You are on page 1of 11

A Theory of Coalition Formation

Author(s): William A. Gamson


Reviewed work(s):
Source: American Sociological Review, Vol. 26, No. 3 (Jun., 1961), pp. 373-382
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/2090664 .
Accessed: 15/11/2012 05:55

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
American Sociological Review.

http://www.jstor.org

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
A THEORY OF COALITION FORMATION *
WILLIAM A. GAMSON
Harvard University

Coalition formation is a pervasive aspect of social life. This paper presents a theory of
coalition formation with a statement of conditions and assumptions. While applicable to
groups of varying sizes, it is shown to be consistent with Caplow's theory of coalitions in
the triad. It successfully handles the experimental results of Vinacke and Arkoff. Finally, the
applicability of various work in n-person game theory is discussed with the conclusion that,
in its present state, it fails to provide a basis for a descriptive theory of coalitions.

ANY novelists as well as political sci- A second tradition has grown entirely
entists have been fascinated by the since the end of the Second World War
intrigues that mark political life. following the publication of von Neumann
When these intrigues involve not only in- and Morgenstern's Theory of Games and
dividuals but also nations we have the stuff Economic Behavior in 1944.3 Articles by
of history. This paper deals in a general Shapley and Shubick and by Luce and
way with a subject that has been treated Rogow on a priori power distributions, the
specifically by historians and journalists for von Neumann-Morgenstern "solution" to
centuries. n-person games and the notion of psi-stabil-
In every historian's description of a revo- ity are relevant examples.4 Both the small
lution, in every political biographer's de- group and the mathematical literature will
scription of the ascent of his subject, there be discussed in detail following the presen-
is a more or less explicit account of the co- tation of the theory.
alitions and alliances which furthered the The third body of work comes from his-
final outcome. Few areas exhibit less ex- torians and journalists and is primarily de-
ternal uniformity. "Politics makes strange scriptive. Not only do these accounts capture
bed fellows" we say to express our bewilder- much of the drama of coalition formation,
ment at some new coalition which belies our
expectations from past knowledge of the (December, 1954), pp. 657-667. F. L. Strodtbeck,
participants. "Family as a Three Person Group," American Socio-
There are three separate streams of work logical Review, 19 (February, 1954), pp. 23-29.
which have been concerned with the theme. T. Caplow, "A Theory of Coalitions in the Triad,"
The sociological tributary flows primarily American Sociological Review, 21 (August, 1956),
pp. 489-493 and "Further Development of a Theory
from Simmel,' and has focused, in partic- of Coalitions in the Triad," American Journal of
ular, on the relatively simple and manage- Sociology, 64 (March, 1959), pp. 488-493. W. E.
able three-person group. The triadic rela- Vinacke and A. Arkoff, "Experimental Study of
tionship has been explored in a series of Coalitions in the Triad," American Sociological
Review, 22 (August, 1957), pp. 406-415.
experimental and theoretical papers by 3 J. von Neumann and 0. Morgenstern, Theory
Mills, Strodtbeck, Caplow, and Vinacke and of Games and Economic Behavior, Princeton, New
Arkoff.2 Jersey: Princeton University Press, third edition,
1953.
*The author is indebted to Dorwin Cartwright 4 L. S. Shapley and M. Shubick, "Method for

of the Research Center for Group Dynamics and Evaluating the Distribution of Power in a Com-
Anatol Rapoport of the Mental Health Research mittee System," American Political Science Review,
Institute for their critical comments and sugges- 48 (September, 1954), pp. 787-792. R. D. Luce and
tions. This paper was completed under a grant A. A. Rogow, "A Game Theoretical Analysis of
from the Social Science Research Council. Congressional Power Distributions for a Stable
1 Georg Simmel, "Significance of Numbers for Two-Party System," Behavior Science, 1 (April,
Social Life" in A. Paul Hare, Edgar F. Borgatta, 1956), pp 83-96. The fine book by R. D. Luce
and Robert F. Bales, editor, Small Groups, New and H. Raiffa, Games and Decisions, New York:
York: Knopf, 1955. John Wiley, 1957, contains excellent summaries of
2T. M. Mills, "Coalition Pattern in Three Per- work in the theory of games of relevance to social
son Groups," American Sociological Review, 19 scientists.
373

