Public Choice 44:7-41 (1984). 1984 Martinus N i j h o f f Publishers, Dordrecht. Printed in the Netherlands.

The pure theory of large two-candidate elections

JOHN O. LEDYARD*

INTRODUCTION A long-standing goal of political theorists has been the development of a coherent, consistent, and nonvacuous theory of elections, particularly of those using majority rule. on rational individual If possible, such a theory is to be based In spite of extensive and group behavior.

effort, recent writings (see, for example, Ordeshook and Shepsle, 1982) reveal that many may now be prepared to give up this research program on the grounds that no such model exists. of participants: (1) the theoretical and (2) There appear to be two main proposition that, given any stumbling blocks to a consistent theory based on the rational behavior realistic assumption about the cost of voting, rational voters will not participate in elections, even i f they vote, majority-rule equilibria rarely exist. The First result is obviously contradicted by

*Northwestern University. The research for this paper has been supported by NSF grant SES8106896 which is gratefully acknowledged. The paper itself was prepared for presentation at the Conference on Political Economy at Carnegie-Mellon University in June 1983. Without that incentive it might still be unwritten. I have had the help and advice of many people--some of whom disagree with the conclusions. Among these were the participants in workshops at Cal Tech, lowa, Tulane, Indiana, Stanford, and Northwestern University. The exacting standards of Howard Rosenthal are responsible for much of whatever quality in presentation exists. I thank him for this help and advice. Peter Coughlin found early lapses. An early version was presented at the Public Choice Society meetings, as my presidential address, in 1981.

the

facts;

the

second means that

the

theory as we know i t

is

fundamentally flawed. directions. "equilibrium" Kramer, 1977.)

Faced with these results, those who have not

given up on political theory altogether have gone in two other obvious They have either given up on "rational" behavior (see, for models and have turned to "process" models (for example, In particular, I example, Hinich et al., 1972; Coughlin, 1979), or they have given up on

I t is my belief that this retreat is premature.

intend to show in this paper that even under assumptions of extremely rational behavior, i t is possible to combine voters, who may or may not vote depending on the benefits and costs, with candidates who game which not only exist but against one another and end with equilibria

Figure I

RATIONAL C OC H IE

PROBABILISTIC VOTING

NO ABSTENTIONS

TRADITIONAL

C U HI O G LN

A S E TO S BTNI N

THIS P P R AE

HINICH, L D A D EYR & O DS O K R EH O

also

have a remarkable

social-welfare

property.

The approach

is

straightforward of elections. voters vote

extension of the now standard spatial competition model Voters have preferences ( u t i l i t y (a point functions) over issues, two-candidate ein the issue space), and then candidate--in

candidates choose a platform for their lections--if the costs. winning). and only i f

most preferred

the expected benefits from so doing outweigh

Given t h i s voter behavior, candidates are assumed to maxiA full general equilibrium occurs when no voter or candidate into the l i t e r a -

mize expected p l u r a l i t y (a very good approximation of the p r o b a b i l i t y of wishes to a l t e r his strategy. To show where the theory posed in this paper f i t s ture on the theory of majority-rule elections, Figure i in which existing theories are divided ditional theory i t is assumed that all I refer the reader to into four "boxes" deIn the t r a and that I t is due to

pending upon the assumptions concerning voting behavior. choice behavior is rational both of these behavioral alienation, while exist (some form of u t i l i t y rarely e x i s t . hypotheses by allowing

vote (no abstentions)

maximization). abstentions

this theory for which e q u i l i b r i a indifference, probabilistic Although

Hinich et al. changed

etc. and by modeling the decision to vote as in this not modification, voting behavior is

leaving the choice of candidate based on u t i l i t y . rational-choice-based. theory. Coughlin

equilibria

somewhat ad hoc and certainly maintained the t r a d i t i o n a l

assumption of no abstentions but removed the Instead, he adopted to u t i l i of Luce (1959, 1977) by assuming that are proportional e x i s t and have no abstentions are

voter's choice of candidate from rational the decision-theoretic-framework choice is p r o b a b i l i s t i c , ty. It is

where p r o b a b i l i t i e s

With this model of voter behavior, e q u i l i b r i a not known what occurs

interesting welfare property albeit d i f f e r e n t allowed. I The model in t h i s paper departs from the

from that in this paper.

in Coughlin's models i f

traditional

by allowing

IStop Press: I have recently seen Coughlin (~983b) in which abstentions but only on an aggregate basis. Individual behavior remains unspecified.

are allowed

10 rational abstention behavior. 2 W begin by recognizing the obvious fact e that when voters make the decision to vote, they do not know how many others have voted, or plan to vote, or, especially, how the others have voted. They face a decision -- or better, a game -- under uncertainty In modeling this simultanesimilar in s p i r i t to a sealed-bid auction.

ous decision problem for a l l voters we impose as much r a t i o n a l i t y as possible -- rational choice and rational expectations - - and arrive at a model in which turnout is neither the 100% nor the 0 that has tra% d i t i o n a l l y been implied by rational-choice models. I t is this model of the voters' behavior which constitutes the "new" component of the theory in this paper. Most of the rest of our model is standard, although the implications derived from this combination of new and old are not. In Section I we describe the behavior of a single voter in much the same way as that posed by Downs (1957), Tullock (1967), and others. Section I I Section I l l we consider the simultaneous behavior of all present the equilibrium concept First introduced in Ledyard (1981). In In voters and

we define and describe both the behavior of candidates and In Section IV we explore the

the equilibrium which arises when a l l actors -- candidates and voters - are combined into a general equilibrium. welfare properties of those e q u i l i b r i a , in Section V we examine the existence of equilibrium, and concluding remarks are added in Section VI.

I.

THE VOTER

The voter is assumed to choose whether to vote or abstain, as well as for whom to vote, consistent with the expected u t i l i t y hypothesis. This model has already received much attention in the l i t e r a t u r e so I w i l l not dwell on i t s rationale but w i l l immediately turn to the notation and definitions. The interested reader can consult Ferejohn and Fiorina (1974) for a good survey.

21 am finally answering the complaints of $1utsky (1975) about the ad hoc and unrealistic nature of our earlier paper (Hinlch e t a l , , 1972). In spite of my efforts he remains unconvinced of the "reality" of the model.

ii We assume now that there are only two candidates, A and B. platform denoted by B. choices and has a u t i l i t y this voter. Candi-

date A chooses a platform which we denote by A and candidate B chooses a We assume that the voter knows the candidates' function over a l l possible platforms, R, given parameters f o r in R. We that u is continuous

by u(R, x) where x represents the appropriate u t i l i t y We assume throughout he w i l l sometimes call to the p o l l s , > u(B, x). voting. x the "type" of t h i s voter.

I f t h i s voter decides to go

cast his vote f o r A over B i f and only i f u(A, x) vote instead of abstaining The expected benefits

We assume the voter receives no consumption benefit from benefit-cost calculation.

