STEVEN J. BRAMS AND PAUL J.

AFFUSO

POWER AND SIZE: A NEW PARADOX 1

AaSTRACr. An analysis of several different indices of voting power reveals that the
voting power of a member of a weighted voting body may increase, rather than
decrease, when new members are added to the original body. Real instances of this
phenomenon, called the paradox of new members, are shown to have occurred when
new states were added to the U.S. Electoral College and new countries to the European
Community Council of Ministers. Conditions for the existence of the paradox, and
probabilities of its occurrence in small and moderate-size voting bodies, are given.
Efficient algorithms for the calculation of the voting power indices, based on generating
functions, are also outlined.

1. INTRODUCTION

'Power' is probably the most suggestive concept in the vocabulary of

political scientists. It is also one of the most intractable concepts,
bristling with apparently contradictory meanings and implications. One
implication of most definitions of power is that the greater the proportion
of resources (e.g., votes) that an actor controls, the greater is his power.
In this paper we shall show that this implication is in general false for
several well-known indices of voting power. In fact, an actor's voting
power may actually be greater when he controls a smaller, not a larger,
proportion of the votes in a weighted voting body (i.e., one in which the
members may cast different numbers of votes).
This property of the voting power indices was first recognized by
Brams, who showed that the addition of a new member to a weighted
voting body could increase the voting power of some of the old members,
despite the fact that their votes, individually and collectively, constitute
a smaller proportion of the total number of votes in the enlarged body.
Moreover, he showed that this phenomenon, which he labeled theparadox
o f new members, could occur whether or not the decision rule in the
enlarged body differed from that in the original body. 2
After explicating the concept of voting power and defining several
voting power indices, we shall review and extend Brams's hypothetical
example of the paradox to illustrate some of its bizarre effects. We shall

Theory and Decision 7 (1976) 29-56. All Rights Reserved

Copyright 9 1976 by D. Reidel Publishing Company, Dordrecht-Holland
30 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

then show that these effects of the paradox are not strictly hypothetical
but have actually occurred, in different forms, in some real voting bodies.
Next, using the mathematics of generating functions, we shall outline
efficient algorithms for computing values for the various power indices
and indicate general theoretical conditions underlying the occurrence of
the paradox. Finally, we shall demonstrate that, for voting bodies of
small and moderate size, the probability of occurrence o f this paradox
when one or more new members are added to a voting body is sur-
prisingly high.
The conclusion we do n o t draw from our analysis is that the paradox
is simply an artifact of the voting power indices that cart be dismissed as
a curious aberration. On the contrary, we argue that the paradox is a
subtle feature inherent in the relationship between power and size; the
power indices, by enabling us to demonstrate its existence and per-
vasiveness in weighted voting bodies, thereby help to render our under-
standing of the nature o f power more precise.

2. THE CONCEPT OF V O T I N G POWER

The different indices o f voting power that we shall describe and analyze
in this paper all relate to the ability of actors to control o u t c o m e s . 3
Although other definitions of power stress the effects that actors have on
other a c t o r s , 4 for the purpose of defining the power of players in voting
games, an outcome-oriented measure is preferable to an actor-oriented
measure. In large voting bodies, wherein the influence of each person on
every other person is for all practical purposes negligible, art actor-
oriented measure would suggest that no one has any power. In fact,
if each person has one vote, each person has an equal chance to influence
the outcome, which seems a more reasonable way to view power in
voting situations. 5
This view is not compatible with defining the voting power of an actor
to be proportional to the number of votes he casts, because votes per se
may have no bearing on outcomes. For example, in a three-member
voting body (a, b, c), in which a has 4 votes, b 2 votes, and c 1 vote,
members b and c are powerless if the decision rule is simple majority
(4 out of 7). Since the fact that members b and c together control ~ o f
the votes is irrelevant to the selection o f outcomes by this body, we call
 POWER AND SIZE: A NEW PARADOX 31

these members dummies. Member a is a dictator, on the other hand, since

his votes by themselves are sufficient to determine the outcome and only
coalitions of which he is a member are winning. Note that there can be
only one dictator in a voting body, whose existence renders all other
members dummies, but there may be dummies without there being a
dictator (an example of this will be given in Section 5).
The votes cast by a member of a voting body are relevant in the selec-
tion of outcomes only in the context of the number of votes east by other
members and the decision rule of a voting body. All the indices of voting
power that we shall define and illustrate in this paper utilize this informa-
tion, albeit in different ways.

3. I N D I C E S OF V O T I N G P O W E R

In this section we shall define, in the reverse order in which they were
originally proposed, several indices of voting power due to James S.
Coleman, John F. Banzhaf III, and L. S. Shapley and Martin Shubik.
For the purpose of defining each of these indices, define the value v of
some subset S of members to be
v(S)=0 if w s < d ,
v(s)=l if ws/>d,
where Ws is the total number of votes (or weight) of all members in S
and d is the decision rule of the voting body (i.e., the minimum number
of votes required for the voting body to take collective action binding
on all its members). Clearly, if subset S is a winning coalition, its value v
is equal to 1 ; otherwise, it is equal to 0.
Given a weighted voting body, Coleman defines the power of member i
to prevent action (CF) to be
Z Iv (s) - v ( s -
p~p = s (1)
Z (s)
s
where ~ s in the numerator and in the denominator represents the summa-
tion across all distinct subsets S of the voting body. 6 Thus, P ~ is the
number of subsets that, with the subtraction of member i, change from
winning coalitions to losing coalitions, divided by the total number of
32 STEVEN J. B R A M S AND PAUL J. A F F U S O

subsets which are winning coalitions. This index can be interpreted as the
proportion of winning coalitions in which member i's defection is critical,
i.e., his defection from a coalition would change its status from winning
to losing.
Banzhaf defines the power of member i to be
Z Iv (s) - v (s - {i})]
p~= s 7 (2)
h-
E Z [~ (s) - ~ ( s - {i})]
i=1 S

Note that the numerator of Banzhaf's index is the same as the numerator
of Coleman's index, giving the number of winning coalitions in which
member i's defection is critical. Since the denominator of Banzhaf's
index gives the number of critical defections of each member i, summed
across all n members of the voting body, P~ can be interpreted as member
i's proportion of critical defections. Manifestly, ~7= ~ P~= 1, i.e., the
Banzhaf power indices for all members of a weighted voting body
necessarily sum to 1.
If we divide the numerator and denominator of Banzhaf's index by
~s v(S), or the number of winning coalitions, we obtain,

Z [v(s)- ~(s- { i } ) ] / Z v(s)

e~ = ~ ~ (3)

Since ~s v(S) is a constant for any voting body, we may include it

within the outer summation ~7=x in the denominator and rewrite
Equation (3) as

e'. = s . (4)