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
374 AMERICAN SOCIOLOGICALREVIEW
but they also serve as a valuable reference his strategic position. In fact, this "influ-
point for a theory with descriptive rather ence of position" is a primary focus of the
than strictly normative ambitions. The ac- theory. The rules of the game provide the
counts of the French National Assembly by manner in which the decision may be made;
Lerner and Aron5 and by Leites a and this includes specification of the resources
descriptions of the rise of Hitler and Stalin which are relevant to the decision.
highlight the dimensions of a theory of A social unit is any individual or group
coalitions. which for the duration of the decision fol-
Coalitions are temporary, means oriented, lows the same coalition strategy. It might
alliances among individuals or groups which be a state delegation to a political conven-
differ in goals. There is generally little value tion, a voting bloc in the United Nations,
consensus in a coalition and the stability of or an association of retail stores. A coalition
a coalition requires tacit neutrality of the is the joint use of resources by two or more
coalition on matters which go beyond the social units. Once formed, a coalition will
immediate prerogatives. This makes the frequently meet the definition of a social
pursuit of power itself, i.e., control over unit from the period of formation until the
future decisions, an ideal basis for coalition decision has been made. A winning coalition
formation since it is an instrument for the is one with sufficient resources to control
achievement of widely ranging and even the decision. The decision point is the mini-
incompatible goals. Two members may real- mum proportion of resources necessary to
ize their mutual goal antagonisms but such control the decision.
decisions lie in the future and the present Conditions of the Theory. A full-fledged
alliance may make both better able to coalition situation is one in which the follow-
achieve a wide range of goals not all of ing conditions are present:
which will be incompatible. Power is the 1. There is a decision to be made and
currency of politics. there are more than two social units at-
tempting to maximize their share of the
THE THEORY payoffs.
2. No single alternative will maximize
Some Definitions. A decision is a selection the payoff to all participants.
among alternatives. When there are several 3. No participant has dictatorial powers,
participants, the selection of any given al- i.e., no one has initial resources sufficient to
ternative will distribute rewards among control the decision by himself.
them in a particular fashion. The reward 4. No participant has veto power, i.e., no
which accrues to any participant or group member must be included in every winning
of participants from a decision is the payoff. coalition.
The payoff may include influence on future The first two of these conditions imply
decisions. that each of the participants has some stake
In any decision, there exists a weight in the outcome-we are not dealing with a
associated with each participant involved null game-and the situation is competitive.
such that some critical quantity of these Together with condition three, we are as-
weights is necessary for the decision to be sured that a full-fledged coalition situation
made. We shall call these weights resources. is an essential game. The portion of condi-
They vary with the situation, from military tion one which states that more than two
force and industrial capacity in a war to social units are involved can easily be de-
votes in a parliamentary situation to verbal rived from the last two conditions. In a
and logical ability in a court of law. One
may be able to influence the decision more 7 An inessential game, write Luce and Raiffa,