Therefore, whether t h i s voter w i l l

depends on a simple

from voting are equal to the p r o b a b i l i t y of a f f e c t i n g the outcome times the gain from so doing. l a r voter w i l l ence is divided Letting P be the p r o b a b i l i t y that t h i s p a r t i c u The u t i l i t y differhe a l t e r the outcome, and assuming that u(A, x) > u(B, x), by 2, since a voter affects the outcome only i f

the expected benefits are (P)(u(A, x) - u(B, x ) ) / 2 . creates a t i e or breaks one. by 2. > 0

Assuming that t i e s are broken by a f a i r d i f f e r e n c e divided linearly. he we

coin toss, the gain from e i t h e r event is the u t i l i t y and that if this cost enters the utility

We assume that the voter faces a known cost of voting equal to c calculation candidate A wins and the voter had gone to the p o l l , model of r a t i o n a l - v o t i n g behavior,

Therefore,

receives u(A, x) - c in u t i l i t y . In order to complete t h i s is the p r o b a b i l i t y one vote. must provide a basis f o r the v o t e r ' s b e l i e f s about Pa and Pb' where Pj that candidate j e i t h e r t i e s the other or loses by We assume, at t h i s point, that the voter knows the p r o b a b i l i (We w i l l see in the next section how these denoted, r e s p e c t i v e l y , I t is it is a standard exercise to

ty that a voter, randomly selected from a l l other voters, w i l l vote f o r A, vote f o r B, or abstain. can be estimated.) Using these p r o b a b i l i t i e s ,

Qa, Qb, and Qo, where Qa + Qb + Qo = i , also easy to calculate the p r o b a b i l i t y

calculate the p r o b a b i l i t y of a t i e when there are n other voters. Adding these we find that Pa = f(Qa' Qb) where f ( z , y) =

that A loses to B by one vote.

zk=O k!k!n-2k! zkyk(1-z-Y)

It,ll

n!

n-2k

li~-~ II n! +s k=O k + l ! k ! n - 2 k - l ! z k y k + l ( l - z - y ) n - 2 k - I

12 A symmetric calculation yields Pb = f(Qb' Qa)" Gathering this all

together we have described the voter. A voter with characteristics (x, c) who is faced with a choice between two candidates, A and B, and who thinks the probability that a randomly selected voter will vote for candidate j is Qj will (a) (b) (c) v o t e for A i f c < (Pa/2)(u(A, x) - u(B, x)), v o t e for B is c < (Pb/2)(u(B, x) - u(A, x)), abstain otherwise,

where Pa = f(Qa' Qb)' Pb = f(Qb' Qa) and f is defined above. This model assumes rational behavior in the form of expected u t i l i ty maximization, no income effects, no candidate specific preferences other than the platform choice, positive costs of voting, and knowledge by the voter of x, c, A, B, Pa' and Pb" At this point most writers reach an unsettling conclusion. "Since the expected benefit from voting is obviously small ( i f Qa = Qb and Qo = 0 then Pa and Pb are of order of magnitude i/2n -- see Chamberlain and Rothschild, 1981), and since the cost of voting is not small, no rational voter will ever vote in large elections. be wrong with the theory." Therefore, something must This is not an unreasonable conclusion but If this voter and others are W turn to that task next. e

the analysis is incomplete, since i t is based on a partial equilibrium view which i s simply not appropriate. al choice to explain voting disappears. embedded in a general equilibrium model, the apparent failure of ration-

II.

PJ~TIONALVOTERS' EQUILIBRIUM

W now explore what happens when voters take into account the fact that e other voters are also rational. The logic is simple and compelling and If everyone is rational is contained in Ferejohn and Fiorina (1974). then, presumably, no one will vote. is 1.

and carries out the partial equilibrium calculus in the previous section But then the probability of a tie If this is true and i f these same rational, partial equilibrium

nonvoters redo their calculus, most will find that i t is now definitely in their interest to vote; they will be able to determine the outcome by

13 themselves. of a t i e voters. And so on. Somewhere between no one voting and everyone in which some vote and in which the p r o b a b i l i t y rational, intermediate situation that we it

voting l i e s a s i t u a t i o n It is t h i s

is consistent with those numbers and with the b e l i e f s of a l l stable,

capture in the voters' equilibrium defined below. To close the p a r t i a l it equilibrium model in the previous section, remains only to specify how a voter estimates Qa and Qb" therefore, that each follows the model of Section I. to each voter, and never w i l l teristics g(x). (x, c). is characteristics probability and h(c) known to We assume that and,

is common knowledge among a l l voters that each voter is rational

What is not known

be, are the values of the others' characall by the density functions h(c) and x,

We do, however, assume that the d i s t r i b u t i o n of these

That is, c and x are indepe,~dently d i s t r i b u t e d , where g(x) is the that a randomly selected voter w i l l have c h a r a c t e r i s t i c is the p r o b a b i l i t y that a randomly selected voter w i l l these densities, one can compute and only i f the probability have a that a

cost of voting equal to c. 3 Given randomly selected voter w i l l the voter w i l l satisfies c < (Pa/2)(u(A, x) - u(B, x ) ) . vote f o r a candidate. We already know that (x, c),

vote for A i f

her c h a r a c t e r i s t i c ,

Using the densities g and h we can compute that the p r o b a b i l i t y of t h i s is 4

qa : fX+(A,B)H((Pa/2)(u(A'x) - u ( B , x ) ) ) g ( x ) d x where X+(A,B) : {x]u(A,x) > u(B,x)} and H(r) = f~ h(c)dc. Writing t h i s

3The assumption of independence is made only f o r e x p o s i t i o n a l convenience. The eager reader can e a s i l y show t h a t c o r r e l a t i o n between c and x in a d e n s i t y f u n c t i o n l i k e g ( x , c) can be accommodated without d e s t r o y i n g the r e s u l t s d e t a i l e d below. 4Note t h a t i f A = B, then Qa = 0 since U(A, x) = U(B, x ) , and H(O) = O.

14 as Qa = t(Pa' A, B; g, h), i t is easy to show that Qb = t(Pb, B, A; g, h) and Qo = I - Qa - Qb" W can thus compute Qa and Qb from Pa and Pb" e section we computed Pa and Pb from Qa and Qb" In the previous

A fully rational voter

with f u l l y rational expectations will require these calculations to be consistent with one another and will be able to compute the values of Q and P for which consistency obtains. Given the densities on characteristics, h and g, and given the candidate platforms A and B, we call <Pa,Pb,Qa,Qb> a RATIONAL VOTERS'

EQUILIBRIUM i f and only i f

Pa : f(Qa'Qb) and Pb = f(Qb'Qa) and Qa = t(Pa'A'B;g'h) and Qb = t(Pb'B'A;g'h)' where f( , ) is define~ in Section I and t(r,s,w;g,h) is defined above. As an aside the reader should note that i f we were to model the voters as playing a game of incomplete information, as is done in modeling auctions, the three pure strategies would be vote A, vote B, I chose the approach above

and abstain, and the Bayes equilibria of that game would be exactly the Rational Voters' Equilibrium defined above. for its expositional simplicity. To complete this section, we consider several properties of the rational voters' equilibrium. PROPOSITION 1: (EXISTENCE). PROOF: If A ~ If H(c)~C (that is, i f H is continuous),

then a rational voters' equilibrium exists. If A = B, then Qa = Qb = O, Pa = Pb = 1 is an equilibrium. B, t h e n define the functions Pa = NI(Pa,Pb) = It

f(t(Pa,A,B),t(Pb,B,A)) and Pb = N2(Pa,Pb) = f(t(Pb'B'A)'t(Pa 'A,B))"

is easy to show that NI and N2 are continuous in (Pa,Pb) since f is polynomial and therefore continuous, while t is continuous in P since H is by assumption. Further, NI and N2 map {0,1]x[0,i] Therefore, Brouwer's fixed-point theorem can be applied. least one pair P* = (Pa*,Pb*) such that P* = N(P*). t(Pa*,A,B ) and Qb* = t(Pb*'B'A)" into i t s e l f . There is at Let Qa* =

Then (P*,Q*) is a rational voters'