But now we can substitute from Equation (1) P~p in the numerator and
denominator of Equation (4) to obtain

e',= e~. (S)

e~p
 POWER AND SIZE: A NEW PARADOX 33

Thus, the Banzhaf index is simply the Coleman index 'normalized'. Put
somewhat differently, if we let
1
k= - - , (6)

i=l

which is constant for any weighted voting body, then

P~ = kP~,. (7)
That is, the Banzhaf index is simply a linear transformation of the
Coleman index, the transforming factor being k.
(We note, parenthetically, that Coleman defines a second index of
individual power, which he calls the power of member i to initiate
action (CI):
Z Iv (s + {i)) - o (s)]
p~r ~ S
E [1 - v (S)]
S

The numerator of this index, though different in appearance from the

numerator of P~p (Equation (1)), counts the number of winning coalitions
in which member i's defection is critical; it is therefore the same as the
numerator of Picp. The denominator, unlike that of Equation (1), counts
the number of subsets which are losing - rather than winning- coalitions,
which is 2 " - ~ s v(S). Thus,
~[v(S)-v(S-{i})] [ s~v(S) "]
/~z s
= 2~ - E ~(s) = eL,. 2 . _ ~ V ( s ) ,
S S

from Equation (1). If we let

2" - E v (S)
k t ~ s
E~(s) '
S
then
~p t i
= kPbr,
which demonstrates that Coleman's power to prevent action (Pc~) is
related to his power to initiate action (P~) by a constant factor k'.
34 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

Substituting this definition of P~p into Equation (7),

p~ = kk ' Pcx,
~

which means that Banzhaf's index is also a linear transformation of

Coleman's second index, the transforming factor being the constant kk'
rather than the constant k for his first index (see Equation (7)). Analogous
to Equation (6), it can be shown that
1
kkt~__.

t=I

so Banzhaf's index can also be expressed in terms of a normalization of

Coleman's second index.)
Finally, Shapley and Shubik define the power of member i to be

P'ss -- ~ [ . ( s - 1 ) , n , ( n - s)!] rv(s) - v ( S - {,})], (8)

where s is the number of members in subset S and n is the total number of

members in the voting body. s Note that the term [ v ( S ) - v ( S - {i})] is
the same as that in the numerator of both the Coleman and Banzhaf
indices. However, it is weighted by the factorial term, [ ( s - 1)! (n-s)!/n!];
thus, the Shapley-Shubik index can be interpreted as the proportion of
permutations of the n members in which member i"s defection from a winning
coalition is critical.
There is no simple analytical relationship between the Shapley-
Shubik index and the Banzhaf or Coleman indices. Like the Banzhaf
index, the Shapley-Shubik index gives normalized power values that sum
to 1 for all members of a weighted voting body. 9 Unlike the Coleman
indices, the Shapley-Shubik index does not distinguish between 'preventive'
and 'initiatory' forms of voting power.
Coalition and bargaining models on which the voting power indices
are based, and calculations involving applications of these indices to both
hypothetical and real voting bodies, have been discussed and illustrated
elsewhere. 1~ We shall proceed directly, therefore, to a demonstration of
the paradox of new members. Although we shall omit Computational
details irt our subsequent discussion of the paradox, we shall describe
later efficient algorithms for generating numerical values for the indices.
 POWER AND SIZE: A NEW PARADOX 35

4. HYPOTHETICAL EXAMPLES OF THE PARADOX

Consider a weighted voting body consisting of three members whose

weights are (3, 2, 2) and for which the decision rule is simple majority
(4 out of 7). Assume that the body is enlarged by one 1-vote member to
(3, 2, 2, 1) and the decision rule remains simple majority (which now
becomes 5 out of 8). What changes in the distribution of voting power
among members are induced by this enlargement in the size of the voting
body?
In Table I we have given values of the Shapley-Shubik (P~s), Banzhaf
(P~), and Coleman power-to-initiate-action (P~x) indices in the original
and enlarged bodies. 11 Observe that while each index attributes some
nonzero voting power to the new member, the combined voting power
of the original members in the enlarged body for the Shapley-Shubik
and Banzhaf indices (0.917) - whose values always sum to 1.000 for a
voting body - is redistributed among the original three members in such
a way that the voting power of the two 2-vote members decreases, the
voting power of the 3-vote member increases. The simultaneous increase
in voting power of the 3-vote member from 0.333 to 0.417 (a 25.09/0
increase), and decrease in his proportion of votes from 0.429 to 0.375
(a 12.59/o decrease), establishes the paradox of new members for the
Shapley-Shubik and Banzhaf indices.
The paradox is also confirmed for Coleman's index, according to which
the 3-vote member's voting power increases from 0.500 to 0.556 (a 11.29/o

TABLE 1
Values of power indices for members of originM and enlarged voting bodies under
simple majority decision rule
Weight of Power indices
member i
Piss P~B P~cr
Original Enlarged Original Enlarged Original Enlarged

3 0.333 0.417 0.333 0.417 0.500 0.556

2 0.333 0.250 0.333 0.250 0.500 0.333
2 0.333 0.250 0.333 0.250 0.500 0.333
1 - 0.083 - 0.083 - 0.111
Total 1.000 1.000 1.000 1.000 1.500 1.333
36 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

increase). Although it would appear that this increase does not match
that attributed to the 3-vote member by the Shapley-Shubik and Banzhaf
indices (11.2~0 for the Coleman index vs 25.0~0 for the Shapley-Shubik
and Banzhaf indices), in fact the increase in this member's voting power
in the enlarged body is the same, in relative terms, by all three indices.
The reason that this is true is that the 'total' power of all members
according to the Coleman index changes from 1.500 to 1.333 when the
body increases from three to four members. Thus, the proportion of
power of the 3-vote member changes from 0.500/1.500=0.333 to
0.556/1.333 = 0.417, which is the same 25.0~o increase as that given by the
Shapley-Shubik and Banzhaf indices. Since, as we showed in the last
section, the power values given by the Coleman indices differ by constant
factors k and kk' from the power values given by the Banzhaf index, we
know in general that the Banzhaf and Coleman indices attribute the same
proportions of voting power to every member of a voting body. Hence,
whenever the paradox of new members is indicated by the Banzhaf index,
it will also be indicated by the Coleman indices if the basis of comparison
is proportions - rather than amounts - of power held by members of the
original and enlarged bodies. If, however, one compares the actual values
of Coleman's indices for members of different voting bodies, the Banzhaf
and Coleman indices may disagree on whether or not there exists a
paradox, as we shall illustrate later.
The fact that these indices do not disagree about the existence of a
paradox in our example demonstrates that even if one's basis of com-
parison is the actual values of the Coleman indices in two different voting
bodies, the paradox may still occur. This fact seems particularly note-
worthy when, as occurs in our example, the total power to irritate action
of all members of the voting body decreases from 1.500 to 1.333 (see
Table I). 1~' Despite this overall decrease, the 3-vote member manages to
increase his initiatory power both relatively (25.0~o) and absolutely
(11.2~) in the enlarged voting body.
Lest one think that the 3-vote member's greater power in the enlarged
voting body is an artifact of a change in the decision rule [from 4 (out
of 7) in the original body to 5 (out of 8) in the enlarged body], consider
what the power of the three original members would be if they had
operated under a decision rule of 5 (out of 7), the same as that assumed
in the enlarged body. In Table II we have given the new values of the
 POWER AND SIZE: A NEW PARADOX 37

power indices for members of the original body and included, as a basis
of comparison, the values of the indices for members of the erdarged body
(which are the same as those given in Table I).
It is immediately apparent that the two 2-vote members in the original
body are now advantaged when a fourth 1-vote member is added to this
body but the decision rule is not changed (5 votes in both bodies). The
voting power of each of these members goes from 0.167 to 0.250 by the
Shapley-Shubik index, from 0.200 to 0.250 by the Banzhaf index, and
from 0.250 to 0.333 by the Coleman power-to-initiate-action index.