than his resources would warrant through op. cit., p. 185, is one in which "no coalition of
players is more effective than the several players
of the coalition operating alone . . . For every
5 D. Lerner and R. Aron, France Defeats EDC, disjoint R and S, V(RUS)=V(R)+V(S) . .
New York: Frederick A. Praeger, 1957. Any game which is not inessential is called essen-
I N. Leites, On the Game of Politics in France, tial." We will call a game strictly essential if the
Rand Corporation: Unedited Advance Copy, 1958. players operating alone always get zero payoff.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
COALITION FORMATION 375
one-man group, the participant has dicta- more complicated than the characteristic
torial powers and, in any dyad, either one function. However, since the theory specifies
member is a dictator or each possesses a that only one coalition wins and the payoff
veto power. to all non-members is zero, in practice we
While the first three conditions merely need know only the payoff associated with
remove trivial situations from considera- each possible winning coalition.8
tion, condition four places much more severe Since the rewards will frequently include
limits on the generality of the theory. Many anticipations of future events, the payoff
interesting situations involving blocking must reflect differences in the probability
coalitions are excluded by this condition for of achieving future rewards. To illustrate,
reasons which will become apparent shortly. the payoff for a coalition at a political con-
However, if the decision point is 50 per vention should reflect the various proba-
cent or less, condition three implies condi- bilities that the coalition's candidate will be
tion four. Then, condition four would be elected. The payoff for a coalition would be
violated if and only if some member con- the expected value of future decisions-the
trolled more than 50 per cent of the re- total payoff from such decisions multiplied
sources; but if this were true, then this by the probability of the coalition's achiev-
member would be a dictator and condition ing them.
three would be violated. 3. Non-utilitarian strategy preferences.
Parameters of the Theory. To predict We must have a rank ordering (with ties
who will join with whom in any specific allowed) of each participant's inclination to
instance, the model requires information on join with every other player exclusive of that
the following: player's control of the resources. The sources
1. The initial distribution of resources. of this non-utilitarian preference will vary
We must know, of course, what the relevant depending on the situation: in a small com-
resources are for any given decision and, at mittee, the primary source would probably
some starting point, how much of these re- be interpersonal attraction. In a political
sources each participant controls. convention, we would expect the relative
2. The payoff for each coalition. Every similarity of others' ideology and beliefs to
alternative coalition is a partition of the be the principal determinant.
players into classes, and for every such par- 4. The effective decision point. The rules
tition we must know the total rewards for of the game will frequently specify an amount
each class. In Game Theory, the character- of resources formally necessary to control
istic function of a game is calculated by the decision. Yet an amount of resources
computing the payoff to any subset of play- less than the formal amount may be suffi-
ers on the assumption that the complemen- cient to control the decision for all practical
tary set of players will form a coalition. In purposes. This may occur through considera-
short, it is postulated that the players as- tions which prevent a potentially winning
sume that every game will reduce to a two- opposition from uniting or through a "band-
person game. This sometimes gives an un- wagon effect."
realistically conservative value for a coali- For example, in a political convention
tion. when a candidate reaches a certain number
We shall include partitions into more of votes, close to but still short of a majority,
than two classes of players in calculating the opposition will "stampede."The decision
the payoffs. The same subset may receive point in which we are interested is the
one payoff when the complementary set is effective rather than the formal decision
partitioned in one manner and an entirely point, although there will be many situations
different payoff when it is partitioned in a in which these are identical.
second way. To illustrate, a coalition may
be losing and have an estimated payoff of
8 Thus, the complete payoff function for any
zero if we assume that its opponents will
particular game maps every possible coalition into
combine but it may be winning on the con- some single value-zero if the coalition is losing
trary assumption. and some positive but variable value if the coali-
The function which we require appears tion is winning.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
376 AMERICAN SOCIOLOGICALREVIEW
If we know the payoff for each coalition, hypothesis specifies the share which A will
then we can logically deduce the effective expect to give to others. Thus, A can assign
decision point. However, in practice the con- to any prospective coalition a personal payoff
struction of the payoff matrix is dependent value-his proportion of the resources in
on our prior knowledge of this value. In the coalition multiplied by the total payoff
other words, to specify the complete payoff for that coalition.
function we must know both whether a coali- These values can be assigned to payoff
tion has sufficient resources to be winning classes of which A will prefer the highest. He
and how much it will receive. Since separate does not recognize payoff differencesbetween
information is required, we have handled coalition strategies (prospective coalitions)
this as an additional constant, but it is not a in the same payoff class. Within any class,
genuinely independent one. he will pursue that coalition strategy whose
Additional Definitions and Assumptions: members have the highest mean rank on his
A minimal winning coalition is a winning scale of non-utilitarian preferences.
coalition such that the defection of any mem- When a player must choose among alter-
ber will make the coalition no longer winning. native coalition strategies where the total
The cheapest winning coalition is that mini- payoff to a winning coalition is constant, he
mal winning coalition with total resources will maximize his payoff by maximizing his
closest to the decision point. A payoff class share. The theory states that he will do this
is a set of payoffs of which the lowest is by maximizing the ratio of his resources to
no more than K per cent less than the high- the total resources of the coalition. Since his
est. The value of K is something which must resourceswill be the same regardlessof which
be determined empirically for a given coa- coalition he joins, the lower the total re-
lition situation. It specifies, in effect, how sources, the greater will be his share. Thus,
large a differencein payoff there must be to where the total payoff is held constant, he
make a difference. will favor the cheapest winning coalition.
The theory applies to full-fledged coalition As an illustrative example, let us say that
situations in which we assume the following A has 30 per cent of the resources, B has
to be true: 19 per cent, C has 30 per cent, and D has
Assumption One: The participants have 21 per cent where the decision point is 51
the same (but not necessarily perfect) in- per cent. For A, the minimal winning coa-
formation about the initial distribution of litions which he must consider are AC and
resources and the payoff to any coalition. AD. In the former, he will expect 1/2 of
Assumption Two: Participants do not dis- the payoff, while in the latter he expects to
tinguish between payoffs in the same payoff get approximately 3/5. If they differ in
class. payoff as well, 1/2 of payoff AC may be
Assumption Three: Every participant has higher than 3/5 of payoff AD. If these two
a rank ordering of non-utilitarian prefer- figures are in the same payoff class, then he
ences for joining with the other players. will choose to join with the one which he
These assumptions and the conditions of ranks higher on non-utilitarian strategy
the full-fledged coalition situation define the preference.
class of games to which the theory is appli- Finally, a coalition will form if and only
cable. We can now state the empirical hypo- if there are reciprocal strategy choices be-
theses of the theory, starting with the general tween two participants. To illustrate, let us
hypothesis: assume that X's desired coalition in some
Any participant will expect others to de- three-persongame is XY, that Y's is XY or
mand from a coalition a share of the payoff YZ, and that Z's favored coalition is XZ.
proportional to the amount of resources Only X and Y have reciprocal strategy
which they contribute to a coalition. choices, i.e. require the other in their pre-
Any participant, A, estimates the payoff ferred coalition, and, thus, the coalition XY
to himself from a prospective coalition as a is predicted by the theory.
product of the total payoff to that coalition The model envisions the process of coali-
and A's expected share of that total. The tion formation as a step-by-step process
total payoff is known to A and the general where the participants join two at a time.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
COALITION FORMATION 377
Once a coalition has been formed, the situa- with Caplow's predicted coalition for the
tion becomes a new one-that is, there is a continuous situation.
fresh distribution of resources-and, in the To make our theory applicable to the
new coalition situation, the original strategies Caplow situations, we must assume (1) that
may or may not be appropriate.If a coalition all winning coalitions have the same payoff,
which forms was predicted by the theory, (2) that there are no differences in non-
then each player's original strategy will re- utilitarian strategy preferences, and (3) that
main the same. Thus, if W's preferred coa- the decision point is a simple majority of
lition was WXY in some game, then if X the resources. It is clear, then, that the pre-
and Y join, W will still necessarily prefer diction from our model will be simply the
the strategy WXY. If, however, a coalition cheapest winning coalition in the applicable
forms which is an "error" in terms of the situations.
theory, the strategy requirements for some Four of the eight types in Table 1 do not
players may change. Thus, if player W meet the conditions for a full-fledged coali-
planned to join with X and Y but Y and Z tion situation. Types Four and Six represent
joined, W might now prefer the group YZ a dictator situation, and Caplow's prediction
to X. for these inessential games is also that no

