li

NOgrH. HOILAND

The Decision Support System G M C R

in Environmental Conflict Management
Keith W. Hipel

Departments of Systems Design Engineering and

Statistics and Actuarial Science
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1

D. Marc Kilgour
Department of Mathematics
Wilfrid Laurier University
Waterloo, Ontario, Canada N2L 3C5
and
Department of Systems Design Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1

Liping Fang
Department of Mechanical Engineering
Ryerson Polytechnic University
350 Victoria Street
Toronto, Ontario, Canada M5B 2K3
and
Department of Systems Design Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1
and
Xiaoyong (John) Peng
Department of Systems Design Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1

APPLIED MATHEMATICSAND COMPUTATION83:117-152 (1997)

© Elsevier Science Inc., 1997 0096-3003//97/$17.00
655 Avenue of the Americas, New York, NY 10010 PII S0096-3003(96)00170-1
118 K. W. HIPEL ET AL.

ABSTRACT

The graph model for conflict resolution is put forward as a unique decision
technology for systematically studying strategic conflicts that can arise in environ-
mental management and elsewhere. This flexible conflict resolution methodology has
solid and realistic mathematical foundations which permit it to model strategic
decisions accurately, forecast compromise solutions, and assist in assessing the
political, economic, environmental, and social viability of alternative scenarios to
resolving environmental conflicts. To permit practitioners to conveniently apply the
graph model for conflict resolution to practical real-world conflict problems, a new
decision support system called GMCR II (Graph Model for Conflict Resolution II) is
designed. To explain the basic structure of GMCR II and how it works in practice,
this paper describes its use to model and analyze an international environmental
management dispute that took place in North America. © Elsevier Science Inc.,
1997

INTRODUCTION

The exploitation of the earth's natural resources and the preservation of

the natural environment are two social goals that are by definition in direct
conflict with each other. To approach the ideal of sustainable development,
decision makers attempt to reach a suitable balance between these two
contradictory objectives. Nonetheless, resource development projects usually
affect a wide range of interest groups that perceive reality from different
perspectives and thereby find themselves in serious conflict. In order to
discover resolutions that are viable both socially and politically, yet sustain-
able from environmental and economic perspectives, leaders and managers
require techniques to assist them in understanding strategic decision mak-
ing. Accordingly, the main objective of this paper is to employ an interna-
tional resource management conflict as an illustrative example to demon-
strate how the graph model for conflict resolution, and the accompanying
decision support system GMCR II (Graph Model for Conflect Resolution II)
can be used to understand environmental disputes, advise decision makers,
and forecast compromise solutions. Using a conflict analysis study, a deci-
sion maker can identify those courses of action that fall within the social and
political constraints of the problem, yet are most likely to lead to a
favorable resolution that is both justifiable and sustainable. For an overview
of sustainable development and a list of some key references on this topic,
refer to Levy et al. [1].
Innovative approaches to the study of a wide variety of conflict situations
that can arise in the real-world include the metagame analysis method of
Howard [2] and the conflict analysis technique of Fraser and Hipel [3, 4].
Environmental Conflict Management 119

The recently developed graph form of the conflict model extends and refines
these methodologies to describe more accurately the behavior of participants
in a strategic conflict [5-7]. The graph form takes states, rather than
individual decisions, as the basic units for describing a conflict. A rich
variety of possible social behavior in a conflict can then be represented by
appropriate solution concepts. This paper focuses upon conflict resolution
methodologies developed by the authors and their colleagues. Overviews of
game-theoretic models and their use in decision making, plus key references
for various decision models, are provided in [6, 8].
In the next section, the theory underlying the graph model for conflict
resolution is outlined. Subsequently, the option form for representing a
conflict is explained using an international resource dispute. This conflict
was occasioned by a proposed open-pit coal mining development in the
province of British Columbia, Canada, along a tributary of the Flathead
River, which flows into the state of Montana. Because effluent from the
mining operation would have adverse effects downstream, especially on
recreational water use, the development faced international opposition from
the federal government of the United States and the state government of
Montana. In accordance with the Boundary Waters Treaty of 1909, the
governments of the United States and Canada requested the International
Joint Commission (IJC) to examine the entire situation.
In the following section, a general design for the new decision support
system GMCR II is presented. Its objective is to allow practitioners to
conveniently and expeditiously apply the graph model approach to an actual
conflict. To demonstrate the efficacy of the graph modeling approach to
conflict resolution in resource development, GMCR II is employed for
modeling and analyzing this international environmental conflict at a crucial
point in its evolution, just before the IJC made its recommendation. In fact,
the conflict study presented in the paper constitutes a revised and expanded
version of earlier analyses carried out in [9, 10]. The conference paper [9]
contains material about GMCR II complementary to what is presented here.

T H E O R E T I C A L FOUNDATIONS OF THE GRAPH MODEL

FOR CONFLICT RESOLUTION

A conflict model is a systematic structure for describing the main

characteristics of a strategic conflict. Subsequent to formulating the model,
one can employ it as a basic structure within which the possible strategic
interactions among the decision makers (DMs) can be extensively analyzed
in order to ascertain the possible compromise resolutions, or equilibria. The
output from this stability analysis as well as related sensitivity analyses can
be useful, for example, to help support decisions made by people who can
120 K . W . HIPEL ET AL.

exercise real power in a conflict. The next two subsections outline some of
the key ideas behind modeling and stability analysis, respectively. Following
a description of how to model a strategic conflict in option form, the
computer implementation of these ideas in the decision support system
G M C R II is described; then G M C R II is applied to the environmental
conflict.

Modeling
The graph model for conflict resolution represents a conflict as moving
from state to state (the vertices of a graph) via transitions (the arcs of the
graph) controlled by the DMs. One inherent advantage of the graph model is
that it can incorporate irreversible moves, whereby a DM can unilaterally
move from state k to state q but not from q to k. Examples of irreversible
moves are represented later in the case study. Another key advantage of the
graph model is its ability to easily describe c o m m o n moves, in which more
than one DM can cause the conflict to move from one particular state to
another common state. For instance, the greenhouse effect can be caused by
greenhouse gases emitted from any one of m a n y countries around the world.
A graph model for a conflict consists of a directed graph and a payoff
function for each DM who can affect the dispute. Let N = {1, 2 , . . . , n}
denote the set of DMs and U = {1, 2 , . . . , u} the set of states or possible
scenarios of the conflict. A collection of finite directed graphs D i = (U, A i),
i ~ N, can be used to model the course of the conflict. The vertices of each
graph are the possible states of the conflict and therefore the vertex set, U,
is common to all graphs. If DM i can unilaterally move (in one step) from
state k to state q, there is an arc with orientation from k to q in A 4. In the
next main section, it is shown how the option form can be used in G M C R II
to conveniently generate the states and the graphs in a model.
DM i's graph can be represented by i's teachability matrix, R i , which
displays the unilateral moves available to DM i from each state. For i ~ N,
R , is the u × u matrix defined by

k, q) = / ! if DM i can move (in onestep) from state k to q

R~(
0 otherwise

where k q: q, and by convention

R,(k, k) = 0.

A more economical expression of DM i's decision possibilities is his or her

reachable list. For i ~ N, DM i's reachable list for state k ~ U is the set
Environmental Conflict Management 121

S i(k) of all states to which DM i can move (in one step) from state k.
Therefore,

S i ( k ) = {q ~ U: Ri(k, q) = 1}.

For each DM i ~ N, a payoff function Pi: U -~ R, where R is the set of

real numbers, is assumed. The payoff function for DM i, Pi, measures how
preferred a state is for i. Thus, if k, q ~ U, then Pi(k) >i Pi(q) iff i prefers
k to q, or is indifferent between k and q. When this inequality is strict for
all pairs of distinct states for every DM, the conflict is called strict ordinal.
Beyond the ordinal information of preference or indifference, nothing can be
inferred from the value of P~. For instance, Pi(k) > P~(q) indicates that i
prefers k to q, but the value of Pi(k) - P,(q) gives no meaningful informa-
tion about the strength of this preference. For convenience, small positive
integers are used as the values of Pi(').
A unilateral improvement from a particular state for a specific DM is any
preferred state to which the DM can unilaterally move. To represent
unilateral improvements, each DM's reachability matrix can be used to
define a matrix R + , according to

1 if Ri(k, q) = 1 and Pi(q) > P~(k)

R +(k,q) = 0 otherwise

Similarly, DM i's reachable list, Si(k), can be replaced by S+(k), defined by

s : ( k ) = {q s,(k): R : ( k , q) = 1}.

