Political Economy and Public Choice: an Introduction

Antoine LOEPER
2017

Chapter I
What are Public Choice and Political
Economy about?
1 Market failures
The starting point of public economics is the existence of market failures. Many markets are not “perfect”
in the economic sense. Below, I list the most common market failure for those of you who have not had a
public economics course before and would like to know a little bit more about the notion of market failure;
you can perfectly skip that list and go directly to the next section if you are already familiar with the notion
of externalities, public goods, information asymmetries, or imperfect economic competition.
1. Externalities: a good is said to impose an externality if its production/exchange/consumption/use
a¤ects the well-being of individuals who are not directly involved in the transaction (that is, someone
who is not the buyer or the seller).
The typical example is pollution: someone using his car a¤ects the air quality in his region (a local
negative externality), he contributes to the greenhouse e¤ect (a global negative externality), and he
increases tra¢ c congestion for other drivers (a congestion externality). Drivers tend not to internalize
these external e¤ects when deciding whether to use their car versus using public transportation. The
price of gasoline only captures the cost of production, not these external costs of consumption. Such
externalities generate a market failure: drivers use too much their car relative to other modes of
transportation (public transportation, or staying at home), as compared to what would be socially
optimal.
Other examples of negative consumption externalities include smoking in public places, antibiotics
(whose systematic use increases the resistance of germs in the long run), building tall buildings (which
blocks the view of the neighbors). Other types of externalities include:
(a) The tragedy of the commons (which can be viewed as a special kind of negative externality). This
problem arises when a resource is jointly used by several agents, its access cannot be restricted,
but consumption is rival so the use of that resource by an individual diminishes the amount or
quality of that resource left for other individuals. The market failure takes the form of overuse,
or even complete depletion.
The typical example of a tragedy of the commons is over…shing: a given …sherman does not
internalize the fact that the more …shes they catch, the less …shes there are left for other …shermen,
and the more likely some species of …sh will go extinct.
(b) Positive externality: in that case, the production/exchange/consumption of a good or service
bene…ts people outside the transaction. The market failure generated by positive externalities is
the underprovision/underconsumption of that good. Typical examples:
These lecture notes have been prepared exclusively for the students of the course Economia Politica at Universidad Carlos
III de Madrid. Please do not circulate.

1
 i. production externality: bee keeping increases the productivity of vegetable growers around the
hive, cleaner energy technologies (e.g., wind/solar energy) reduce greenhouse gas emission,
R&D (more generally, the production of ideas) bene…ts the company who is …nancing the
R&D, but also other …rms via knowledge spillovers
ii. consumption externality: vaccination (limits the spread of a disease, which decreases the risk
of infection for other individuals)
(c) Network externality (the opposite to a congestion externality): a network externality is a positive
externality imposed by an individual A on an individual B when A makes the same choice as
B. The market failure induced by network externalities is that choices are too heterogenous and
incompatible.
i. Choice of technical and technological standards: when countries use the same railway gauge,
or power outlet voltage, they reduce switching and adaptation costs.
ii. Choice of software: two individuals using the same software can interact more easily.

2. Public goods: a public good is a good whose consumption is not rival (i.e., an individual’s consump-
tion of that good does not reduce the amount that other individuals can consume) and not excludable
(the access to that good cannot be restricted). A public good can be viewed as an extreme form of
positive externality, one that a¤ects everyone in the society.
When public goods are privately provided, the market failure takes the form of underprovision of the
public good, often no provision at all. The reason is that a provider of the public good does not
internalize the fact that the extra amount of public good she generates will bene…t other people as
well. Examples of public goods:

(a) Fundamental research: without patents, research …ndings are not excludable, and their consump-
tion is clearly not rival. Everybody can use Pythagore’s theorem.
(b) Protection from foreign invasions: if a foreign army is prevented from entering in a country by
force, everybody within the borders bene…ts from the protection
(c) An independent court of justice that enforces contract: when contracts are credibly enforced, all
transactions become more e¢ cient, because e¢ cient and credible contracts reduce transaction
costs, and thus grease the wheel of the economy.
(d) Redistribution: a more equal society is less prone to violence and theft, which bene…ts everyone.
(e) Biodiversity, culture...

3. Informational imperfection: when consumers and producers have asymmetric information (for in-
stance, producers have better information than the consumers about the quality of their product),
some bene…cial transaction may not occur because information cannot be transmitted in a reliable way
between the buyer and the seller, and the market may not work as e¢ ciently as if buyers and seller had
the same the information. A typical example of a market with asymmetric information is the market
for used cars: the seller knows more about the quality of the car than the buyer.