TABLE 1. PREDICTED COALITIONS IN TRIADS OF VARYING INITIAL STRENGTH

Predicted Coalition

Type No. Distribution of Resources Caplow Gamson


1 A=B=C any any
2 A>B, B=C, A<(B+C) BC BC
3 A<B, B=C AB or AC AB or AC
4 A>(B+C), B=C none none
5 A>B>C, A< (B+C) BC or AC BC
6 A>B>C, A>(B+C) none none
7 A>B>C, A= (B+C) AB or AC Inapplicable
8 A=(B+C), B=C AB or AC Inapplicable

We can now explain why we have ex- coalition will form. Types Seven and Eight
cluded games in which some member pos- fail to meet our fourth condition that no
sesses veto power, i.e., in which condition memberhave veto power for in each of these
four of the full-fledged coalition situation isA must be included in any winning coalition.
violated. The bargaining situation which is In Type One, any coalition will have the
essential for the general hypothesis to be same total resources and thus, under the
correct is one in which every participant previous assumptions that other things are
has alternatives. Where one member has veto equal, any coalition would have equal prob-
power, there is no alternative to his inclu- ability. In Type Two, B and C will form a
sion; he could no longer be expected to winning coalition, and since A is greater than
demand only a proportional share of the either B or C, the coalition BC must be
payoff. cheaper than either AB or AC. Therefore,
our prediction for Type Two agrees with
COALITIONS IN THE TRIAD Caplow's.
In Type Three, where A's position is ideal
Caplow has published two papers on a for the role of tertius gardens, the coalitions
theory of coalitions in the triad including AB and AC are equal in strength and both
an evaluation of the experimental evidence. are cheaper than the coalition BC. Once
He specified eight types of coalition situa- again, our prediction corresponds to Cap-
tions based on the initial distribution of re- low's.
sources. Table 1 reproduces the eight types Type Five is the only situation in which
the two theories differ in their consequences.
9 Caplow, 1956, op. cit. Caplow finds the following assumptions