15 equilibrium. Q.E.D. PROPOSITION 2: (SYMNETRY). (Pa,Pb,Qa,Qb) is a rational voters' equilibrium given (A,B) i f and only i f equilibrium given (B,A). PROOF: Immediate. Q.E.D. This is the f i r s t of several propositions concerning the symmetry of the model in this paper. The primary reason for symmetry is that we have assumed that voters care about the platform which candidates adopt and not the name of the candidate. The next property is of interest for i t s implications about the voting probabilities in equilibrium. PROPOSITION 3: PROOF: Pa In any rational voters' equilibrium, (Pa - Pb) = (Qb Pb = f(Qa,Qb) - f(Qb,Qa) = (Pb,Pa,Qb,Qa) is a rational voters'

Qa)F where F > = O.

ll-n-~IJ n! xkyk(l_x_y)n-2k-1 (Qb-Qa) Z k=O k!k+1!n-2k-l! Q.E.D. I t should be noted for completeness that F = 0 i f and only i f the number of voters is even and Qa + Qb = 1, ( i . e . , turnout is 100%). Another interesting property of equilibrium is uniqueness, or lack thereof. W have two propositions to present, both of which depend on e the turnout probabilities. Definition: (Maximum Turnout Probability). Given the

candidates platforms, A and B, and given the distribution of voters' characteristics, we can compute an upper limit on turnout which is independent of the particular voter equilibrium arrived at. In particular, let

M(A,B,g,h) = ~H((I/Z)IIu(A,x) - u(B,x)U)g(x)dx. W call M( ) the maximum turnout probability. e

16 W have defined M( ) this way since M is the probability that a e randomly selected voter w i l l go to the polls i f he thinks the probability of a t i e is i. To see this, remember that

Qa+ Qb = fX+(A,B)H((Pa/2)(u(A'x)

- u(B,x)))g(x)dx

+ fX+(B,A)H((Pb/2)(u(B,x ) - u(A,x)))g(x)dx.

Let Pa = Pb = I.

The observation follows immediately, ~ O.

since H is a

distribution function and H'

M( ) = 0 i f no rational voter will go to the polls even when the probability of influencing the election is 1. H(O) = 0 and let A = B. For an example, assume

PROPOSITION 4: (UNIQUENESS 1).

I f M(A,B,g,h) = 0 then (1,1,0,0) is the

unique rational voters" equilibrium. PROOF: Under the hypothesis, Qa = Qb = 0 for all values of Pa and Pb since H(c) ~ H(c') whenever c > c ' . But i f Qa and Qb are 0 i t follows Q.E.D. I t would be helpful i f we were also able to exhibit a proposition l i s t i n g sufficient conditions for the uniqueness of the voter e q u i l i b r i um when the maximum turnout probability is positive. have not yet discovered such a result. Unfortunately, I I t is true, however, that i f the that Pa = Pb = i.

candidates' platforms are close enough, then M is near 0 and the equilibrium will be both unique and continuous in (A,B).

PROPOSITION 5: (UNIQUENESS2).
all x, and H(c)~C1 for all c. for example, i f A is near B),

Suppose M(A,B,g,h) > O, u( ,x)~CI for I f M(A,B,g,h) is near O, (which is true, the equilibrium (Pa,Pb,Qa,Qb) is unique

17 and is C1 in A and B (for A ~ B). 5 Let Qa(Pa) = t(Pa,A,B) and Qb(Pb) : t(Pb,B,A ). equilibrium i f and only i f P solves
PROOF:

(P,Q) is an

Pa

f(Qa(Pa),Qb(Pb)) : 0 and The Jacobian of this system of

Pb - f(Qb(Pb)'Qa(Pa )) = O. equations is

I
i

fl(Qa,Qb)Q~ f2(Qb,Qa)Qa

f2(Qa,Qb)Qb i - fl(Qb,Qa)Q~

Q~ = BQa/BPa : ~fx+hi(Pa/2){u(A,x)

- u(B,x)}l g(x) dx/BPa

= fx+h[(Pa/2){U(A,x)

- U(B,x)}] g(x) {U(A,x) - U(B,x)} () dx

Since this integral is taken over X its value is positive. +, for Q~. FromLedyard (1981) we know that

Similarly

fl(X,y)= (y)

ZIIIIk + l ! k -n!l ! n - 2 k - l !
k=l

k-lyk(l_x_y)n-2k-i

k=l

n! k!k+l!n-2k!

kyk(l_x_y)n-2k-1

_ n(l_x_y) n-I

5Qa and Qb may have discontinuous derivatives at A : B, not{ng this in an earlier version.

I thank Peter Coughlin for

18 and

f2(x,y) : (x-y) Z

k-1

n! xk-lyk-l(l_x_y)n-2k. k!k-l!n-2k!

From these i t can be seen that the signs of f i are: fl(Qa,Qb) i f Qa>Qb i f Qa=Qb i f Qa<Qb Since f l ' ? f2(Qa,Qb) + 0

f1(Qb,Qa) ?
-

f2(Qb,Qa)

f2 are continuous i f Qa is near Qb then f1(Qa,Qb) < 0 and

fl(Qb,Qa) < O. Therefore, for Qa>Qb J = :

for Q a:Qb

~ 0

for Qa<Qb

Now we know that any solution must satisfy 0 ~ Qa = Qb ~ M(A,B), and Qa,Qb ~ O. Therefore, IIQ - Qbll ~ M(A,B) and i f M(A,B) is small a enough, Qa is always near Qb" J is positive definite for all such Qa,Qb. Therefore, the equilibrium is unique. (Gale-Nikaido, 1965) Continuity follows from the Implicit Function Theorem Q.E.D. To summarize, i f the maximum turnout probability is small enough (or i f the candidates' platforms are close enough), the voter e q u i l i b r i um is unique and CI in the platforms. I do not know how close is "enough." I f A is not near B, i t seems that multiple e q u i l i b r i a may be

19 possible. A final bility comment seems in order about the amount of turnout preW have seen that i f the maximum turnout probae Since we dicted by this model.

is 0 or i f A = B, then turnout is predicted to be O,

have adopted a rational behavior hypothesis, one might suspect that, in fact, turnout is never positive. PROPOSITION 6: Such a suspicion would be false. If the maximum turnout

(POSITIVE EXPECTED TURNOUT).

probability is positive, given A and B, then expected turnout is positive in any rational voters' equilibrium. PROOF: Rememberthat expected turnout is (n + I)(Q a + Qb)" Suppose that Qa + Qb = O. Pb = 1. It Then Qa = Qb = O. But f(O,O) = i. 'Therefore, Pa = follows that expected turnout is then (n + I)M(A,B) > 0 Q.E.D. Corollary: I f there is a set of x, with positive measure, such that be positive in equilibrium. M( ) u(A,x) - u(b,x) # 0 and i f H(c) > 0 when c > 0 ( i . e . , h(c) > 0 for all c _ = 0), then expected turnout will > gives an upper-bound on expected turnout. Thus, contrary to naive expectations based on partial equilibrium analysis, a rational general equilibrium consideration of voting behavior yields positive turnout in equilibrium unless each voter refuses to vote even when he knows he is the only voter.

which is a contradiction.

III.

THE CANDIDATEAND THE ELECTION EQUILIBRIUM

In this section, we model how candidates determine their platforms and, therefore, the outcome of the election. motivates the candidates. have been assuming that platforms w i l l W begin by considering what e be implemented and that the it Since this is a static model and since we

extent of implementation does not depend on the margin of victory, about winning.

seems reasonable to assume that these candidates care, ex post, only The appropriate outcome space then is simply the twoThe simultaneous choice of platforms by candidate A w i l l choose point set [W,L} = [win,lose]. and the rational,

the candidates determines a probability distribution (Ra,Rb) on this set expected utility-maximizing

20 the platform to maximize Ra*V(W + Rb*V(L). ) Thus, this candidate will W assume that e

always choose to maximize the probability of winning. as i f they know i t . outcome function, is possible

these candidates know the model of the previous section, or at least act From that model they can determine an election(A,B) into sets of 4-tuples (Pa,Pb,Qa,Qb). It or, more properly, an outcome correspondence, which implications of their

maps pairs of platforms

for candidates to compute various

choices such as the probability of winning. The probability A wins Given A,B, h(c), and g(x), and a rational voters' equilibrium of a twocandidate election, the probability that A wins is:

[~] zn-2k+l n+l! (QA)k+r(QB)k(I_QA_QB)n-2k-r+I Ra = Zk=O r=l k+r! k! n-2k-r+1!