Thus, S + ( k ) is called the unilateral improvement list of DM i from state k.

Stability Analysis
The stability analysis of a conflict is carried out by determining the
stability of each state for every DM. A state is stable for a DM iff that DM
has no incentive to deviate from it unilaterally, under a particular stability
definition or solution concept. A state is an equilibrium or possible resolu-
tion under a particular solution concept iff all DMs find it to be stable under
that stability definition.

Overview of Solution Concepts

A solution concept constitutes a mathematical description of a behavior
pattern. Because DMs can react to conflict situations in m a n y ways, there
122 K. W. HIPEL ET AL.

are many different solution concepts. Fang et al. define [6, Chapter 3] and
mathematically compare [6, Chapter 5] the solution concepts listed in Table
1. Furthermore, they demonstrate how the graph model and its associated
solution concepts can be equivalently expressed as extensive form games
[6, Chapter 4].
The solution concepts in Table 1 can be applied to conflicts having two or
more DMs. The first column gives the names of the solution concepts and
their acronyms, while the second one provides original references. The last
three columns furnish qualitative characterizations of the solution concepts
according to the three criteria of foresight, disimprovements, and knowledge
of preferences. Foresight refers to the ability of a DM to think about
possible moves that could take place in the future. A DM who has high or
long foresight can imagine many moves and countermoves into the future
when evaluating where the conflict could end up because of an initial
unilateral move on his or her part. Notice, for example, that in Nash
stability foresight is low, whereas, it is very high for nonmyopic stability.
The criterion of strategic disimprovements in the fourth column means that
a particular DM may temporarily move to a worse state in order to reach a
more preferred state eventually. Disimprovements by opponents indicate
that the other DMs may put themselves in worse positions in order to block
unilateral improvements by the given DM. The last criterion, knowledge of
preferences, refers to the preference information used in a stability analysis.
For example, under R, GMR, or SMR, a DM need not know the preferences
of all other DMs, although their ability to move unilaterally to other states

TABLE 1
SOLUTION CONCEPTS AND HUMAN BEHAVIOR

Knowledge of
Solution Concepts References Foresight Disimprovements Preferences
Nash Stability (R) [11, 12] Low Never Own
General [2] Medium By opponents Own
metarationality
(GMR)
Symmetric [2] Medium By opponents Own
metarationality
(SMR)
Sequential stability [3, 4] Medium Never All
(SEQ)
Limit-move stability [7, 13, 14] Variable Strategic All
(Lh)
Nonmyopic stability [7, 13, 15, 16] High Strategic All
(NM)
Environmental Conflict Management 123

is taken into account. These solution concepts remain useful in situations

where there is high uncertainty by a particular DM over the preferences of
the others.

Unilateral Moves
For a conflict involving more than two DMs, one must consider sequen-
tial moves involving more than one DM. Let H ___N be any subset of the
DMs. Then SH(k) denotes the set of all states that can result from any
sequence of unilateral moves by some or all of the DMs in H, starting at
state k. In this sequence, the same DM may move more than once, but not
twice consecutively. If all of the DM's graphs are transitive, there is always
an equivalent single move to any sequence of moves in which one DM moves
twice consecutively. The formal definition of a unilateral move by H c N is

UNILATERAL MOVE. Let k ~ U and H c N , H ¢ Q. A unilateral move

by H is a m e m b e r of Sg(k) C U, defined inductively by
1. if j ~ H and k I ~ Sj(k), then k 1 E SH(k) ,
2. i l k 1 ~ SH(k) , j ~ H, and k 2 ~ Sj(kl) , then k 2 ~ SH(k).
For computational implementation, the induction stops as soon as no new
state (k 2) is added to SH(k).
By replacing S H ( k ) , Sj(k), and Sj(k 1) by S~(k), S~(k), and S+(kl),
respectively, in the above definition, one obtains the definition of a unilat-
eral improvement by H when all DMs' graphs are transitive. More general
definitions are given in [6, Section 3.4] for the case when the graphs can be
intransitive.
SH(k) and S ~ ( k ) can be thought of as H ' s reachable list and unilateral
improvement list, respectively. Specifically, the sets Sg_i(k) and S~v_i(k)
represent the results of all possible "response sequences" by i's opponents
against a move by i to k.

Illustrative Definitions of Solution Concepts

To provide the reader with an appreciation of possible types of strategic
interactions that can take place among the DMs involved in a dispute, the
first four solution concepts listed in Table 1 are defined for conflicts with
two or more DMs. Definitions for all of the solution concepts along with
illustrative examples are provided in [6, Chapter 3].

NASH STABILITY. Let i ~ N . A state k ~ U is Nash Stable ( R ) f o r D M

i i/f S ~ ( k ) = ~ .
124 K . W . HIPEL ET AL.

Under Nash stability, DM i chooses his or her actions without consider-

ing possible responses.

GENERAL METARATIONALITY. F o r i ~ N , a state k ~ U is general

metarational ( G M R ) for D M i iff for every k 1 ~ S+(k) there is at least one
state k s ~ SN_i(k 1) with Pi(k~) <_ Pi(k).
Under general metarationality, DM i expects that the other DMs (N - i)
will respond to hurt i, if it is possible for them to do so, hy any sequence of
unilateral moves. DM i anticipates that the conflict will end after the DMs
of N - i have responded. Note that DM i's opponents axe assumed to
ignore their own payoffs when making their sanctioning moves.

SYMMETRICMETARATIONALITY. For i ~ N , a state k ~ U is symmetric

metarational ( S M R ) for D M i iff for all k I E S+(k), there exists k~
SN_i(kl) , such that Pi(k~) < Pi(k) and Pi(ka) < Pi(k) for all k a ~ Si(kx).
The SMR solution concept postulates that DM i expects that he or she will
have a chance to counterrespond (k a) to the other DMs' response (k~) to i's
original move (kl). DM i anticipates that the conflict will end after this
counterresponse.

SEQUENTIALSTABILITY. For i ~ N, a state k ~ U is sequentially stable

(SEQ) for D M i iff for every k 1 ~ S~(k) there is at least one state
k~ ~ S~v_i(k 1) with Pi(k~) < P~(k).
The difference between GMR and SEQ stability is SEQ's requirement that
sanctions be credible (by being unilateral improvements for the sanctioning
DMs), as well as feasible, which is all that is required by GMR.

Advantages of the Graph Model

Fang et al. [6, pp. 21-23] list many benefits of the graph model for
conflict resolution when used to solve practical real-world problems. Specific
advantages over other approaches to representing a conflict include the
graph model's ability to handle conveniently:

IRREVERSIBLEMOVES. An irreversible move occurs when a DM can cause

a conflict to move from state k to q by a unilateral move, but cannot make
the transition back from q to k. Irreversible moves axe often present in the
real-world applications.
Environmental Conflict Management 125

COMMON MOVES. A common moves takes place when either of two (or
more) DMs can independently make a unilateral choice that causes the game
to move from one or more initial states to one specific final state.

FORCING MOVES. A forcing move occurs when a DM makes a move that

implies that one or more other DMs have no choice but to make a certain
response; effectively, the original DM achieves the final state "unilaterally".

PATH-DEPENDENT TRANSITIONS. Path-dependent transitions take place

when there are several ways (paths) for two or more DMs to move jointly
from one state to another. The specific details of such sequences are useful
when assessing possible coalitions in a model with more than two DMs.