(a) Consumer good quality: a food producer has no incentive to reveal that its product contains some
harmful additive. Likewise, a toy manufacturer has no incentive to reveal that its product can be
dangerous for babies.
(b) Skills of service providers (doctors, lawyers, mountain guides): a lousy doctor has no incentive
to reveal his degree of expertise if he is below average, and skilled doctors may not have reliable
ways to prove how skilled they are to their potential patient.
(c) Workplace safety: a …rm might not be willing to reveal its information about the safety of the
working conditions in its plants
(d) Financial information about publicly traded …rms (asymmetric information between managers
and shareholders)

2
 (e) Insurance markets: each person has information about his health that a¤ects the cost of provision
for his health insurance. This asymmetry of information generate important frictions in the
market for health insurance. The same applies to car insurance.

4. Imperfect competition: an e¢ cient market requires a perfectly competitive environments. That is,
…rms should not have signi…cant market power so as to manipulate prices. In some cases, this is hard
to sustain. For instance:

(a) Natural monopoly: in the presence of large …xed costs, or decreasing marginal cost, economic
competition typically leads to the monopoly of a single producer, which can then use its monopoly
power to increase prices above marginal cost. Examples: network providers, railway operators,
plane manufacturers...
(b) Predatory pricing: some …rms might be tempted to temporarily reduce price so as to kill their
competitors, and latter on abuse their monopoly position.
(c) Price discrimination: …rms can increase their pro…t by charging di¤erent prices to di¤erent cus-
tomers. In some cases, letting them that possibility can be detrimental for social welfare, because
…rms might be able to charge a very high price to consumers who are in desperate need of a
particular product.

2 How to deal with market failures? The public economics ap-

proach
For each of the above market failure, in theory, there exists an intervention that correct the failure and lead
to an e¢ cient allocation of resources. A public economics approach advocates the creation of a public sector
with the policy instruments needed to implement that intervention (power to tax, to produce public goods,
to regulate economic competition...). The rationale for that approach is that the public sector is not pro…t
driven, and thus will not be prone to the same failures as the private sector. A benevolent social planner
advised by an omniscient advisor should therefore be able to optimally regulate the market and restore
e¢ ciency.
The type of public intervention needed to correct the above market failures depends on the nature of the
market failure. Below, I list some examples of public intervention advocated by public economists to correct
the aforementioned market failures. Again, the list is there only for those of you who have never taken a
public economics course, but it is not crucial to know that list by heart; the important point here is that
the type of intervention depends a lot on the market failure, and hence the regulator needs to have a lot of
discretion about the tools he can use, and he needs to know a lot about economy to take the appropriate
action.

1. Externalities:

(a) the solution to externalities can take the form of an "incentive mechanism" or a "command and
control" mechanism. An incentive mechanism does not restrict the discretion of the economic
agents (they can produce/consume how much they see …t), but it shapes their incentives, usually
by changing the market prices. The most prominent examples of incentive mechanisms are the
following:
i. Corrective taxes (also called Pigouvian taxes) for negative externalities: a tax increases the
price of a good, which decreases demand, and can thus correct for the market failure of
overconsumption/overproduction that occurs in the presence of negative externalities.
ii. Corrective subsidies for positive externalities: a subsidy decreases the price of a good, which
increases demand, and can thus correct for the market failure of underconsumption/underproduction
that occurs in the presence of positive externalities.
iii. Emissions permits with a Cap and trade mechanisms to limit pollution emissions.

3
 (b) A "command and control" mechanism is a mechanism which does not distort prices, but im-
poses/restrict certain actions related to the external e¤ect. The most prominent examples of
"command and control" mechanism are the following:
i. Imposing a maximal level of provision/consumption in the case of an activity generating
positive externality (e.g. regulating the level of car emission)
ii. Imposing a minimal level of provision/consumption in case of an activity generating positive
externality (e.g., requiring that energy producers produce a certain percentage of their energy
from clean sources)
iii. Imposing a technological constraint or the use of certain equipment that produce less exter-
nalities (e.g., imposing car manufacturer to put a particle …lter on all cars), or restricting the
use of certain goods (antibiotics are available only with a prescription from a certi…ed doctor)

2. Public good: for pure public goods, the most common solution is to have the public sector provide
the public good. For instance, the state provides national defense, public courts, national parks, public
universities, many cultural services (Cervantes Institute, Confusius Institute, Alliance francaise)
3. Imperfect information: there are various forms of regulation that address information imperfection,
among others:

(a) imposing producers to provide some information (food producers are required to display some
information about their product on the packaging)
(b) prohibiting producers to some information: in some countries, health insurers cannot use infor-
mation about previous conditions to price their services.
(c) occupational licensing: some services can be provided only by people with certain public certi…-
cation: doctors, taxi drivers, notaries, teachers.