[ n+.__~l1
+ k20 J k! k~+~-2k+l (QA)k(QB)k(I-QA-QB)n-2k+l I

where Qa(A,B) and Qb(A,B) are the appropriate parts of a rational voter equilibrium for A,B. Although the analysis can be carried out using i t , this is a remarkably unwieldly function. To simplify, let us use an approximation (See Hinich, 1977.) If of R which is appropriate for large elections. a place of R a. i

n is large, then Qa-Qb is a good approximation for a candidate to use in To see this, let Si = 1 i f voter i votes for A, Si = 0 i f Then A wins i f and only i f abstains, and Si = -I i f i votes for B.

n+l Z Si > O. 11

This is true i f and only i f

( l l ( n + I))

n+l Z S. : S > O.
i= ~ 1

Since the Si are independently and

21 identically distributed, i t follows from a Law of Large Numbers that 1 lim Pr (S > O) : n + ~ 0 i f Qa > Qb i f Qa < Qb the probability

1/2 i f Qa = Qb

Therefore, maximizing Qa - Qb maximizes (in the limit) that A wins. the candidates.

Using this approximation, we posit the following model of

An alternative j u s t i f i c a t i o n for the use of expected p l u r a l i t y , QA - QB' in place of R is that the equilibrium described below is the same a in both cases because there are two Candidates and symmetry. Basically, we can show that R a will same equilibrium. ~ 1/2 i f and only i f Qa ~ Qb" Since equilibria occur only at R = 1/2 or Qa = Qb, the two objectives produce the a W adopt Qa - Qb for simplicity. e

The candidates' objectives

In a large, two-candidate election, each candidate will try to maximize expected p l u r a l i t y . A is In particular, the objective function of candidate

W(A,B) = Qa(A,B) - Qb(A,B) and that of candidate B is V(A,B) = Qb(A,B) - Qa(A,B)

where (Pa,Pb,Qa,Qb) is a Rational Voters Equilibrium for the platform choices A,B. 6 tential not be unique The observant reader w i l l a n d , therefore, the have already noticed a povoters equilibrium may be a d i f f i c u l t y with this model--a rational

mapping W(A,B) may not

6The Rational Voters t Equilibrium is defined in Section II.

22 function. unique, W do have to confront this problem, but i f W( ) and V( ) are e the above objective functions would point instantly to the

appropriate behavior for the candidates in their choice of a platform, since this is a two-person zero-sum game for which game theorists are in agreement about a solution concept. W adopt the consensus solution e concept with modifications because of the nonuniqueness. The candidates' behavior In a two-candidate (A*,B*) which satisfy min W(A*,B*) ~ max W(A,B*) min V(A*,B*) ~ max V(A*,B) for all A for all B large election, candidates will choose platforms

where W(A,B) = [w lIQa - Qb = w for some rational voters' equilibrium}, and V(A,B) = [viIQb - Qa = v for some rational voters' equilibrium]. call (A*,B*) a (STRONG) RATIONAL ELECTION EQUILIBRIUM. I f W( ) and V( ) are single-valued, this definition corresponds to the noncooperative equilibrium (or maximin solution) of this game. Due to the modification, we have called this a strong equilibrium since, i f the candidates choose these strategies, then even i f candidate A could choose from the multiple set W( ) after changing her strategy, she could do no better than now. Weaker equilibria may also exist since a riskaverse candidate might choose to play a strategy A*, even though max W(A,B*) > min W(A*,B*) for another A, in order to avoid a possible loss if min W(A,B*) < max W(A*,B*). I have chosen the stronger version, since a more strategic candidate would notice that even i f such a loss occurred he could regain at least a payoff of 0 by choosing A = B*. Thus, no outcome which yields survive as an equilibrium. each player receives O. PROPOSITION 7: (VALUE). If (A*,B*) is a strong rational election less than 0 to some candidate should A strong equilibrium has the property that We

equilibrium then W(A*,B*) = V(A*,B*) = 0 PROOF: Given any B, since candidate A can always choose the same

23 platform, V(A*,B*) B, it must be true But that it min W(A*,B*) is easy to > O. see Also, that min if

> O. _

wEW(A*,B*) then -w~V(A*,B*) .

= O.

Therefore, min W(A*,B*)

> max W(A*,B*) Q.E.D.

This means that (A*,B*) is a strong equilibrium i f and only i f max W(A,B*) < 0 for all A and max V(A*,B) < 0 for all B. Even i f there are "weaker" equilibria, one suspects that only strong equilibria are

permanent. We, therefore, concentrate on them.

IV.

EQUILIBRIUMAND OPTIMALITY
functions are concave and have

In this section we show that i f u t i l i t y

continuous derivatives in A, and costs are distributed from zero, then all equilibria can be characterized in a remarkably simply manner; the candidates choose the same platform, the chosen platform maximizes We give fu(A,x)g(x)dx, and no one votes. Thus, i f an equilibrium exists there

is a simple maximization problem by which i t can be computed. several examples at the end of this section. To show these properties intermediate results. of except their platforms. the model; candidates of equilibrium,

we need to establish respects

The f i r s t of these occurs because of the symmetry are essentially anonymous in all

PROPOSITION 8: PROOF:
v if if

(SYMMETRY).

If

(A,B) is a strong rational

election

equilibrium then so are (A,A), (B,A), and (B,B). S i n c e (A,B) is an equilibrium, W(A,B) = 0 = V(A,B) from For all D and w i f if and only if w~W(D,B) then w ~ O. w~V(B,A). For all D and Further, we know that w~W(A,B) i f and only Now suppose that w~W(D,B) and proposition 4. -w~V(A,B) w ~ O.

vEV(A,D) then v ~ O.

w~V(B,D) for some D. therefore um.

Then, -w~V(O,B) which implies that

Thus, (B,B) is a strong rational election e q u i l i b r i Q.E.D.

The rest follows in a similar manner. Now we take up two lemmas which allow us to use calculus in the

analysis of equilibrium.

24 L M A 1: E M (A*,B*) is a strong equilibrium if and only if

fH((Pa/2)ilDil)l(D)g(x)dx ~ 0 for all A where D = u(A,x) - u(B*,x) and where I(D) = 1 i f D > O, I(D) = 0 i f D = O, and I(D) = -1 i f D < O. PROOF: By definition (A*,B*) is a strong equilibrium i f and only i f ~x+H((Pa/2)(liDll))g(x)dx - ~x_H((Pb/2)(IIDil))g(x)dx ~ 0 for all A. This is true i f and only i f (1)

~H((Pal2)(ItDIL))I(D)g(x)dx + fx_[H((Pal2)(liDil)) H((Pb/2)(LiDII))] g(x)DX ~ 0

(2)

for all A.

This in turn is true i f and only i f

~H((Pa/2)(llD11))I(D)g(x)dx < 0

(3)

for all A. Statement (I) Statement follows (2) from the remark after by I will the adding Proposition and side of (1). Assume 7 above. subtracting To esthat It must

follows

~Hx_((Pa/2)(llDll))g(x)dx leave the converse

to and from the l e f t to reader.

tablish (3) takes more work. ~H((Pa/2)(llDll))g(x)d then be true that

prove that (2) implies (3) and

> 0 and that (2) is correct for some A.