R E P R E S E N T A T I O N OF C O N F L I C T S IN O P T I O N F O R M

The graph model for conflict resolution outlined theoretically in the

previous section provides the foundation stones for the decision support
system G M C R II described in the next main section. At the modeling stage,
it is often convenient to represent a given conflict in option form, from
which the states in the dispute can be readily calculated and used in the
ensuing stability analysis. The option form can be conveniently employed
with small, medium, and large conflicts. Moreover, in m a n y conflict situa-
tions the option form provides a natural way to express some of the key
aspects of a conflict, including any finite number of DMs, the courses of
action available to each DM, and each DM's preferences. In order to clearly
explain how option form works in practice, the environmental dispute
described in the next subsection is utilized.

Background of the Flathead River Resource Development Dispute

A brief summary of the background of the Flathead River resource
development dispute is given here, based on [17-19]. The Flathead River
Region is shown in Figure 1. The Flathead River flows from its source in the
southeastern corner of British Columbia south across the international
boundary into Montana, then into Flathead Lake, and eventually into the
Columbia River. The discovery of coal in the British Columbia portion of
the Flathead valley was first reported in 1910. In 1970, a new company
called Sage Creek Coal Limited was formed to explore and develop the site.
Its Stage I proposal for developing the mine was approved by the province of
British Columbia in October 1976. In February 1984, the government of
126 K . W . HIPEL ET AL.

• e

BRITISH
COLLIMB1,4

,CANADA
U.S.A.

FIG. 1. Developmentlocation in the Flathead River drainage basin.

British Columbia granted Sage Creek Coal Limited approval-in-principle for

Stage II of the p r o p o s a l - - a 2.2 million tonnes per year thermal coal mine
located on Howell and Cabin Creeks, tributaries of the Flathead River, 10
km upstream from the international boundary. The mine plan was based on
21 years of planned operation at the proposed extraction rate, but coal
reserved would likely justify a further 20 years of mining at this same rate.
The governments of the United States and Montana were concerned
about the possible effects of this proposed mine on the Flathead River
system, Glacier National Park, and Flathead Lake. In response to these
concerns, the United States and Canadian governments requested that the
International Joint Commission examine the possible impacts of the pro-
posed mine on water quality and quantity, fisheries, and other water uses
associated with the Flathead River, and make recommendations.
Environmental Conflict Management 127

Option Form Model of the Flathead River Resource Development Dispute

The Flathead River Resource Development Dispute is modeled at a time
just prior to the IJC decision in December, 1988. The main DMs were:
1. Sage Creek Coal Limited (Sage Creek). Sage Creek Coal Limited was
owned by Rio Algom Mines (60%) and Pan Ocean Oil Limited (40%). As of
1988, Sage Creek already had substantial financial and other commitments
to the Flathead River Development project. Sage Creek hoped for an IJC
recommendation that would favor the continuation of the Cabin Creek coal
development, which would pave the way for the province of British Columbia
to issue a full Stage II license.
2. Province of British Columbia (British Columbia). The provincial
government had the authority to grant the license for mining but had to be
concerned with the environmental impact assessment. It would be subject to
pressure from the federal government of Canada and from the U.S.
3. State of Montana (Montana). The Montana government feared that
the proposed mining development would result in transboundary pollution
and environmental degradation. Environmental groups and the U.S. Depart-
ment of Interior strongly agreed with the Montana state government on this
issue.
4. International Joint Commission (IJC). The IJC is defined under
Section IX of the Boundary Waters Treaty of 1909 and is composed of three
members from Canada and three from the U.S. Its mandate includes making
recommendations and judgments on conflicts falling under this treaty.
Other DMs, such as the federal governments of Canada and the U.S.,
were considered not to be directly involved in the dispute at this stage.
Environmental groups, including some from Canada, were so closely identi-
fied with the Montana state government that they could be included with it.
The DMs and their options are listed in Table 2. As of December, 1988,
Sage Creek could continue the original project, modify it to reduce environ-
mental impact, or stop it all together by not selecting either the option to
continue or to modify. The government of British Columbia could license
the original development, or a project having appropriate modifications, or
it could deny a license for the project by not granting any license. Montana
could continue to oppose any development or drop its active opposition by
not opposing. The IJC could support either the original, or a modification,
or no project at all.
The option representation of a state is shown as a column of Y's and N's,
as in Table 2. A " Y " indicates "Yes" the option is taken by the DM
controlling it, whereas, an "N" means "No", the option is not selected. A
strategy is a choice made when a given DM decides which of his or her
options to invoke. For example, in Table 2, Sage Creek is selecting the
128 K . W . HIPEL ET AL.

TABLE 2
THE DMS AND OPTIONS OF THE FLATHEAD RIVER
RESOURCE DEVELOPMENT DISPUTE

DMs and Options Status Quo

Sage Creek
1. Continue: Continue original development Y Strategy for
2. Modify: Modify to reduce impact N Sage Creek
British Columbia
3. Original: Support original project Y Strategy for
4. Modification: Require modification N British Columbia
Montana
5. Oppose: Oppose any development Y Montana Strategy
IJC
6. Original: Recommend original project N Strategy
7. Modification: Recommend modification N for
8. No: Recommend no project N IJC

option of "Continue" and rejecting its other option "Modify." Possible

strategies for the other three DMs are also shown in Table 2.
A state is formed when each DM selects a strategy. Written horizontally
in text, the state (YN YN Y NNN) given in Table 2 is formed by Sage
Creek, British Columbia, Montana, and the IJC following strategies (YN),
(YN), (Y), and (NNN), respectively. This state is actually the status quo
situation at the time of modeling.
In the Flathead dispute, there is a total of 8 options. Because each option
can be either selected or rejected, there is in principle a total of 2 s = 256
states. However, m a n y of these states are infeasible in the actual conflict, for
a variety of reasons. For example, Sage Creek's two options, to continue the
original project and to modify the project, are mutually exclusive, so
selecting both options at the same time is infeasible. When infeasible
combinations are removed, there remain 55 states, numbered 1, 2 . . . . ,55,
which appear as columns in Table 3.
The reachable lists for this dispute are given in Table 4, while some of the
possible state transitions are indicated graphically in Figure 2. As shown by
the horizontal lines in Table 4 and the three groupings to the right in Figure
2, states 2 to 55 can be divided into three sets, each containing 18 states,
according to the recommendation of the IJC. State 1 in Tables 3 and 4 is the
status quo position. To produce the detailed graph in Figure 2, for states 2
Environmental Conflict Management 129

TABLE 3
STATES OF THE F L A T H E A D R I V E R R E S O U R C E D E V E L O P M E N T D I S P U T E

S~eCreek

1. Continue Y Y N N Y N N Y N N Y N N Y N N Y N N
2. M o d ~ N N Y N N Y N N Y N N Y N N Y N N Y N
British
Columbia

3. Original Y Y Y Y N N N N N N Y Y Y N N N N N N
4. M o d ~ c a t i o n N N N N Y Y Y N N N N N N Y Y Y N N N
Montana

5, Oppose Y N N N N N N N N N Y Y Y Y Y Y Y Y Y
IJC

6. Original N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
7. M o d ~ c a t i o n N N N N N N N N N N N N N N N N N N N
8. No N N N N N N N N N N N N N N N N N N N

State Numbers 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Sage Creek

1. Continue Y N N Y N N Y N N Y N N Y N N Y N N
2. Modify N Y N N Y N N Y N N Y N N Y N N Y N
British
Columbia

3. Original Y Y Y N N N N N N Y Y Y N N N N N N
4. Modification N N N Y Y Y N N N N N N Y Y Y N N N
Montana

5. Oppose N N N N N N N N N Y Y Y Y Y Y Y Y Y
IJC

6. Original N N N N N N N N N N N N N N N N N N
7. Modification Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
8. No N N N N N N N N N N N N N N N N N N

State Numbers 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

Sage Creek

1. Continue Y N N Y N N Y N N Y N N Y N N Y N N
2. Modify N Y N N Y N N Y N N Y N N Y N N Y N
British
Columbia

3. Original Y Y Y N N N N N N Y Y Y N N N N N N
4. Modification N N N Y Y Y N N N N N N Y Y Y N N N
Montana

5. Oppose N N N N N N N N N Y Y Y Y Y Y Y Y Y
IJC

6. Original N N N N N N N N N N N N N N N N N N
7. Modification N N N N N N N N N N N N N N N N N N
8. No Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

State Numbers 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
130 K . W . HIPEL ET AL.

