7

SOCIAL CHOICE THEORY AND

CONSTITUTIONAL DEMOCRACY

William Riker

^° metnod of voting can be said to amalgamate individual judgments

truly and fairly because every method violates some reasonable canon of
fairness and accuracy. All voting methods are therefore in some sense
morally imperfect. Furthermore, these imperfect methods can produce different
outcomes from the same profile of individual judgments. Hence it follows that
sometimes—and usually we never know for sure just when—the social choice is
as much an artifact of morally imperfect methods as it is of what people truly
want. It is hard to have unbounded confidence in the justice of such results.
It is equally hard, as I will show in this essay, to have unbounded confidence
in the meaning of such results. Individual persons presumably can, if they think
about it deeply enough, order their personal judgments transitively. Hence their
valuations mean something, for they clearly indicate a hierarchy of preference that
can guide action and choice in a sensible way. But the results of voting do not
necessarily have this quality. It is instead the case that no method of voting can
simultaneously satisfy several elementary conditions of logical arrangement.
Hence, not only may the results of voting fail to be fair, they may also fail to
make sense. It is the latter possibility that will be analyzed in this essay.

Arrow's Theorem
Kenneth Arrow published Social Choice and Individual Values in 1951. Although
his theorem initially provoked some controversy among economists, its profound
political significance was not immediately recognized by political scientists.1 In
the late ip6os, however, a wide variety of philosophers, economists, and political
scientists began to appreciate how profoundly unsettling the theorem was and
how deeply it called into question some conventionally accepted notions—not
only about voting, the subject of this work, but also about the ontological validity

161
162 Philosophy and Democracy

of the concept of social welfare, a subject that, fortunately, we can leave to meta-
physicians.
The essence of Arrow's theorem is that no method of amalgamating individ-
ual judgments can simultaneously satisfy some reasonable conditions of fairness
on the method and a condition of logicality on the result. In a sense this theorem
is a generalization of the paradox of voting, for the theorem is the proposition
that something like the paradox is possible in any fair system of amalgamating
values. Thus the theorem is called the General Possibility Theorem.
To make the full meaning of Arrow's theorem clear, I will outline the situ-
ation and the conditions of fairness and of logicality that cannot simultaneously
be satisfied.2 The situation for amalgamation is:

1. There are n persons, n ^ 2, and n is finite. Difficulties comparable

to the paradox of voting can arise in individuals who use several
standards of judgment for choice. Our concern is, however, social
choice, so we can ignore the Robinson Crusoe case.
2. There are three or more alternatives—that is, for the set X = (x\, . . . ,
xm), m ^ 3. Since transitivity or other conditions for logical choice
are meaningless for fewer than three alternatives and since, indeed,
simple majority decision produces a logical result on two alternatives,
the conflict between fairness and logicality can only arise when m ^
3-
3. Individuals are able to order the alternatives transitively: If x R,y and
y Rf z, then x Rt z. If it is not assumed that individuals are able to
be logical, then surely it is pointless to expect a group to produce
logical results.

The conditions of fairness are:

i. Universal admissibility of individual orderings (Condition U). This is the

requirement that the set, D, includes all possible profiles, D, of in-
dividual orders, D,. If each D, is some permutation of possible or-
derings of X by preference and indifference, then this requirement is
that individuals can choose any of the possible permutations. For
example, if X = (x, y, z}, the individual may choose any of the
following 13 orderings:

1. x y z 7. x (y z) 10. (x y) z 13. (x y z)
2. y z x 8. y (z x) U.(yz) X
3. z x y 9. z (x y) 12. (z x) y
4. x z y
5. z y x
6. y x z (7-1)

The justification for this requirement is straightforward. If social

outcomes are to be based exclusively on individual judgments—as
seems implicit in any interpretation of democratic methods—then to
restrict individual persons' judgments in any way means that the social
 Social Choice Theory and Constitutional Democracy 163

outcome is based as much on the restriction as it is on individual

judgments. Any rule or command that prohibits a person from choos-
ing some preference order is morally unacceptable (or at least unfair)
from the point of view of democracy.
2. Monotonidty. According to this condition, if a person raises the val-
uation of a winning alternative, it cannot become a loser; or, if a
person lowers the valuation of a losing alternative, it cannot become
a winner. Given the democratic intention that outcomes be based in
some way on participation, it would be the utmost in perversity if
the method of choice were to count individual judgments negatively,
although some real-world methods actually do so.
3. Citizens' sovereignty or nonimposition. Define a social choice as imposed
if some alternative, x, is a winner for any set, D, of individual pref-
erences. If x is always chosen, then what individuals want does not
have anything to do with social choice. It might, for example, happen
that x was everyone's least-liked alternative, yet an imposed choice
of x would still select x. In such a situation, voters' judgments have
nothing to do with the outcome and democratic participation is
meaningless.
4. Unanimity or Pareto optimality (Condition P). This is the requirement
that, if everyone prefers x to y, then the social choice function, F,
does not choose y. This is the form in which monotonicity and cit-
izens' sovereignty enter all proofs of Arrow's theorem. There are only
two ways that a result contrary to unanimity could occur. One is that
the system of amalgamation is not monotonic. Suppose in D' every-
body but i prefers x to y and y P\ x. Then in D, i changes to x P, y
so everybody has x preferred to y; but, if F is not monotonic, it may
be that x does not belong to F ({x, y}, D). The other way a violation
of unanimity could occur is for Fto impose y even though everybody
prefers x to y. Thus the juncture of monotonicity and citizens' sov-
ereignty implies Pareto optimality.
Many writers have interpreted the unanimity condition as purely
technical—as, for example, in the discussion of the Schwartz method
of completing the Condorcet rule. But Pareto optimality takes on
more force when it is recognized as the carrier of monotonicity and
nonimposition, both of which have deep and obvious qualities of
fairness.
5. Independence from irrelevant alternatives (Condition I). According to this
requirement, a method of amalgamation, F, picks the same alternative
as the social choice every time F is applied to the same profile, D.
Although some writers have regarded this condition simply as a re-
quirement of technical efficiency, it actually has as much moral con-
tent as the other fairness conditions. From the democratic point of
view, one wants to base the outcome on the voters' judgments, but
doing so is clearly impossible if the method of amalgamation gives
different results from identical profiles. This might occur, for example,
164 Philosophy and Democracy

if choices among alternatives were made by some chance device.

Then it is the device, not voters' judgments in D, that determines
outcomes. Even if one constructs the device so that the chance of
selecting an alternative is proportional in some way to the number
of people desiring it (if, for example, two-thirds of the voters prefer
x to y, then the device selects x with p = 2/3), still the expectation is
that, of several chance selections, the device will choose x on p se-
lections and y on i — p selections from the same profile, in clear
violation of Condition I. In ancient Greece, election by lot was a
useful method for anonymity; today it would be simply a way to by-
pass voters' preferences. Another kind of arbitrariness prohibited by
the independence condition is utilitarian voting. Based on interper-
sonal comparisons of distances on scales of unknown length, utilitarian
voting gives advantages to persons with finer perception and broader
horizons. Furthermore, independence prohibits the arbitrariness of
the Borda count (to be discussed later).
6. Nondictatorship (Condition D). This is the requirement that there be no
person, i, such that, whenever x P, y, the social choice is x, regardless
of the opinions of other persons. Since the whole idea of democracy
is to avoid such situations, the moral significance of this condition is
obvious.

Finally, the condition of logicality is that the social choice is a weak order,
by which is meant that the set, X, is connected and its members can be socially
ordered by the relation, R, which is the transitive social analogue of preference
and indifference combined. (This relation, as in x R y, means that x is chosen
over or at least tied with y.) In contrast to the method of amalgamation or choice,
F, which simply selects an element from X, it is now assumed that F selects
repeatedly from pairs in X to produce, by means of successive selections, a social
order analogous to the individual orders, D;. And it is the failure to produce such
an order that constitutes a violation of the condition of logicality.2
Since an individual weak order or the relation R, is often spoken of as in-
dividual rationality, social transitivity, or R, is sometimes spoken of as collective
rationality—Arrow himself so described it. And failure to produce social transi-
tivity can also be regarded as a kind of social irrationality.
Arrow's theorem, then, is that every possible method of amalgamation or
choice that satisfies the fairness conditions fails to ensure a social ordering. And
if society cannot, with fair methods, be certain to order its outcome, then it is
not clear that we can know what the outcomes of a fair method mean. This
conclusion appears to be devastating, for it consigns democratic outcomes—and
hence the democratic method—to the world of arbitrary nonsense, at least some
of the time.
Naturally there has been a variety of attempts to interpret and sidestep this
conclusion. One line of inquiry is to raise doubts about its practical importance;
another is to look for some theoretical adjustment that deprives the theorem of
 Social Choice Theory and Constitutional Democracy 165

its force. The rest of this essay is devoted to a survey of both branches of this
huge and important literature.
I will begin with inquiries about the practical importance of the theorem.
One such inquiry is an estimate of the expected frequency of profiles, D, that do
not lead to a transitive order.