[H((Pa/2)IDi) - H((Pb/2)iDi)] g(x)dx < O. Therefore, Pa < Pb" Qa" Referring to lemma 3 in Section II we see that Qb < is Qa " Qb" This in

But this implies that (1) is > 0 since (i)

turn implies that (2) is > 0 which contradicts our i n i t i a l assumption. L M A 2: E M Q.E.D. Given (A*,B*,Pa) where Pa is a voters' equilibrium for

25 A*,B*. I f A* is "near" B* and i f u~C I and H~CI and i f their deriva-

r i v e s are bounded, then d~H((Pa/2)IDll(D))g(x)dx ~h((Pa/2)ID)) /dA : g(x)dx.

{(dPa/dA)(D/2 ) + (Pa/2)(dD/dA)]

PROOF: For any A and x such t h a t I(D) ~ O, we f i n d that d(H((Pa/2)IDIl(D))/dA =

h((Pa/2) IDl)[D/2}(dPa/dA ) + (Pa/2)(dD/dA)}.

I t f o l l o w s that e q u a l i t y of the previous

is also true i f for

I(D) = 0. 7 all x,

From Proposition 5 Theorem.

section dH/dA e x i s t s

since A* is near B*. Q.E.D.

The Lemma then f o l l o w s from the Lebesque Dominated Convergence Lemma 2 is v a l i d even i f A is n-dimensional A i f o r i = i . . . . . n. We now have a l l of t h i s section. Theorem I : that u~C I , Given the d i s t r i b u t i o n H~CI , their x and s t r i c t l y of v o t e r s ' the t o o l s needed to e s t a b l i s h

where A is replaced by the main p r o p o s i t i o n such is a

types, g(x) and h ( c ) , h(O) I f (A*,B*)

derivatives

are bounded,

> O, and u is

concave in A f o r a l l strong r a t i o n a l
:

concave f o r some x.

election equilibrium,

then A* = B*, Pa = Pb = i , Qa = Qb (A*,B*) is an e q u i l i b r i u m From Lemma for all A. i t must be

O, and A* maximizes f u ( A , x ) g ( x ) d x . PROOF: From Proposition 8 we know that i f We concentrate on the l a t t e r . that J = Suppose that (A*,A*) _< 0

then so is ( A * , A * ) . is an e q u i l i b r i u m . i it must be true that

We know t h a t max W(A,A*) < 0 f o r a l l A. fH((Pa/2)llDil)l(x)g(x)dx c o n d i t i o n s f o r maximization

From Lemma 2 and the f i r s t - o r d e r true, therefore,

7For f'(O) =

arbitrary a. Let

functions B so

f(x), D ~

if f1(x) O.

a as x + dH/dA +

for

all

sequences

of x, all

then such

A +

that

Then

h(O)(~)(dU(A,x)/dx)

for

sequences.

26
dJ/dA = 0 at A = A*. I f A = A* then D = O, Pa = 1, and

h(O) ~ (du(A*,x)IdA)g(x)dx = 0

Since

~u(A,x)g(x)dx

is a s t r i c t l y concave function, A* maximizes that (A*,B*) is an equiB* are maximizers

function. To finish the proof, we need to show that i f librium then A* = B*. of ~u(A,x)g(x)dx. Suppose not. and (B*,B*) are equilibria. From Proposition 8, both (A*,A*)

Therefore, both A* and

But u is s t r i c t l y concave for some x which implies Q.E.D.

that there is a unique maximizer; that is, A* = B*. Theorem I f u l l y characterizes the rational-election equilibrium i f it exists. In that equilibrium, even though no one votes -- thus avoiding all the nonproductive costs of voting -- candidates are led to select a platform which maximizes a social-welfare function, the sum of voters' u t i l i t i e s . The existence of voters who are on the margin of Because of the l i n e a r i t y of u t i l i t y is "locally" provoting, those with costs near O, leads candidates to take the preferences of these voters into account. in the costs of voting, the change in the probability that a voter w i l l vote, due to a change in a candidate's position, portional to the extra u t i l i t y is elected. utility of It received by the voter i f that candidate

is always in the interest of the candidates to change voters." This leads them inexorably to a

their position in the direction which maximizes the "aggregate marginal the marginal position which maximizes the aggregate u t i l i t y of the voters whose costs are minimal. 8 Because of welfare the similarity of this theorem to the fundamentalequilibrium allocations are theorem that competitive-market

Pareto-optimal, I am finding i t d i f f i c u l t to refrain from phrases like

81f types and costs are correlated~ that is if the density is g(x,c) instead of g(x)h(c), then candidates will choose the platform A which maximizes fu(a,)g(x,O)dx.

27

"the i n v i s i b l e

hand of the electorate."

However, the fact that equifunction" should not lead are also allocations

librium platforms maximize a "social u t i l i t y Pareto-optimal.

the reader to conclude that election-equilibrium

implications of the model. Example I: -IA-xl. this

The next few examples help to i l l u s t r a t e this and other

Suppose there is a one-dimension issue space and that the For

class of u t i l i t y functions which any voter can have is given by u(A,x) = x is usually interpreted to be voter x's ideal platform. theory t e l l s platform Let of will be the A* will ideal the type of example, traditional where ~X*g(x)dx = i/2. us that the election median voter, the rational-e-

equilibrium A : x*

us calculate

lection equilibrium.

maximize ~u(A,x)g(x)dx = ~-IA-xlg(x)dx.

I t is easy to see that A* w i l l also be the median of the density g(x). The two theories yield the same predicted-equilibrium platform, although turnout is predicted to be 100% by the traditional theory but 0% by this theory.

Example 2: utility still voter's platform.

Let us now look at a well-used u(A,x) = -(A-x) 2. the platform That is, will

example.

Suppose that theory ideal

preferences over a one-dimension issue space are given by the Type-i functions, predicts ideal In this case t r a d i t i o n a l be the median voter's is, however, [-(A that

The r a t i o n a l - e l e c t i o n platform.

equilibrium

the mean

A* maximizes

x) 2 g(x)dx.

Differentiating,

one gets [-2(A - x)g(x)dx = O. that there

From t h i s , we know that is absolutely nothing

~Ag(x)dx = [xg(x)dx or a : [xg(x)dx, the mean of g(x). This simple example i l l u s t r a t e s sacred about the median voter. 9 One might just as easily be concerned The predicted

about the mean or, indeed, any other moment. For example, i f u = -(Ax) b then the ( b - l ) s t moment is the equilibrium platform. functions. equilibrium platform depends on the composition of the class of u t i l i t y An important implication of this and the p r i o r example is forms are important. The functions - I x - A I and -(x that functional

gHinich (1977), Coughlin and Nitzan [ l g 8 1 ) , and Coughlin (1983a) also find the median to be unimportant when u n c e r t a i n t y i5 included in the voting model,

28 A) 2 each represent the same ordinal risk-free preferences on the set of A. However, they do represent different attitudes towards risk and different indifference surfaces between c and A. reflected in different equilibria. result. One other fact to note in this example: a multiple issue space will not eliminate this equilibrium. then g(x). the equilibrium is I f A and x are, say, n-dimensional, the multivariate distribution and (b) the mean of functional These differences are

The intensity of preference for A as

opposed to c, as measured by the willingness to vote, is what drives the

These remarks reinforce the insights of Hinich (1978) that (a) in uncertainty forms are important

the presence of

quadratic-loss functions can imply that the mean ideal platform is the two-candidate equilibrium. perceptions of candidates. Hinich bases his model on voters' errors in My model shows that his conclusions hold

even where errors are not present. Example 3: Finally, let us look at a simple application of this theory the election is held to decide the a l l o -

and consider what happens i f for that good.