4. Imperfect competition:

(a) nationalization, or price regulation of monopolies

(b) anti-trust policies: merger regulation, legal sanctions in case of collusion
(c) predatory pricing: in many countries, …rms are not allowed to price below their marginal cost

3 The political economy/public choice approach: market failures

versus political failures
The premise of the public economic approach is that the policy makers in charge of correcting the market
failures have all the information necessary to correct the market failure, and more importantly, that they
have the appropriate incentives to do so. If regulators were able to impose Pigouvian taxes and subsidies on
all externalities, regulate all markets in which there is imperfect competition, adequately address problems
of information imperfections, life would be great! However, reality suggests that this premise is questionable.
The political economy/public choice approach argues that comparing a malfunctioning market to an ideal
government intervention is not very informative. In practice, politicians and civil servants often have to
make decision with inadequate information. Most of the time, they know less about the market than market
participants do. And most problematically, their incentives are not aligned with the interests of the citizens
they are supposed to serve. They do not necessarily maximize their own pro…t as private …rms do, but they
typically do not behave as benevolent planner whose only goal is to maximize social welfare, as a standard
public economy approaches assumes. In fact, citizens typically disagree on what social welfare is, and so they
typically disagree what is the socially optimal response to a market failure. Therefore, when deciding when
the public sector should intervene, and which activity should be left to the private sector, we should compare
how markets and governments function in reality. In other words, we should compare market failures and
political failures!

4
 This approach raises a host of questions: how do public institutions function in practice? What are their
most important ine¢ ciencies? How can we compare these political failures to market failures? What type of
rules and institutions can minimize the likelihood and cost of such political failures? Of course, the answers
to these questions depend a lot on the type of public institutions that we are dealing with. In this course,
for the most part, we will focus on democratic institutions.

5
Chapter II
How should we take collective decisions?
Our …rst task as political economists is to investigate how to make collective decisions for a group of in-
dividuals with heterogeneous preferences. This question sounds a little abstract, but it is the essence of
democracy. There are many ways to think about collective decision making. One way is to consider each
collective choice problem separately, try and take the perspective of an impartial but benevolent planner,
and decide given the speci…cities of the collective choice problem what is the socially optimal decision. For
instance, if the choice under consideration is how to tax income in a given country, the optimal tax law
should presumably depend on the degree of inequality in that country, the level of social mobility, its cost
of living, how meritocratic its labor market is, the geographical mobility of its citizens, the …scal need of
its government, which itself depend on environmental conditions, external threats, the demand for public
education and other public goods....
It should be clear that it is impossible to agree on simple and transparent decision rules that would take all
these parameters into account in a socially optimal way. Likewise, leaving such a complex and controversial
decision to the discretion of a representative would open the door to arbitrariness, no matter how benevolent
and impartial that representative is. That problem was …rst raised by the political philosophers of the
“utilitarianism” tradition (e.g., Jeremy Bentham and John Stuart Mill). They proposed instead a less
discretionary, more systematic and transparent way to choose in groups: collective decisions should depend
only on what people want. In more formal terms, collective decision making should depend only on preferences
reported by the individuals a¤ ected by the decision (or individual utilities, hence the term “utilitarianism”)
and not on some arbitrary criterion invoked by the ruler of the day. This way to make collective decisions
takes the speci…cities of the choice under consideration into account, but only via individual preferences.
Following the utilitarian approach, in this chapter, we describe the collective decision process as a function
that maps individual preferences into a collective decision. In less abstract terms, this function maps the
potentially con‡icting interests of the citizens into a decision of society in a mechanical and transparent way.
We will call this map a preference aggregation rule. The most famous preference aggregation rule is the
majority rule. According to this rule, when comparing two alternatives, society should choose the one that is
preferred by the greatest number of individuals. Variants of this rule are typically used when an important
decision is submitted to a popular referendum. As we will see, there are many alternative ways to aggregate
individual preferences into a collective choice.
The utilitarian approach is essentially a normative approach in the sense that it describes how collective
decisions should be made. In actual democracies, collective decisions are rarely taken simply by asking
citizens to state their preferences among the di¤erent alternatives and by mechanically aggregating these
preferences. The functioning of modern democracies requires countless daily decisions and one cannot consult
all citizens on each of these decisions. Most decisions are delegated to political representatives, and these
representatives may not perfectly know the preferences of their citizens, or may have incentives not to take
these preferences into account. In this chapter, we abstract away from the possible political failures that
come from political delegation (we will come back to this issue latter on in the course). The goal of this
chapter is not to describe in a realistic way how actual democratic systems work. Instead, we will take the
seat of a constitutional designer whose goal is to …nd the best possible mechanism to aggregate individual
preferences, and we will discuss the trade o¤s that he faces. To make his task as easy as possible, we will
assume that the individuals know what is best for them, and that the designer can observe their preferences.
Despite these somewhat heroic assumptions, we will show that it is impossible to design a collective decision
process which, at the same time, (i) takes into account the preferences of all citizens in a democratic way and
(ii) lead to rational decisions for the group. This tension between democratic values and group rationality can
be viewed as a political failure because it imposes limits on what democratic decision making can achieve,
even when one abstracts away from all the problems related to political delegation.
This chapter is fairly abstract, but no advanced mathematical tools will be used, only basic logic. The
exercises and the end of each sections will help you familiarize yourself with the formalism.

