A Decision Making Model Its Structure and Form
A Decision Making Model Its Structure and Form
To cite this article: Bruce Bueno de Mesquita (1997) A decision making model: Its structure and
form, International Interactions, 23:3-4, 235-266, DOI: 10.1080/03050629708434909
I explain the logic and real-time track record of the expected utility model used for
forecasting the outcome and dynamics of political choices. The model represents an
example of applied modeling. It is a tool designed for practical application. As such,
some sacrifices in theoretical or analytic purity are made to gain empirical leverage. At
the same time, the model remains faithful to the rational choice perspective that suggests
that decision makers do what they believe is in their best interest. Decision makers are
modeled as individuals with bounded rationality, who are not able to look ahead over
an unbounded time horizon, but instead see only one move ahead of their current choice.
Furthermore, while they update information, they do so crudely. I also attempt to
explain limitations of the model and areas for further improvement.
<p£\x],xk)=(f:XsiWxJ-Jx& (1)
Thus, the "vote" or power mobilized by actor i in a comparison of two
alternatives (xj and xk) is equal to the potential capabilities of i
discounted by how important the issue is to i (i.e., si) and by how much
i prefers one proposal to the other (u'xj—u'xk).
The "voting" scheme reflects, if you like, what takes place "inside
the smoke-filled room" before the formal, visible decision-making pro-
cess occurs. It assumes that any formal process echoes the agreements
reached beforehand. Put differently, it assumes that stakeholders
240 B. BUENO DE MESQUITA
anticipate future action in the immediately next stage of decision
making, using backward induction. They anticipate how the formal
decision making setting will influence all actions and pick proposals in
the "smoke-filled room" that they believe will survive the formal
process. Stakeholders are bounded in their rationality, however, in that
they are unable to look farther ahead than the next stage of decision
making. So, their choices are locally rational, but may turn out to be
inefficient two or more steps down the road.
The prospect that a proposal will succeed is assumed to depend on
how much support can be mustered in its favor as compared to the
feasible alternatives. This is calculated as the sum of "votes" across all
actors in a comparison between Xj and xk. This sum equals vJk, with
(2)
If vjli is greater than zero that implies that x} defeats xk because the
tacit coalition in favor of / s proposal is more motivated and powerful
than the coalition supporting k's proposal. If vjk is less than zero, xs is
expected to be defeated by xk and if vjk equals zero the competing
interests are collectively indifferent between the two alternatives.
In any negotiation, there are likely to be many more than two pro-
posed settlements. By pitting all alternatives against one another two
at a time, the outcome preferred by the median voter (weighted by
power, salience and intensity of preference) is found. Barring percep-
tions or beliefs that lead decision makers to switch their position, the
median voter position is the predicted outcome (Black, 1958). In the
original version of the expected utility model, the median voter
position was always the predicted outcome. However, the current
version allows decision makers to switch positions in response to
proposals and pressures from one another. By doing so, the model now
provides a quasi-dynamic element.
In practice, perceptions or beliefs often lead decision makers to
grant concessions or to give in to a rival's point of view, sometimes
needlessly. Such concessions or capitulations can change the location
of the median voter. For now we only note that the initial median
voter outcome can be the predicted outcome provided key actors do
not switch positions in a way that alters the location of the median
voter. Of course, providing an accounting of when such switches in
position are expected to take place is crucial.
A DECISION MAKING MODEL 241
The basic, median voter prediction is not the final prediction of the
model. The beliefs and perceptions of the relevant actors frequently
suggest compromises and concessions that one or another actor is
willing to pursue and that other actors are willing or compelled to
accept. These beliefs and perceptions may influence the array of
interests sufficiently to require re-estimating the median voter, perhaps
several times, until perceptions and positions stabilize around the
"dominant" outcome. To undertake such before-the-fact analytic
updating it is necessary first to develop the means to estimate the
relevant beliefs and perceptions.