cation of a public good and the assignment of the taxes needed to pay Let u ( y , l , x ) be the u t i l i t y of voter x for the publicW assume that x and I e and s(1). Platforms good level, y, when that voter's income is I. are independently distributed according to r(x) w i l l be of the form ( y , t ( ) ) to be paid i f income is I.

where the function, t ( 1 ) , indicates the tax I am assuming that taxes cannot be placed I f the cost of the public good is C(y), = C(y) for all platforms -no is allowed. Given this model, we know Letting L be the i t follows from

directly on the unobservable x. we require that deficit

ff

t(1)r(x)s(1)dldx financing

or surplus

that, in a rational-election equilibrium, y and t ( ) maximize f f u ( y , I t(1),x)r(x)s(1)dldx subject to the above constraint. Lagrangian multiplier associated with the constraint, f i r s t - o r d e r conditions that d[~f u ( y , l - t ( 1 ) , x ) r ( x ) s ( 1 ) d l d x ] / d y - L[dC(y)/dy] = O, (1) f ( d u ( y , l - t ( 1 ) , x ) / d l ) r ( x ) d x s + L(1)fr(x)dx s = 0
for all I and

29 C(y) = [ [ t(1)r(x)s(1)dldx. Let I* solve f ( d u ( y , I * , x ) / d I ) r ( x ) d x = L fr(x)dx. is, t(I) = I I*. Everyone's after-tax income will income is I* = (RM The second ebe identical --

quation above implies that in equilibrium I - t(1) = I* for all I; that income is redistributed towards the mean. Using the constraint, we find that everyone's after-tax
fS(1)dl, if R = ~r(x)dx conditions, first-order

C(y))/NR where N =
of the

and M = f l s ( 1 ) d l .

Turning to the f i r s t

i t can be shown t h a t i f d ( d u / d l ) / d x = O, t h a t is

d u / d l i s constant over a l l x, then L = dU(y,l*,x)/dl

and
N f (du/dy)/(du/dl) r ( x ) d x : dC/dy.

This

is

simply

the

Samuelson-Lindahl f i r s t - o r d e r

condition

for

the the

Pareto-optimal

allocation of the public good: the sum of marginal rates Thus, we conclude that i f allocate resources

of substitution equals marginal cost. then large two-candidate elections

post-tax marginal u t i l i t y of income is independent of the voter's type, efficiently. There are no "free riders" in this situation. I Examples of u t i l i t y given:
U = v(y,x) + I v(y,x) + w(u,l), + in I. and, as a special case, U =

functions for which d(du/dl)/dx = 0 can be

u = x iny

I leave i t to the interested reader to show that i f income and type are not independent, then the efficiency disappears and redistribution w i l l no longer require equal post-tax income. One can also show that i f costs of voting and income are positively correlated, as is sometimes argued, then low-income types w i l l have a larger impact on r e d i s t r i -

IOA side issue: since this case covers utility functions without income effects, it covers a l l situations covered by the Demand-Revealing Mechanisms, Therefore, i t dominates that method for social choice,

30 bution. These three examples are only a small indication of the powerful use one can make of the rational-election equilibrium. eager reader can provide many more. To prove that a l l the above is not vacuous, we move next to the question of existence. I am sure the

V.

EQUILIBRIUMAND EXISTENCE
of majority-rule equilibrium with no aban unusual occurrence. One

In the traditional theory stentions, existence of

equilibrium is

implication is that we cannot rely on theorems which assume existence. For example, local public-goods theories using the median voter should be highly suspect; the results are l i k e l y to be vacuous. The e q u i l i b r i um described in this paper, on the other hand, exists in many cases. These e q u i l i b r i a can potentially provide the foundation for many models which make social choices by majority-rule elections. In the last section we proved that i f A* were a rational-election equilibrium, then A* maximized aggregate u t i l i t y . I f we could prove the converse, that i f A* maximizes aggregate u t i l i t y then A* is a rational election equilibrium, we would be done since the appropriate compactness and continuity conditions which ensure the existence of a maximum (the Weierstrauss Theorem) are well-known. Unfortunately, the converse is It not true without additional conditions on the densities g and h. I~ for which the following is true: (S) if A* solves max ~u(A,x)g(x)dx, then A* is a rational-election equilibrium. I f we knew for which (g,h) A and convex in B, with the function W(A,B) were concave in

is our task to delineate as much as possible the set of distributions

V(A,B) behaving symmetrically, we would be game-theoretic solution to the

done, since under these conditions the

lIEconomists will, notice that this phenomenon welfare theorems about competitive equilibria.

also arises

when considering the

31 candidates' problem is known to exist. this approach. Remember that Unfortunately, one cannot take Qb where Qa =

W(A,B) = Qa -

~x+H((Pa/2)(D))g(x)dx and Qb = ~x-H((Pb)(-D))gix)dx and where D is concave in A and convex in B from the concavity of u (leaving aside the behavior of Pa and Pb for the moment). I f H is a concave function of c I f H is convex, then we then Qa is concave, but we cannot t e l l about Qb which, in this instance, is a concave function of a convex function. about Qa" have a symmetric problem since -Qb is concave, but we can say nothing Only i f H is linear, both concave and convex, can we discuss W capture this intuition in the next e the concavity properties of W. proposition.

PROPOSITION 9:
is true. PROOF: f(i/k)(Pa/2)

i f h( ) is the uri~orm density on [O,k}, k > O, then iS)

Let J = fH(Pa/2)IDl)l(D)g(x)dx = f(I/k)(Pa/2)IDII(D)g(x)dx = DG(x)dx. At A*, the maximizer, letting D = u(A,x) At any other A, J ~ O. Referring to lemma Q.E.D.

u(A*,x), we see that J = O.

I we now conclude that A* is a rational-election equilibrium. Absolutely no conditions have been placed on g. not worry about single-peakedness, mensionality. commodated. about h. uniform. That is, we need or unidi-

symmetry, unimodality,

Any density over concave-utility functions can be acThe second thing to notice is that we have been precise

An obvious question is whether (S) is true when h is not The answer is "no" i f we require all g to be accommodated. (S) is true for all g i f and only i f h is uniform. is simply Proposition 9. W prove the "only i f " e

PROPOSITION 10:
PROOF: statement.

The " i f "

Suppose we have a nonuniform h, an A, a g and an A* such that A* solves max~u(A,x)gdx and such that ~Hi(Pa/2)IDl)l(D)gdx ~ 0 when D = u(A,x) ~ u(A*,x). I f there are no such g, h, and A then we are done, I f there are such be since there will then be no g for which (S) is true. another position. g for which iS) is not true,

a g, h, and A for which (S) is true, we can show that there w i l l

which would prove the pro-

Thus, i f we show that we can perturb g to g' such that ~g'Ddx

0 and ~HiiPa/2)IDl)l(D)g'dx > O, then we w i l l have proven that (S) is

32 false and the proposition is true. The perturbation works as follows. Let SI be the set of x for which D > 0 and H(c) > c/k, let S2 be those x for which D > 0 and H(c) < c/k, let S3 be the x for which D < 0 and H(c) > c/k, and let S4 be the set of x such that D < 0 and h(c) < c/k. W make g larger on S1 and S3 e such and smaller on S2 and S4 by letting g'(x) = g(x) + ei when x~Si ,

that ireI dx + ~e2 dx + ~e3 dx + ~e4 dx = O, such that e1~Ddx = e2~Ddx = e3~Ddx + e4fDdx s O, and such that el~H((Pa/2)D)I(D)dx + e2CHldx + e3HIdx + e j H I d x > O. The careful reader can check to see that as long this perturbation w i l l be possible, since the Q.E,D. Proposition 10 informs us that if we want a simple-existence possible preference theorem and we want i t costs. to be applicable to all as H is is not uniform, sets Si will be nonempty.

patterns, we must r e s t r i c t our attention to uniform distributions of Suppose, instead, we want a theorem applicable to all d i s t r i The answer is similar to that in Proposition 10 - cost distributions if and only i f we In order to butions of costs.

statement (S) is true for all

severely r e s t r i c t the possible preference distribution g. entials.

see why, let us f i r s t define the derived distribution of u t i l i t y d i f f e r Let J(r) = ~X(r) g(x)dx where X(r) = [xlu(A,x)-u(A*,x) < r ] , Finally let 1(r) = j ( r ) - j ( - r ) for all r > O. and let j ( r ) d r = dJ(r).