TABLE II
Values of power indices for members of original and enlarged voting bodies under
s a m e decision rule

Weight of Powerindices
member i
P~ss P~B P~cx
Original Enlarged Original Enlarged Original Enlarged

3 0.667 0.417 0.600 0.417 0.750 0.556

2 0.167 0.250 0.200 0.250 0.250 0.333
2 0.167 0.250 0.200 0.250 0.250 0.333
1 - 0.083 - 0.083 - 0.111

Total 1.000 1.000 1.000 1.000 1.250 1.333

These increases, of course, are at the expense of the 3-vote member, who
had previously benefitted when the decision rule - assumed to be simple
majority - increased when the body was enlarged.
This second example illustrates that the paradox of new members may
occur with or without a change in the decision rule. Furthermore, different
members of the original body may be helped (or hurt), depending on what
changes, if any, are made in the decision rule. We may, for purposes of the
subsequent analysis, distinguish three patterns associated with its
occurrence:
(I) One or more dummies are empowered.
(2) One or more other members, excluding the largest (in terms of
weight), are advantaged.
(3) The largest member (in terms of weight) enhances his dominant
position.
38 STEVEN J. BRAMS AND PAUL J. AFFUSO

We have already illustrated the second and third patterns; in the next
section, we shall give a real-life example of the first pattern.
In terms of practical politics, the third pattern is probably the most
interesting. Conventional political wisdom suggests that a stratagem for
diluting the power of the dominant member (or coalition) in a voting
body is to increase the size of the body. There are numerous examples of
this maneuver's being employed, ranging from Franklin Roosevelt's
(unsuccessful) attempt to pack the U.S. Supreme Court in the 1930s to
(successful) efforts by Democrats in the U.S. House of Representatives
to enlarge the membership of the Rules Committee in the 1960s and
1970s. Granted, part o f the rationale behind this sort o f maneuver is
usually the conviction that the new members will be positioned ideo-
logically and behaviorally in opposition to the previously dominant
faction - that is, they will join only certain coalitions, which violates the
assumptions underlying the power indices. Nonetheless, it seems safe to
assume that the advocates of increased size expect to benefit from changes
in the formal structure of the voting body as well.
The third pattern of the paradox of new members, though, should
cause us to think twice about the wisdom of this ploy. A striking example
of this pattern, whereby the largest member does better when new mem-
bers are added to a voting body, is given in Table III using the Banzhaf
index and assuming simple majority rule irt all cases. The member with
weight 6, and 4 6 ~ of the total votes, has 50~ of the voting power in the
original body (case 1). Yet, when a new member of weight 1 is added to

TABLE III
Values of Banzhaf index for four cases under decision rule of simple
majority
Weight of Power values of Banzhaf index (P~n)
member i
Case 1 Case 2 Case 3 Case 4

6 0.500 0.565 0.542 0.636

3 0.167 0.130 0.125 0.091
2 0.167 0.130 0.125 0.091
2 0.167 0.130 0d25 0.091
1 - 0.043 0.042 -
1 - - 0.042 -

2 - - - 0.091
 POWER AND SIZE: A NEW PARADOX 39

this body, decreasing the largest member's percentage of the total votes
to 43~, the largest member's voting power increases from 50 to 57~
(case 2). While the introduction of a second new member of weight 1
decreases everyone's power (case 3), his addition still gives an advantage
to the largest member over and above that which he enjoyed in case 1
(54Y/ovs 50~ of the voting power). Incredibly though, if instead of adding
one or two members of weight 1 (cases 2 and 3), we add a single member
of weight 2 (case 4), the largest member's power jumps from 50~ to 64~.
Evidently, one does not necessarily countervail the power of a previously
dominant member by adding either several new members or a single
member of greater weight.
Given the possibility of such dramatic shifts in unexpected directions,
it is useful to explore both the conditions for the existence of the paradox
and the probability of its occurrence. First, however, to establish that
our explorations are not merely an academic exercise, we shall describe
two different real cases of the paradox.

5. E M P I R I C A L E X A M P L E S OF T H E P A R A D O X

Article II, Section 1 of the United States Constitution provides for the
election of a President by electors from each state equal to its number of
Senators and Representatives; and, as modified by Article XII (ratified
in 1804), for the election by the electors, on a separate ballot, of the Vice
President. Although there is no provision in the Constitution that the
electors from each state must vote as a bloc, in fact since the beginning
of the Republic the states, in an apparent effort to maximize their voting
power, have almost invariably east their votes as blocs. 13 This feature of
voting in the Electoral College is known as 'unit rule'.
Given unit-rule voting, the Electoral College can, in effect, be considered
a weighted voting body, with the states - casting different numbers of
votes - as its members. This body grew from its 13 original members
when the Constitution was ratified in 1789 to 51 members in 1964, when
the District of Columbia (casting three electoral votes) was admitted to
membership. With the expansion of the Electoral College as new states
were added to the Union, are there any instances in which the voting
power of old members increased?
Eleven times in its history the Electoral College has increased in size
 40 STEVEN J. B R A M S AND PAUL J. A F F U S O

while the original assigned weights of members remained intact. 14

Normally, this was the result of the admission of a state or a set of states
to the Union during the interval between decennial apportionments. In
two of these eleven cases, the computation of the voting power of the
states according to tile various indices showed up occurrences of the
paradox o f new members.
The first case involved the admission o f Louisiana (3 electoral votes)
in 1812, which increased the voting power of Connecticut (9 electoral
votes) and Massachusetts (22 electoral votes) - according to all the power
indices - over what they had had in 1810 (see Table IV). In turn, the
admission of Indiana (3 electoral votes) in 1816 increased the voting