PERCEPTUAL ANALYSIS
EiuiAxj\d=QiutAx<j\d+(l-Qi)
t S d ] . (3)
i can challenge j's position on issue a by proposing a change in j's
position. In doing so, actor i presumably takes into account the
probability that j does not care enough about the issue to resist the
proposed settlement by i (1 — sj). i also considers the possibility that j
will resist i's proposal (sj). If there is resistance, then there is some
likelihood that i will succeed in enforcing its wishes onj (P') and some
probability that it will fail (1—P'). Should i succeed, then i will derive
the utility associated with compelling or convincing j to switch from
A DECISION MAKING MODEL 243
its current policy stance to that supported by /. This is denoted by
i/Axf\d, which equals u'fxi—Xj). Should i fail, then it confronts the
prospect of having to abandon its objectives in favor of those pursued
by j , denoted by u'Axj" \d=u'(Xj—JC(). The expected utility for challeng-
ing/s proposed resolution of the multilateral dispute (EVAxj\d) is
EVAxj=EyAxj\(l-EVAxj\2. (5)
Equations (3) and (4) reflect each actor's effort to look ahead and
estimate the consequences of alternative actions. The difference between
the two initial actions—challenge or not challenge—is represented in
equation (5). If equation (5) is greater than zero, then i believes that
challenging/s position is superior to not challenging it. In that case,
i is assumed to make a proposal of its own. If equation (5) is less than
zero, then not challenging is preferred and i is said to be deterred.
If (5) equals zero, then i is indifferent between challenging and not
challenging/s proposed settlement. Each actor evaluates equation;(5)
vis-a-vis each other actor. In doing so, actors consider the expected
actions of third parties. The estimates of P', the subjective probability
that i will be successful, include calculations of how i expects all other
parties to respond to a dispute over policy settlements between i and j .
In particular, Pl places each other actor in i's coalition, / s coalition,
or in a neutral position as indicated by each third party's preference
for i's policy proposal or j's. j makes a comparable calculation (as
does each keN). Because equation (5) includes such subjective ele-
ments as utilities and subjective probabilities, estimating a complete
matrix of expected utilities that capture all possible confrontations,
compromises, and capitulations among all the participants in the
relevant political arena is possible.
Once the expected utility values are estimated, we can denote each
relationship between pairs of stakeholders. If equation (5) is positive
for i and negative for./ then the relationship implies either compromise
or coercion. If the value of (5) for i is greater than the absolute value
of (5) as calculated for/ then both players agree that i has the upper
hand. In this instance, j is expected to be willing to offer concessions
244 B. BUENO DE MESQUITA
to i, although the concessions are not likely to be as large as what i
would like. The likely resolution of their exchange is a compromise
reflecting the weighted average of f's expectations and / s . If equation
(5) is positive for i and negative for/ and the absolute value of (5) from
/ s point of view is larger than the value of (5) from i perspective, then
j is compelled to accept I'S wishes unless someone else offers j an
enforceable compromise that spares it from having to capitulate to i.
If both i and j believe that they have the upper hand in the
relationship, then conflict is likely and that conflict has an uncertain
outcome. In international disputes, this situation is highly correlated
with the probability of a war (Bueno de Mesquita, 1985). Should both
/ and j believe that equation (5) is negative for them, then there may
be blustering and bluffing. Nevertheless, the expectation is that the
relationship is a stalemate. The most likely outcome is that the status
quo will continue to prevail between / and j .
u'xj^l-lxi-xjf, (6)
1
1 \ " " " ! Willing to trade about
• >. ! equally in each
1
-SLcategory.
Sfi
security.
t-H
Political Satisfaction —*• More
SI S2 S3
Political Satisfaction - More
„
and r i =l-(/? i -l/3)/(i? i + l/3) so that rt ranges between 0.5 and 2.
248 B. BUENO DE MESQUITA
The expected utility values summarized in (1) and (2), and in (3)
and (4), respectively, describe each actor's perception of its relation-
ship vis-a-vis each other actor. With Banks's monotonicity of escalation
result in mind, these relationships can be described in continuous
form. According to Banks's theorem, the probability with which an
actor anticipates confronting a given rival increases with its expected
utility for challenging the rival's proposal. This means that the higher
A DECISION MAKING MODEL 249
i's expected utility from persuading/ to accept i's position, the higher
the likelihood that./ will confront j .
The likelihood with which confrontation or concessions occur can
be easily displayed in a polar coordinate space. For ease of presenta-
tion, we divide such a space into six sectors, with the boundary between
each reflecting a fundamental turning point in the probability func-
tions. Figure 3 displays such a polar coordinate space, along with rele-
vant labels for each of the six sectors, reflecting the general likelihood
of alternative outcomes according to Banks's monotonicity theorem.
When the expected utility values (as perceived by either or both
actors) favor a challenge by both i and/ then a confrontation is likely
in which neither actor is inclined to offer concessions or to bargain.
Such conflictual situations involve high political costs and great
uncertainty regarding the ultimate outcome. When, however, one actor
j ' s EU
\Compromise Conflict Zone
\ 9 = Angle +
i'sE
\
N£ompron
Stalemate Zone j AcquiscesN.