The Practical Relevance of Arrow's Theorem:

The Frequency of Cycles
One meaning of Arrow's theorem is that, under any system of voting or amal-
gamation, instances of intransitive or cyclical outcomes can occur. Since, by def-
inition, no one of the alternatives in a cycle can beat all the others, there is no
Condorcet winner among cycled alternatives. All cycled alternatives tie with re-
spect to their position in a social arrangement in the sense that x y z x, y z x y,
and z x y z have equal claims to being the social arrangement. Borda voting
similarly produces a direct tie among cycled alternatives. Hence a social arrange-
ment is indeterminate when a cycle exists. When the arrangement is indetermi-
nate, the actual choice is arbitrarily made. The selection is not determined by the
preference of the voters. Rather it is determined by the power of some chooser
to dominate the choice or to manipulate the process to his or her advantage.
Every cycle thus represents the failure of the voting process. One way to inquire
into the practical significance of Arrow's theorem is, therefore, to estimate how
often cycles can occur.
For this estimate, a number of simplifying assumptions are necessary. For one
thing, majority voting (rather than positional voting or any other kind of amal-
gamation) is always assumed. This assumption of course limits the interpretation
severely. For another thing, only cycles that preclude a Condorcet winner are of
interest. Voting may fail to produce a weak order in several ways:

1. With all three alternatives, there may be a cycle: xRyRzRxor

simply x y z x.
2. With four or more alternatives, there may be
a. A Condorcet winner followed by a cycle: w x y z x
b. A cycle among all alternatives: w x y z u>; or intersecting cycles: s
twxyzwvs
c. A cycle in which all members beat some other alternative: x y z
xw

If one is interested in social welfare judgments involving an ordering of all alter-

natives, then all cycles are significant no matter where they occur. But if one is
interested in picking out a social choice, as in the voting mechanisms discussed
here, then the significant cases are only i, 2(b), and 2(0), where there is no unique
social choice. (These are often called top cycles.) Attempts to estimate the signifi-
cance of Arrow's theorem by some sort of calculation have all been made from
the point of view of social choice rather than welfare judgments and have
therefore concerned the frequency of top cycles.
166 Philosophy and Democracy

For Arrow's theorem, Condition U allows individuals to have any weak

ordering, R{, of preference and indifference, as in (7.1). Calculation is simpler,
however, based on strong orders—that is, individual preference orders, P,, with
indifference not allowed.
With m alternatives, there are ml (i.e., i • 2 • . . . • m) such linear orders
possible; and, when m — 3, these are:
x y z, x z y, y x z, y z x, z x y, z yx

Each such order is a potential D,.. When each of n voters picks some (not nec-
essarily different) D,, a profile, D, is created. Since the first voter picks from ml
orders, the second from ml, . . . , and the last from ml, the number of possible
different profiles, D, is (ml)", which is the number of members of the set, D, of
all profiles, when voters have only strong orders.
A calculation that yields some estimate of the significance of cycles is the
fraction, p(n, m), of D in D without a Condorcet winner:
Number of D without a Condorcet winner
p(n, m) = ——
(ml)"
If one assumes that each D is equally likely to occur (which implies also that, for
each voter, the chance of picking some order is i / ml), then p(n, m) is an a priori
estimate of the probability of the occurrence of a top cycle. Several calculations
have been made, as set forth in display y-i 3 As is apparent from the display, as
the number of voters and alternatives increases, so do the number of profiles
without a Condorcet winner. The calculation thereby implies that instances of
the paradox of voting are very common. Most social choices are made from many
alternatives (though often we do not realize this fact because the number has been
winnowed down by various devices such as primary elections and committees
that select alternatives for agendas) and by many people, so the calculations imply
that Condorcet winners do not exist in almost all decisions.
But, of course, there are a number of reasons to believe that such calculations

DISPLAY 7.1 Values of p(n, m): Proportion of Possible Profiles Without a Condorcet
Winner

n = Number of Voters
m = Number
of Alternatives 3 5 7 9 n Limit

3 .056 .069 .075 .078 .080 .088

4 .111 .139 .150 .156 .160 .176
5 .160 .200 .215 .251
6 .202 .315
Limit 1.000 1.000 1.000 1.000 1.000 1.000

The entry in the row for four alternatives and in the column for seven voters—namely, .150—is the ratio of the
number of profiles without a Condorcet winner to the number of profiles possible when seven voters order four
alternatives.
 Social Choice Theory and Constitutional Democracy 167

are meaningless. People do not choose an ordering with probability i / m\.

Rather, at any particular moment, some orders are more likely to be chosen than
others. The six strong orders over triples generate two cycles:

"Forward Cycle" "Backward Cycle"

1. x y z 4. x z y
2. y z x 5. z y x
3. z x y 6. y x z (7-2)

The entry in the row for four alternatives and in the column for seven
voters—namely, .150—is the ratio of the number of profiles without a Condorcet
winner to the number of profiles possible when seven voters order four alterna-
tives.
Cycles occur when voters concentrate on one or the other of these sets of
three orders. But suppose voters are induced by, for example, political parties, to
concentrate heavily on, say, (i), (2), and (5). Then there is no cycle. Furthermore,
there is good reason to believe that debate and discussion do lead to such fun-
damental similarities of judgment. Calculations based on equiprobable choices
very likely seriously overestimate the frequency of cycles in the natural world.
On the other hand, it is clear that one way to manipulate outcomes is to
generate a cycle. Suppose that in display 7-2 profile D exists and that person 2
realizes that his or her first choice, y, will lose to the Condorcet winner, x. Person
2 can at least prevent that outcome by generating a cycle (or a tie) by voting as
if his or her preference were y z x as in D'.
The tendency toward similarity may thus reduce the number p(n, m), while
the possibility of manipulation may increase the number. It seems to me that
similarity probably reduces the number of profiles without Condorcet winners on
issues that are not very important and that no one has a motive to manipulate,
while the possibility of manipulation increases the number of such profiles on
important issues, where the outcome is worth the time and effort of prospective
losers to generate a top cycle. Neither of these influences appears in the calcu-
lations and thus renders them suspect from two opposite points of view.

DISPLAY 7.2. The Generation of a Cycle

D D'
D,: x yz D[: x y z
D2: yx z D'2: yz x
D3: z x y D'^. z x y

Note. Majoritarian or- Note. Cycle in D' un-

dering of D: x Py P z. der majoritarian vot-
ing: x P Y P z P x.

In D' person 2 has reversed z and x from D, thereby

generating a cycle.
168 Philosophy and Democracy

The Practical Relevance of Arrow's Theorem:

Conditions for Condorcet Winners
Another approach to estimating the practical significance of Arrow's theorem is
to inquire into what kinds of profiles are certain to produce a Condorcet winner.
As in the previous approach, only majoritarian voting is considered, which limits
the relevance of the inquiry to the theorem but does say something about its
practical effect on this kind of decision process. For example, as can be seen in
Display 7-1, for m = n = 3, the number of elements of D = (ml}" = 216 and
p(n, ni) — 12/216 = .056. It is natural to look for the features that guarantee a
Condorcet winner for 204 of the profiles in D. If one can generalize about the
sets of preference orders that produce these results, then it may be possible to
estimate the practical significance of the theorem for majoritarian voting.
To give a simple example: If each voter chooses the same preference order,
D,, then under majoritarian rules the social order for the profile D will be identical
with the chosen D,, and the unique social choice will be the first alternative in
that social order.
The goal of this approach is to identify kinds of preference orders, D,, such
that when the whole profile, D, is composed of such orders, then D will lead by
majoritarian methods to a weak order and a Condorcet winner as a social out-
come.
Even before Arrow's theorem was uttered, Duncan Black observed one such
pattern of orders in D—namely, that the profile can be expressible as a set of
single-peaked curves.4 A preference order can be graphed as in figure 7.1. On the

FIGURE 7.1. A single-peaked curve with the linear order z v y w x

 Social Choice Theory and Constitutional Democracy 169

FIGURE 7.2. Single-peaked curves

vertical axis is measured the degree of preference from lowest at the origin to
highest at the top. On the horizontal axis is placed some ordering of the alter-
natives in X, an ordering appropriately chosen to depict one particular D, as a
single-peaked curve. This is always possible if D, is a strong order (with indiffer-
ence not allowed). The general definition of single-peaked curves (with
indifference permitted at the top) is, as displayed in figure 7.2, reading from left
to right: (i) always downward sloping, (2) always upward sloping, (3) sloping
upward to a particular point and then sloping downward, (4) sloping upward to
a plateau and then sloping downward, (5) horizontal and then downward sloping,
(6) upward sloping and then horizontal. Curves that are not single-peaked are
shown in figure 7.3.
A profile, D, is single-peaked if some ordering of alternatives on the hori-
zontal axis allows every D,, in D to be drawn as a single-peaked curve. As already
observed, for a single D,, it is always possible to find such an ordering. But with
three or more D,, an ordering that renders D;, single-peaked may preclude that
Dk be single-peaked. Indeed, it is exactly when cycles exist that single-peakedness
cannot be attained for D. In Figure 7.4 assume there are three persons who have
chosen different preference orders in the forward cycle (7-2). Then all possible
orderings of X = (x, y, z) on the horizontal axis result in a set of curves that fail
to be single-peaked, as in Figure 7.43-4^ where the axes are all the m\ permu-
tations of {x, y, z}. The same is true of the backward cycle. So to say a profile,
D, is single-peaked is to say it does not admit of cycles. In general, if D is single-
peaked, then:
 FIGURE 7.3. Non-single-peaked curves