TABLE IV
Changes in voting power in the electoral college
Member i Power indices

P~ss PtB Ptc~

1810 1812 1816 1810 1812 1816 1810 1812 1816

Indiana - 0.01282 - 0.0134 - - 0.0387

Louisiana - 0.01279 0.01282 0.0131 0.0134 - 0.0369 0.0387
Connecticut 0.0395 0.0402 0.0384 0.0393 0.0410 0.0386 0.1109 0.1157 0.1116
Massachusetts 0.1030 0.1036 0.1015 0.1015 0.1022 0.0998 0.2865 0.2882 0.2887
T o t a l o f 15 other
states 0.8575 0.8434 0.8345 0.8592 0.8437 0.8348 2.4239 2.3787 2.4137

power of Louisiana over what it had had in 1812 according to both the
Shapley-Shubik and Banzhaf indices.
Astounding as it may seem, this latter change, according to the Cole-
man power-to-initiate-action index, increased not only the voting power
of Indiana, but that of every other state (except Connecticut) as well.
We have shown this increase only for Massachusetts in Table IV; the
power values of the fifteen other states have been combined in a single
total.
If there are any empirical correlates of the paradox of new members in
the two instances of its occurrence that we have discovered in the Electoral
College, they would be difficult to find. The first instance is academic, for
no presidential election was held with the 1810 voting scheme in effect.
Since only one election was held with the 1812 scheme in effect, and one
 POWER AND SIZE: A NEW PARADOX 41

with the 1816 scheme, the only differences in voting behavior in the
Electoral College that could possibly be attributed to the paradox occur
in just one case. We doubt, however, that the small power changes we
have uncovered in this case have any empirical import.
Despite our misgivings about trying to validate the effects of the para-
dox in the Electoral College, its discovery in this institution underscores
the fact that any new apportionments that might be made with the
admission of new states to the Union could result in its occurrence again.
To be sure, the paradox is less likely to occur, as our later calculations
suggest, in a large voting body than a small voting body; but its re-
currence, in principle, cannot be ruled out. Certainly no method of
apportionment we know of precludes its existence.
A more significant, if less pure, case of the paradox of new members
occurred with the recent expansion of the European Community from
six to nine members. Comprising the European Coal and Steel Com-
munity (ECSC), the European Economic Community (EEC), and the
European Atomic Energy Commission (Euratom) since July 1, 1967, this
international organization has had as its governing body a Council of
Ministers, representing the national viewpoints of its members. The
numbers of votes of each country in the original Council of the EEC,
established by the Treaty of Rome in 1958, and the new Council, which
replaced the old Council in 1973, are shown in Table V. 15

TABLE V
Values of power indices for members of 1958 and 1973 councils on proposals from commission
Member i Weights Power Indices
P~ss Inn P~cz
1958 1973 1958 1973 1958 1973 1958 1973

France 4 10 0.233 0.179 0.238 0.167 0.204 0.122

Germany 4 10 0.233 0.179 0.238 0.167 0.204 0.122
Italy 4 10 0.233 0.179 0.238 0.167 0.204 0.122
Belgium 2 5 0.150 0.081 0.143 0.091 0.122 0.067
Netherlands 2 5 0.150 0.081 0.143 0.091 0.122 0.067
Luxembourg 1 2 0.000 0.010 0.000 0.016 0.000 0.011
Denmark 3 - 0.057 - 0.066 - 0.048
Ireland 3 - 0.057 - 0.066 - 0.048
United Kingdom - 10 - 0.179 - 0.167 - 0.122
42 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

In Table V we have also indicated the voting power of each member

of the old and new Councils when it considers policy proposals of the
European Commission. (The Commission is a collegiate body of
thirteen individual members, chosen by the member states, which serves
as the administrative arm of the Council, the main decision-making body.)
Action by the 1958 Council on proposals of the Commission required a
qualified majority of 12 out of 17 votes; action by the 1973 Council
requires a qualified majority of 41 out of 58 votes.
As is clear from Table V, the admission of Denmark, Ireland, and the
United Kingdom to membership in the European Community in 1973
diluted the voting power of all the members on the original Council
except Luxembourg. By all three indices of voting power, Luxembourg
exercised absolutely no influence on decisions of the original Council.
The subtraction of its one vote could, under no circumstances, render a
winning coalition losing. 16 Unless its representative was able, in discus-
sion, to influence the voting decisions of representatives from other
countries, he might as well have not attended Council meetings.17
In any event, when Luxembourg was given a second vote on the new
Council, its voting power increased to 0.01 by the Shapley-Shubik index,
0.02 by the Banzhaf index, and 0.01 by the Coleman power-to-initiate-
action index, which would seem to confirm the existence of a paradox of
new members. Strictly speaking, however, this increase in Luxembourg's
voting power on the new Council does not meet the conditions of the
paradox illustrated earlier. In our previous examples of the paradox, the
weights of the original members did not change when new members were
added to a voting body, though we did allow the decision rule to change.
Yet, in the case of the Council, the weights of the old members did change
when new members were added.
We do not think, however, that this fact violates the spirit of the
paradox. For while Luxembourg's votes were doubled on the new
Council, the votes of all the other original members were increased by a
factor of two and one half. The fact that Luxembourg still managed to
increase its voting power, despite its having proportionally fewer votes
on the new Council - relative to the other original members - than on
the old Council, seems to us to be an even more striking manifestation
of the inverse relationship between power and size than is illustrated by
the examples of the paradox discussed earlier.
 POWER AND SIZE: A NEW PARADOX 43

Although this fact is not related to the paradox, it is interesting to note in

passing that the next-smallest countries on the new Council- Denmark and
Ireland, which were each given three votes (one more vote than Luxem-
bourg)- have four to six times the voting power of Luxembourg according
to the various power indices. Clearly, weight and voting power may diverge
significantly. Given that 'the weighting of votes takes some account of the
differences in population of the various Member States...,'ls this divergence
will obviously affect the correlation between voting power and population.
When the Council does not act on proposals that it receives from the
European Commission but instead develops its own proposals, the
decision rule for the 1958 Council required not only a qualified majority
of 12 votes but also the assent of at least four of the six members. For
the 1973 Council, the assent of at least six of the nine members is re-
quired, in addition to a qualified majority of 41 votes. 19 This latter provi-
sion has ensured that 'the ['four] large countries, voting en bloc, cannot
dominate the smaller nations on any issue'. 20
These slightly more stringent decision rules afforded Luxembourg
some nonzero voting power on the 1958 Council as well as on the 1973
Council. Nonetheless, Luxembourg still gained power on the 1973
Council, according to the Banzhaf index, raising its proportion from
0.024 to 0.026. However, its power dropped, like that of all the other
original members, according to the two other power indices, falling from
0.017 to 0.016 by the Shapley-Shubik index and from 0.022 to 0.018 by
the Coleman power-to-initiate-action index.
This example illustrates that the different power indices may not agree
on the existence of a paradox of new members. In general, though, it
seems that the disagreement of the power indices on the existence of a
paradox occurs only when changes in the voting power of the original
members from the old to the new body are slight. Under these circum-
stances, we would expect it would be difficult to validate any empirical
consequences associated with the existence (or nonexistence) of the
paradox, anyway. The paradox may have more dramatic consequences
when a formerly disenfranchised member like Luxembourg is enfranchised
(under the less stringent decision rule allowing the Council to act on
proposals from the Commission by only a qualified majority), but we
have found no evidence that Luxembourg's role as a dummy member on
the original Council has heretofore been recognized.
44 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

