SOLVING THE THREE PERSON GAME

IN GAME THEORY USING EXCEL

William P. Fox
Naval Postgraduate School

Abstract solution techniques. Our last lesson in the block

extends the two-person total conflict games to
We built templates to assist in the solving of the three-person total conflict games.
the three person game for both total and partial
conflict games. These solutions find any pure In the three-person games, we find Nash
strategie solutions for the players playing alone equilibrium via movement diagrams and then
and without communication. Then every break the game down into possible coalitions.
combination of coalitions between players is This pits two players versus the third player.
found and solved. Users must interpret the All possible coalitions are evaluated and their
results to determine if any coalition is more results are used to look for likely forming
likely than others to be formed. Additionally, coalitions.
the user must consider the use of bribes or side
payments to change the outcomes if any player Next, we visit the partial conflict games. After
would prefer a different outcome than the covering the techniques for finding equilibrium
results found. and negotiated solutions, we return to the three
person games. We cover the solution techniques
Key words: Total conflict games, partial for finding the Nash equilibrium and all the
conflict games, zero sum, nonzero sum games, possible coalitions to attempt to determine what
game theory, Nash equilibrium, prudential might happen.
strategies, security levels, Nash arbitration,
linear programming, nonlinear programming, Our students must complete a course project of
MS-Excel their own choice using one of the modeling
techniques from class. Students use the
Introduction modeling process in their project: they identify
the problem; they list the appropriate
In our interdisciplinary Department of Defense assumptions with justifications; they explain
Analysis at the Naval Postgraduate School, we why their modeling technique is selected; they
teach a three course sequence in mathematical solve the model; interpret the solution; perform
modeling for decision making. In the first sensitivity analysis (if applicable); and they
course, we usually teach basic linear discuss strengths and weaknesses of their
programming both using the two-variable modeling approach. The three person games add
graphical simplex technique and the Excel more reality to many of their projects. Here is a
Solver using SimplexLP. short list of some of the game theory projects:

In this 3rd course, we teach the basic concepts • Game Theory with US and Non-State
and solution techniques for game theory. In our actors.
class we use the Straffin text [8] as well as • Game Theory in Cameroon-Nigeria
Chapter 10 from Giordano, Fox, and Horton [4]. dispute.
We do not cover the basic solution techniques in • Game Theory in PMI and US military
this paper other than to illustrate the movement tasks.
diagram. In the course for total conflict games • US-China Game.
we first present the zero-sum and constant sum • The Somali Pirates game.
two-person games. We cover many of the • US-Afghanistan drug dilemma.
COMPUTERS IN EDUCATION JOURNAL 65
 • US-Afghanistan Regional Game. In a three person total conflict game (zero-sum
• US Coin Operations Game. or constant sum), the values in each triplet, (Ri,
• Dealing with Safe Havens as a Game. Ci, Li), sum to either zero or the same constant.
• IEDS and Counter-IEDS as a Game.
• Game theory for Courses of Combat In a three person nonzero-sum game the
Actions values in each triplet, (Ri, Ci, Li), do not all sum
• Game Theory and Dark Money Networks to zero nor do they sum to the same constant.
• Dealing with ISIS
• Russia, US, and Ukraine We also make the following assumptions
• Dealing with Snowden as a three person about the game:
game: Snowden, US Government, &
Russia. Games are simultaneous
Players are rational meaning they want the best
In the past, our coverage did not cover much outcome possible versus their opponents.
linear programming or nonlinear programming, Games are repetitive.
so our solution processes were limited to two- Players have perfect knowledge about their
person, two strategy games using the algebraic opponents.
method or other short cut methods because of
the complexity of the solution mechanics. 3-Person Total Conflict Games
Recently, we have added more applications of
linear programming and a non-linear template as The solution methodology of the three person
a solution technique so students might add more total conflict games involves several steps. First,
reality to the number of possible strategies we use the movement diagram, as we will
available to the players. describe, to find all the Nash equilibriums. The
Nash equilibrium is defined when no player
Let’s define a generic simultaneous three would unilaterally change their outcomes.
person game theory payoff matrix as shown in
Table 1. We give Larry two strategies {L1, L2}, Consider the following three person (total
Colin two strategies {C1, C2} and Rose two conflict) zero-sum game between Rose, Colin,
strategies {R1, R2}. and Larry (from Straffin, Chapter 19) shown in
Table 2.