TABLE 4
REACHABLE LISTS [ S i ( ' ) ] AND PAYOFFS [ P~(-)] FOR THE FLATHEAD RIVER
RESOURCE DEVELOPMENT DISPUTE

Sage Creek British Columbia Montana IJC

k Sa(k) Pl(k) S2(k) P2(k) S3(k) P3(k) S4(k) P4(k)
1 50 25 8 11,29,47 1

2 3, 4 55 5, 8 55 11 1 2
3 2, 4 48 6, 9 41 12 20 2
4 2, 3 36 7, 10 45 13 39 2
5 6,7 18 2,8 12 14 3 2
6 5, 7 42 3, 9 4 15 22 2
7 5, 6 30 4, 10 6 16 41 2
8 9, 10 16 2, 5 24 17 5 2
9 8, 10 17 3, 6 14 18 24 2
10 8, 9 24 4, 7 18 19 42 2
11 12, 13 54 14, 17 52 2 2 2
12 11, 13 47 15, 18 39 3 21 2
13 11, 12 35 16, 19 43 4 38 2
14 15, 16 15 11, 17 10 5 4 2
15 14, 16 41 12, 18 2 6 23 2
16 14, 15 29 13, 19 5 7 40 2
17 18, 19 13 11, 14 22 8 6 2
18 17, 19 14 12, 15 13 9 25 2
19 17, 18 23 13, 16 16 10 43 2

20 21, 22 53 23, 26 3 29 7 2
21 20, 22 46 24, 27 11 30 26 2
22 20, 21 34 25, 28 8 31 45 2
23 24, 25 12 20, 26 40 32 10 2
24 23, 25 40 21, 27 54 33 28 2
25 23, 24 28 22, 28 49 34 47 2
26 27, 28 10 20, 23 27 35 12 2
27 26, 28 11 21, 24 37 36 30 2
28 26, 27 22 22, 25 33 37 49 2
29 30, 31 52 32, 35 1 20 9 2
30 29, 31 45 33, 36 9 21 27 2
31 29, 30 33 34, 37 7 22 44 2
32 33, 34 9 29, 35 38 23 11 2
33 32, 34 39 30, 36 51 24 29 2
34 32, 33 27 31, 37 47 25 46 2
35 36, 37 7 29, 32 26 26 13 2
36 35, 37 8 30, 33 35 27 31 2
37 35, 36 21 31, 34 31 28 48 2
Environmental Conflict Management 131

TABLE 4
(CONTINUED)
38 39,40 51 41,44 17 47 14 2
39 38,40 44 42,45 20 48 32 2
40 38,39 32 43,46 23 49 51 2
41 42,43 6 38,44 29 50 16 2
42 41,43 38 39,45 32 51 34 2
43 41,42 26 40,46 36 52 53 2
44 45,46 4 38,41 44 53 18 2
45 44,46 5 39,42 48 54 36 2
46 44,45 20 40,43 53 55 55 2
47 48,49 49 50,53 15 38 15 2
48 47,49 43 51,54 19 39 33 2
49 47,48 31 52,55 21 40 50 2
50 51,52 3 47,53 28 41 17 2
51 50,52 37 48,54 30 42 35 2
52 50,51 25 49,55 34 43 52 2
53 54,55 1 47,50 42 44 19 2
54 53,55 2 48,51 46 45 37 2
55 53,54 19 49,52 50 46 54 2

State 1 = Status Quo

States 2-19 = IJC recommends original project.
States 20-37 = IJC recommends modified project.
States 38-55 = IJC recommends no project.

to 19 as well as states 20 to 37, simply subtract the number 36 and 18,

respectively, for each state number in the lower detailed graph.
Recall that a DM's payoff function indicates that DM's preferences. In
this model, payoff functions were determined by assessing the relative
preferences of the DMs. For example, Sage Creek most prefers the maximum
possible project consistent with British Columbia's license, with IJC's
support, and no opposition from Montana. British Columbia's preference
can be characterized by its priorities for matching with IJC's recommenda-
tion, for Sage Creek not building a project larger than IJC's recommenda-
tion, and for no opposition from Montana. Montana's preference is deter-
mined by the size of Sage Creek's project, IJC's decision, and British
Columbia's license. Because the IJC is required by treaty to make a
recommendation on the project using only technical (rather than strategic)
criteria, its only preference is to make a recommendation. Hence, all
recommendations are equally preferred for IJC. The four payoff functions for
the four DMs are shown in column P~(k) in Table 4, where a higher number
means a more preferred state for the DM. Following a description of GMCR
132 K. W. HIPEL ET AL.

()
DECISION MAKERS

SC Sage Creek
BC British Columbia
MT _ _ Montana
IJC International
Joint Commission

UC
* ' C STATES 2 - 19

IIC
C ' ~ S 20 - 37 ~.~

UC

.................. MT ..... • .......

F
M.X,, ~............ M:I: . . . . . . . . . .
A, ........ . ............. ~'::,,:::... ~i~"............. " ..........

"" - - - :_ B C - - -" "" . . . . '.. B C - - -"

•...... .~ •... ,....~:: . . . . ... , •...,.

...... MT . . . . . . M . c ......
•. " ~ . ....... ~:::'" : ..... i............

. . . . . . . . BC . . . . - . . . . . BC- --"

F r o . 2. The graph model for the Flathead River resource development dispute.

II in the next main section, the Flathead environmental dispute is analyzed

using GMCR II.

T H E DECISION S U P P O R T SYSTEM GMCR II

Other Systems for Analysis of Strategic Conflicts

To conveniently apply a formal decision model to practical problems, a
flexible decision support system (DSS) is required [20]. A wide variety of
decision-making models and associated DSSs have been developed in fields
Environmental Conflict Management 133

which include operations research, management sciences, systems engineer-

ing, and statistics. The purpose of this section is to present the design of a
new DSS, GMCR II, which implements the graph model for conflict resolu-
tion approach outlined earlier to study systematically real-world conflicts
arising in resource management, international trade, and elsewhere.
A number of DSSs has been developed for modeling decision situations
involving more than one DM. A DSS designed for use in negotiations is
referred to as a negotiation support system. Overviews and comparison of
existing negotiation support systems are found in [21-23]. Papers describing
the theory and application of existing negotiation support systems include
[24-31]. Reference [32] provides an overview of DSSs more specific to the
context of conflict resolution, and [9] outlines the main features of DSSs that
can be utilized for aiding decision making under conditions of strategic
conflict. These DSSs include CONAN (cooperation-or-conflict analysis) [33,
34], I N T E R A C T [35, 36], DeeisionMaker [37], SPANNS [38], DSA (Decision
Systems Analysis) [39], and GMCR I [6]. Of particular importance here is
the earlier implementation of GMCR called GMCR I.
The DSS GMCR I was developed by Fang, Hipel, and Kilgour and
included with the textbook Interactive Decision Making: Graph Model for
Conflict Resolution [6]. The analysis component of the Graph Model for
Conflict Resolution was implemented in a powerful engine written in the C
language. As the only DSS based on the graph model, GMCR I reflects
major recent achievements in the field of conflict management and can take
advantage of the strength of graph model (outlined earlier). One of the
attractive features of GMCR I is its ability to provide stability results for a
wide range of solution concepts (refer to Table 1). However, CMCR I lacks a
user-friendly interface, so the user has to prepare an input file off-line
relative to the decision environment. GMCR I emphasizes only analysis, and
most modeling and interpretation procedures must be carried out off-line.
Furthermore, interpretation may be prone to mistakes, of which the user
might be unaware. Because of memory constraints and other factors, GMCR
I can deal only with conflicts of limited size: no more than 200 states for a
2-player conflict and 100 for an n-player conflict. GMCR I is essentially a
set of analysis programs and has no user-system interaction. GMCR II is
specifically designed to overcome these and other limitations of GMCR I. In
addition to being a flexible and user-friendly DSS, some of the specific
advantageous features incorporated into GMCR II include an informative
graphical interface based on a Windows environment, model input flexibility
using either option form, a graphical representation, or an input file, the
capacity to use the powerful GMCR engine to analyze large conflicts using
all of the solution concepts given in Table 1, and the availability of
automatically generated interpretations and advice based on the output
from a model analysis.
134 K.W. HIPEL ET AL.

