage. This is usually trivial, such as 𝑉𝑉𝑡𝑡 = 0 7. A recursive relationship identifies the optimal policy for stage 𝑛𝑛, given an optimal policy for stage 𝑛𝑛 + 1 Called the value functions or optimality equations or Bellman equations or Hamilton-Jacobi-Bellman equations (c) Laura Albert 2019Two point conversion DP
  • 13. Dynamic Programs as shortest or longest paths • Almost any DP problem can be formulated as a shortest or longest path (by maximizing) in a graph with directed arcs (digraph) • Nodes of the digraph correspond to states in an incomplete solution • Arcs of the digraph correspond to decisions. They link each state to a subsequent state to which the decision leads. Implication for our football problem: • The nodes represent the triplet: (1) who has the possession, (2) the current score differential, (3) how many possessions are left. • We solve a longest path problem to maximize win probability (not distance) 13 (c) Laura Albert 2019Two point conversion DP
  • 14. Dynamic programming Let • 𝑉𝑉𝑛𝑛 𝐴𝐴 (𝑝𝑝) = the probability that team A wins the game if they are 𝑝𝑝 points ahead, they have just gotten the ball, and there are 𝑛𝑛 remaining possessions. • 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝) = the probability that team A wins the game if they are 𝑝𝑝 points ahead, their opponent team B has just gotten the ball, and there are 𝑛𝑛 remaining possessions. • 𝑛𝑛 = 0 means the game is over. (c) Laura Albert 2019 14Two point conversion DP
  • 15. Boundary conditions • 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 1 for 𝑝𝑝 > 0 • 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 0 for 𝑝𝑝 < 0 • 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 𝑉𝑉0 𝐴𝐴 𝑝𝑝 = 0.5 for 𝑝𝑝 = 0 [a tie] • Goal: maximizing 𝑉𝑉𝑛𝑛 𝐴𝐴(𝑝𝑝) for some 𝑛𝑛 and 𝑝𝑝 – Find by conditioning on the possession outcome and applying the law of probability (c) Laura Albert 2019 15Two point conversion DP
  • 16. Law of total probability • Let 𝑋𝑋 denote a random variable. – 𝑋𝑋 is a binary variable denoting the outcome of the game (win/loss) • We can find the probability of 𝑋𝑋 given some additional information 𝑌𝑌 – 𝑌𝑌 is a random variable denoting the outcome of the current possession (touchdown, field goal, or no score). [ ] ( | ) ( ) y Y P X P X y P y ∈ = ∑ (c) Laura Albert 2019 16Two point conversion DP
  • 17. Finding an expected value • Let 𝐼𝐼 be an indicator variable if the team wins: – 𝐼𝐼 = 1 if they win and 𝐼𝐼 = 0 if they lose. • The expected value of 𝐼𝐼 is the probability that the team wins 𝐸𝐸[𝐼𝐼] = 𝑃𝑃(𝐼𝐼 > 0) • 𝑃𝑃(𝐼𝐼 > 0) = 𝑉𝑉𝑛𝑛 𝐴𝐴 (𝑝𝑝) – This can be found by computing the expected value of an indicator variable 𝐸𝐸[𝐼𝐼] – This is the solution to a longest path problem! (c) Laura Albert 2019 17Two point conversion DP
  • 18. Law of total probability • 𝑃𝑃𝑇𝑇𝑇𝑇 = probability of a touchdown • 𝑃𝑃𝐹𝐹𝐹𝐹 = probability of a field goal • 𝑃𝑃𝑁𝑁𝑁𝑁 = probability of no score • 𝑃𝑃𝐴𝐴𝐴𝐴 = probability of scoring a point after a touchdown • 𝑃𝑃2𝑃𝑃𝑃𝑃 = probability of a two point conversion after a touchdown Three outcomes of the current drive: TD, FG, or NS ( 𝑃𝑃𝑇𝑇𝑇𝑇 + 𝑃𝑃𝑃𝑃𝑃𝑃 + 𝑃𝑃𝑃𝑃𝑃𝑃 = 1) 𝑉𝑉𝑛𝑛+1 𝐴𝐴 𝑝𝑝 = 𝑃𝑃𝑁𝑁𝑁𝑁 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝)  No score + 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝 + 3)  Field Goal + 𝑃𝑃𝑃𝑃𝑃𝑃 max{𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝 + 7) + (1 − 𝑃𝑃𝐴𝐴𝐴𝐴) 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝 + 6) , TD with AT 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝 + 8) + (1 − 𝑃𝑃2𝑃𝑃𝑃𝑃) 𝑉𝑉𝑛𝑛 𝐵𝐵 (𝑝𝑝 + 6) }  TD with 2PT Note: define 𝑉𝑉𝑛𝑛+1 𝐵𝐵 𝑝𝑝 in an analogous way (c) Laura Albert 2019 18Two point conversion DP