6. GENERATING FUNCTIONS AND COMPUTATION OF T H E

POWER INDICES

The formulas for all the power indices involve summations over all
subsets of members S of a voting body. Since, as we showed in Section 3
there are 2" subsets S in a n n-member voting body, the computation of
the indices quickly becomes very tedious as n increases. Even with the
aid of the computer, this computation is practically unmanageable for a
weighted voting body of 20 members, not to mention a body like the
Electoral College with 51 bloc members.
Fortunately, there is a way to count the number of coalitions in which
each member's defection is critical that does not involve checking all 2"
subsets S, or even the subsets that are winning coalitions. This procedure
involves the use of generating functions, which are functions of the form

f ( x ) = ao + a l x + a2x2 .k- . . . . ~ ajx ~.

y=O

A generating function generates an infinite sequence {aj}, whose elements

are real.
As an illustration of such a function, let

x
f (x) = - - = 1 + x + x 2 +....
1--x

T h e n f ( x ) generates the infinite sequence (1, 1, 1.... ); that is, a j = 1 for

each j. The variable x has no significance of its own and is introduced
solely to identify a~ as the coefficient of xa in the expansion o f f ( x ) .
If the sequence {aj} is bounded, f ( x ) converges for [xl < 1.zt But the
question of whether the series converges is not germane if x is inter-
preted as just a formal variable.
Let the number of votes (or weights) of each member i in an n-member
weighted voting body be wi, where ~ = 1 wi = W. Assume, as before, that
the decision rule of the voting body is d. If we cart find a generating func-
tion f ( x ) whose coefficients aj give the number of winning coalitions
containing members with exactly 1 vote, 2 votes ..... d votes, d + 1 votes,
.... W votes, then the task of computing the Banzhaf and Coleman
il~dices will be considerably eased. (The Shapley-Shubik index presents
 POWER AND SIZE: A NEW PARADOX 45

a complication that will be discussed below.) For by summing the co-

efficients that give the number of losing coalitions (with less than d votes)
which would be rendered winning (with at least d votes) by the addition
of member i with w~votes,
d-1
ad-w, + aa-w,+1 + ' " + an-1 = ~., aj, (9)
j=d-wt

we can determine the number of critical defections of each member i

with wi votes. This expression can then be substituted for the expression,

Y, Iv (S) - v (S - {i})],
8

in the Banzhaf and Coleman indices and values for these indices readily
obtained.
The problem of computing the coefficients aj giving the number of
distinct coalitions with members whose combined weight is j votes is very
similar to the problem of computing the number of partitions of a
positive integer j into summands wl ..... w., subject to the restriction that
each of these summands enters a partition at most once - but possibly not
at all - and the order of the summands in a partition is disregarded. To
solve this problem, 2~ we form the product

(1 + x " ) (1 + x wS)-.. (1 + x TM)= f i (1 + xW'). (10)

1=1

If we remove the parentheses in the product, then we can obtain a sum

of the form
IV
a l x 1 + az x2 + . . . + aIVx ve = ~ aix j, (11)
j=l

where the coefficients aj correspond to the number of coalitions with

members whose combined weight isj votes. Equations (10) and (11) give,
therefore, the generating functionf(x) for the unordered partitions of an
integerj into summands wl ..... w,.
The coefficients a i given by Equation (11) do not correspond to the aj
given by Equation (9), however, for the former include member i, whose
wi votes we want to exclude as a summand in the coefficients of Equa-
tion (9). (Otherwise, member i would already be included in some losing
coalitions with at least d-w~ votes.) Accordingly, we define a generating
46 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

function for member i to be

fi (x) = 1-I (1 + xW0, (12)

where the product ranges over all membersj ( j = 1, ..., n) except member i.
As an example, consider a three-member weighted voting body where
w, = 1, wz =2, and w3--3. The generating function for member 3 is
fa (x) = (1 + x x) (1 + x2),
which expands to
fa (x) = 1 + x* + x 2 + x a .
Assuming a decision rule of simple majority (4 out of 6), from Equation
(9) the number of critical defections of member 3 is a, + a z + a a =
1 + 1 + 1 = 3. Procedures for successively computing the various aj from
recurrence relations can be readily found. =a
Generating functions, to the best of our knowledge, have never been
used before to compute the Banzhaf or Coleman indices. In contrast, a
generating function has been used to compute values of the Shapley-
Shubik index in large weighted voting bodies. 24 This function,

fi(x, y) = I I (1 + xW,y), (13)

likefi(x) in Equation (12), is specific to member i; also, as in Equation

(12), the product term on the right-hand side of Equation (13) ranges
over aU members j ( j = 1.... , n) except member i. Its second argument in
y, which is equal to the constant 1 in Equation (12), distinguishes subsets
of members S whose combined weight is the same but whose number of
members s is different. Although it was not necessary to make this
distinction in computing the Banzhaf and Coleman indices, it is necessary
to make this distinction to compute the ( s - 1 ) t ( n - s ) ! / n l different
orderings of members in the Shapley-Shubik index (see Equation (8)).
The coefficients ajk in the expansion of Equation (13),
n - - I W--w~
f,(x,y)= y. y, aj sf ,
k-1 j=l

yield values for [ v ( S ) - v ( S - { i } ) ] that can be obtained in a manner

analogous to that defined for the Banzhaf and Coleman indices by
 POWER AND SIZE: A NEW PARADOX 47

Equation (9). It is also worth noting that the Shapley-Shubik index can
be efficiently computed from another generating function, given by
Vilenkin, that counts the ordered partitions of an integer with summands
W t ~ . . . , Wn.25

7. CONDITIONS F O R T H E E X I S T E N C E OF T H E P A R A D O X

In this section we develop, with the aid of generating functions, a con-

dition for the existence of the paradox of new members based on the
Banzhaf index. Similar conditions can be developed for the Coleman
indices using the generating function given by Equation (12), and the
Shapley-Shubik index using the generating function given by Equation
(13).
The Banzhaf index defined by Equation (2) can be expressed as
d-1
aj
p~= j=a-w, -, (14)
d-1

i=1 j=d-wf

where the coefficients aj are generated by the function

fi(x) = l--[ ( l + x W ' ) = a o + a l x + ' " + " w - w , ,, x w-w, 9 (15)

This index gives the number of critical defections for member i (see
Equation (9)), divided by the total number of critical defections for all n
members of the voting body.
If an (n + 1)st member of weight w. + 1 is added to the voting body, and
the decision rule changes from d to d', then

g , ( x ) = (1 + x~"+~)f~(x) =
= bo + b t x + ' " + " w - ~ + w . + l x w-w'+w"§ (16)

is the function that generates new coefficients bj for members in the

enlarged body. But since the coefficient bj of M in Equation (16) is

from Equation (15), the Banzhaf index for member i in the enlarged body
48 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

(indicated by a prime in the subscript) will be

d'- 1

p~,= n+l
~---d'-.,
d'-I
(17)
~ (aj Jr aj-wn+,)
iffil jfd'-wt
IfP~,>P~ for at least one member i (i=1 ..... n), then there will be a
paradox of new members. However, this inequality, as defined by Equa-
tions (14) and (17), does not appear after simplification to have any ready
interpretation, even when d = d' (i.e., the decision rule in the original and
enlarged bodies is the same), not to mention the more complex inequality
engendered when more than one new member is added to the original
body.
Even if we were able to offer a simple interpretation for conditions
leading to various forms of the paradox, they would not necessarily tell
us anything about its expected frequency of occurrence. It is to this
question that we turn in the next section.