FIGURE 7.4. Non-single-peakedness for the forward cycle

 Social Choice Theory and Constitutional Democracy 171

1. If all D, are strong orders and n is odd, the social ordering is strong.
2. If all D, are weak orders, n is odd, and no D, involves complete
indifference over a triple, the social ordering is a weak order.5
So single-peakedness implies transitivity and hence ensures the existence of a
Condorcet winner.
It is furthermore a remarkable fact that, if D is single-peaked and n is odd,
the Condorcet winner is immediately identifiable as the alternative on the hori-
zontal axis beneath the median peak.6 (If n is even, the winner is some alternative
between the n/2th peak and the (n/2) + ith peak, if such an alternative exists.
And, if none exists, the alternatives at these peaks tie.) In Figure 7.5, with five
peaks, the alternative beneath the median peak (3) is identified as xmed. If xmed is
put against some alternative to its left, say x t , then xmed wins because a majority
consisting of voters 3, 4, and 5 prefer xmed to x, (that is, their curves are upward
sloping from x, to xmed). Similarly, xmed can beat any alternative to its right, say
x4, with a majority consisting of voters i, 2, and 3, whose curves are downward
sloping from xmed to x4, which means they prefer xmed to x4. Hence xmed can beat
anything to its right or left and is a Condorcet winner.
Single-peakedness is important because it has an obvious political interpre-
tation. Assuming a single political dimension, the fact that a profile, D, is single-
peaked means the voters have a common view of the political situation, although
they may differ widely on their judgments. Person i may choose D,. = x y z, and
person j may choose Df — z y x; yet they agree that x is at one end of the scale,
z at the other, and y in the middle, which means they agree entirely on how the
political spectrum is arranged. This kind of agreement is precisely 'what is lacking
in a cycle, where voters disagree not only about the merits of alternatives but
even about where alternatives are on the political dimension.

FIGURE 7.5. Single-peaked curves with Condorcet winner

172 Philosophy and Democracy

If, by reason of discussion, debate, civic education, and political socialization,

voters have a common view of the political dimension (as evidenced by single-
peakedness), then a transitive outcome is guaranteed. So if a society is homoge-
neous in this sense, there will typically be Condorcet winners, at least on issues
of minor importance. This fact will not prevent civil war, but it will at least ensure
that the civil war makes sense.
A number of other kinds of restrictions on preference orders, D,, that guar-
antee that D will produce a transitive outcome have been identified. Like single-
peakedness they minimize disagreement over the dimensions of judgment. Con-
sider "value-restrictedness," which is an obvious development from the forward
and backward cycles of (7.2). One property of those cycles (observable by in-
spection) is that each alternative in X appears in first place in some D,, in second
place in another, and in third place in a third. So, if, for strong orders in D,, some
alternative is never first in a D,, or never second, or never last—if, in short, an
alternative is "value-restricted"—then no cycle can occur and transitivity is guar-
anteed.
A number of other such provisions for transitivity have been identified. They
have been exhaustively analyzed by Peter Fishburn.7 They are important because
they indicate that quite a wide variety of rather mild agreement about the issue
dimension guarantees a Condorcet winner. Furthermore, not all voters need dis-
play the agreement to obtain the guarantee. Richard Niemi has shown that the
probabilities of the occurrence of top cycles, by calculations similar to those set
forth in display 7-1, reduce to tiny proportions (e.g., .02 to .04) when as few as
three-fourths of 45 or 95 voters agree on the issue dimension while disagreeing
on orders.8 This result implies that agreement about dimensions probably renders
uncontrived cyclical outcomes quite rare. So I conclude that, because of agree-
ment on an issue dimension, intransitivities only occasionally render decisions by
majoritarian methods meaningless, at least for somewhat homogeneous groups and
at least when the subjects for decision are not politically important. When, on the
other hand, subjects are politically important enough to justify the energy and
expense of contriving cycles, Arrow's result is of great practical significance. It
suggests that, on the very most important subjects, cycles may render social out-
comes meaningless.

The Theoretical Invulnerability of Arrow's Theorem:

Independence
Assuming that the practical significance of Arrow's theorem increases with the
political importance of the subject for decision, it is then reasonable to inquire
whether the theorem is too demanding. Does it overstate the case by stressing
the possibility of intransitivity and its consequent incoherence when perhaps this
is too extreme an interpretation?
To weaken the force of Arrow's theorem, it is necessary to question the
conditions of either fairness or logicality. Most of the fairness conditions seem
intuitively reasonable—at least to people in Western culture—so most of the
 Social Choice Theory and Constitutional Democracy 173

attack has been focused on logicality. One fairness condition, independence, has,
however, often been regarded as too strong.
The independence condition has at least three consequences:
1. It prohibits utilitarian methods of choice.
2. It prohibits arbitrariness in vote-counting, such as lotteries or methods
that work in different ways at different times.
3. It prohibits, when choosing among alternatives in a set 5, which is
included in X, reference to judgments on alternatives in X — S.
It seems to me that one can defend the independence condition for each of
these consequences. As for consequence I, since interpersonal comparisons of
utility have no clear meaning, the prohibition of utilitarian methods seems quite
defensible, although a weaker form of Condition I might accomplish the same
result. With respect to consequence 2, earlier in this chapter it was shown that
arbitrary counting is just as unfair as violations of Conditions U, P, and D. It is
difficult to imagine that any weaker form of Condition I would accomplish what
I does, because the arbitrariness must be prohibited for any set.
Most attention has been given to consequence 3, because many people be-
lieve that judgments on alternatives in X — 5 are germane to judgments on 5
itself.9 In a presidential preference primary, for example, choice among several
candidates may depend on judgments of still other candidates. For example, in
the 1976 Democratic primaries, in thinking about a decision between Carter and
Udall as if they covered the whole spectrum of party ideology, a mildly left-of-
center voter might prefer Udall. But if the voter thought about Jackson also, so
that Udall appeared as an extremist, that same voter might have preferred Carter
to Udall. So "irrelevant" alternatives (here, Jackson) may really be "relevant."
The question is whether there should be some formal way to allow judgments
on the "irrelevant" alternatives to enter into the choice. And the difficulty in
answering is: How can one decide which nonentered candidates are relevant?
Why not allow consideration of still other, even a hundred, irrelevant alternatives?
But if no irrelevant alternatives are considered, then y might beat x\ but with
such consideration, if there is no Condorcet winner, x might beat y. Thus mean-
ing and coherence depend on variability in the voting situations (on the size, that
is, of X and 5) as much as on voters' judgment.
There seems, unfortunately, no wholly defensible method to decide on de-
grees of irrelevance.10 In the absence of such a method, Condition I seems at least
moderately defensible. Furthermore, while some might argue about the desira-
bility of consequence 3, Condition I seems necessary because consequence 2 is
indispensable for fair decision.

The Theoretical Invulnerability of Arrow's Theorem:

Transitivity
If the fairness conditions survive, then the only condition left to attack is transi-
tivity. The sharpest attack is to assert that transitivity is a property of humans, not
of groups. Hence the individual relation, JR.,, should be transitive, but it is simple
174 Philosophy and Democracy

anthropomorphism to ask that the social relation, R, be transitive also.11 Still, there
is some reason to seek transitivity for outcomes.
Without transitivity, there is no order; and without order, there is no coher-
ence. Social outcomes may in fact be meaningless, but one would like to obtain
as much meaning as possible from social decisions. So the obvious question is:
Can one, by modifying the definition of coherence, obtain some lesser coherence
compatible with fairness? Unfortunately, the answer is mainly negative.
The social relation, R, which generates a weak order in Arrow's logicality
condition, combines social preference, P, and social indifference, /. And R is useful
for the purpose Arrow had in mind—namely, social judgments involving com-
parisons and ordering of all feasible social policies, such as distributions of income.
Suppose, however, that one does not require quite so general a result. For pur-
poses of making a social choice, one does not need to impose a complete order
on the whole set X merely to find a best alternative in X. We can think of a best
element in X as one that is chosen over or tied with every other alternative.12
The best alternative is then the choice from X or C(X).13
A requirement, weaker than transitivity, that nevertheless ensures the exis-
tence of one best alternative is quasi-transitivity—that is, the transitivity of P, but
not of R or /. This means that, if x P y and y P z, then x P z; but if the antecedent
does not hold (e.g., if x I y), then the consequent need not hold either. For
example, quasi-transitivity allows (as in note 12) y P z, z I x, and x I y, which
is clearly intransitive in both R and /, although it is enough to establish that the
choice from X = (x, y, z) is C(X) = (x, y).
Another, even weaker requirement for a choice, is acydicity, which is the
requirement that alternatives in X can be arranged so that there is no cycle.14 It
turns out that, by using the logical requirement of acyclicity rather than transi-
tivity, it is possible to find social choice that satisfies all of Arrow's fairness con-
ditions as well as the revised condition of logicality. A. K. Sen offers an example
of such a method: For a set X = (a, b, . . . , ) , let a be chosen for C(X) over b if
everybody prefers a to b and let a and b both be chosen if not everybody prefers
a to b or b to a.15 This rule satisfies Condition U because all individual orders are
allowed. It satisfies Condition P because it is based on the principle of unanimity.
It satisfies Condition I because the choice between any pair depends only on
individual preferences on that pair, and it satisfies Condition D because the only
way a can be better socially than b is for everyone to prefer a to b. Finally, it is
always acyclic. So even if one cannot guarantee an order with fair procedure, it
appears that one can at least guarantee a best choice.
Unfortunately, however, something very much like dictatorship is required
to guarantee quasi-transitivity or acyclicity. Quasi-transitive social outcomes can
be guaranteed only if there is an oligarchy.16 (An oligarchy is a subset of choosers
who, if the members agree, can impose a choice, or, if they do not agree, enables
all members individually to veto the choice.) If one modifies Condition D from
no dictator to no vetoer, then even a quasi-transitive social outcome cannot be
guaranteed.17 As for acyclicity, Donald Brown has shown that acyclic choice re-
quires a "collegium" such that alternative a is chosen over b if and only if the
whole collegium and some other persons prefer a to b. Thus, although a collegium
 Social Choice Theory and Constitutional Democracy 175