Overview of GMCRII--The Next Generation

The objective of GMCR II is to provide the next generation DSS for the
Graph Model for Conflict Resolution in a way that significantly expands the
capability of the previous generation, GMCR I. Figure 3 depicts the overall
design of GMCR II. The model subsystem in Figure 3 uses the GMCR I
engine for calculating stability results for all of the solution concepts listed
in Table 1. However, the memory management for the engine is being
improved to allow larger conflicts to be analyzed expeditiously. The user
interface for both input and output and data subsystem are completely new
designs that are currently being incorporated into GMCR II.
Figure 4 shows how stability results in the data subsystem of Figure 3 are
organized and are thereby easily retrieved for purposes such as displaying
appropriate findings, updating the model, and executing various sensitivity
analyses. Figure 4, which originally appeared in [6, p. 195], can be presented
in a variety of ways including the following:

1. For each DM, the decision maker's plane (parallel to the S T A T E /

STABILITY TYPE plane) indicates the stability types or solution concepts
(if any) under which each state is stable for that specific DM.

MODEL SUBSYSTEM
Algorithms for
Modeling,
Individual Stabilities,
Solution Concep~, ~ f ] and Equilibria

DATA SUBSYSTEM
• Transform Input Decision Makers
• Retrieve Data Options
• Interpret Output Feasible States
State Transitions
Preferences
Stability Results

GMCR II

FIG.3. Designof GMCRIf.

Environmental Conflict Management 135

DECISION MAKERS

decision moker's
~/~~plan

--stability-type
,,°,. 0,o°.

555 J /
P°7-//
STATES
FIG.4. Organizationof stability results in the data subsystem.

2. For each stability type, the stability-type plane (parallel to the DECI-
SION M A K E R / S T A T E plane) provides a complete analysis of the model
according to that stability type.
3. For each state, the state plane (parallel to the DECISION M A K E R /
STABILITY T Y P E plane) identifies the DMs for whom the particular state
is stable (s), or unstable (u) under each possible stability type.
4. The STABILITY T Y P E / S T A T E plane itself, referred to as the
equilibrium plane, contains the projection of stability results for each DM ( E
if s for all DMs, blank otherwise), indicating all equilibria for each stability
type.
GMCR II is being developed using Borland C ++ within the Microsoft
Windows 3.1 environment. In this way, the system will be widely accessible,
be user-friendly, and possess powerful graphical capabilities. Table 5 fur-
nishes the menu for the User Interface for GMCR II. As can be seen, GMCR
II can employ the option form for defining states, as illustrated below. The
option form is especially useful when the analyst can express the specific
courses of actions as options or general choices available to each DM that
may be selected more or less independently. Because a large number of
states can be automatically calculated by GMCR II using a relatively small
number of options, the option form can be used with very large conflicts.
In some situations, one may prefer to discuss a conflict in terms of overall
states and not worry about the detailed definition of each state. For
example, senior decision makers may be interested in possible final results,
136 K . W . HIPEL ET AL.

TABLE 5
MENUTREEFOR GMCR II
1st Level 2nd Level 3rd Level
File New
Open
Save
Save As ...
Exit
Edit Undo
Cut
Copy
Paste
Model Input Option Form Input DMs/Options
State Elimination
State Combination
Common Moves
Irreversible Moves
Forcing Moves
Graph Form
Preview Conflict Description
Feasible State List
Reachable Lists
Preferences Option Based Preference Tree
Option Weighting
Further Ranking
Graph Based
Stability Output Option Form
Graph Form
Interpretation Status Quo Analysis
Graphical Evolution Path
Further Analysis Sensitivity Analysis
Hypergame Analysis
Coalition Analysis
Help Contents
About...

and in general, how the more preferred results can be achieved. The
graphical input of states mentioned in Table 5 may be very helpful in these
cases, especially for smaller conflicts at early stages of development. A circle
or node can be drawn on a screen to represent a state while possible moves
between states are depicted using arrows. This graphical information can
then be converted by GMCR II into a form which can be processed by the
analysis engine. For output, the user will be able to select different aspects
Environmental Conflict Management 137

of the analysis to be illustrated graphically. It is also desirable to show

possible paths for a status quo state to an equilibrium. For both the input
and output interface, aesthetically pleasing drawings of graphs [40] are
especially useful.
Table 5 also furnishes the structure of the main menu for the user
interface o f GMCR II. At each end branch of the menu tree, one or a
sequence of dialog boxes is invoked to fulfill a particular task.

Option-based Modeling
GMCR II can employ the option form to input states and state transi-
tions for a model. This information is transformed to become the input for
the state-based analysis engine.
This section presents the design of option-based modeling and shows how
the input information that the analysis engine needs, such as the feasible
states, reachable lists, and preferences, can be generated from the modeling
information. State removal and preferences trees were implemented in
DecisionMaker [37]. These algorithms will be redesigned using the data
structure of GMCR II. Features like state combination, irreversible moves,
common moves, and forcing moves have never previously been implemented
in option form. As much as possible, option form input will be introduced in
the context of the Flathead River dispute, described above.
DMs and Options. Figure 5 shows the prototyping dialog box for DM
and option input, where "British Columbia" in the Decision Makers listbox
is highlighted and, accordingly, the lower input area is ready for British
Columbia's options. The system will read in the information appearing in
this dialog box and store it in a proper format in the data subsystem. If, for
example, the user deletes the highlighted item "British Columbia", the
options belonging to "British Columbia" will also disappear from the option
list as well as from the inner storage, and the label for the rest of the
DMs/Options will be updated accordingly.
Data Structure. The option form can cause problems in memory and
execution time. For example, the DSS I N T E R A C T mentioned above has
difficulty obtaining a list of all feasible scenarios, even in models with only
10 options [35]. Efficient algorithms and data structures based on the
preference tree concept have been devised [41], creating significant savings in
memory and execution time during a stability analysis using the sequential
stability solution concept. In GMCR II, a 32-bit DOUBLEWORD is used to
represent a specific option selection defining a state. This approach is
efficient and can be used in analyses employing any of the solution concepts
given in Table 1. Based on this data structure and the modeling algorithms
introduced below, all feasible states in a model can be generated very
quickly.
138 K.W. HIPEL ET AL.

GMCR - FLATHEAD.GMC

FIG. 5. The DM/Option input dialog box.

For each digit or bit, a " 0 " or "1" indicates whether the option is
selected by the DM controlling it, or not selected, respectively. Because
there are 32 bits, this format can handle up to 32 options, which is more
than sufficient for use in most real-world applications. If there are n options
in the model, GMCR II automatically works with the resulting 2 n mathe-
matically possible states.
For example, state 1 of the 8-option Flathead River conflict shown in
Tables 2 and 3 is represented in binary form as

32 b~ts

. . . . . . 000 1 01 01 (binary).
DM4 DM3 DM2 DMI

Recall that there are four DMs in this model. The brackets indicate which
options each DM controls.
This data representation is straightforward and analogous to option form
itself. As well, this data structure utilizes the C language's bit-wise operator
features.
Environmental Conflict Management 139

GMCR-FLATHEAD.GMC 1

FIG. 6. The State Elimination dialog box.

State Removal. In practice, there is usually a large number of mathe-

matically possible option combinations that cannot occur and therefore
must be eliminated from the conflict model in option form. Figure 6 provides
a prototyping design of the dialog box for state removal.
Because of the inner data structure, ambiguity in state elimination can be
eliminated only if the syntax requirement for input containing option
numbers follows formats such as

- 3 , 2 & - 1, or ( 3 & - 5)1(2& 8)1 - 4,

etc. Input such as

( - 312) & (714)

is not allowed. In other words, "l" must have lowest priority. Here "&"
means "and", and "l" means "or".
140 K. W. HIPEL ET AL.