6
4 Preliminaries and notations
In this section, we introduce the formalism and notations that we will use thoughout this chapter.
Let N be a …nite set of individuals, i.e. all the individuals whom the social planner care about for the
collective decision under consideration. With a slight abuse of notation, N will also refer to the number of
individuals, so N = f1; ::; N g.
Let X = fa; b; c:::g be a …nite set of group alternatives, and let jXj denote the number of alternatives in
X. These alternatives are assumed to be mutually exclusive so the group has to choose one and only one of
these alternatives.
To make things more concrete, consider the following examples:
N is the population of a country, and X contains di¤erent ways to design the income tax scheme.
These schemes can vary in terms of their degree of redistribution, the among of money they raise, how
distortive they are, their tax base (i.e., which types of income are tax free)...
N is the population of a country, and X is the set of possible allocations of legal rights for homosexual
couples: right to marry or not, right to adopt children or not, …scal treatment...
N is a group of friends who wants to go to the movie theater together on a given day, and X is the set
of movies that show at the theater on that day.
N is the population of Spain, and X is the set of possible constitutional arrangements that govern
the relation between Catalunia and the rest of Spain: Catalunia remains part of Spain with the same
level of autonomy as the other regions, Catalunia remains part of Spain but with a greater …scal and
cultural autonomy, complete secession...
N is the population of a presidential democracy, and X is the set of candidates for a given presidential
election. Note that as argued earlier, in this chapter, we ignore the issues related to the con‡ict of
interests between the citizens and their political representatives once they are in o¢ ce. However, we
will study the possible political failure that are related to the voting rules that are used to select
political representatives. In other words, we will talk about elections, but not about political agency
costs.
N are the countries that are members of the E.U., and X is whether Turkey should become a member
of the E.U or not.
Our goal is to …nd a collective decision process that is nondiscretionary, that is, that does not depend on
the speci…cities of the collective decision under consideration. For this reason, in this chapter, N and X will
remain abstract sets, but we will illustrate our results in concrete examples whenever possible.
Notation 1 The variables x; y; z refer to any arbitrary alternatives in X, while the variables a; b; c refer to
speci…c alternatives in X. In other words, x; y; z will be used to make generic statement of the sort "for all
alternatives x and y, then blabla..." while a; b; c will be used to make speci…c statement such as "suppose that
alternative a is that we tax only consumption, and we tax it at a rate of 50%, whereas alternative b is that
we tax only income, and we tax it at 50%, then blabla..."

4.1 Preferences and choices

We will …rst de…ne formally the concept of preferences, which we call a preference relation in this course.
De…nition 1 A preference relation speci…es for every pair of distinct alternatives x and y in X whether
x y or x y (one and only of these statement is true for each pair x; y 2 X).
For instance, if X = fa; bg (i.e., there are only two distinct alternatives), the following two conditions
characterize a preference relation:
a b, and b a: (1)
Suppose that a preference relation represents the preferences of an individual i over X: Then for any
two distinct alternatives x; y 2 X, there are four distinct possibilities:

7
Case 1 x y and y x. We say that i prefers x to y.

Case 2 x y and y x. We say that i prefers y to x.