cannot unilaterally impose a choice, unlike an oligarchy it can always at least

veto.18
Furthermore, if one strengthens Arrow's conditions just a little bit by re-
quiring not just the monotonicity that enters into Condition P, but a condition
of positive responsiveness (Condition PR), then quasi-transitivity again involves
dictatorship. (Monotonicity requires merely that, if a voter raises her or his val-
uation of an alternative, the social valuation does not go down. In contrast, pos-
itive responsiveness requires that, if a voter raises her or his valuation, society does
so as well, if that is possible.) It is then the case that any quasi-transitive social
result that satisfies Conditions U, P, I, and PR must violate Condition D if there
are three alternatives; and, furthermore, someone must have a veto if there are
four or more alternatives.19
Weakening transitivity into some logical condition that requires only a social
choice but not a full ordering does not gain very much. This brief survey indicates
there is a. family of possibility theorems of which Arrow's theorem is a special case.
And in the whole family there is still some kind of serious conflict between
conditions of fairness and a condition of logicality. In general, the only effective
way to guarantee consistency in social outcomes is to require some kind of con-
centration of power in society—a dictator, an oligarchy, or a collegium. So fairness
and social rationality seem jointly impossible, which implies that fairness and
meaning in the content of social decisions are sometimes incompatible.

The Theoretical Invulnerability of Arrow's Theorem:

Conditions on Social Choice
Of course, one can abandon entirely the effort to guarantee some kind of ordering
for social "rationality," whether it be transitivity or merely acyclicity. One can
simply provide that a social choice is made and impose no kind of ordering
condition. The reason, however, that transitivity or even less restrictive ordering
conditions are attractive is that they often forestall manipulation by some partic-
ipants either of agenda or of sets of alternatives to obtain outcomes advantageous
to the manipulator. As Arrow remarked at the conclusion of the revised edition
of Social Choice and Individual Values, "the importance of the transitivity condition"
involves "the independence of the final choice from the path to it."20 "Transitiv-
ity," he said, "will ensure this independence," thereby ensuring also that the pref-
erences of the participants (rather than the form of or manipulation of the social
choice mechanism) determine the outcome. He went on to point out that both
Robert Dahl and I had described ways in which intransitive social mechanisms
had produced "unsatisfactory" results. So Arrow concluded that "collective ra-
tionality" was not merely an "illegitimate" anthropomorphism, "but an important
attribute of a genuine democratic system." Consequently, if one gives up on social
transitivity or some weaker form of ordering, one is in effect abandoning the
effort to ensure socially satisfactory outcomes.
To ensure satisfactory outcomes without imposing an anthropomorphic col-
lective rationality, one might impose consistency conditions on the social choice
mechanism—conditions that could have the same effect of forestalling manipu-
176 Philosophy and Democracy

lation that transitivity does, but that would not attribute to society the ability to
order possessed only by persons. Hopefully, one would thereby avoid all the prob-
lems of the possibility theorems put forth by Arrow and his successors. Unfor-
tunately, however, it turns out that these consistency conditions also cannot be
satisfied by social choice mechanisms that satisfy the fairness conditions. Conse-
quently, although the problem can be elegantly restated in terms of choice rather
than ordering, the main defect of the methods of amalgamation is unaffected by
the new language. Just to say, for example, that x Pl y and x P2 y lead to C(x, y)
= x rather than to say that they lead to x P y does not solve the problem of
amalgamation. Some kind of inconsistency is ineradicable.
Consistency requirements on choice have been discussed in two quite dif-
ferent ways, which, however, turn out to be substantially equivalent in this con-
text. I will discuss both ways here, despite their equivalence, because their verbal
rationales are complementary.
A. K. Sen and subsequently many others have imposed on social choice con-
ditions of logicality that were originally devised as standards for individual choice
behavior. This procedure has the advantage of relating consistency in groups to
consistency in persons, but it is subject to the same charge of anthropomorphism
that was leveled against the use of ordering conditions. Charles Plott, however,
has devised a consistency condition for social choice itself, one that could not
easily be applied to persons but captures the spirit of Arrow's insistence that the
final choice ought to be independent of the path to it. It is interesting and re-
markable that Sen's and Plott's conditions turn out to be closely related and almost
equivalent.21
Looking first at Sen's conditions, let 5 and T be sets of alternatives in X =
(x^, x2, . . . ,xm) and let S be a subset of T. Sen's conditions are restrictions on the
choice sets from these two sets of alternatives, C(S) and C(T):
1. Property OC: For sets 5 and T, with 5 a subset of T, if x is in both
C(T) and 5, then x is in C(S).
2. Property j3 +: For sets 5 and T, with 5 a subset of T, if x is in C(5)
and y is in S, then, if y is in C(T), so also is x in C(T).
The meaning of these conditions is easily explained: Property OC requires that, if
the choice from the larger set is in the smaller set, then it is in the choice from
the smaller set as well.
To see the rationale of OC, consider a violation of it: A diner chooses among
three items on a menu, beef (B), chicken (Q, and fish (F), which are the set {B,
C, F}. The diner chooses beef (B); then the restaurant runs out offish (F). The
new menu is the set {B, C}, whereupon the diner chooses chicken (C) in vio-
lation both of property OC and of apparent good sense.22
Property OC guarantees consistency in choices as the number of alternatives
is contracted because in going from T to 5 the choice does not change if it is in
both sets. Property (3+, on the other hand, guarantees consistency in choices as
the number of alternatives is expanded. It requires that, if any element in the
smaller set is the choice from the larger set, then all choices from the smaller set
are choices from the larger set. Thus, in going from S to T, if any choices
 Social Choice Theory and Constitutional Democracy 177

from S continue to be chosen from the larger set, all such choices continue to
be chosen.
The rationale of (3+ can be appreciated from a violation of it: For a seminar
with students S = (a, b, c, d), a teacher ranks d best. Then another student en-
rolls making T = (a, b, c, d, e), whereupon the teacher ranks c best. Doubtless
student d discerns an inconsistency and believes that if he is the best or among
the best in 5 and if some other member of S is best in T, then he (d) ought to
be among the best in T also.
As I have already noted, property OC and property (3+ apply as well to in-
dividuals as to society. Plott, however, attempted to embody Arrow's notion of
"independence of the final result from the path to it" directly in a condition on
social choice. Plott justified his condition, which, appropriately, he called "path
independence," thus:

[Tjhe process of choosing, from a dynamic point of view, frequently proceeds in

a type of "divide and conquer" manner. The alternatives are "split up" into
smaller sets, a choice is made over each of these sets, the chosen elements are
collected, and then a choice is made from them. Path independence, in this case,
would mean that the final result would be independent of the way the alternatives
were initially divided up for consideration.23

The definition of path independence is that, for any pair of sets 5 and T, the choice
from the union of the sets is the same as the choice from the union of the separate
choices from each set.24 Manifestly, if S and T are any ways of breaking up the
set of alternatives, X, then to equate the choices from their union with the choice
from the union of their choice sets is to say that it makes no difference to the
final outcome how X is divided up for choosing.
Path independence (PI) can be broken up into two parts—PI* and *PI:

1. PI* is the condition that the choice from the union of 5 and T be
included in or equivalent to the choice from the union of their choice
sets.
2. *PI is the converse of PI*. Specifically, *PI is the condition that the
choice from the union of the choice sets of S and T be included in
or equivalent to the choice from the union of S and T.25

It is a remarkable and important fact that PI* is exactly equivalent to property

OC.26 Furthermore, a choice function satisfying property (3+ satisfies *PI, so that,
though not equivalent, *PI is implied by (3+.27
These standards of consistency in choice turn out to be quite similar in effect
to ordering principles.28 Although property a does not guarantee transitivity, it
does guarantee acyclicity in choices from X. So also, therefore, do PI and PI*.
Consequently, social choice methods satisfying these conditions are dictatorial or
oligarchic, just as are those satisfying ordering principles.
On the other hand, property (3+ does not guarantee even acyclicity when
choices from Xare made in a series ofpairwise comparisons. Consequently, meth-
ods satisfying [3+ and *PI do not imply dictatorship or oligarchy or any other
kind of concentration of power. If one is willing to give up consistency in con-
i y8 Philosophy and Democracy

tracting alternatives—and this is quite a bit to give up—then reliance on simple

consistency in expanding alternatives might be a way around all the difficulties
discovered by Arrow. Unfortunately, however, methods of choice satisfying (3+
and *PI violate another fairness condition—namely, unanimity or Pareto opti-
mality.29
Suppose a choice is to be made by three people with these preference orders:
(i) x y z w, (2) y z w x, (3) z w x y. This leads to a cycle in simple majority rule,
xPyPzPwPx, so that the choice set is all the alternatives: C(w, x, y, z) =
(w, x, y, z). But everyone prefers z to w, although there is a path by which w
can be chosen. Let 5, = (y, z) and C(S^) = y; S2 = (x, y) and C(S2) = x; 53 =
(x, w) and C(S3) = w. Using 5, at step i, 52 at step 2, and 53 at step 3, w is
selected even though z, eliminated at step i, is unanimously preferred to w. This
result is generalized by Ferejohn and Grether.30 It tells us that, even if we rely
solely on an expansion consistency condition and thus avoid concentrations of
power, we still do not achieve fairness. So, in a quite different way, we are back
where we began. Nothing has been gained except an elegant formalism that
avoids anthropomorphizing society.