8. E X P E C T E D F R E Q U E N C Y OF T H E P A R A D O X

With the aid of the computer, we generated all partitions of the integers 2
through 18 and used these partitions to define a sample space of weighted
voting bodies with between 2 and 18 members. 26 The partitions of the
integer 5, for example, define six weighted voting bodies with between
2 and 5 members: (4, 1), (3, 2); (3, 1, l); (2, 2, 1); (2, 1, 1, 1); (1, 1, 1, 1, 1).
Subject to the restriction that the combined weight of all members is
never greater than 18 in the enlarged voting body, we then tested whether
the addition of one or more weighted members to all 2-member, 3-
member .... ,17-member voting bodies in our sample produced the
paradox of new members. We assumed that a decision rule of simple
majority was operative in both the original and enlarged voting bodies
in computing the power indices.
In Table VI we present an abbreviated table of results of this computa-
tion using the Banzhaf index. (The Coleman and Shapley-Shubik indices
produce similar results.) For bodies whose number of original members
varies between 2 and 7, Table VI shows the probabilities of occurrence of
the paradox when between 1 and 6 members are added that increase the
 P O W E R A N D SIZE: A N E W P A R A D O X 49

bodies sizes to between 3 and 13 members. (Given in parentheses below

these probabilities are the numbers o f cases sampled for relevant partitions
o f the integers 2 t h r o u g h 18.) The results reveal that the probability o f
the p a r a d o x tends to decrease - with qualifications, to be spelled out
subsequently - as more new members are added (reading across the rows)
and as the size o f the b o d y increases (reading d o w n the columns).

TABLE VI
Probability of the paradox of new members a
No. of No. of new members added
original
members 1 2 3 4 5 6

2 0.459 0.620 0.666 0.669 0.661 0.646

(444) (960) (1245) (1197) (959) (690)
3 0.492 0.483 0.446 0.405 0.371 0.344
(711) (1245) (1387) 0187) (871) (581)
4 0.578 0.548 0.475 0.416 0.374 0.332
(797) (1197) (1187) (932) (636) (400)
5 0.561 0.442 0.357 0.290 0.248 0.204
(719) (959) (871) (636) (410) (244)
6 0.529 0.389 0.302 0.237 0.184 0.!43
(567) (690) (581) (400) (244) (139)
7 0.463 0.290 0.194 0.126 0.086 0.054
(412) (462) (365) (237) (138) (74)

a Based on the application of the Banzhaf index to all partitions of the integers 2
through 18. Simple majority rule is assumed in the calculation of the Banzhaf index;
the numbers in parentheses indicate the numbers of cases examined.

A l t h o u g h these trends confirm our expectations that the p a r a d o x is

less likely to occur in large bodies or when several new members are
added, we do find it surprising that the probability values are as high
as they are. All the probabilities, for example, exceed 0.45 when one new
m e m b e r is added to each o f the original bodies (with between 2 and 7
members). Even when as m a n y as 5 members are added to a 5-member
b o d y - thereby doubling its size - the probability o f the p a r a d o x is still
a relatively high 0.25.
50 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

Very small bodies with between 2 and 4 original members that are
enlarged by 1 to 5 additional members contribute some nonuniformity
to the trends mentioned earlier. The reason is that for bodies containing
very few members, their enlargement does not generate many additional
power distributions. For example, in a 2-member body, there are by the
Banzhaf index only two distinct power distributions (0.50, 0.50) or (1, 0);
that is, either the members have equal voting power or there is a dictator.
The addition to this body of one member may result in three different
power distributions: (1, 0, 0), (0.33, 0.33, 0.33), or (0.60, 0.20, 0.20).
While two of these three distributions for a 3-member body will always
produce a paradox if there is a dictator in the original 2-member body,
when this original body is enlarged progressively by the addition of 2, 3,
and 4 new members, a greater proportion of power distributions gener-
ates the paradox. As can be seen in the progression of probability values
in the first row of Table VI, when there are more than 4 new members,
the general trend of decreasing probabilities takes hold.
We have not computed the expected frequency of the paradox for large
weighted voting bodies like the Electoral College, which would seem to
require the use of approximation methods based on Monte Carlo tech-
niques for the sample space as we have defined it. Although we suspect
that the probability of the paradox approaches zero as the number of
original members approaches infinity (and the number of new members
remains finite), we would conjecture that the probability of the paradox
is never zero for any finite weighted voting body (trivial exceptions ex-
cluded, such as bodies in which all members have equal weight). That is,
we hypothesize that it is always possible to add one or more new members
to any (nontrivial) finite weighted voting body and create a paradox of
new members.
Whether there exists a voting body invulnerable to the paradox is of
less practical import than the probability of occurrence of the paradox.
At least in small and moderate-size bodies, this probability is reasonably
high. (Of the 24744 changes in voting body size included in Table VI,
11352, or 45.8~, result in the paradox). Since most weighted voting
bodies we know of tend to be relatively small, we think that the paradox
cannot be dismissed as an unlikely aberration. Certainly the real in-
stances of its occurrence that we described earlier demonstrate that its
existence is not strictly hypothetical.
 P O W E R A N D SIZE: A N E W P A R A D O X 51