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
378 AMERICAN SOCIOLOGICALREVIEW
equally plausible 10: "The 'chooser' in a situation. Thirty triads played each game
triad seeks the maximum advantage or mini- three times with the order arranged to vary
mum disadvantage of strength relative to systematically the position of the situation
his coalition partner" or, "The 'chooser' in in each series.
a triad seeks to maximize the strength of the Vinacke and Arkoff suggested a "game
coalition in relation to the excluded mem- theory" prediction for each of these situa-
ber." He reasons that the weak man, C, in tions as well, which we shall call the Strict
a Type Five situation, would be sought as a Rationality Theory. They reason that the
coalition partner by both of the others and strictly rational player must realize in a
could choose on either basis. situation such as Type Five that any pair
Our theory clearly implies the first of will win, and that if he fails to form a coa-
these two assumptions. C will prefer the coa- lition, he can expect his opponents to do so.
lition BC to the coalition AC because he This reasoning will hold whether one has a
expects that the stronger A will demand a weight of two, three, or four, and there is
larger share of the payoff in accordance with no reason to expect, on rational grounds,
his superior resources. The coalition BC is, that any coalition will form with greater
of course, the cheapest coalition. frequency than any other. In fact, this
Caplow has discussed several experimental
TABLE 2. VINACKE AND ARKOFF DESIGN FOR Ex-
studies of triads with the conclusion that PERIMENTAL TEST OF SIX CAPLOW SITUATIONS
they lend some support to his analysis al-
though designed with other purposes in mind. Weights
For example, Mills discovered that a subject Type No. A B C
Description
who was the "odd man" (A) in a Type Three
situation tended very slightly to make more 1 A=B=C 1 1 1
2 A>B, B=C, A<(B+C) 3 2 2
efforts to disrupt the coalition between the 3 A<B,B=C 1 2 2
equals than the subject placed in A's role 4 A>(B+C), B=C 3 1 1
in a Type Two situation.1 Caplow concludes 5 A>B>C, A<(B+C) 4 3 2
that "we would expect less resistance to the 6 A>B>C, A> (B+C) 4 2 1
'inevitable' coalition of BC in Type Two
than to the improbable, and, therefore, un- reasoning holds for the first three types as
stable coalition of BC in Type Three." 12 well. In the non-essential types four and six,
The most crucial and significant evidence there is nothing to be gained by forming a
on coalitions in the triad comes from Vinacke coalition and the prediction is that none will
and Arkoff who, stimulated by Caplow's first take place.
paper, designed an experiment to test the This experiment, then, gives us a chance
first six of his situations.'3 This experiment to compare its results with three different
is certainly a test of our theory as well, theoretical predictions-Caplow's, Strict
given the earlier predictions of Table 1. Rationality, and our own. In Table 3 we
Furthermore,in testing the Type Five situa- compare these predictions with the actual
tion, it provides a comparison at the only results of the experiment.
point where our predictions differ. In situations one, four, and six, where
The experimenters had subjects play a there are no differences between theories,
parchesi game in which each player's moves each is supported. Coalitions do not usually
were weighted by a numbered counter which occur in the latter two situations, and in
he drew from a hopper at the beginning of situation one, they seem to occur approxi-
the game. The weights on these counters mately at random.
represented the six different initial distribu- In situations two and three, the results
tions of resources specified by Caplow's provide negative evidence for the strict
theory. Table 2 gives the weights for each rationality predictions and positive evidence
for the other two theories. Apparently, the
10
ability to perceive the necessity for a coali-
Caplow, 1959, op. cit.
" Mills, op. cit. tion in these situations is more difficult from
12 Caplow, 1956, op. cit. certain positions than from others. "It is
13 Vinacke and Arkoff, op. cit. harder," Vinacke and Arkoff write, "for an

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
COALITION FORMATION 379
TABLE 3. RESULTS OF VINACKE-ARKoFF EXPERIMENT WITH THREE THEORETICAL PREDICTIONS

1 2 3 4 5 6
A=B=C A>B, B-C, A<B, A>(B+C), A>B>C, A>B>C,
Predicted A<(B+C) B-C B__C A<(B+C) A>(B+C)

Caplow any BC AB or AC none AC or BC none


Strict rationality' any any any none any none
Gamson any BC AB or AC none BC none
Actual
AB 33 13 24 11 9 9
AC 17 12 40 10 20 13
BC 30 64 15 7 59 8
Total 80 89 79 28 88 30
No coalition 10 1 11 62 2 60
Probability' NS .01 .01 NS .01 NS
a These are also the predictions made by Caplow for the "episodic" situation.