Larry Larry
L1 L2
Colin
C1 C2 C1 C2
Rose R1 (R1,C1,L1) (R1,C2,L1) R1 (R1,C1,L2) (R1,C2,L2)

R2 (R2,C1,L1) (R2,C2,L1) R2 (R2,C1,L2) (R2,C2,L2)

Table 1. Generic three person game between Rose, Colin, and Larry.

66 COMPUTERS IN EDUCATION JOURNAL

 Larry L1
Colin
C1 C2
Rose R1 (1,1,-2) (-4,3,1)
R2 (2,-4,2) (-5,-5,10)

Larry L2
Colin
C1 C2
Rose R1 (3,-2,-1) (-6,-6,12)
R2 (2,2,-4) (-2,3,-1)
Table 2. Three person game example (Source: Straffin, Chapter 19).

Movement Diagram value 2 in R2 is greater than the value 1 in R1 so

the arrow goes from R1 to R2. For Colin,
We define a movement diagram as follows for arrows are drawn horizontally between C1 and
each player’s possible outcomes R1 or R2, C1 C2 from smaller values to larger values. For
or C2, and L1 or L2, draw an arrow from the Larry, arrows are drawn diagonally to represent
smallest to the largest value. For Rose arrows the two games, L1 and L2 with arrows drawn
are drawn vertically from smaller to larger. For from corresponding positions. This is illustrated
example, under Larry L1 and Colin C1, the in Figure 1.

Larry L1
Colin

C1 C2

Rose R1 (1,1,-2) (-4,3,1)

R2 (2,-4,2) (-5,-5,10)

Larry L2
Colin
C1 C2

Rose R1 (3,-2,-1) (-6,-6,12)

R2 (2,2,-4) (-2,3,-1)

Figure 1. Movement diagram for three person zero sum game.

COMPUTERS IN EDUCATION JOURNAL 67
 We follow the arrows. If any set or sets of Coalitions Possible
arrows bring us to a point where no arrow
leaves that point or points then we have an Let’s consider communications with the
equilibrium point or points. Result: The ability to form coalitions. Assume first that
movement diagram reveals two pure strategy Colin and Larry form a coalition against Rose.
Nash equilibriums at R1C1L2 (3,-2,-1) and at The following steps are helpful in the setting up
R2C1L1 (2,-4, 2). These are not equivalent and and analysis of the coalition.
not interchangeable. Going for one equilibrium
point over another by either player may lead to a Step 1. Build a payoff matrix for Rose against
non-equilibrium outcome because of player’s the Colin-Larry coalition using Rose’s values
preferences. from the original payoffs as follows:

Colin-Larry
C1L1 C2L1 C1L2 C2L2
Rose R1 1 -4 3 -6
R2 2 -5 2 -2

Step 2. Try to find a solution for the Nash The graph, Figure 2, shows that the Maximin
equilibrium using either: a) Saddle points solution is found by using the following values
(maximin) or b) Mixed strategies. for Rose versus the Coalition. We can easily
find the solution.
a) No saddle point solution RowMin {-6,
-5} ColMax {2, -4 ,3, -2} If the game has a saddle point solution, those
values are the value of the game for all three
b) players. Since we have a mixed strategy then we
must find the value for each of our three players.
Rose R1 Rose R2

Figure 2. William's graphical method to

eliminate strategies (rows) not used to obtain
the solution.
68 COMPUTERS IN EDUCATION JOURNAL
 Colin-Larry
C2L1 C2L2 Oddments
Rose R1 -4 -6 2 3/5
R2 -5 -2 3 2/5
Oddments 1 4
4/5 1/5 Value -22/5 or
-4.4

Step 3. Finding the values of the game for each Larry: Max {2, -1.43, 5.04) is 5.04 so Larry
player. prefers a coalition with Colin.