The Absence of Meaning

The main thrust of Arrow's theorem and all the associated literature is that there
is an unresolvable tension between logicality and fairness. To guarantee an or-
dering or a consistent path, independent choice requires that there be some sort
of concentration of power (dictators, oligarchies, or collegia of vetoers) in sharp
conflict with democratic ideals. Even the weakest sort of consistency ((3+ or *P7)
involves a conflict with unanimity, which is also an elementary condition of
fairness.
These conflicts have been investigated in great detail, especially in the last
decade; but no adequate resolution of the tension has been discovered, and it
appears quite unlikely that any will be. The unavoidable inference is, therefore,
that, so long as a society preserves democratic institutions, its members can expect
that some of their social choices will be unordered or inconsistent. And when
this is true, no meaningful choice can be made. If y is in fact chosen—given the
mechanism of choice and the profile of individual valuations—then to say that x
is best or right or more desired is probably false. But it would also be equally
false to say that y is best or right or most desired. And in that sense, the choice
lacks meaning. The consequence of this defect will be explored in what follows.

The Rejection of Populism

The social amalgamations of individual values are, for reasons just summarized,
often inadequate—indeed meaningless—interpretations of public opinion. Fur-
thermore, we seldom know whether to assign any particular social choice to the
adequate category. Hence all choices can be suspected of inadequacy. What does
this conclusion imply for two views of voting in democracy—Can either popu-
lism or liberalism stand up?
 Social Choice Theory and Constitutional Democracy 179

Clearly populism cannot survive. The essence of populism is this pair of

propositions:
1. What the people, as a corporate entity, want ought to be social policy.
2. The people are free when their wishes are law.
Since social choice theory is a device to analyze moral (and descriptive) propo-
sitions, not an ethical theory to choose among them, social choice theory cannot
illuminate the question, implied in proposition i, of what ought to be public
policy. But, if the notion of the popular will is itself unclear, then what the people
want cannot be social policy simply because we do not and cannot know what
the people want. An election tells us at most which alternative wins; it does not
tell us that the winner would also have been chosen over another feasible alter-
native that might itself have a better claim to be the social choice. Hence falls
proposition i. And if we do not know the people's wishes, then we cannot make
them free by enacting their wishes. Thus falls proposition 2.
Populism as a moral imperative depends on the existence of a popular will
discovered by voting. But if voting does not discover or reveal a will, then the
moral imperative evaporates because there is nothing to be commanded. If the
people speak in meaningless tongues, they cannot utter the law that makes them
free. Populism fails, therefore, not because it is morally wrong, but merely because
it is empty.
In the history of political ideas, a similar rejection of a moral imperative on
the ground that it was uninterpretable occurred in the sixteenth and seventeenth
centuries when religious directives on politics lost their presumed clarity. So long
as the spiritual authority of the pope was unquestioned, he could state the political
content of moral and divine law. With the success of the Reformation, however,
there were many conflicting voices speaking for God, no one of which was more
clearly vested with divine quality than another. Thus, even though no one seri-
ously questioned the existence of the Divinity or the authority of divine direction
of human politics, the direction nevertheless failed simply because no one could
be sure what the Divinity said. Modern secular political thought begins with that
uncertainty. Similarly I believe that in the next generation populist claims will be
rejected simply because it will be realized that, however desirable they might be,
they are based on a flawed technique that renders populism unworkable.
Perhaps it will be said in defense of populism that, although there is always
doubt about the meaning of any particular electoral outcome, a series of elections
over time establishes a rough guide to policy. Yet even that defense cannot be
sustained. It is possible that alternative x (say, some political platform) repeatedly
beats alternative y (another platform) so that one is fairly certain that x has a good
majority over y. But suppose x wins only because z was eliminated earlier or was
suppressed by the Constitution or by the method of counting or by manipulation.
What then is the status of xt If x is as precise as a motion, then one can still be
fairly sure that x at least beats y. But if x is as vague as an ideology, it is far from
certain that a clear decision is ever made.
In the history of American presidential elections, three long-enduring clusters
of ideas have repeatedly been more or less endorsed by the voters. From 1800 to
i8o Philosophy and Democracy

1856, the agrarian expansionism of Jefferson and Jackson won most of the time;
indeed it was clearly beaten only in 1840. From 1860 to 1928, the Republican
program of commercial development won most of the time; it was clearly beaten
(in popular vote, but not in the election itself) only in 1876. From 1932 to the
present, Democratic welfare statism won most of the time; it was clearly beaten
only in 1952, 1956, 1972, and 1980.31 Can it not be said that these repeatedly
endorsed clusters of ideas have been, though rather vague, the true and revealed
popular will? In a very narrow sense they have indeed been approved because x
(one of these three clusters) has beaten some y (not always the same y) a number
of times, frequently in two-party contests. But x has never won over all relevant
alternatives. No party program has ever been approved in general.
The Jefferson-Jackson program of agrarian expansion was approved two-
thirds of the time from 1800 to 1856; but it failed to obtain a majority whenever
it was effectively shown that agrarian expansion entailed approval of slavery: in
1840, 1844, 1848, 1856, and 1860. The election of 1852 and its aftermath are
instructive. In 1852, Whigs, themselves in office, did not raise the slavery issue;
the Free Soil party was weak; and the Democratic party, again united on its
traditional economic platform, won by a clear majority. Democratic leaders, as-
suming they had a mandate, then produced the Kansas-Nebraska Act (1854),
which, though ostensibly mere agrarian expansion and free soil at that, actually
promoted the expansion and approval of slavery. Thereafter, and directly as a
consequence, the Democratic party did not get a national majority for more than
20 years. Though this party with its policy of approval of slavery for the sake of
expansion was apparently endorsed overwhelmingly in 1852, only a slight varia-
tion in its program resulted in savage repudiation in the next five elections. The
point is clear: x (agrarian expansion and slavery) was approved against y (com-
mercial development of Federalism and Whiggery) but lost by a huge margin to
z (commercial development and free soil, ultimately the Republican combina-
tion) .
In the next period, between 1860 and 1928, the Republican platform of
commercial development usually triumphed against a dying agrarianism. But
again, there was no general mandate for commercialism alone. Initially it suc-
ceeded (except in 1876, when, however, technically it did win) because it was
associated with free soil; in its heyday it succeeded because, I believe, its com-
mercial ideal was clearly linked to social welfare. When the welfare aspects were
forgotten in an excess of commercialism (especially around the i88os, when the
old Democratic combination of agrarianism and repression of blacks was not yet
fully rejected), the Republican party failed to maintain its majority: in 1876, 1880,
1884, 1888, 1892, 1912, and 1916. Notice that even the interlude of Wilson's
presidency was occasioned by an internal Republican split that presumably re-
flected Theodore Roosevelt's distaste for Taft's pure emphasis on commerce. Thus
the mandate often thought to be overwhelming was at best conditional on an
appeal to interest wider than commerce alone.
Finally, in the third period, from 1932 to the present, welfare statism appeared
to receive a huge mandate under Franklin Roosevelt and thereafter as long as
welfare rather than statism was emphasized. But whenever statism dominated, as
 Social Choice Theory and Constitutional Democracy 181

in the wars undertaken by Truman and Johnson—wars involving huge statist

compulsion of citizens in unpopular causes—then welfare statism lost to a com-
bination of commerce and civil liberties as expressed in Eisenhower's mitigation
of the military draft and Nixon's elimination of it. As the statist component of
welfare statism becomes increasingly apparent in other ways besides military ad-
ventures, it may well be that welfare statism as a whole will be rejected. The
victory of Ronald Reagan in 1980 may well be the beginning of a more general
rejection of welfare statism.
In all three periods, then, a dominant program has been dominant only when
it has been tested in one dimension. When more than one dimension has been
salient to voters, majorities have disappeared, even when the second dimension
is closely related to the first. Voters rejected agrarian expansion when its slavery
component was emphasized. They rejected commercial development when mere
commerce was emphasized at the expense of development, and they rejected
welfare statism when the statist feature of militarism exceeded the welfare com-
ponent of redistribution. Throughout all three periods, one cluster of ideas was
repeatedly endorsed; but there was no clear Condorcet winner, and indeed there
were probably always covert, unrevealed cycles of popular values. The inference
is clear: The popular will is defined only as long as the issue dimensions are
restricted. Once issue dimensions multiply, the popular will is irresolute. Slight
changes in dimensions induce disequilibrium. Thus it is indeed difficult to speak
of a popular will so narrowly construed, and that is why populism is an empty
interpretation of voting and why the populist ideal is literally unattainable.32