'From Vinacke and Arkoff: Chi Square with two degrees of freedom.

initially stronger member to reach the con- occasions when the strategic position of a
clusion that the relative strengths are irrele- member gives him power out of proportion
vant than for the other one or two to arrive to his rank order of status or resources."
at this interpretation. In effect, the weaker There is apparently some basis for the as-
members can immediately understand the sumption that the size of the demands which
necessity for forming a coalition, whereas the a participant will make reflects the propor-
stronger member must go through more tion of resources which he controls-or, at
complex reasoning to do so." least, will affect what others will expect him
Situation five is perhaps the most crucial to demand. In a situation where participants
since each theory makes a different predic- meet with~eachother sequentially rather than
tion for the outcome. While Caplow predicts simultaneously, these expectations of others'
that either of the coalitions AC or BC are bargaining demands become even more
equally likely, the cheapest coalition BC crucial.
actually takes place three times as frequently In short, the small group studies of coali-
as the alternative! tions in the triad uniformly support the
This confirmation is interpreted by Vina- theory presented here. While Caplow's pre-
cke and Arkoff in a manner which echoes dictions are in most cases consistent with
the general hypothesis of our theory: ". the predictions from our theory under the
the weakest member was found to be most special case where payoffs and non-utilitarian
often a member of the winning coalition; strategy preferences are held constant, in
furthermore, his share of the winnings was one crucial difference, the results of Vinacke
larger than his strength might seem to war- and Arkoff support the prediction made here.
rant mainly because there was competition
for him, because the other players saw him MATHEMATICAL LITERATURE
as weaker, hence more readily to be induced
into partnership."14 The mathematical theory of games of
Willerman draws similar conclusions from strategy as it presently exists is a rich source
his study of coalitions in a fraternity coun- of ideas, but it can only provide orientations
in situations of the type with which we are
cil.15". . . Distribution of control within the
concerned here. This is true for several
coalition was isomorphic with the relative
reasons. The most powerful mathematical
status and resources of the members outside
developments of the theory are in the area
of the coalition. However, there seem to be of two-person, zero-sum games. The theory
of games involving many players is, to quote
14Ibid., emphasis mine.
15B. Willerman, A Final Report: Research on Abraham Kaplan, "in a very unsatisfactory
Cohesive and Disruptive Tendencies in Coalition state." 16
Type Groups, University of Minnesota, Mimeo-
graphed, 1957. 16 In M. Shubik, Readings in Game Theory and