3 4 31 24 21 In two of these cases we find that Colin-Larry

⋅ R1C 2 L1 + R1C 2 L 2 + R 2C 2 L1 + R 2C 2 L 2 is the preferred coalition so we might expect
5 5 55 55 55
that the Colin-Larry coalition will naturally be
We now substitute the values from the original the coalition formed. We note that we may or
payoff matrix. may not be able to determine which coalition
might be formed. We also note that there are
3 4 3 1 both bribes and side payments allowed. These
· (−4,3,1) + · (−6, −6, 12) +
5 5 5 5 bribes or payments entice a coalition to either
change or keep the coalition together.
2 4 2 1
· (−5, −5, 10) + · (−2, 3, 01)
5 5 5 5 Characteristic function: The number v(S),
called the value of S, is to be interpreted as the
= (-4.4, -0.64, 5.04) amount S would win if they formed a coalition.
We assume that the empty coalition (none are
We find the payoffs are to Rose -4.4, to Colin - formed) value is zero, v(∅) =0
.64, to Larry 5.04
Colin versus Rose-Larry (2,-4,2)
Step 4. Redo steps 1-3 for Colin versus a Larry versus Rose-Colin (2.12,-.69,-1.43)
coalition of Rose-Larry and then redo steps 1-3 Rose versus Colin-Larry (-4.4,-.64,5.04)
for Larry versus a coalition of Rose-Colin.
We can build the functions:
Results are as follows: Empty set: v(∅) =0
Alone: v(Rose) = -4.4, v(Colin) = -4,
Colin versus Rose-Larry: Value of (2, -4,
v(Larry) = -1.43
2) and this was the saddle point solution.
Coalition by two(s):
Larry versus Rose-Colin (2.12, -0.69, -
v(Rose-Colin) = 1.43 v(Rose-Larry) = 4
1.43)
v(Colin-Larry) = 4.4
Rose versus Colin-Larry (-4.4, -0.64, 5.04)
from before. We add the payoff for the coalition’s
partners in the associated games.
Step 5. Determine which coalition, if any,
yields the best payoff for each player. Coalitions by three: These are zero-sum
games so adding all payoffs together =0
Rose: Max { 2, 2.12, -4.4} is 2.12 so Rose v(Rose-Colin-Larry) = 0
prefers a coalition with Colin.
Colin: Max {-4,-0.69,-0.64} is -0.64 so Thus,
Colin prefers a coalition with Larry. Larry versus Rose-Colin (2.12,-.69,-1.43)

COMPUTERS IN EDUCATION JOURNAL 69

 Rose- No saddle point exists since Max of {-2,-4} is
Colin -2 and Min of {-1,12,2,10} is -1. We move on to
R1C1 R1C2 R2C1 R2C2 find the mixed strategies.
Larry L1 -2 1 2 10
L2 -1 12 -4 -1

Larry L1 Larry L2

Maximin

Subgame:

Rose-Colin
R1C1 R2C1 Oddments
L1 -2 2 4 3/7
Larry L2 -1 -4 3 4/7
Oddments 1 6
6/7 1/7 Value is -10/7

(3/7)*(6/7)* (1,1,-2) + (3/7)*(1/7)*(2,-4,2)+(4/7)*(6/7)*(3,-2,-1)+(4/7)*(1/7)*(2,2,-4) = (104/49, -

34/49 , -10/7)

=(2.12,-.069,-1.43) rounded to 2-decimal places.

Although the mathematics is not difficult the Technology Assistant with EXCEL
number of calculations is quite tedious.
Therefore, we built a technology assistant We developed a technology assistant to assist
for student use. the students with the many calculations
involved. Instructions are provided within the
template, which is a macro-enhanced Excel
worksheet. These instructions include:

70 COMPUTERS IN EDUCATION JOURNAL

 (1) Put the R,C,L entries into the blocks to the left

(2) Go to Coalition_R_CL and execute the Solver

(3) Go to Coalition_C_RL and execute the Solver

(4) Go to Coalition_L_RC and execute the Solver

(5) List the equilibrium values if the players play alone and the equilibriums in the three coalitions

(6) Determine if any coalition naturally forms

(7) Is there a legitimate bribe to change the coalition?

In Figure 3, we find the results or outcomes of user must then interpret the results and make
the calculations made to find the pure strategies conclusions about those results as to what is
equilibrium and the results of the coalitions. The likely to occur.

Figure 3. Screen shot of 3 person game template with instructions.

N-Person Games with Linear Programming illustrate with a three person zero-sum game
that we just saw in the previous example.
The coalition’s solution on each worksheet Recall, we created the game payoffs for the
uses the Solver, specifically SimplexLP. We potential coalitions:

COMPUTERS IN EDUCATION JOURNAL 71

 Colin-Larry
C1L1 C2L1 C1L2 C2L2
Rose R1 1 -4 3 -6
R2 2 -5 2 -2

This coalition game is a zero sum game. We Larry vs Rose-Colin (2.12,-0.69,-1.43)