I t can be shown that A* maximizes ~u(A,x)g(x)dx i f and only i f ~0 r l ( r ) d r < 0 for all A. I t can also be shown that expected p l u r a l i t y W(A,A*) = ~0 {H((Pa/2) r ) / r } r l (r)dr"
co co

(5.1)

(5.2)

Statement (S) is true when (5.1) equal to O.

implies that (5.2) is less than or statement (S) to be true, the

Therefore, in order for

function H((Pa/2)r)/r cannot weight r r e l a t i v e l y more heavily when 1(r) > 0 than when l ( r ) < O. Notice that an equal relative weighting occurs I f we require that (S) be true for all exactly when H is uniform.

possible h, then we must not allow l ( r ) > O, for otherwise there will be

33 at least one H which weights l ( r ) in the following: PROPOSITION 11:

PROOF:

incorrectly.

W capture a l l of this e

(S) is true for a l l h i f and only i f # z l ( r ) d r ~ 0 for Let (5.1) be true and l e t zI = sup { r l l ( r ) > 0}. all r, then we are done since #(H/r)r1(r)dr If If

a l l z ~ O, and a l l A at A*. (IF) l(r) ~ 0 for ~ O.

Therefore, we consider the case for which l ( r ) > 0 for some r < ~. zI = ~, then there is a z' such that # z , l ( r ) d r > O. possible by hypothesis.

But this is im-

Thus, we need only consider cases for which z1

Let z2 = sup {rlO ~ r ~ zI , l ( r ) > 0}. (z2,zl). Since # z l ( r ) d r ~ O, i t

Let I I = [ a l , ~ l and 12 = that #i21(r)dr ~ O.

follows

Further, since H is a distribution function, H(kz) ~ H(kzl) i f z ~ z I

and H(kz) s H(kZl) i f z ~ z I . ~H(pr)l(r)dr = ~ zI H(pr)l(r)dr

Therefore, l e t t i n g p = Pa/2, f~ H(pr)1(r)dr ~ H(P(Zl))~l(r)dr ~ O. + zI

One can iterate this proof for a l l zi until z i = O. (only i f ) = 1 if z Suppose that f z , l ( r ) d r > 0 for some z* e O. ~ z* and = 0 if z < z*. let H"(z) Then

f { H " ( ( P a / 2 ) r ) / r } l ( r ) d r = f l ( r ) d r > O. fu(A,x)g(x)dx.

But then A* cannot maximize Q.E.D.

In the proof of Proposition 11, we used a distribution of costs, H" which is very discontinuous at z*. continuous H ' " which is near to H" W could, however, have found a e and which is also appropriate.

Thus, the above proof would remain applicable with minor adjustments. 12 In this section, we have proven results only about the extreme l i m i t s of the set of (g,h) for which rational-election equilibrium

12As a side note, the cost d i s t r i b u t i o n , H ' ' , used in this proof, which assumes equal costs of voting known to a l l , is the same distribution used in Ledyard (1981), The fact that this d i s t r i b u t i o n causes the most d i f f i c u l t i e s for existence partly explains the weak theorem in that paper,

34
exists. T h a t is, we have required existence to occur either for all g I f we are willing to consider only some g or h, we should In particular, if h is almost uniform or i f g is I suspect that there One can show that there is an open set of (g,h) for which then equilibrium will exist.

or for all h. do better.

existence obtains. almost "symmetric," an open question.

is a large set of such (g,h), but its precise characterization remains Since the results of Coughlin; of Hinich, Ledyard, and Ordeshook; of Hinich; and of this paper all point to the fact that multidimensional election equilibria exist more often than suspected and that they rarely involve the median voter, results. 13 one might speculate whether i t is my assumption of r a t i o n a l i t y or the role of uncertainty which derives these I suspect uncertainty is the key to existence and that some This remains a future form of r a t i o n a l i t y is the key to "optimality." research issue.

VI.

VARIATIONSON A THEME

As I have presented this paper in many places, a number of issues have been raised which seem to be easily handled within the framework of the above model. Let me address these variations.

(I)

Income effects
function linearly. This assumption is not let u(A,O,x) be the

In the analysis of the rational voter I assumed that the cost of voting entered the voter's u t i l i t y utility necessary and can be eliminated. In particular,

received by the voter i f A wins and this voter did not vote.

Let u(A,c,x) be the u t i l i t y i f A wins and this voter voted where x and c are as in the original model. Assume that du/dc exists and is less than zero (that is, an increase in the cost of voting lowers x's u t i l i t y , Although the analysis is messier than above, one can For example, at an equilibrium ceteris paribus).

derive similar results.

13These remarks are motivated by an i n s i g h t f u l

o b s e r v a t i o n of Howard Rosenthal.

35

h(O)~UA{(A,O,x)/-Uc(A,O,x)}g(x)dx = O. This is identical to the e a r l i e r result if we "normalize" marginal

u t i l i t y by the marginal u t i l i t y of voting costs at O. equilibrium exists A* = B* and A* maximizes f[U(A,Q,x)/UA(A*,O,x)] g(x) dx. I do not yet know how other results translate.

T h a t is, i f an

For example, es-

tablishing existence appears to be more d i f f i c u l t .

(2)
or

Negative voting costs

how close the election. In almost every election there are

I am suspicious of anyone who claims to vote no matter what the issues f r i c t i o n s , or other phenomena ignored by this model, which cause d i f f e r ences among the candidates and which might l e a d low-cOst voters to vote. As far as I can t e l l there is s t i l l no common agreement on the facts about voter behavior. for completeness of of voting i t s e l f . is negative. In spite of my skepticism i t is important

the theory to explore what would happen to the

equilibrium i f there were indeed voters who derive u t i l i t y from the act I t is easiest to model these as voters whose cost, c, With this in mind If A = B In the model of the calculus of voting, a voter with c < 0

w i l l always vote for his most preferred candidate.

consider now the equilibrium in which those voters with c < 0 always vote and those voters with c > 0 behave as described above. then only the voters with c < 0 w i l l vote and, therefore, i f A = B in equilibrium i t must be true that A is the ideal platform of the median voter, the median of those who always vote, i f one exists. from standard theory that existence can be problematical.) are uncorrelated and if that median platform also fu(A,x)g(x)dx, then A w i l l be the equilibrium. utility, is. I then we must look elsewhere. It (We know I f c and x maximizes

However, i f the median

either does not exist or does not equal the maximizer of aggregate is an open question as to whether an equilibrium even exists in this situation and, i f so, what i t am not even sure whether candidates would choose the same All I can conclude so far is that " i r r a t i o n a l " platform in equilibrium.

voters who derive u t i l i t y from the act of voting create an externality

36
which interferes with the selection by the election of a socially desirable outcome. Perhaps we should educate voters not to be "citizens" but to be selfish?