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
380 AMERICAN SOCIOLOGICALREVIEW
We do not object to the theory of games 'should.'" This would seem to offer some
on the grounds that its assumptions are "un- promise for our purposes in spite of the fact
realistic." They are, at least, clearly stated that the emphasis is on distribution of pay-
and we may substitute more plausible ones offs. An imputation where some values are
if we can find some which are workable. positive and others are zero defines an im-
Luce and Raiffa write: ". . . it is crucial that plicit coalition between the positive entries.
social scientists recognize that game theory Unfortunately, we are not given a single
is not descriptive but rather (conditionally) imputation as a solution but rather a set of
normative. It states neither how people do these, and furthermore, a set in which all
behave nor how they should behave in an possible coalitions are allowed. As if this
absolute sense, but how they should behave difficulty were not sufficient, the solution to
if they wish to achieve certain ends." 17 Our the three-man game given above is not the
own object is descriptive but a normative only solution; in fact, there is an infinity
theory often provides a useful starting point of solutions. "In their theory," write Luce
for a descriptive theory. and Raiffa, ". . . freedom to cooperate leads
One attempt to handle the problem of to vast numbersof 'solutions' with no criteria
the n-person game is the von Neumann- to select among them. They are forced . . . to
Morgenstern "solution" theory.18A solution the ad hoc assumption that in practice there
generally consists of a set of imputations exist social standards which determine the
(an imputation is an n-tuple giving the pay- solution which actually occurs, but no at-
off to each player and satisfying certain con- tempt is made to exhibit a theory of these
ditions) having the following two properties: standards."
(1) no imputation in the set dominates any There has been an attempt by Vickrey 20
other imputation, and (2) every imputation to narrow down the number of solutions to
not in the set is dominated by one in the be considered. "Roughly, a solution is called
set.19 strong if the sequence- 1) an imputation
I shall illustrate this with a solution to the in the solution, (2) a change to a non-con-
three-man game: (1/2, 1/2, 0), (1/2, 0, forming imputation, and (3) a return to an
1/2), (0, 1/2, 1/2). No imputation among imputation in the solution-always means
the three dominates another since only one that at least one of the players participating
player could improve his position by switch- in the original deviation ultimately suffers
ing from one to another. Any particular im- a net loss. Thus, a strong solution has an
putation in the solution such as (0, 1/2, 1/2) inherent stability not possessed by other
is dominated by amputationsoutside of the solutions, and so it might be expected to
set-for example, (1/6, 2/3, 1/6), but this, occur rather than one of the weaker solu-
and in fact any imputation outside of the tions."
solution set, is dominated by a member of It turns out, encouragingly, that the only
the solution (in this case by 1/2, 0, 1/2). strong solution for the three-person game is
Luce and Raiffa argue that a "solution the symmetric one given earlier: (1/2, 1/2,
must be interpreted as a description of a set 0), (1/2, 0, 1/2), (0, 1/2, 1/2). Since none
of possible payments, any of which might of these amputations dominates any other,
arise if the players choose strategies and again extra-theoreticalreasons will determine
form collusive arrangements as they which imputation in the set is chosen.
Vinacke and Arkoff21 present data on the
Political Behavior, Garden City, New York: division of spoils made by their subjects
Doubleday, 1954. which allow us to examine the frequency
17 Luce and Raiffa, op.
cit.
18This is discussed in relatively non-technical
with which the strong solution did occur. In
fashion in Luce and Raiffa, op. cit. In a sense, we the Type One situation where all players
are proposing in this paper a new definition of had an equal share of the resources origi-
solution for certain classes of n-person essential nally, the final imputation was a member of
games, but because of the specific meaning of
"solution" in the von Neumann-Morgenstern sense,
the solution set 60 per cent of the time, but
we have refrained from using the word. in the Type Three situation (A<B, B - C)
19 For a formal definition of imputation and
domination, see Luce and Raiffa, op. cit., p. 193 20Ibid., p. 213.
and p. 201. 21Vinackeand Arkoff, op. cit.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
COALITION FORMATION 381
only 39 per cent of the time. Many of these triadic situation, any two-man coalition will
may have occurred on those occasions (19 meet the conditions of psi-stability regard-
per cent) in which the two strong, equally less of the initial distribution of resources.
powerful members joined. If our aim is uniqueness, we are no better
We may tentatively conclude that where off here than under solution theory. Attempts
the initial distribution of resources differs by Milnor 24 to describe n-person games in
among the three members of the triad, not terms of "reasonable outcomes" involve the
only are the various amputations in a solu- same type of difficulties ascribed to solution
tion set not equally probable, but the tend- and psi-stability theory.
ency to divide the rewards symmetrically is Shapley 25 gives a method for evaluating
considerablyless than when participants have the worth of an n-persongame for any player
equal power. that should help us to determine the relative
A second mathematical concept, that of bargaining positions of the several players in
psi-stability, would seem to be more appro- a game. He lists three apparently weak con-
priate for our purposes since here a game is ditions and then shows that these uniquely
-describedby both an imputation and a coali- determinean evaluation function. Ultimately,
tion structure. The basis of this notion, which he arrives at an explicit formula for calculat-
has been developed by Luce,22 is that a ing the value for a player, i. "It amounts,"
pair-an imputation and a given coalition to quote Luce and Raiffa, "to a weighted
structure-is stable when no admissable sum of the incremental additions made by
change in the coalition structure is immedi- i to all the coalitions of which he is a mem-
ately profitable. ber." In the full-fledged triadic situation,
An important addition here is the recog- the values are 1/3 for each player. This
nition that from any given coalition struc- suggests the symmetric solution.
ture, every possible coalition is not admis- In an article by Shapley and Shubik,26
sable. The concept of non-utilitarian strategy the authors attempt to apply the Shapley
preferences developed earlier is, in part, an value to certain "simple" games (in a simple
attempt to define the admissible changes game, every coalition has as its payoff either
between any two stages of the process of coa- one or zero, i.e. it is either winning or losing).
lition formation. They argue that the value gives an a priori
The implications of psi-stability for the estimation of relative power in many com-
theory presented here are less important mittee or parliamentary situations. An indi-
than one might hope. First, we are concerned vidual's power is given by the index, P/N,
primarily with the process of coalition forma- where N is the total number of permutations
tion rather than coalition stability. Although among the players and P is the number of
problemsof stability can frequently be trans- permutations in which his resources are
lated into the terms of the theory, essenti- pivotal in turning a losing coalition into a
ally the game is over when a winning coali- winning one. Luce and Rogow 27 applied
tion has been formed for a particulardecision, this to an analysis of coalitions between the
and the next decision involves a new game.23 President and the parties in the two houses
Secondly, psi-stable pairs, like solutions, are of Congress.
not generally unique and the problem of how The calculations involved in the Shapley-
to select just one still exists. Shubik power index are relatively simple to
A full-fledged coalition situation is a make, especially when the N is small. In the
strictly essential game and it is not diffi- triad, there are 3! or six permutations, and
cult to see that any losing coalition will be in the full-fledged coalition situation each
psi-unstable. However, in the full-fledged person will be pivotal twice, giving rise to
the earlier figure of 1/3.
22 R. D. Luce, "A Definition of Stability for Suppose, however, we did not assume that
N-Person Games," Annals of Mathematics, 59
(May, 1954), pp. 357-366. 24Described in Luce and Raiffa, op. cit.
23This is not a criticism of Luce since coalition 25L. S. Shapley, "A Value for N-Person Games,"
stability is obviously an important problem in its Annals of Mathematics Studies, 28 (1953), pp. 307-
own right. Eventually, a satisfactory theory of 317.
coalitions should be able to handle both coalition 26 Shapley and Shubik, op. cit.