The Survival of Liberalism and Democracy

Given that social choice theory reveals populism to be inconsistent and absurd,
how does the liberal interpretation of voting fare? For democracy, this is a crucial
question. Since populism and liberalism, as I have defined them, exhaust the
possibilities and since populism must be rejected, then, if liberalism cannot survive,
democracy is indefensible. Fortunately, liberalism survives, although in a curious
and convoluted way. Liberalism does not demand much from voting, and hence
the restrictions placed on the justification of voting by social choice theory do
not quite render the liberal ideal unattainable.
The essence of the liberal interpretation of voting is the notion that voting
permits the rejection of candidates or officials who have offended so many voters
that they cannot win an election. This is, of course, a negative ideal. It does not
require that voting produce a clear, consistent, meaningful statement of the pop-
ular will. It requires only that voting produce a decisive result: that this official
or this party is retained in office or rejected. This very restricted expectation about
voting can, I believe, easily coexist with all the defects we have observed in the
voting mechanism. If so, then liberalism survives.
Let us investigate systematically the relation between liberalism and the dis-
coveries of social choice theory. To do so, I will begin by assuming the existence
of what we already know does not exist—namely, a fair and accurate amalga-
mation of voters' values. That this assumption is unrealistic does not taint the
182 Philosophy and Democracy

analysis because I will use it only as an initial standard, not as an instrument of

interpretation. With this initial assumption, let us then consider the following
cases

1. Suppose an official or candidate has not offended enough voters for

them to reject him or her in a fair and true amalgamation of their
values. If, in the actual voting under any particular method, the of-
ficial or candidate is not rejected, then the method is working ade-
quately in this case.
2. Suppose, however, that the actual voting does lead to the rejection
of an unoffending official or candidate. Such rejection might occur
in many ways. A Condorcet winner (assume him or her to be a "true
winner") might lose in a plurality or approval or Borda election, or
clever opponents might beat him or her by strategic voting or by
manipulation of the agenda or by the introduction of additional, di-
visive candidates. If this happens, has the ideal of liberalism been
violated? In our populistic era, even many liberals might say yes.
Madison himself would have been troubled by this case. But if the
liberal ideal is strictly interpreted, it has not been violated. First and
foremost liberalism requires the rejection of the offending, not the
retention of the unoffending. If a system admits rejection at all, then
there is no a priori reason why it may not sometimes work imper-
fectly. One should expect, therefore, that some times an official or
candidate will be rejected "wrongly."
This means that we must modify to some degree our expectations
from liberalism. Liberalism has been said to mitigate the oppression
caused by the failure of officials to act as agents of voters' participa-
tion. It is clear now, however, that the expectation of such an agency
is a populist fantasy, not a tenet of liberalism (which is a wholly neg-
ative kind of control). This means that an official in a liberal regime
may indeed abandon any effort to ensure voters' participation through
reading the voters' will, not because the official is sophisticated
enough to know that there is nothing there to read, but merely be-
cause he or she knows by experience that voters' rejection may be
random. Such randomness, however, does not really matter for the
liberal hope of preventing an official's abuse of office and authority.
The threat of the next election retains its force. Indeed, an official
who faces an electorate knowing that it sometimes works randomly
and may "unfairly" reject him or her has a powerful motive to try
even harder to avoid offending voters.
3. Suppose, on the other hand, that an official or candidate has—in our
imaginary true and fair amalgamation—offended enough voters to be
rejected. If he or she is in fact rejected, then, clearly, the voting
method is working adequately.
4. But what if the offending official or candidate wins? This might hap-
pen because he or she successfully manipulates the agenda or invents
 Social Choice Theory and Constitutional Democracy 183

additional issues or sets up spurious opponents. Has the ideal of lib-

eralism then been violated? Again, most liberals in this populist era
would say yes. I imagine that Madison would have believed this case
impossible. But still, if the liberal ideal is strictly interpreted, there is
no violation. Liberalism requires only that it be possible to reject a
putatively offending official, not that the rejection actually occur. We
know from social choice theory that those who should be winners
in our imaginary true and fair amalgamation can be defeated. We also
know that those who should be losers can also be defeated, as, of
course, they should be. Consequently, the voting system does not
prevent the rejection of offenders—and that is precisely the condition
liberalism requires. Of course, it may happen that an uninformed or
unsophisticated or well-manipulated electorate fails to operate the
voting system as its members would wish. But the fact that in partic-
ular instances people fail to make the system work well does not alter
the fact that they can make it work. And if success is even sometimes
possible, then the liberal interpretation can be sustained.33

It seems worthwhile to point out just how little is contained in the liberal
interpretation of voting. In some sense it is very close to the cynical view that
counting heads is better than breaking heads to solve the problem of succession.
Let me point out some of the things that the liberal interpretation is not, simply
to show how easy it is for liberalism to survive the criticisms of social choice
theory. For one thing, liberalism does not require that society itself act. In the
liberal interpretation, society is an anthropomorphized entity that cannot order or
choose anything either consistently or inconsistently. Rather it is thought that
individual people in the society choose, and what they individually choose is
whether to support or oppose candidates. The social amalgamation of these
choices need not be fair or just. It may even be part of a social cycle. But if it
results in a decision on candidates, it is, from the liberal point of view, adequate.
Since social decisions are not, in liberal theory, required to mean anything,
liberals can cheerfully acknowledge that elections do not necessarily or even usu-
ally reveal popular will. All elections do or have to do is to permit people to get
rid of rulers. The people who do this do not themselves need to have a coherent
will. They can be—and often are—strange bedfellows. Voters on the far right and
the far left, for example, can combine to throw out a ruler in the center. The
liberal purpose is then accomplished, even though one could not make a coherent
ideological statement about what these voters did and even though their majority
might be cyclical. The Indian rejection of Indira Gandhi in 1997 was accom-
plished by just such a coalition, the Janata party, which was, however, so inco-
herent in its combination of right and left that Mrs. Gandhi was able to win again
in 1979.
The liberal interpretation of voting thus allows elections to be useful and
significant even in the presence of cycles, manipulation, and other kinds of "er-
rors" in voting. Since it is precisely cycles, manipulation, and "error" that render
populism meaningless, the fact that liberalism can tolerate them demonstrates that
184 Philosophy and Democracy

liberalism can survive the revelations of social choice theory, while populism
cannot.
The kind of democracy that thus survives is not, however, popular rule, but
rather an intermittent, sometimes random, even perverse, popular veto. Social
choice theory forces us to recognize that the people cannot rule as a corporate
body in the way that populists suppose. Instead, officials rule, and they do not
represent some indefinable popular will. Hence they can easily be tyrants, either
in their own names or in the name of some putative imaginary majority. Liberal
democracy is simply the veto by which it is sometimes possible to restrain official
tyranny.
This may seem a minimal sort of democracy, especially in comparison with
the grandiose (though intellectually absurd) claims of populism. Still, modest
though liberal democracy may be, it fully satisfies the definition of democracy I
favor. To begin with, it necessarily involves popular participation. Though the
participation is not the abstract self-direction of a corporate people, it is the con-
crete self-direction of individuals who vote and organize voting to make the
democratic veto work. Furthermore, since the veto does exist—even when ma-
nipulated or cyclical—it has at least the potential of preventing tyranny and ren-
dering officials responsive.
Since officials are not responsive to some imaginary popular will, this popular
participation is not the act of making policy. At best officials are responsive to a
(possibly random) threat of expulsion from office. But this may lead them to avoid
gross offense to groups of citizens who can eject them from office. Participation
in this sense is then the act of placing a curb on policy, a veto at the margin.
Nevertheless it is participation. Furthermore, it can engender that self-direction
and self-respect that democracy is supposed to provide because candidates, trying
to construct winning platforms in the face of that potential veto, also try to
generate majorities, at least momentary ones.
Furthermore, the liberal veto generates freedom because of the very fact that
it is a curb on tyranny. Whether one thinks of freedom as the absence of restraint
("negative liberty" in Isaiah Berlin's terms) or as the ability to direct one's own
life ("positive liberty"), it is apparent that oppression by rulers eliminates either
kind of freedom. Suppose freedom is simply the absence of governmental restraint
on individual action. Then the chance to engage in vetoing by rejecting officials
and the chance that the rejection actually occur are the very essence of this
freedom, which is substantially equivalent to liberal democracy. Suppose, how-
ever, freedom is defined as the ability to use government to work one's will—
the populist expectation. The agent of this (imaginary) will is government—that
is, rulers who can oppress both the minority and the very majority whose will
they are supposed to work. An extreme example is socialist rulers who, in order
to free workers from the supposed bondage to owners, subject them to the own-
ership of the state and the terrorism of the police. But conventional populists
turned dictator (for example, Vargas, Gandhi, and Peron) are often just as op-
pressive. Liberal democracy, insofar as it allows people to restrain and reject, is
the main sanction against this majoritarian oppression also.
Sometimes populists argue that the true meaning of democracy is not to be
 Social Choice Theory and Constitutional Democracy 185

found in voting, party organization, and the like, but rather in the democratic
ideals of civil liberties, tolerance, and humane concern for popular rights. No one
doubts that these ideals are indeed truly central in democratic thought (although
I suspect that efforts to distinguish them from voting are often the first step—a
la Marx—to some kind of coercive enterprise). Still, these democratic ideals de-
pend on a vigilant citizenry. What permits a citizenry to be vigilant is the liberal
method of regular elections. In that sense liberal democracy is a necessary part of
what populists claim to want.
In the same way, liberal democracy promotes a kind of equality. Equal chances
to restrain, to reject, and to veto inhere in the very idea of using votes to control
officials. This is the notion of the equality of the right of democratic participation,
which is the essence of the idea of juristic equality. Equality has many other
derived meanings, including notions such as equal shares of the national treasure
and equality before the law (that is, in the courtroom). But equality in these
derived senses cannot occur unless juristic equality exists. In that way, juristic
equality is primary, and that is what liberal democracy provides.
So the liberal interpretation of voting, however much it admits of "unfair"
voting methods, manipulation, cycles, and the like, still contains the essential
elements of democracy. It may, from the populist view, be a minimal kind of
democracy; but this is the only kind of democracy actually attainable. It is the
kind of democracy we still have in the United States; and it is the kind of de-
mocracy so much admired by those who live in closed societies.