9. S U M M A R Y AND C O N C L U S I O N

We began our analysis by distinguishing power as control over outcomes

from power as control over other actors, arguing that the former con-
ception better captures the idea of the power of members in weighted
voting bodies. We then defined several indices of voting power, each of
which measured an actor's ability to control the selection of outcomes in
a weighted voting body. We showed that the Banzhaf and Coleman
indices, which are linear transformations of each other, counted the
number of winning coalitions in which an actor's (critical) defection
would render it losing, whereas the Shapley-Shubik index took account
of different permutations of the members of a voting body in which an
actor's defection from a winning coalition is critical.
All the power indices were shown to be vulnerable to the paradox of
new members, which occurs when new members are added to a voting
body and the voting power of, at least, one of the original members
increases, rather than decreases, as a result of the body's enlargement.
This phenomenon seems paradoxical, we argued, because we commonly
think that the smaller the proportion of votes that a member controls,
the less will be his voting power.
We showed that the paradox of new members cart occur whether the
decision rule changes or stays the same when the original body is en-
larged. Moreover, it is not necessarily the case that adding more or
weightier new members will mitigate the effects of the paradox, as one
of our hypothetical examples demonstrated.
We described two real instances of the paradox in the Electoral College
when new states were added to the Union and the voting power of some
of the original states increased. Elections were held in one of these
instances, but because the power changes were practically negligible, we
questioned whether they could have had any noticeable empirical con-
sequences (even if they were perceived, which seems highly unlikely).
On the other hand, although Luxembourg also increased its voting power
only marginally on the European Community Council when the Coun-
cil's membership was enlarged from six to nine members, the fact that it
started out as a dummy - according to one of the decision rules commonly
used by the Council - makes this case much more interesting and signif-
icant. For Luxembourg did benefit from a qualitative change in its power
52 S T E V E N J. B R A M S A N D P A U L J. A F F U S O

position; that this change seems to have gone undetected underscores the
capricious nature of constitution writing - done mostly by lawyers
uninformed as to the significance of the weights and decision rules they
set down - even today.
We next showed how generating functions can be used to compute
values for the various power indices that otherwise could only be ap-
proximated in large weighted voting bodies. These functions also pro-
vided a convenient way to express conditions for the existence of the
paradox, but these conditions did not appear to have any simple
interpretation.
Using the Banzhaf index, we then developed expected frequencies for
the occurrence of the paradox based on partitions of the integers 2
through 18. Although the probabilities for the paradox's occurrence
decrease as the size of the original voting body and the number of new
members that are added increase, they are surprisingly high in relatively
small weighted voting bodies.
If one cannot dismiss the paradox of new members as either contrived
or improbable, one might still argue that it is a creation of indices that
only imperfectly reflect power realities in actual voting bodies. But to
make this argument cogent, one must spedfy what the 'realities' are.
Although political, historical, ideological, and other constraints have
an evident effect on one's ability to control outcomes in an actual
weighted voting body, we believe that the power indices nonetheless
provide very useful measures of a priori voting power based on members'
weights and the decision rule of the body. The fact that they are founded
on more or less plausible coalition models, and have been widely applied
to the analysis of numerous real weighted voting bodies, would seem to
indicate their general appeal as intellectual constructs. 27 At a more
practical level, the Banzhaf index has been accepted by the New York
State Court of Appeals as a basis for assigning weights to representatives
on that state's County Boards of Supervisors. 2s
Given the widespread use and acceptance of the power indices, we
believe that an aberration they show up must be taken seriously. Instead
of thinking of the paradox of new members as 'aberrant', however, we
prefer to view it as an aspect of voting power whose existence would have
been difficult to ascertain in the absence of precise quantitative concepts.
These concepts, by throwing into bold relief the subtle relationship
 POWER AND SIZE: A NEW PARADOX 53

b e t w e e n t h e p o w e r o f m e m b e r s in a n original an d in an enlarged v o t i n g
b o d y , enable us t o c o m p r e h e n d m o r e fully different d i m e n s i o n s o f t h a t
power. It is a l i m i t a t i o n in o u r t h i n k i n g a n d models, n o t all a b e r r a t i o n
in the p h e n o m e n o n , t h a t has h e r e t o f o r e led us to e q u a t e p o w e r a n d size.

N e w York University

NOTES

t An earlier version of this paper was presented at the Annual Meeting of the Public
Choice Society, Chicago, April 3-5, 1975.
2 Steven J. Brains, Game Theory andPolitics (New York, Free Press, 1975), pp. 178-180.
a Note that this definition excludes the preferences of actors, though information about
preferences - insofar as it is available - can certainly be incorporated in a formal
definition of voting power, as Peter C. Fishbum illustrates in The Theory of Social
Choice (Princeton, N.J., Princeton University Press, 1972), pp. 53-55. For a lucid
analysis of power as the causation of outcomes by preferences, see Jack C. Nagel,
The Descriptive Analysis o f Power (New Haven, Conn., Yale University Press, 1975).
4 For a good collection of readings on the concept of power, see Political Power:
A Reader in Theory and Research, Roderick Bell, David V. Edwards, and R. Harrison
Wagner (eds.), (New York, Free Press, 1969).
5 John F. Banzhaf Ill, 'Weighted Voting Doesn't Work: A Mathematical Analysis',
Rutgers Law Review, 19 (Winter 1965), pp. 329-330, No. 31.
6 James S. Coleman, 'Control of Collectivities and the Power of a Collectivity to
Act', in Social Choice, Bernhardt Lieberman (ed.), (New York, Gordon and Breach,
1971), pp. 269-300. Note that in a voting body of n members, each member either is
or is not a member of subset S. Thus, there are 2 n distinct subsets S (including that
containing 0 members and that containing n members) across which the summation
?Es ranges, which is also the number of partitions of the voting body into two subsets.
7 Banzhaf, 'Weighted Voting Doesn't Work: A Mathematical Analysis'; John F.
Banzhaf IH, 'Multimember Electoral Districts - Do They Violate the "One Man,
One Vote" Principle?' Yale Law Journal 75 (July 1966), pp. 1309-1338. Combined
with a measure of citizen voting power in states, Banzhaf's index has been applied to
the calculation of the power of states in the U.S. Electoral College in John F. Banz-
hay III, 'One Man, 3.312 Votes: A Mathematical Analysis of the Electoral College',
Villanava Law Review 14 (Winter 1968), pp. 304-332. Approximate values for both the
Shapley-Shubik and Banzhaf indices have recently been found for the Electoral College
in which individual citizens are assumed to be players in Gtfillermo Owen, 'Evaluation
of a Presidential Election Game', in American Political Science Review, forthcoming.
s L. S. Shapley and Martin Shubik, 'A Method of Evaluating the Distribution of
Power in a Committee System', American Political Science Review 48 (September
1954), pp. 787-792. The exclamation point (l) in Equation (8) indicates a 'factorial'
and means that the number it follows is to be multiplied by every positive integer
smaller than itself (e.g., 4[ = 4-3.2.1 = 24).
9 For a proof, see L. S. Shapley, 'A Value for N-Person Games', in Annals of Mathe-
matics Studies (Contributions to the Theory of Games, H. W. Kuhn and A. W. Tucker
54 STEVEN J. BRAMS AND PAUL J. AFFUSO