MODEL OUTPUT
TABLE I
Data Inputs
TABLE lib
Verbal Summary for Group SULTAN
TABLE m
Opportunity Summary for Group SULTAN
can see, quite detailed, often running to many iterations and hundreds
of pages of information based on very few inputs. The particular issue
illustrated here is used to predict oil prices three months in advance.
The article correctly reports that I said that political forecasts made
with a model based on the "spatial theory of voting" were accurate
about 90 percent of the time... The forecasting model about which I
lectured at the University of Utah was developed by Professor Bruce
Bueno de Mesquita, now at Stanford University... Since 1982 a
colleague and I have used Bruce's models to analyze and identify policy
choice scenarios for over 1,000 issues in scores of countries around the
world... At the end of 1985 we did a systematic analysis of the accu-
racy of forecasts made with the policy choice model. That assessment
shdwed the policy decision model with inputs provided by recognized
country or issue experts correctly identified the configurations of
political forces that would lead to specific, well defined policy decisions
over 90 percent of the time. The model made it possible to identify
easy-to-observe differences among alternative political situations and
to forecast correctly the policy decision associated with each... [The
models] provide specific forecasts, something few other methods or
pundits can do with more than a moderate degree of accuracy.
[O]ne of the last (and most successful projects) in the political metho-
dologies program was the expected utility theory work of Professor
Bruce Bueno de Mesquita of the University of Rochester. The theory
is both exploratory and predictive and has been rigorously evaluated
through post-diction and in real time. Of all quantitative political fore-
casting methodologies of which I am aware, the expected utility work
is the most useful to policy makers because it has the power to predict
specific policies, their nuances, and ways in which they might be
changed" (emphasis in original).
he does not mention a set of related streams of research and theory that
justifies, we believe, a more optimistic evaluation of the field's ability to
deliver accurate predictions. The streams of research to which we refer
are, specifically: a rational choice approach to political forecasting...
The origins of the political forecasting model based on rational choice
theory can be traced to The War Trap by Bruce Bueno de Mesquita. The
theory introduced there was refined in 1985, and served in turn as
the basis for a model designed to produce forecasts of policy decisions
and political outcomes in a wide variety of political settings... This
'expected utility' forecasting model has now been tried and tested exten-
sively. [T]he amount of publicly available information and evidence
regarding this model and the accuracy of its forecasts is sufficiently sub-
stantial, it seems to us, to make it deserving of serious consideration as a
'scientific' enterprise... [W]e would argue in a Lakatosian fashion that
in terms of the range of issues and political settings to which it has been
applied, and the body of available evidence regarding its utility and
A DECISION MAKING MODEL 263
validity, it may be superior to any alternative approaches designed to
offer specific predictions and projections regarding political events.16
NOTES
1. Let N= {1,2,3,...,n} be the set of actors or stakeholders trying to influence a
multilateral decision. An actor might be a government representative, an official from
a faction within a political party or a bureaucracy, a leader of some interest group,
an influential private citizen, and so forth.
Let M={a,b,c m} be the set of issues in a multilateral negotiation and let Ra
be the line segment that describes the unidimensional policy continuum for any
individual issue a selected from among the larger set of issues M.
Let each actor i, ieN (i.e., i is a member of the set of actors trying to influence the
decision), have its own preferred resolution of issue a, with that preferred resolution
denoted as j £ . x'^ is the outcome actor i has revealed to be preferred on issue a. It
may or may not be i's true ideal point. We generally do not know for sure what
another actor's true ideal point is as there are strategic incentives for an actor to
misrepresent his or her ideal point. Because the model as applied here assesses policy
decisions on one issue at a time, we drop the issue-denoting subscript (a, or b, etc.)
from the notation so that henceforth x\ is the preferred position of actor i on the issue
being evaluated at the moment.
For any feasible proposed outcome on issue a, sayfc'sproposal, xt, i's utility for
xk, u'xt, is a decreasing function of the distance between the proposal and i's preferred
resolution, so that u'xk=f(—\xt—x*|). This means that proposals farther away from
actor i's preferred outcome are of less value to i than are proposals closer to i's
preferred outcome.
2. Again I will drop the "a" subscript from the notation throughout, but the reader is
alerted to the fact that the model does not assume that an actor's capabilities or
potential power is the same on all issues.