Are Liberal and Populist Interpretations Compatible?

To say that the populist interpretation of voting cannot survive criticism, but the
liberal interpretation can, does not necessarily imply that the two interpretations
are incompatible. It may be that if voting permits the rejection of officials (the
liberal goal), then from the liberal point of view it makes no difference whether
people attempt, however fruitlessly, to use voting to embody the supposed popular
will in law (the populist goal). Since the populist method is not valid, it will
usually fail. But it may sometimes succeed. And if its failure or success are irrel-
evant by liberal standards, then there is no necessary incompatibility between
liberalism and populism.
The case can be put thus: Let liberal rejectability or LR stand for "citizens
are able to reject some rulers," and let populist incorporability or PI stand for
"citizens are able to embody popular will in law." Then, if we guarantee at least
LR and sometimes also get PI, we have perhaps obtained an extra benefit. This
happy result can occur, however, only if LR and PI are complementary. If they
are inconsistent, if, for example, PI is equivalent to the negative of LR, then
populism is certainly incompatible with liberalism. It is important to discover,
then, whether LR and PI can be simultaneously affirmed.
In the view of liberalism set forth by Berlin populism transforms liberalism
into tyranny when a coercive oligarchy claims to enforce an imaginary popular
will. But that transformation assumes an oligarchy that refuses to submit to the
liberal discipline of the next election. Suppose, however, that free and regular
186 Philosophy and Democracy

elections are preserved. Then, even though the populist ideal may be unattainable,
populism need not destroy freedom. So the question really is: Can liberal insti-
tutions (that is, the ability to reject rulers) be maintained when the populist in-
terpretation of voting and populist institutions are adopted?
To answer, I will first define briefly the institutions appropriate for each
interpretation of voting.

Populist Institutions. The populist ideal requires that rulers move swiftly and
surely to embody in law the popular decision on an electoral platform. Consti-
tutional restraints that retard this process are populistically intolerable. The appro-
priate institutions are those that facilitate speedy embodiment, and the simplest
such institutions are so-called constitutional dictatorships, by which I mean dic-
tatorial executives who submit to real elections or plebiscites but rule by decree
or through a complaisant legislature. Latin American governments often have this
form. Mexico is perhaps an example, though, since Mexican presidents never
stand for reelection, the one-party plebiscites may not be real elections.
More complex populist institutions are those that develop out of parliamen-
tary governments when they degenerate into a sovereign self-regenerating legis-
lature run by a leader (who is also the executive) of a disciplined majority party.
For the legislature to be sovereign, there cannot be external restraints like an
externally selected executive and independent courts; nor can there be internal
restraints such as multiple houses with different constituencies. For the legislature
to be disciplined members of the majority party dare not defy the party leader.
Furthermore, at least one party must be a majority alone. With such arrangements
the party leader can then ensure that party platforms are quickly adopted in true
populist fashion. Finally, since the legislature is self-regenerating, this leader can
manipulate the time and conditions of elections to facilitate his or her own con-
tinuance in office.
The closest current approximation to this ideal is Great Britain as it has
developed over the last generation. External checks on the House of Commons
have been removed. The House of Lords was effectively countered by the Par-
liament Act of 1911, and the last attempt at ever personal freedom in the Crown
was repulsed at the abdication of Edward VIII in 1937. Strong third parties, a
characteristic feature of nineteenth century Parliaments (for example, the Radicals,
Irish, and Labour), were eliminated with the decline of Liberalism after the 19205,
though they may be reappearing in the new Liberalism or Social Democracy. The
national leaders of the two main parties acquired control of nominations so that
the members became disciplined. And finally the national leadership of the ma-
jority party was lodged in the prime minister, who, once in office, could usually
control all factions of his or her party and, within wide limits, ensure his or her
continuance in office. What has so far saved the British system from constitutional
dictatorship is, I believe, a three-century-old liberal tradition of free and regular
elections. But even that tradition now seems threatened. The ideal of new elec-
tions when the government loses in the House of Commons has degenerated
into the government's selection of a propitious time within the limits of a five-
year term. (And that term has twice been doubled with the excuse of war.) More
 Social Choice Theory and Constitutional Democracy 187

significant evidence of the populist elimination of electoral restraint, however, is

the confiscatory and truly oppressive taxation and inflation by which rulers have
financed their reelection, thereby acting against the society that they are supposed
to serve.

Liberal institutions. Liberalism, as here defined, simply requires regular elec-

tions that sometimes lead to the rejection of rulers. It is often thought that lib-
eralism also involves additional constitutional restraints, and it is indeed historically
true that liberal regimes have always had them. This association, however, may
be no more than a historical accident, owing to the fact that liberal democracy
developed out of the imposition of constitutional restraints on monarchies. Per-
haps, in the abstract, liberal methods do not need to be supplemented with these
restraints because liberalism has only one stipulated sanction on rulers—namely,
the threat of the next election. Nevertheless, in practice, liberal democracy prob-
ably does not work without the additional restraints always heretofore associated
with it—multicameral legislatures, decentralized parties, and so on.
Having defined populist and liberal institutions, it is now possible to inves-
tigate the main question: Is the liberal interpretation of voting compatible with
populist institutions? The answer depends on whether rulers in a populist system
can be expected to maintain the electoral arrangements essential for liberal de-
mocracy.
It is difficult, I believe, for rulers of any kind to maintain free elections. The
function of an election is to put at stake the rulers' jobs, even their lives. In
nations where the democratic tradition is fragile or externally imposed, it is com-
mon for rulers who have won office in elections to prohibit elections they might
lose. Even adherents of the democratic ideal hesitate to hold elections when they
fear the electorate may do something "foolish" or "wrong." Dictators from Crom-
well to Peron, who have possible genuinely regarded themselves as caretakers until
democracy can be resumed, have always been reluctant to trust an electorate that
might install an alternative demagogue.
In general, therefore, the electoral system is only precariously maintained.
But in populist systems both the temptation and the ability to weaken the electoral
sanction are especially strong. For one thing, with a populist interpretation of
voting it is easy for rulers to believe their programs are the "true" will of the
people and hence more precious than the constitution and free elections. Populism
reinforces the normal arrogance of rulers with a built-in justification for tyranny,
the contemporary version of the divine right of rulers.
The main threat to democracy from populism is not, however, the excep-
tional temptation to subvert elections but the exceptional ability to do so. Populist
institutions depend on the elimination of constitutional restraints, and the populist
interpretation of voting justifies this elimination. With the restraints removed, it
is easy to change electoral arrangements, which is why populist democracies so
often revert to autocracies. Perhaps the leaders of some future populism will be
so thoroughly imbued with liberal ideals that they will never meddle with free
elections. But since even in Britain, where liberal ideals originated, the populist
elimination of constitutional limitations has begun to produce attacks on the in-
188 Philosophy and Democracy

tegrity of elections, it seems unlikely that the liberal sanction can survive populist
institutions. Indeed this empirical regularity suggests to me that there is a profound
theoretical reason that populism induces rulers to ensconce themselves in office.
At any rate, on the practical level at least, the answer is clearly negative to the
main question of this section: Is liberal rejectability compatible with populist in-
corporability? No: because the constitutional restraints practically associated with
liberalism must be destroyed to achieve populism.

The Preservation of Liberal Democracy

It seems clear to me that democracy cannot be preserved simply with the liberal
interpretation of voting. Suppose one had a liberal regime without the constitu-
tional limitations typically associated with liberalism. Would the regime long re-
main liberal? Would it not operate for all the world as if it were populist? Would
it not move certainly toward some kind of oppression by rulers? I think the
answers to these questions are yes and so did Madison. Although he argued that
the necessary and sufficient condition for republicanism (that is, democracy) to
exist was simply regular, popular elections, it is still true that he did not think this
condition was sufficient to preserve the liberal system. Instead he argued the ne-
cessity of constitutional limitations. And the constitution he was justifying in The
Federalist contained a variety of very real restraints that together have successfully
kept rulers from subverting regular, popular elections for 200 years. These re-
straints have the effect of preventing any single ruler or any single party from
getting enough power to subvert:

1. A multicameral legislature (really three "houses": the president, Senate,

and House of Representatives) based on different divisions of the
people into constituencies. The different constituencies have typically
kept the interests of rulers separate and thus forestalled the fusion of
their ability to rule into a tyranny.
2. A division between legislative and executive authority. Though constitu-
tional interpreters from Montesquieu on have believed this a funda-
mental limitation, in the American system it is no more than an ex-
tension of the differing constituencies of the multicameral legislature.
3. A division of authority between national and local governments. This is the
famous American federalism, copied over half of the world. The con-
stitutional restraint is not, however, the legal division of duties be-
tween central and local governments but rather the resultant locali-
zation of political parties that renders national leadership of them
impossible.
4. An independent judiciary. American lawyers have typically believed this
to be the main limitation. Indeed, the separation of the judiciary from
the rest of government does render blocking possible. But the judi-
ciary has no independent constituency and so always loses in a crisis.
Hence, this limitation is much less important than the multi-cameral
legislature or the decentralized parties.
 Social Choice Theory and Constitutional Democracy 189

5. Limited tenure and regular elections. This is the fundamental restraint, but
it is not a self-enforcing limitation. It depends rather on the force of
tradition and on the other restraints previously listed.