  • 19. Series of equations Team A: 𝑉𝑉𝑛𝑛+1 𝐴𝐴 𝑝𝑝 = 𝑃𝑃𝑇𝑇𝑇𝑇 max{ 𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 7 + 1 − 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 6 , 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵(𝑝𝑝 + (c) Laura Albert 2019 19Two point conversion DP
  • 20. Team A’s win probability Note: the points P are before the touchdown is scored V^A_n(p) Possessions remaining n P 0 1 2 3 4 5 6 7 8 9 10 -16 0 0.000 0.000 0.003 0.002 0.014 0.010 0.030 0.022 0.047 0.036 -15 0 0.000 0.000 0.006 0.004 0.020 0.014 0.038 0.029 0.058 0.044 -14 0 0.000 0.000 0.015 0.010 0.035 0.025 0.056 0.042 0.077 0.059 -13 0 0.000 0.000 0.024 0.016 0.050 0.036 0.075 0.056 0.097 0.075 -12 0 0.000 0.000 0.026 0.018 0.057 0.042 0.086 0.066 0.112 0.088 -11 0 0.000 0.000 0.035 0.026 0.074 0.056 0.108 0.084 0.136 0.108 -10 0 0.000 0.000 0.045 0.034 0.092 0.070 0.131 0.102 0.161 0.129 -9 0 0.000 0.000 0.062 0.046 0.119 0.090 0.162 0.126 0.193 0.154 -8 0 0.046 0.031 0.108 0.079 0.159 0.122 0.198 0.155 0.226 0.182 -7 0 0.091 0.062 0.154 0.113 0.199 0.153 0.234 0.186 0.260 0.212 -6 0 0.186 0.127 0.252 0.183 0.283 0.216 0.303 0.240 0.318 0.258 -5 0 0.190 0.135 0.271 0.202 0.311 0.242 0.334 0.268 0.350 0.288 -4 0 0.190 0.141 0.282 0.218 0.332 0.265 0.362 0.296 0.382 0.319 -3 0 0.255 0.198 0.362 0.288 0.409 0.332 0.430 0.356 0.441 0.371 -2 0 0.320 0.243 0.427 0.336 0.465 0.378 0.480 0.398 0.485 0.410 -1 0 0.320 0.251 0.440 0.357 0.489 0.405 0.509 0.430 0.518 0.444 0 0.5 0.660 0.500 0.622 0.500 0.601 0.500 0.588 0.500 0.579 0.500 1 1 1.000 0.749 0.805 0.643 0.714 0.594 0.667 0.569 0.639 0.555 2 1 1.000 0.757 0.817 0.662 0.736 0.621 0.695 0.600 0.670 0.588 3 1 1.000 0.803 0.849 0.713 0.774 0.668 0.733 0.644 0.708 0.628 4 1 1.000 0.859 0.896 0.782 0.832 0.734 0.789 0.703 0.758 0.681 Team A’s last possession Team A’s second to last possession(c) Laura Albert 2019 20Two point conversion DP Boundary conditions
  • 21. How do we determine whether to go for two points? Test the condition in the max statement to see what was selected: For Team A who gets the ball with 𝑛𝑛 + 1 possessions to go and is up by 𝑝𝑝 points 𝑉𝑉𝑛𝑛+1 𝐴𝐴 𝑝𝑝 𝑃𝑃𝐴𝐴𝐴𝐴 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 7 + 1 − 𝑃𝑃𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 6 < 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 8 + 1 − 𝑃𝑃2𝑃𝑃𝑃𝑃 𝑉𝑉𝑛𝑛 𝐵𝐵 𝑝𝑝 + 6 Then go for two points is better than going for one point. A similar approach works for Team B by solving 𝑉𝑉𝑛𝑛+1 𝐵𝐵 𝑝𝑝 (c) Laura Albert 2019 21Two point conversion DP
  • 22. When should team A go for 2 if they score a touchdown Note: the points P are before the touchdown is scored Possessions remaining n P 0 2 2 3 4 5 6 7 8 9 10 -10 1 1 1 1 2 2 2 2 2 2 1 -9 1 1 1 1 1 1 2 2 2 2 1 -8 1 1 1 1 2 2 2 2 2 2 1 -7 1 1 1 1 2 2 2 2 2 2 1 -6 1 1 1 1 1 1 1 1 1 1 1 -5 1 1 1 1 2 2 2 2 2 2 1 -4 1 1 1 1 2 2 2 2 2 2 1 -3 1 1 1 1 2 2 2 2 2 2 1 -2 1 1 2 2 2 2 2 2 2 2 1 -1 1 1 1 1 2 2 2 2 2 2 1 0 1 1 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 2 1 3 1 1 2 2 2 2 2 2 2 2 1 4 1 1 1 1 1 1 1 1 1 1 1 5 1 1 1 1 1 1 1 1 1 1 1 6 2 2 2 2 2 2 2 2 2 2 2 7 1 1 1 1 2 1 2 1 2 1 1 8 1 1 1 1 1 1 1 1 1 1 1 9 1 2 2 2 2 2 2 2 2 2 1 10 1 1 2 2 2 2 2 2 2 2 1 Team A’s last possession Team A’s second to last possession(c) Laura Albert 2019 22Two point conversion DP