3. Denote the salience of issue a for actor i as su, with 0<J t a <l. Each actor is described
by the value of u'xt for all i, keN, c,, and *( on each issue. j f is assumed to be greater
than zero because if it were equal to zero for more than one stakesholder then it is
A DECISION MAKING MODEL 265
possible for division by zero to arise in the computation of the model. Strictly speak-
ing, then, the model can tolerate one actor with a salience of zero. Still, this is an
odd concept in that it implies that there is a stake holder who does not care at all
about the issue in question. In that case, the actor in question really does not have
a stake in the decision. Each actor is described by the values of i^xk for all i.fceN,
ciy and Si on each issue.
4. Say/s proposal (u'xj), to another proposal, say k's (u'xk).
5. That is, selecting xt such that i/x'^u'xt.
6. That is, to alter xt so that K*;C* ^xk.
7. To do so requires a focus on the three characteristics: u'xj, for all UjeN, sb and ct.
8. Player i makes a proposal if, in Fig. 3, the conjunction of /'s expected utility and I'S
estimate of/s expected utility falls between zero degrees from the horizontal axis
and 45° or falls between 270° and 360° from the horizontal axis. That is the domain
within which i believes it has an advantage over j and i expects more gains than
losses from challenging/s position.
9. A proposed acquiescence or capitulation by j to i's wishes is made if i locates the
conjunction of the respective expected utilities in the wedge that falls between 270°
and 315° below the horizontal axis in Fig. 3 or in the wedge, between zero degrees
and 45°. In the latter instance, i expects resistance from j , but i believes it can
enforce its demand. In the former case, i expects no resistance from j .
10. A compromise is proposed if i believes the conjunction of the relevant expected
utilities falls between 315° and 360° from the horizontal axis in Fig. 3.
11. For software based on this model, interested parties should contact Decision
Insights, International, at 11708 Bowman Green Drive, Reston, Virginia 22090.
12. A training manual for defining issues and generating data estimates is available from
the author.
13. S. Feder, "Factions and Policon: New Ways to Analyze Politics," in H. Bradford
Westerfield, Inside CIA's Private World: Declassified Articles from the Agency's
Internal Journal, 1955-1992 (New Haven, Conn.: Yale University Press, 1995).
14. Within the commercial setting, the expected utility model is sometimes called
Policon and sometimes called Factions.
15. James Lee Ray and Bruce M. Russett, "The Future as Arbiter of Theoretical
Controversies: Predictions, Explanations and the End of the Cold War," British
Journal of Political Science 25 (forthcoming, 1996).
16. Ray and Russett, p. 1569.
17. Gaddis e-mail to James Ray, February 8, 1995 and quoted in Ray and Russett,
p. 1578.
18. B. Bueno de Mesquita, "Forecasting Policy Decisions: An Expected Utility Ap-
proach to Post-Khomeini Iran," PS (1984), 226-236. This article contains several
other predictions that also have been borne out by subsequent events.
19. B. Bueno de Mesquita and B. Berkowitz, "How to Make a Lasting Peace in the
Middle East," Rochester Review (1979), 12-18, B. Bueno de Mesquita, "Multilateral
Negotiations: A Spatial Analysis of the Arab-Israeli Dispute," International Organ-
ization, 44 (1990), 317-40, B. Bueno de Mesquita and D. Beck, "Forecasting Policy
Decisions: An Expected Utility Approach," in S. Andriole, ed., Corporate Crisis
Management"(Princeton, 1985), 103-22, S. Wu and B. Bueno de Mesquita,
"Assessing the Dispute in the South China Sea: A Model of China's Security Decision
Making," International Studies Quarterly, 38 (1994), 379-403, J. D. Morrow, B.
Bueno de Mesquita and S. Wu, "Forecasting the Risks of Nuclear Proliferation:
266 B. BUENO DE MESQUITA
Taiwan as an Illustration of the Method," Security Studies, 2 (1993), 311-31, A. F.
K. Organski and B. Bueno de Mesquita, "Forecasting the 1992 French Referen-
dum," in R. Morgan, J. Lorentzen and A. Leander, eds. New Diplomacy in the
Post-Cold War World (New York, 1993), 67-75. The executive summary of a study
with Jacek Kugler predicting the shape of a peace agreement in Cambodia is
available from the author. That study was completed for the State Department in
November 1989. The peace agreement was reached in November 1991 and fit quite
well with the predictions that were made, including the holding of successful
elections and Khmer Rouge participation. Even days before the election these events
were considered unlikely by the New York Times.
CONTRIBUTOR