The input information can be entered in two different ways. A "short

cut" for frequent users is to type in directly from the upper left editbox in
Figure 6. The longer way is to use the mouse to push the logical operation
buttons and double-click items in the lower right listbox.
There is an "error preventing" feature when this dialog box is used with
mouse input. In Figure 6, the right area of the list item for option 6 has just
been highlighted, so that the " N O T " button, the " ( " button and the lower
right listbox are disabled (grey), because according to the syntax require-
ment, they axe illegal operations at this moment. Table 6 shows this kind of
error-preventing design. When a row item has just been input, there are
always some column items that axe disabled. In this way, input errors can be
avoided.
Consider the Flathead River conflict presented above. The information
shown in the lower left edit box is interpreted as in Table 7.
Obviously, the representation of the pattern to be eliminated in this
prototyping dialog box is not suitable for a naive user. Also, the system
should have the capability to do more input interpretation and transforma-
tion, in addition to simply applying syntax requirements. The objective here
is to show one way of specifying infeasible states. The error-preventing idea
is followed throughout the design.
Several types of option constraints identify infeasible states, including:
• M u t u a l l y exclusive, e.g., " a and b cannot be chosen at the same time";
• c o m p l e m e n t a r y , e.g., "one and only one from a and b and c can be
chosen";
• conditional, e.g., " a can be selected only if b is chosen".
These representations are more suitable for naive users. The dialog box will
be designed by including corresponding verbal statements and "form-filling"
features as in the above prototype.
A user m a y specify information such as "option 5 can be selected only if
option 1 is chosen". A logic interpretation program then transforms this

TABLE 6
ERROR PREVENTING FEATURES WITH STATE ELIMINATION DIALOG BOX

AND OR NOT ( ) list

AND disabled disabled disabled disabled
OR disabled disabled disabled
NOT disabled disabled disabled disabled disabled
( disabled disabled disabled disabled
) disabled disabled disabled disabled disabled
list disabled disabled disabled
Environmental Conflict Management 141

TABLE 7
INFEASIBLESTATESPECIFICATIONSFOR THE FLATHEADRIVERCONFLICT
Input
Information Interpretation
1&2 Sage Creek cannot do both "Continue" and "Modify"
3&4 British Columbia cannot do both "Original" and "Modification"
6&7 IJC cannot do both "Original" and "Modification"
7&8 IJC cannot do both "Modification" and "No"
8&:6 IJC cannot do both "No" and "Original"

information and puts an "(option 1) and (not option 5)" into the "Infeasible
Situation" list, which in turn requires all states with the pattern
"" . . . . . . . . 1 - - - i f ' be eliminated. Here, the dash ..... indicates that the
entry can be either a 0 or 1.
This idea of pattern is actually used to represent a group of states
conveniently and compactly. But a pattern cannot be conveniently recorded
in a simple data structure. However, thanks to the C language's low-level
bit-wise operation, a pair of masks, which are extremely simple to record
and process in the program, can be used to equivalently represent a pattern.
The following two " m a s k s " are used to represent the pattern
......... 1 - - - 0 : m0 . . . . . . . 11111110; m 1 . . . . . . . 00010000. Here
"equivalently" means that if a state k satisfies

(1) k]m o = mo, and

(2) k& ml = ml,

then k must belong to the above pattern. (Here "l" and "&" mean bit-wise
OR and bit-wise AND, respectively.) This kind of pattern and the equiva-
lent mask pair will be used throughout the design.
From the state elimination dialog box, a list of patterns can be recorded
and will be used as "filters" for state removal. Infeasible state removal has
been called "one of the most tedious tasks when performing a conflict
analysis" [4, p. 159], and several programs have been developed to deal with
it based on different data structures. In this design, state removal is simple,
efficient, and straightforward.
State Combination and Common Moves. Sometimes in a conflict model,
a group of states with a common pattern is indistinguishable and, thus,
should be treated as a single state. To specify which states are to be
coalesced to form a state combination, users are required to provide the
common characteristics of the states.
142 K . W . HIPEL ET AL.

Common moves happen when either of two (or more) DMs can unilater-
ally force the model to change from one state to another. (Common moves
do exist in the real world. An example in a simplified model of a superpower
nuclear confrontation is given in [6, Chapter 2].) Common moves can be
specified in almost the same way as state combinations. Compared with the
specification of indistinguishable states, the only difference is that common
move specification allows the use of the logic operator OR on cross-DM
options, which is not available for state combinations.
Generation of the Feasible State List. After a user provides the input
information consisting of the DMs, their options, the infeasible states, the
indistinguishable states, and the common moves, GMCR II automatically
generates the feasible state list, which is then available to the user under the
Preview menu. Figure 7 shows how the feasible state list is obtained, where
the filtering is accomplished as described in the section of state removal.
Irreversible Moves. In an irreversible move, a DM can unilaterally cause
a conflict to move from state k to q by changing his or her option selection,
but cannot make the reverse transition from q to k. Accordingly, within the
Irreversible Move dialog box, a user can highlight the option or options
which cause the irreversible move. Figure 8a shows a prototyping design of
this dialog box to specify irreversible moves based on single option status
shift constraints. The information shown in the dialog box is sufficient for
many models. But, in some situations, status shift constraints based on

Interface
generate All mathemaucaUy
I DM/OptionInput
Dialog Box possiblestates I
State Elimination interpret J Infeasible State]
Filter all the states

User
DialogBox "] PaRems ] matching the pattern
.........................

I State Combination
Dialo~Box
Filterall the states matching
I ~. - ~ Indistinguishable] the pattern from the rest
State Patterns I ..........................
Common Moves
Dialo~Box Addeach pattern
into the list as
representative of its group

List of
feasible states

FIG. 7. Generatingthe feasible state list.

Environmental Conflict Management 143

(a)

¢o)
Fic. 8. Irreversible Move dialog box: (a) single option; (b) multiple options.
144 K . W . HIPEL ET AL.

multiple-option combinations might be required to specify irreversible moves.

Therefore, the dialog box is to be designed as in Figure 8b.
Forcing Moves. Sometimes, when a DM makes a certain move, one or
more other DMs have no choice but to respond with the "forced" move.
Thus, the DM achieves a new state by moving as if "unilaterally". An
example of a forcing move appears in [9]. A user can specify a forcing move
by providing a "forcing p a t t e r n " and a "forced pattern".
Calculation of Reachability. The information elicited from the specifica-
tion of DMs, options, irreversible moves, common moves, and forcing moves
permits teachability from any state to any other to be calculated for each
DM. Figure 9 shows the flowchart for determining reachability. The reach-
able lists thus generated are essential inputs for the analysis engine.
Option-based Preferences Specification. The preference tree is an ordi-
nal preference representation approach suggested and developed in [41, 42].
The preference tree ranks the states for a given DM based upon lexico-
graphic preference statements about the options. Because the preference
information for a particular DM appears as a tree when portrayed graphi-
cally, the preference information is called a preference tree. As defined in
[43], a preference statement can be a simple one:
• an option o,, or
• oi A N D P ' , o r
• o, O R P',or
• NOT P'
where P ' is a simple preference statement. (The definition is thus recursive.)
It can also be a conditional preference statement:
• P ' IF P"

where P ' and P" are simple preference statements. Obviously, the input
information for a set of preference statements can be elicited in a similar
way, as discussed in the section on state removal. Each preference statement
expresses a logical relationship among options t h a t can be used to bifurcate
a set of states. Therefore, a set of lexicographic preference statements can
produce a ranking of the states. An algorithm modified from the QuickSort
algorithm [44] is designed to carry out preference tree ranking based on
G M C R II's d a t a structure. To obtain the ranking based on n preference
statements, states t h a t match the first statement are assigned a weight of
2 ~, the second 2 ~- 1 and so on; states are then ranked according to their
total weights, producing ordinal payoffs which are consistent with corre-
sponding preference tree rankings.
Another general and flexible method is the option weighting method, in
which weights (positive, negative, or zero) are assigned to each option, and
Environmental Conflict Management 145

l ieN;k,qeUI

Doesk
sCOntrol,are the N match a forcing
tatus of k and q pattern

the irreversible N
moves' startpattern

Does q match the

corresponding
forced pattern

Does q match the Y

corresponding Y
end pattern

I q~si~k~ I I q~si~ I
I 1

F~C.9. Flowchart for determining reachability.

the preference level for each state is d e t e r m i n e d b y its t o t a l weight relative

to other states.
T h e F u r t h e r R a n k i n g dialog box can be used to refine the r a n k i n g list for
a given DM b y allowing the user to move a state from one location to
another, or to highlight a range of contiguous states to create a n indifference
block. Recall t h a t some states could result from the coalescing of indistin-
146 K. W. HIPEL ET AL.

guishable states, where certain option choices are considered to be irrele-

vant. Because these states could easily be misplaced in the preference
ranking, the opportunity for further ranking is usually helpful and some-
times necessary.