About the only important restraint commonly found in other constitutional de-
mocracies and not included in this one is a system of more than two political
parties so that no single party is ever a majority by itself. Something of that
restraint, however, is provided in the American system by decentralized parties.
Madison was correct, I believe, in his assertion that constitutional limits pre-
serve liberal democracy. Fortunately the limits he helped provide retain most of
their original force. Yet in the American political tradition there has always been
a strong strand of populism, usually expressed as the notion that the winners of
an election ought to be able immediately to embody their platform in law and
policy. This was, for example, the basis for the attack by Jacksonians on bureau-
cratic tenure, for the attack by populist political philosophers (such as Charles
Beard and J. Allen Smith) on constitutional limitations of all sorts, for the per-
sistent advocacy of a rigid system of two disciplined political parties (thus allowing
a "majority" immediately to enact its program), and finally in recent decades for
the idealization of "presidential leadership"—a euphemism for transcending con-
stitutional limitations by the domination of a populistically endorsed, quasi-
monarchical president.
Along with the populist notion of an unfettered agent (whether party or
president) of the popular will, there is also the notion that the popular will can
express itself directly, as in legislation by referenda and even by public opinion
polls. The device of referendum was developed in the progressive era—an epoch
of the populist spirit—to provide a "truer" expression of the popular will than
statutes produced by legislatures, which were, it was argued, merely distorting
intermediaries of the popular will. Since that time there has been considerable
disillusionment with referenda because they have produced both inconsistent and
bizarre legislation. Still, the populist belief in direct democracy dies hard, even
though it can in no wise escape the defects of manipulation. So now there is
considerable enthusiasm for using cable television to conduct elections on the
content of statutes and even of administrative policy. Presumably citizens will
listen to debate and then vote by push button, supplanting thereby the need for
any kind of legislature.
Many current proposals for institutional reform are populistically intended
to nullify constitutional limitations. In spite of disasters with the "imperial"
presidencies of Johnson and Nixon, people continue to search for ways to en-
hance presidential leadership—by tightening presidential control of the bureauc-
racy and by elaborating presidential influence in the party. The decentralization
of political parties, the fundamental restraint in federalism, is under constant at-
tack with proposals for frequent policy-making conventions to be dominated by
national leaders and proposals for centralized national financing of campaigns.
While at the moment proposals to eliminate legislatures with direct law-making
over cable television or to substitute (a la Marcus Raskin) instructions from
grand juries for legislative judgment seem bizarre, I have no doubt that, as
ipo Philosophy and Democracy

technology and opportunity combine, such populist proposals will be taken se-
riously.
The present situation in the United States is, therefore, that, although the
fundamental constitutional limitations remain, populists persistently seek to un-
dermine them. Since the twentieth century is a populist era worldwide, our
homegrown populists may well succeed.
Populism puts democracy at risk. Democracy requires control of rulers by
electoral sanctions; the spirit of populism and populist institutions allows rulers to
tamper with this sanction, thereby rendering it a weak defense against the tyranny
of officials. The maintenance of democracy requires therefore the minimization
of the risk in populism.
How can we minimize the risk? This is the great question of political pru-
dence forced on us by the revelations of social choice theory. I will conclude this
survey with some remarks on this practical problem.
The main defense against populist excesses is the maintenance of the consti-
tutional limitations inherited from eighteenth-century Whiggery. It would prob-
ably help also to have a citizenry aware of the emptiness of the populist inter-
pretation of voting. And surely a wide dissemination of the discoveries of social
choice theory is a desirable additional defense. But the dissemination of a rather
arcane theory is a task for generations. (It took me a score of years of reflection
on Black's and Arrow's discoveries to reject the populism I had initially espoused.)
Consequently, the fundamental method to preserve liberty is to preserve ardently
our traditional constitutional restraints—decentralized parties and multicameral
government.
Almost everyone who has written about American politics has emphasized
its peculiar style. Except for the one great disaster of the Civil War, the nastiest
features of political scarcity have seldom flourished. One group has not persistently
tried to do in another, probably because political coalitions are always shifting and
constitutional restraints make it difficult to organize zero-sum situations. Conse-
quently, there are no elaborately rationalized ideologies that carry with them a
logically clear and intricately arranged set of public policies. Instead, political
leaders are almost always engaged in constructing petty and pragmatic compro-
mises for marginal adjustments in policy. Only in a few great crises have great
leaders emerged, and great leaders have always disappeared as the crises have
subsided.
The reason for this style is, I believe, the existence of constitutional limita-
tions. In his famous defense in The Federalist, No. 51, of the notion of the sepa-
ration of powers, Madison remarked:

A dependence on the people [by which he meant democracy, or regular elections

and limited tenure of office] is, no doubt, the primary control on the govern-
ment; but experience has taught mankind the necessity of auxilliary precautions
[by which he meant precisely the constitutional restraints here enumerated].

The experience of the subsequent two centuries has reinforced the teaching to
which Madison referred. And the reason is this: Liberal democracy almost guar-
antees some circulation of leadership so that great power is usually fleeting and
 Social Choice Theory and Constitutional Democracy 191

no vested interest lasts forever. The constitutional restraints have always reinforced
this style. Multicameralism and federalism have enforced localism in parties, and
this in turn has forced rulers to persuade rather than to control. The total effect
is that policy does not change either rapidly or sharply enough to hurt anyone
very badly, which is why we have usually—except in the Civil War—avoided
the worst features of political scarcity. That the system has worked in this moderate
way is due partly to luck; but it also is due to that happy mixture of liberal
democracy and constitutional restraints that has so far preserved us from the ha-
treds and oppression implicit in populism.

Notes
1. William H. Riker, "Voting and the Summation of Preferences," American Polical
Science Review, Vol. 55 (December 1961), pp. 900-911.
2. Since many proofs are easily available to those who wish to follow up the subject,
I will not reiterate the proof here. For Arrow's proof, as revised by Julian Blau, see Kenneth
Arrow, Social Choice and Individual Values, 2nd ed. (New Haven: Yale University Press,
1963), pp. 96-100. A refined form of Arrow's proof is to be found in Amartya K. Sen,
Collective Choice and Social Welfare (San Francisco: Holden-Day, 1970), chap. 3. See also
Peter C. Fishburn, The Theory of Social Choice (Princeton: Princeton University Press, 1973),
p. 206. Bengt Hansson, "The Existence of Group Preference Functions," Public Choice, Vol.
28 (Winter 1976), pp. 89-98, contains a topological proof. An informal sketch of Arrow's
proof is contained in William H. Riker and Peter C. Ordeshook, An Introduction to Positive
Political Theory (Englewood Cliffs, N.J.: Prentice-Hall, 1973), pp. 92-94.
3. See Duncan Black, The Theory of Committees and Elections (Cambridge: Cambridge
University Press, 1958), pp. 50-51, where the calculation was first proposed. Calculations
are set forth in Richard Niemi and Herbert Weisberg, "A Mathematical Solution for the
Probability of the Paradox of Voting," Behavioral Science, Vol. 13 (July 1968), pp. 317-323;
Mark Gorman and Morton Kamien, "The Paradox of Voting: Probability Calculations,"
Behavioral Science, Vol. 13 (July 1968), pp. 306-316; Frank DeMeyer and Charles Plott, "The
Probability of a Cyclical Majority," Econometrica, Vol. 38 (March 1970), pp. 345—354. See
also R. M. May, "Some Mathematical Remarks on the Paradox of Voting," Behavioral
Science, Vol. 16 (March 1971), pp. 143-151; Peter C. Fishburn and William V Gehrlein,
"An Analysis of Simple Two-Stage Voting Systems," Behavioral Science, Vol. 21 (January
1976), pp. 1-12; Fishburn and Gehrlein, "An Analysis of Voting Procedure with Nonranked
Voting," Behavioral Science, Vol. 22 (May 1977), pp. 178-185; Gehrlein and Fishburn, "The
Probability of the Paradox of Voting," Journal of Economic Theory, Vol. 13 (August 1976),
pp. 14-25; and the essays cited in note 8 of this chapter.
4. Duncan Black, "On the Rationale of Group Decision Making," Journal of Political
Economy, Vol. 56 (February 1948), pp. 23-34; Black, The Theory of Committees and Elections,
chaps. 4 and 5.
5. Fishburn, The Theory of Social Choice, p. 105.
6. Black, The Theory of Committees and Elections, chap. 4.
7. Fishburn, The Theory of Social Choice, pp. 111-144. See also Sen, pp. 166-186. A
somewhat different kind of condition for transitivity (not involving restrictions on D,) has
been identified by Rubin Saposnik, "On Transitivity of the Social Preference Relation
Under Simple Majority Rule," Journal of Economic Theory, Vol. 10 (January 1975), pp. i—
7, where it is shown that, if the number of voters in D having orders constituting the