(3) Minimax-regret voters

Suppose some voters in the electorate use the minimax regret c r i t e r i a of Savage (made popular by Ferejohn and Fiorina 1974, 1975). The analysis remains much the same, but the conclusions are s l i g h t l y altered. showed in Ledyard (1981), regret voter. see that, if we replace Pa by I/2 As I in the model of

rational-voting behavior we will have modeled the behavior of a minimaxI f one then Follows the model to its conclusion, one w i l l turn out, and A* w i l l is the density of the in equilibrium A* = B*, no one will voters and g*( )

maximize fu(A,x)[g(x) + (i/2)g*(x)Idx where g( ) is the density of the expected utility-maximizing minimax-regret voters. I t appears that because minimax voters do not At the margin, when A is

care about closeness, they end up being weighted at half that of u t i l i t y maximizers in their effect on the outcome. their effectiveness. near B, they react more slowly to changes in platforms and, thus, lose

(4) Vote-maximizing candidates

I t is sometimes argued that candidates care about other things than just winning. theory. dates. This is another of those areas of disagreement in p o l i t i c a l There is no agreement on the Factors which motivate candiAlthough i t is obviously of l i t t l e use to a candidate to have a Several that candidate does not win, some argue that candidates First, candidates w i l l (Of

large vote i f

should want to maximize votes, not the probability of winning. of our conclusions change i f that is the case. not choose the same platform.

I f they did, one of them could increase In equilibrium, i f one

his votes (from O) by simply moving away from the other candidate. course, this could lead to an election loss.) exists, turnout will occur with vote-maximizing candidates.

(5)

Turnout
in equilibrium. While I see this as good (the

A major issue raised by many who see this model for the f i r s t time is the lack of turnout deadweight loss of voting costs is avoided), many see this as a pre-

37 diction of the model clearly contradicted remembered that, lections w i l l because of rarely match this theory. by the facts. I t must be actual e-

the many possible f r i c t i o n s ,

Among other things, most e-

lections are held to decide several contests simultaneously and p o l i t i cal a c t i v i s t s , ignored in my model, operate to interfere with the natural forces. I t is true that single-issue elections with few a c t i v i s t s Examplesabound but the normal In a New Hampshire town an and with low stakes have l i t t l e turnout. school-tax election is the obvious one. election was held to f i l l ballot. No one voted.

the school board. Only one slate was on the I am not sure why the judges did not write in

t h e i r own names, but the moral is clear; when there is no choice i t pays not to go to the polls. 14 I am not sure what an appropriate example is for the model in this paper, but the following provides ease of computation. Let u = -(A-x) 2, H( ) = l-exp(-ac), and let g( ) = (R)(b) exp(-bx). appropriate Going through the manipulations one can, somewhat tediously, discover that

given the platforms A and B, the maximum turnout M( ) is 1 + (b/(aD-b))((exp-2aDS) - ((2aD/(aD+b))exp - bS)). Here, D = (A-B)/2 and S = (A+B)/2. I t can be easily shown that M is near 1 i f D,S,a,b are large. near 0 i f D,S,a,b are near O. of these parameters are. Does anyone want to guess? M is

I have no idea what "reasonable" values

I f one wishes to estimate equilibrium turnout, given A and B, one must solve the following equation; l e t M(a,b,D,S) by the equation above, then solve N = M(a/N,b,D,S) for N. N/R w i l l then be an estimate of the percentage turnout since 1/N estimates Pa" These are but a few of the p o s s i b i l i t i e s for refinement of the model. Others follow which I think are as important, but of which I know l i t t l e :

]4Stop Press: For those who b e l i e v e the p r o b a b i l i t y of a t i e d e l e c t i o n is e m p i r i c a l l y zero, let me r e p o r t the outcome of the 1983 e l e c t i o n f o r the Board of Trustees of Pasadena Community College: Gertmenia, 2592 and M i e l e , 2592. This followed a recount. Both candidates argued against drawing l o t s , the l e g i s l a t e d a c t i o n , as being undemocratic.

38 (a) (b) (c) (d) (e) (f) (g) Three-candidateelections (and multiple-candidate) Political activists and parties Candidatechoice and the role of primaries Intertemporal considerations Representativedemocracy and the responsiveness of the system Multiple, simultaneous elections Empirical estimation

39 REFERENCES

Chamberlain, G. and Rothschild, M. (1981) A Note on the Probability of Casting a Decisive Vote,
Journal of Economic Theory, 25: 152-162.

Coughlin, P.J. (1982) Pareto Optimality of Policy Proposals with Probabilistic Voting, Public Choice, 39: 427-433.

(1983a)

Davis-Hinich Conditions and Median Outcomes in

Proba-

b i l i s t i c Voting Models, Journal of Economic Theory, (forthcoming).

(1983b) Social U t i l i t y Functions for Strategic Decisions in Probabilistic 275-292. Coughlin, P.J. and Nitzan, S. (1981) Electoral Outcomes with Probabilistic Voting and Nash and Social Welfare Maxima, Journal of Public Economics, 15: 113121. DeGrazia, A. Voting Models, Mathematical Social Sciences, 4:

(1953)

Mathematical Derivation of an Election System, ~is, 44: 42-51.

Downs, A. (1957)
An Economic Theory of Democracy.

New York: Harper and

Row. Ferejohn, J.A. and Fiorina, M.P. (1975) Closeness Counts Only in Horseshoes and Dancing, The
American Political Science Review, 69: 920-925.

40 Ferejohn, J.A. and Fiorina, M.P. (1974) The Paradox of Not Voting: A Decision Theoretic Analysis, The American Political Science Review, 63: 525-536.

Gale, D. and Nikaido, H. (1965) The Jacobian Matrix and Global Univalance of Mappings,
Mathematische Annalen, 159.

Hinich, M. (1978) The Mean Versus the Median in Spatial Voting Games, Game
Theory and Political Science, (ed.) Peter Ordeshook.

New

York: New York University Press.

(1977)

Equilibrium in Spatial Voting: The Median Voter Result is an Artifact, Journal of Economic Theory, 16: 208-219.

, Ledyard, J., and Ordeshook, P. (1972) Non-Voting and the Existence of Equilibrium under Majority Rule, Journal of Economic Theory, 4: 144-153. Kramer, G.H. (1977)

A Dynamic Model of Political Equilibrium, Journal of Economic Theory, 16: 310-334.

Ledyard, J.O. (1981) The Paradox of Voting and Candidate Competition: A General Equilibrium Analysis, Essays in Contemporary Fields of Economics, Purdue Research Foundation, (eds.) George Horwich

and James Quirk, 54-80. Luce, R.D. (1959)

Individual Choice Behavior: A Theoretical Analysis.

New York:

Wiley.

41 Luce, R.D. (1977) The Choice Axiom After Twenty Years, Journal of Mathematical Psychology, 15: 215-233.

McKelvey, R.D. and Ordeshook, P.C. (1982) Elections with Limited Information: A Fulfilled Expectations Model Using Contemporaneous Poll and Endorsement Data as Information Sources. Social Science Working Paper 434, California Mellon University. Institute of Technology and Carnegie-

(1982)

Rational

Expectation

in

Elections:

Some Experimental

Results Based on a Multidimensional Model, Public Choice, this issue. Ordeshook, P.C. and Shepsle, K.A. (Editors) (1982)
Political Equilibrium.

Amsterdam: Kluwer-Nijhoff Publishing

Company. Palfrey, T.R. and Rosenthal, H. (1983) Slutsky, S. A Strategic Calculus of Voting, Public Choice, 41: 7-55.

(1975)

Abstentions and Majority Equilibrium, Journal of Economic Theory, 11: 292-304.

Tullock, G. (1967)
Towards a Mathematics of Politics,

Ann Arbor: University of

Michigan Press.