Stability Output and Further Analyses

As can be seen in Table 5, a variety of output can be generated by the
engine for G M C R II after the user specifies which solution concepts to
employ for each DM. Examples of sensitivity anMyses are presented with
the analysis of the Flathead River Resources Development conflict in the
next section.

ANALYSIS OF T H E F L A T H E A D R I V E R R E S O U R C E
DEVELOPMENT DISPUTE

The background and modeling for the Flathead River environmental

conflict are presented in the section on representing conflicts in option form.
Moreover, some screens used in inputing this model to G M C R II are shown
in the previous section. In this section, G M C R II is used to carry out a
complete stability analysis for the Flathead River model given in Tables 2
to 4.
For each state and each DM in the Flathead conflict, G M C R II calculates
stability for each of the solution concepts listed in Table 1. When a state is
stable according to a specified solution concept for each DM in a conflict, the
state constitutes an equilibrium for that solution concept. The equilibria for
the model are given in Table 8, where the acronyms for the solution

TABLE 8
EQUILIBRIA FOR THE FLATHEAD RIVER RESOURCE DEVELOPMENT DISPUTE

k Equilibrium
3 GMR, SMR
4 GMR, SMR
11 R, GMR, SMR, SEQ, L1, L2, L~, NM
12 GMR, SMR
13 GMR, SMR
24 GMR, SMR
25 GMR, SMR
33 R, GMR, SMR, SEQ, L1, L2, L3, NM
34 GMR, SMR
46 R, GMR, SMR, SEQ, L1, L2, L3, NM
55 GMR, SMR
Environmental Conflict Management 147

concepts are taken from Table 1. This table indicates that states 11, 33, and
46 are equilibria for all solution concepts and possess a much greater degree
of stability than any other states. In Tables 8 and 1, the symbol L~ means
that the corresponding state is an equilibrium according to the limited move
stability for a horizon of h steps.
The status quo at the time of analysis is state 1, shown in Tables 2 and 3,
in which Sage Creek is proceeding with the original project, the government
of British Columbia has approved-in-principle Stage II of the project by
supporting the original project, Montana is opposing the project, and the
IJC has not yet made a recommendation. Equilibria 11, 33, and 46 corre-
spond to the IJC's three possible recommendations. If it recommends
stopping the project, the likely resolution is state 46. The actual outcome
was that IJC recommended stopping the project and the immediate result
was state 47. The sequence of state transitions from the status quo state 1
through the transition states 47, 53, and 55 to the equilibrium 46 is shown in
Table 9, where an arrow indicates which strategy selection changes at each
transition.
A sensitivity analysis was conducted by using GMCR II to analyze
alternative models to the one defined in Table 4. These models differ only in
certain aspects of the DMs' preferences. The objective was to assess the
effects of these preference variations on the results of the analysis.
First, British Columbia's preferences were changed to reflect the hypoth-
esis that even if the IJC recommends stopping the project, British Columbia

TABLE 9
STATE TRANSITIONS FROM STATUS QUO TO FINAL OUTCOME

Sage Creek
1. Continue Y Y Y ) N N
2. Modify N N N N N
British Columbia
3. Original Y Y N N N
4. Modification N N N N N
Montana
5. Oppose Y Y Y Y -~ N
IJC
6. Original N N N N N
7. Modification N N N N N
8. No N --* Y Y Y Y

State Numbers 1 47 53 55 46
148 K. W. HIPEL ET AL.

would prefer to grant a license for a modified project, as long as there is no

opposition from Montana, i.e., (P2(41) = 44, P2(42) = 48, P2(43) = 53,
P2(44) = 29, P2(45) --- 32, P2(46) = 36) instead of (P2(41) = 29, P2(42) =
32, P2(43) = 36, P2(44) = 44, P2(45) = 48, P2(46) = 53). The equilibria for
this case are summarized in Table 10.
By comparing Tables 8 and 10, one can see that there is no change in
equilibria for the first two sets of states (states 2 to 37), but the changes in
last group (states 38 to 55) are substantial. First, there is no Nash equilib-
rium from states 38 to 55. Second, the two forms of limited-move and
nonmyopic equilibria defined in [6, 7] make different predictions. In Case I of
limited move stability, the original DM takes part in the sanctioning: here
Montana can move from state 55 to 46 after 55 is reached, explaining why
state 55 is a nonmyopic (high foresight) equilibrium only in Case II. As well,
there is another nonmyopic equilibria 46. But Sage Creek can stop the
project in both of these high foresight equilibria. This means that, even if
British Columbia wants some kind of project to proceed, no project will be
built, provided the DMs behave nonmyopically.
Another sensitivity analysis involves changing British Columbia's prefer-
ence to reflect that even if the IJC were to recommend stopping the project

TABLE 10
EQUILIBRIA FOR THE SENSITIVITY ANALYSIS

k Equilibrium
3 GMR, SMR
4 GMR, SMR
11 R, GMR, SMR, SEQ, L1, L2, L3, NM
12 GMR, SMR
13 GRM, SMR
24 GMR, SMR
25 GMR, SMR
33 R, GMR, SMR, SEQ, Lx, L2, L3, NM
34 GMR, SMR
42 GMR, SMR
43 GMR, SMR, SEQ
46 GMR, SEQ, L3, L4, Lb, L6, NM
51 GMR
52 GMR
55 GMR, SMR, SEQ, L2, L3, LI4t, L~l, LIIb,NMH

Two cases for limited move stability:

Case I has a fLxed point at length 6;
Case II has a fLxed point at length 4. See [6, 7].
Environmental Conflict Management 149

and opposition from Montana continued, British Columbia would prefer to

grant a license for a modified project, i.e., ( P 2 ( 4 1 ) = 44, P2(42)= 48,
P2(43) = 53, P2(44) = 29, P2(45) = 32, P2(46) -- 36, P2(50) -- 42,
P2(51) = 46, P2(52) -- 50, P2(53) = 28, P~(54) = 30, P2(55) - 34) instead
of (P2(41) = 29, P2(42) = 32, P2(43) = 36, P2(44) = 44, P2(45) = 48,
P2(46) = 53, P2(50) -- 28, P2(51) = 30, P2(52) = 34, P2(53) = 42,
P2(54) = 46, P2(55) --- 50). The resulting equilibrium for the last group of
states is state 51, for which Sage Creek will build a modified project. If this
case were to occur, the federal governments of Canada and the U.S. would
likely become directly involved in this dispute.
As a consequence of the original analysis and the two sensitivity analyses,
it can be concluded that the pressure from Montana (and elsewhere) on
British Columbia was as crucial to the final outcome as the IJC's recommen-
dation. If British Columbia had been willing to face national and interna-
tional opposition in order to achieve at least a modified project, then the
analysis implies that such a project would likely have been built. The
analysis indicates that British Columbia's sensitivity to Montana's continu-
ing pressure (or the threat of pressure) was decisive.

CONCLUSIONS

GMCR II constitutes the next generation of a DSS for the efficient and
effective management of strategic conflicts. Such conflicts can take place in
land, water, and environmental management, as well as many other areas of
human endeavor. As demonstrated by the real-world resource management
conflict in the previous sections, GMCR II can provide decision makers and
analysts with decision advice, structural insights and answers to what-if
questions. With this enhanced understanding, analysts can better explain
strategic relationships and assist DMs, who may thus gain the opportunity
to direct the evolution of the conflict toward more favorable results. In order
to approach solutions to resource management problems that axe both
economically and environmentally sustainable, it is essential that the view-
points of competing interest groups be taken into account through a decision
technology such as GMCR II.