solve for Rose’s solution in Excel with
SimplexLP. We get the Colin-Larry coalition’s It is still up to the user to interpret and analyze
results from the sensitivity column in Excel. these results. These procedures work for
Note there are some negative entries as payoffs constant sum games as well.
so we let v = V1-V2 (see Winston 1995). We
formulate the LP. A 3-person Game that is a Strict Non-Zero
Sum Game Using Technology
Maximize v= V1-V2
x1+2x1-V1+V2>0 We also developed an assistant for the partial
-4x1-5x2-V1+V2>0 conflict game. This technology assistant
3x1+2x2-V1+V2>0 requires the use of the Solver six times in the
-6x1-2x2-V1+V2>0 spreadsheet since each player or side in a
x1+x2=1 coalition requires a linear programming
x1<1 solution. The instructions are listed inside the
x2<1 template.
non-negativity
The results here are as follows:
We find the LP solution to this game for Rose is
v =-4.4, when x1 = 0.6 and x2 = 0.4. We find Pure strategy by movement diagram finds an
from the reduced costs (the dual solution for equilibrium at R1C1L2 with values (2,1,1)
Colin & Larry coalition), is Vcl=4,4, when Equilibrium
y1=y3=0, y2 = 0.8 and y4=0.2. R1C1L1 No
R1C2L1 No
Although this gives us a Coalition value, we
R2C1L1 No
must use all the probabilities for the players to
obtain the values to each of our players R2C2L1 No
separately. We only have to use the strategies R1C1L2 Yes
with probabilities greater than 0: R1C2L2 No
R2C1L2 No
(.6)(.8) R1C2L1 +(.4)(.8) R2C2L1 + (.6)(.2) R2C2L2 No
R1C2L2 + (.4)(.2) R2C2L2
We easily see a better set of values as an
.48 (-4,3,1) + .32 (-5,-5,10) +.12 (-6,-6,12) output of (4,2,3) at R1C2L1. We analyze all
+ .08 (-2,3,-1) = (-4.4, -0.64,5.04) coalitions to see if that solution rises from any
coalitions.
Rose loses -4.4 (as shown before) and the
Coalitions 4.4 is broken down as -0.64 for Colin
and 5.04 for Larry.

We repeat this process for each Coalition to

obtain these results:

Colin vs Rose-Larry (2, -4, 2)

72 COMPUTERS IN EDUCATION JOURNAL
From the linear programming solution of the coalitions, we find:

To Players
Rose Colin Larry
Coalition One 1.5 1 1
Rose VS Colin-
Larry

Coalition Two 1.75 0.5 0.75

Colin vs Rose-
Larry

Coalition Three 1.5 1.5 1

Larry vs Rose-
Colin

Rose prefers a coalition with Larry, Colin There is no preferred coalition and none gets us
prefers a coalition with Rose, and Larry prefers to the better value.
either a coalition with Colin or being alone.

COMPUTERS IN EDUCATION JOURNAL 73

 Perhaps all the players should just all agree to Biographical Information
play the strategies that provide the best solution.
William P. Fox is a professor at The Naval
Conclusions Postgraduate School in Monterey, California.
He obtained his Ph.D. degree in Industrial
We have described the use of Excel templates Engineering and Operations Research from
to assist in the solution to the three person Clemson University and his M.S. degree in
games. We remark that users must still analyze Operations Research from the Naval
the numerical values to determine what will Postgraduate School. His research interests
most likely happen. The author will provide include modeling, optimization, game theory,
these templates upon request. Email requests to and simulation. He has many conference
[email protected]. presentations including: INFORMS,
Mathematical Association of America Joint
References Annual Conference, Military Application
Society (MAS), and the International
1. Fox, W.P. (2010). Teaching the Conference of Technology in Collegiate
applications of optimization in game Mathematics (ICTCM). He has coauthored
theory’s zero-sum and non-zero sum several books and over one hundred articles. He
games, International Journal of Data has previously taught at both West Point and
Analysis Techniques and Strategies Francis Marion University. He was the Director
(IDATS), 2(3), pp. 258-284. of both the High School Mathematical Contest
in Modeling (HiMCM) and the collegiate
2. Fox, W. (2012). Mathematical modeling Mathematical Contest in Modeling (MCM) and
with Maple, Cengage Publishers, is currently the Past-President of the Military
Boston: MA, pp. 221-227. Applications Society of INFORMS.

3. Fox, W. P. (2015). The Partial Conflict

Game Analysis without Communication
in EXCEL, Computers in Education
Journal, 6(4), pp. 2-10.

4. Straffin, Philip (2003). Game theory and

strategy. The Mathematical Association
of America. Washington: DC. , Chapter
19.

5. Winston, Wayne (1995). Introduction to

mathematical programming:
Applications and algorithms. Duxbury
Press, Belmont: CA.

74 COMPUTERS IN EDUCATION JOURNAL