formation and stability. 27 Luce and Rogow, op. cit.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions
382 AMERICAN SOCIOLOGICALREVIEW
resourcesare used as a bloc. Instead of ask- Our general hypothesis stated that partici-
ing which person is pivotal in the permuta- pants will expect others to demand from a
tion, we might ask which resource unit is coalition a share of the payoff which is pro-
pivotal. Referring back to Table 3, we can portional to the amount of resources which
see that for the Type One situation, the they are contributing to it. Each participant
Shapley value remains as 1/3. However, for will estimate the value of any coalition
Type Two, there are 7! permutations in- strategy as the total payoff to a coalition
stead of 3!. Any given resource unit will be multiplied by his share. He estimates this
pivotal in 6! ways since the other six units latter figure by the ratio of his resources
can be permuted that many ways while it to the total resources of the coalition. Every
remains fixed in the pivotal spot. It follows player will pursue strategies in the highest
that A's three resource units will be pivotal payoff class, but among alternative strate-
3 x 6! times while B and C will have the gies in the same class, he will choose that
pivotal unit in 2 x 6! ways each. The result- one which maximizes his non-utilitarian
ant Shapley values are 3/7, 2/7, 2/7 respec- strategy preference.
tively or exactly proportional to the share A coalition will form between two players
of resources. if and only if there are reciprocal choices
Finally, suppose we make the assumption of coalition strategy between them. Thus,
that within any coalition, a player can expect the model envisions the process of coalition
to share in the payoff proportionally to his formation as a step-by-step process until
Shapley value. It then follows that one will by successive pairing, the decision point
maximize his share in a simple game if he has been reached.
can maximize his power relative to the other The theory was compared with Caplow's
members of his winning coalition. In short, predictions for coalitions in the triad and
we are led to predict that the cheapest coa- we found that in the special case where pay-
lition will form! offs and non-utilitarian strategy preferences
It is certainly possible to question the are constant, the two theories make identical
assumptions by which we used the Shapley predictions with one exception. In an experi-
value to yield the predictions of our theory. mental test by Vinacke and Arkoff, the
Why should the resource rather than the results supported Caplow's and our own
player be considered the unit and why theory where they were opposed to the pre-
should the Shapley value determine the dictions of a strict rationality theory. At the
proportion of payoff within a coalition? one point where Caplow's theory differed
Certainly these are not Shapley's assump- from the one presented here, Vinacke and
tions and his analysis strongly suggests the Arkoff's evidence supported the latter.
symmetric solution to the triadic situation, In examining the mathematical literature,
but it is interesting to note that we can we found that the von Neumann-Morgen-
reach the same theoretical predictions by this stern solution theory was inadequate for our
slightly different pathway. purposes because of its profusion of solu-
tions for many games. An attempt by Vickrey
SUMMARY to limit these somewhat by defining a strong
We have presented a theory of coalition solution still left the crucial difficulty of
formation to apply to a full-fledged coali- determining which member of a set of im-
tion situation defined by four conditions. putations would actually occur. The concept
It is intended to apply where several parties of psi-stability also left the unique specifi-
are competitively attempting to determine a cation of a coalition to extra-theoretical de-
decision and in which no participant has termination.
either dictatorial or veto powers. The theory Finally, we explored the Shapley value and
requires information on the initial distribu- showed that it still suggests the equal prob-
tion of resources, the payoff for each coali- ability of coalitions despite initial differences
tion, the non-utilitarian strategy preferences, in resources.However, by the addition of two
and the effective decision point. Three addi- not unreasonable assumptions, it will lead
tional assumptions further defined the situa- to the same predictions as the theory pre-
tion to which the model applies. sented here.

This content downloaded by the authorized user from 192.168.82.206 on Thu, 15 Nov 2012 05:55:05 AM
All use subject to JSTOR Terms and Conditions

You might also like