REFERENCES

J. K. Levy, K. W. Hipel, and D. M. Kilgour, Sustainable development of life

support systems. In: M. Jamshidi, (Ed.), Encyclopedia of Life Support Systems:
Water, Energy, Environment and Global Food Security. 1996. Sponsored by the
International Foundation for Water Science and Technology.
150 K . W . HIPEL ET AL.

2 N. Howard, Paradoxes of Rationality, MIT Press, Cambridge, Mass., 1971.

3 N. M. Fraser and K. W. Hipel, Solving complex conflicts, IEEE Transactions
on Systems, Man and Cybernetics SMC-9:805-817 (1979).
4 N. M. Fraser and K. W. Hipel, Conflict Analysis: Models and Resolutions,
North-Holland, New York, 1984.
5 L. Fang, K. W. Hipel, and D. M. Kilgour, Conflict models in graph form:
Solution concepts and their interrelationships, European Journal of Operational
Research 41(1):86-100 (1989).
6 L. Fang, K. W. Hipel, and D. M. Kilgour, Interactive Decision Making: The
Graph Model for Conflict Resolution, Wiley, New York, 1993.
7 D. M. Kilgour, K. W. Hipel, and L. Fang, The graph model for conflicts,
Automatica 23(1):41-55 (1987).
8 K.W. Hipel, L. Fang, and D. M. Kilgour, Game theoretic models in engineering
decision making, Journal of Infrastructure Planning and Management, Japan
Society of Civil Engineers 470(IV-20):1-16 (1993).
9 K.W. Hipel, D. M. Kilgour, L. Fang, and X. Peng, Using the decision support
system GMCR for resolving conflict in resource management. In: Proceedings of
the 1st International Conference on Multiple Objective Decision Support Sys-
tems for Land, Water, and Environmental Management, Honolulu, Hawaii,
July 1995. To be published in 1996.
10 D. M. Kilgour, L. Fang, and K. W. Hipel, Analyzing the Flathead River
resource development dispute using the graph model for conflicts. In M. G.
Singh and L. Trav&Massuy~s (Eds.), Decision Support Systems and Qualitative
Reasoning, Proceedings of the IMACS International Workshop on Decision
Support Systems and Qualitative Reasoning, Toulouse, France, pp. 101-110,
(1991).
11 J.F. Nash, Equilibrium points in n-person games, Proceedings of the National
Academy of Sciences of the U.S. A., 36:48-49 (1950).
12 J. F. Nash, Noncooperative games, Annals of Mathematics, 54(2):286-295
(1951).
13 D.M. Kilgour, Anticipation and stability in two-person noncooperative games,
In: M. D. Ward and U. Luterbacher (Eds.) Dynamic Models of International
Conflict, pp. 26-51. Lynne Rienner Press, Boulder, Colorado, 1985.
14 F. C. Zagare, Limited-move equilibria in 2 × 2 games, Theory and Decision,
22:1-19 (1984).
15 S.J. Brams and D. Wittman, Nonmyopic equilibria in 2 × 2 games, Conflict
Management and Peace Science 6(1):39-42 (1981).
16 D. M. Kilgour, Equilibria for far-sighted players, Theory and Decision
16:135-157 (1984).
17 J.E. Carroll, Environmental Diplomacy: An Examination and a Prospective of
Canadian-U.S. Transboundary Environmental Relations, The University of
Michigan Press, Ann Arbor, Michigan, 1983.
18 International Joint Commission. Flathead River International Study Board
Report. Ottawa, Ontario-Washington, D.C., 1988.
Environmental Conflict Management 151

19 International Joint Commission. Impacts of a Proposed Coal Mine in the

Flathead River Basin. Ottawa, Ontario-Washington, D.C., 1988.
20 A.P. Sage, Decision Support Systems Engineering, Wiley, New York, 1991.
21 D.M. Kilgour, L. Fang, and K. W. Hipel, GMCR in negotiations. Negotiation
Journal, 11(2):151-156 (1995).
22 E.M. Thiessen and D. P. Loucks, Computer-assisted negotiation and multiob-
jectivc water resources conflicts, Water Resources Bulletin, 28(1):163-177 (1992).
23 M . T . Jelassi and A. Foroughi, Negotiation support systems: An overview of
design issues and existing software, Decision Support Systems 5:167-181 (1989).
24 J. Angus, Negotiate! This software teaches you the fine art of negotiations,
Portable Computing, 4(3):56 (1990).
25 R. Anson and M. T. Jelassi, A development framework for computer-support
conflict resolution, European Journal of Operational Research, 46:181-199
(1990).
26 S. Gauvin, G. L. Lilien, and K. Chatterjee, The impact of information and
computer based training on negotiator's performance, Theory and Decision,
28:331-354 (1990).
27 M. Jarke, M. T. Jelasi, and M. F. Shakun, MEDIATOR: Towards a negotiation
support system, European Journal of Operational Research 31(3):314-334
(1987).
28 S.S. Nagel and M. K. Mills, Multicriteria dispute resolution through computer
aided mediation software, Mediation Quarterly, 7(2):175-189 (1989).
29 J. F. Nunamaer, Jr., Experience with and future challenges in GDSS (group
decision support systems), Decision Support Systems, 5:115-118 (1989).
30 M.G. Singh, J. B. Singh, and M. Corstjens, MARK-OPT-A negotiating tool for
manufacturers and retailers, IEEE Transactions on Systems, Man and Cyber-
netics, 15(4):483-495 (1985).
31 F.W. Winter, An application of computer decision tree models in management-
union bargaining. Interfaces, 15(2):74-80 (1985).
32 K.J. Radford, K. W. Hipel, and L. Fang, Decision making under conditions of
conflict, Group Decision and Negotiation, 3:169-185 (1994).
33 N. Howard, The CONAN Manual, 1986. Copyright by Nigel Howard Systems,
10 Bloomfield Road, Birmingham, England B13 9BY.
34 N. Howard, CONAN 3.0. Nigel Howard Systems, 10 Bloomfield Road, Birming-
ham, England, B13 9BY, 1989.
35 P . G . Bennett, A. Tait, and K. Macdonagh. INTERACT: Developing software
for interactive decisions. Working Paper 35, Department of Management Sci-
ence, University of Strathclyde, United Kingdom, 1992.
36 P . G . Bennett, A. Tait, and K. Macdonagh, INTERACT: Developing software
for interactive decisions, Group Decision and Negotiation 3(4):351-372 (1994).
37 N.M. Fraser and K. W. HipeL DecisionMakcr: The conflict analysis program.
University of Waterloo, Waterloo, Ontario, Canada, 1993.
38 D.B. Meister and N. M. Fraser, Conflict analysis technologies for negotiation
support, Group Decision and Negotiation, 3(3):333-345 (1994).
152 K . W . HIPEL ET AL.

39 J-P. Langlois, Decision Systems Analysis. San Fransisco State University,

September 1994. Version 3.3.
40 G. Di Battista, P. Eades, R. Tamassia, and I. C. Tollis. Algorithms for drawing
graphs: An annotated bibliography. Available via anonymous ftp from
wilma.cs.brown.edu, file/pub/gdbiblio.tex.Z, 1993.
41 N.M. Fraser and K. W. Hipel, Decision support systems for conflict analysis. In:
M. G. Singh, K. Hindi, and D. Salassa, (Eds.) Managerial Decision Support
Systems, Proceedings of the IMA CS/ IFORS 1st International Colloquium on
Managerial Decision Support Systems, pp. 13-21, North-Holland, Amsterdam,
1988.
42 K . W . Hipel and D. B. Meister, Conflict analysis methodology for modeling
coalitions in multilateral negotiations, Information and Decision Technology
19:85-103 (1994).
43 N.M. Fraser, Ordinal preference representations, Theory and Decision, 36:45-67
(1994).
44 A.V. Aho, J. E. Hopcroft, and J. D. Ullman, Data Structures and Algorithms,
Addison-Wesley Publishing Company, 1987.