  • 23. When should team A go for 2 if they score a touchdown Two point conversion DP (c) Laura Albert 2019 23 Team A’s last possession Team A’s second to last possession
  • 24. When should Team A go for 2? On their last possession (from 𝑉𝑉1 𝐴𝐴 𝑝𝑝 ): When 𝑝𝑝 = −8 i.e., when Team A is down 8 and then scores a touchdown Team A would be tied or down 2 after this possession. On their second to last possession (from 𝑉𝑉3 𝐴𝐴 𝑝𝑝 ): When 𝑝𝑝 = −16, −14, −11, −8, −5, −4, −2, −1 (Or 𝑝𝑝 = −10, −8, −3, −2, 1, 2, 4, 5 including the 6 points from the touchdown) Note: the points 𝑝𝑝 are before the touchdown is scored (c) Laura Albert 2019 24Two point conversion DP
  • 25. When should team A go for 2 if they score a touchdown Two point conversion DP (c) Laura Albert 2019 25 Go for it when down 2 Go for it when down 8 Go for it when up 1, 2, 4, 5 Go for 1 here, otherwise the other team could win with a FG
  • 26. When should Team A go for 2? When we set 𝑃𝑃𝐴𝐴𝐴𝐴 = 0.95 (from 0.96) the results change slightly: • Team A should go for 2 if they score a touchdown when down 15 before the touchdown • Note: the win probability does not change very much On their second to last possession (from 𝑉𝑉3 𝐴𝐴 𝑝𝑝 ): When 𝑝𝑝 = −16, −𝟏𝟏𝟏𝟏, −14, −11, −8, −5, −4, −2, −1 Note: the points 𝑝𝑝 are before the touchdown is scored How much does going for 2 improve the win probability? (c) Laura Albert 2019 26Two point conversion DP
  • 27. How much does going for 2 improve the win probability? Answer: not by much. Rationale: if you are down by 15, you have to score two touchdowns while keeping your opponent from scoring. • If down by 8 (before the 6 TD points) with 1 or 3 possessions to go, the strategy improves the win probability by 4.6% or 4.5%. • If down by 1, 5, or 11 (before the 6 TD points), the strategy improves the win probability by 0.6 – 1.8% depending on the remaining number of possessions • If down by 14, 15, or 16 (before the 6 TD points), the strategy improves the win probability by <1% In some scenarios, the win probability gets worse (!) • Your opponent is also finding the best strategy The win probability can be further improved by improving other decisions • E.g., Going for it on fourth down, better play calling Two point conversion DP (c) Laura Albert 2019 27
  • 28. Why would a team go for 2 when they are down by 8? Let’s say a team is down by 14 and scores a touchdown, putting them down by 8. Let’s say there will be 4 possessions left after this possession. • Traditional decision: go for 1 point. If they succeed (with probability 0.96), they will have a 11.3% probability of winning. If they do not succeed, they will have a 7.9% probability of winning Overall: 11.2% win probability • Going for a two point conversion If they succeed (with probability 0.48), they will have a 18.3% probability of winning. If they do not succeed, they will have a 7.9% probability of winning Overall: 12.9% win probability Two point conversion DP (c) Laura Albert 2019 28