(eds.)), 28 (Princeton, N.J., Princeton University Press, 1953), pp. 307-317; and
Pradeep Dubey, 'On the Uniqueness of the Shapley Value', Technical Report, Depart-
ment of Operations Research (Ithaca, N.Y., Cornell Ulfiversity, June 1974). An
axiomatization of the Banzhaf index is given in Pradeep Dubey and Lloyd S. Shapley,
'Some Properties of the Banzhaf Power Index' (unpublished paper, February 1975).
Axioms that all the power indices satisfy are given in M. G. Allingham, 'Economic
Power and Values of Games', Zeitschrift fiir National~konomie, forthcoming.
10 Brams, Game Theory and Politics, Chap. 5; and William F. Lucas, 'Measuring
Power in Weighted Voting Systems', Technical Report No. 227, Department of
Operations Research (Ithaca, N.Y., Cornell University, September 1974). We have
not included in our discussion of the various power indices an index and model, based
on the notion of bribery, recently proposed in H. P. Young, 'Power, Prices, and In-
comes in Voting Games', Mathematical Programming, forthcoming.
11 As we showed in Section 3, values of Coleman's power-to-initiate-action index
(P~cz), when multiplied by a constant factor k', give values of his power-to-prevent-
action index (P~cp). In the case of the original three-member body, k ' = 1 (i.e.,
P~cr =picz); for the enlarged four-member body, k" = 1.286 [i.e., Ptcp= (1.286)P~cz].
1~ This decrease is consonant with the increase in the 'power to act' of the voting body,
which is a measure of voting power proposed by Coleman that is applicable to the
voting body as a whole. It is defined to be

Xv(s)
P(A)- s

and increases from 0.500 for the three-member body to 0.563 for the four-member
body. See Coleman, 'Control of Collectivities and the Power of a Collectivity to Act',
pp. 277-280. [Incidentally, like P (A), the total power of all members to prevent action
(E;*=I/nc~) also increases from 1.000 to 1.714.]
13 See Nelson W. Polsby and Aaron B. Wildavsky, Presidential Elections: Strategies
of American Electoral Politics, 3d ed. (New York, Charles Scribner's Sons, 1971), p.
108, No. 51.
14 See Svend Petersen, A Statistical History o f the American Presidential Elections
(New York, Frederick Ungar Publishing Co., 1963), pp. 3-6, Table 1.
15 See Traitds instituant les communautds europdennes (Communaut6s europ6ennes,
1971), pp. 292-293, Article 148, for voting weights of members of the original Council;
and Treaties Establishing the European Economic Communities (Luxembourg, European
Communities, 1973), pp. 294-295, Article 148, for voting weights of members of the
new Council. We are grateful to Glenda Rosenthal for calling our attention to these
changes in voting weights.
18 Ronald Rogowski's calculation of Luxembourg's 'probability of unique determina-
tion' on the original Council is not zero, but the substantive meaning of this concept -
at least in the context of measuring votingpower - is not at all clear. See Ronald
Rogowski, Rational Legitimacy: A Theory of Political Support (Princeton, N.J.,
Princeton University Press, 1974), p. 135, Table 3.4.
17 For another example of a voting body with dummy members that ultimately led
to the reassignment of weights to members when this fact was recognized and challenged
in the courts, see Banzhaf's discussion of the voting power of members of the Nassau
County Board of Supervisors in Banzhaf, 'Weighted Voting Doesn't Work: A Mathe-
matical Analysis', pp. 338-340. For a report on the continuing litigation concerning
 POWER AND SIZE: A NEW PARADOX 55

representation on the Nassay County Board, see New York Times, November 17, 1974,
p. 43; and New York Times, June 15, 1975, Sec. 4, p. 6.
t8 Encyclopedia of European Community Law: European Community Treaties (New
York, Matthew Bender, 1974), vol. BI1, pt. B10, p. B10118.
19 There are, in addition, some decisions on which the Council may take action by
simple majority (at least five out of nine members since 1973) and some which require
unanimity (all nine members). In a comment on the Treaty article on voting procedures,
it is noted that 'in practice unanimity is still generally sought by the Council and voting
occurs rarely'. Encyclopedia of European Community Law: European Community
Treaties, p. B10118. For further information on the Council of Ministers, see Gerhard
Mally, The European Community in Perspective: The New Europe, the United States,
and the Worm (Lexington, Mass., D. C. Heath and Co., 1973), pp. 105-107; and
European Community: The Facts (Washington, D.C., European Community Informa-
tion Service, 1974), p. 4. Both of the latter publications contain erroneous information
on voting weights and decision rules in the Council. Conversation with staff members
of the European Community Information Service, New York City, February 25, 1975.
Correct information on weights and rules can be found in Keesing's research report,
The European Communities (New York, Charles Scribner's Sons, 1975), pp. 22-24;
and Uwe Kitzinger, Diplomacy and Persuasion (London, Thames and Hudson, 1973),
p. 95.
2o European Community: The Facts, p. 4.
21 William Feller, An Introduction to Probability Theory and Its Applications, Vol. 1,
3d ed. (New York, John Wiley and Sons, 1968), p. 264.
22 See N. Ya. Vilenkin, Combinatorics (New York, Academic Press, 1971), pp. 167-171.
as For details, see Vilenkin, Combinatorics, p. 169.
24 It was first used in Irwin Mann and L. S. Shapley, 'Values of Large Games VI:
Evaluating the Electoral College Exactly', Memorandum RM-3156-PR (Santa Monica,
Calif., RAND Corporation, May 1962); previous to the discovery of this function,
Mann and Shapley had used computer simulation techniques to compute approximate
power values of states in the Electoral College in Irwin Mann and L. S. Shapley,
'Values of Large Games, IV: Evaluating the Electoral College by Montecarlo Tech-
niques', Research Memorandum RM-265 (Santa Monica, Calif., RAND Corporation,
September 1960). These RAND memoranda are published in abbreviated form in
Irwin Mann and L. S. Shapley, 'The A Priori Voting Strength of the Electoral College',
in Game Theory and Related Approaches to Social Behavior, Martin Shubik (ed.),
(New York, John Wiley and Sons, 1964), pp. 151-164.
25 See Vilenkin, Combinatorics, pp. 169-171.
26 The partitions of a positive integer are the ways of writing that integer as a sum of
one or more positive integers. Thus,
3=2+1=1+1+I
give the three partitions of the integer 3.
27 Brains, Game Theory andPolitics, Chap. 5. For an attempt to validate the Banzhaf
and Shapley-Shnbik indices as applied to the analysis of congressional-presidential
power in the United States, see Steven J. Brams and Lee Papayanopoulos, 'Legislative
Rules and Legislative Power' (Paper presented at the Seminar on Mathematical Models
of Congress, Aspen, Col., June 16-23, 1974).
28 Ronald E. Johnson, 'An Analysis of Weighted Voting as Used in Reapportionment
of County Governments in New York State', Albany Law Review 34 (Fall 1969),
56 STEVEN J. BRAMS AND PAUL J. AFFUSO

pp. 317-343; Robert W. Imrie, 'The Impact of Weighted Voting on Representation in

Municipal Governing Bodies of New York State', Annals of the New York Academy
of Sciences (Democratic Representation and Apportionment: Quantitative Methods,
Measures, and Criteria, Lo Papayanopoulos (exi.)), 219 (New York, New York Academy
of Sciences, 1973), pp. 192-199.