Case 3 x y and y x. We say that i is indi¤erent between x and y.

Case 4 x y and y x. We say that i cannot compare x and y.

Clearly, a preference relation for which Case 4 occurs is weird. We will rule out this pathological case in
Section 4.2 (see De…nition 3).

Remark 1 A preference relation on a …nite set X can be conveniently represented by a matrix denoted
M (a table) with jXj column and jXj rows. For all x; y 2 X, the term of M in row x and column y is 1
if x y and 0 if x y. For any preference relation , one can de…ne such a matrix M , and conversely,
given any square matrix M of size jXj of 1s and 0s, one can de…ne the corresponding preference relation
M
. So these two representations are equivalent.

For instance, if denotes the preference relation characterized in (1), then

0 1
a b
M = @ a : 1 A:
b 0 :

Exercise 1 Suppose X = fa; b; cg (i.e., X consists of three alternatives)

1. How many comparisons do you need to specify to completely de…ne a preference relation on X?
2. Let be such that 8
< a b, and b a,
a c, and c a, (2)
:
c b, and b c,
What do the above conditions imply for M ?

Solution.

1. To completely characterize a preference relation on X, one must specify for every pair of distinct
alternative x; y 2 X whether x y or x y. There are jXj (jXj 1) ordered pairs of alternatives in
X, so there are jXj (jXj 1) comparison to specify. In this case, jXj (jXj 1) = 6, so the 6 comparisons
speci…ed in (2) completely characterize the preference relation.
2. The conditions in (2) imply the following for M :
0 1
a b c
B a : 1 1 C
B C:
@ b 0 : 0 A
c 0 1 :

The following de…nition relates the notion of preference relation to the notion of choice. Presumably,
if an individual is given the choice between any alternative he wants in some set X; he should choose the
alternative that is the best in X according to his preferences (provided there exists such an alternative, more
on that issue in Section 4.2).

De…nition 2 Let be a preference relation. An alternative x in X is a most preferred alternative (or a

stable choice) for if there exists no alternative y in X such that y x. Equivalently, x is most preferred
if for all y 2 X; y x:
For amy subset S of X; an alternative x is a most preferred alternative (or a stable choice) in S for if
there exists no alternative y in S such that y x.

8
Remark 2 Observe that when a preference relation is represented by a matrix M as explained in Remark
1, one can easily check whether an alternative x is a stable choice in X for . One simply has to check
whether the column of that matrix which corresponds to alternative x contains only 0s. If that is the case,
then x is a stable choice (or a most preferred alternative) in X for . If there is a 1 in some line of the
column of x, x is not a stable choice because it is dominated by the alternative that corresponds to this line.
Preference relations will be used to describe both individual preferences and social preferences.
Notation 2 The preferences of an individual i 2 N will typically be denoted i, whereas the social prefer-
ences will typically be denoted S .
An individual preference relation has a straightforward meaning: x i y means if individual i had the
choice between x and y only, she would choose x. A social preference relation S can be interpreted as
follows: x S y means that according to S , alternative x is deemed socially superior to y. Therefore, if
y is the status quo prevailing in society N and if the collective choice under consideration is whether to
replace y by x, S would prescribe to repeal y and implement x instead. Given this interpretation, if an
alternative x is a stable choice/most preferred alternative in X according to a social preference relation S
(see De…nition 2), then there is no alternative y 2 X that should replace x according to S , which explains
the term “stable choice”.

4.2 Properties of preferences

De…nition 1 is fairly general. As the next example shows, it allows for a large class of preferences, some of
them quite exotic.
Example 1 Consider the preference relation represented by the following matrix:
0 1
a b c
B a : 1 0 C
B C
@ b 0 : 1 A: (3)
c 1 1 :
This preference relation has the following questionable properties:
1. Alternatives b and c are ranked above each other: b c and c b. This implies that is unable to
make a stable choice in fb; cg!1
2. This preference relation also induces a three way cycle: a b, b c, and c a! In particular,
that preference relation would be unable to choose a most preferred alternative among fa; b; cg because
any alternative in fa; b; cg is dominated by another alternative in fa; b; cg: a is dominated by c, b is
dominated by a, and c is dominated by b.
To rule out these pathological patterns, we can require a preference relation to satisfy the following
condition:
De…nition 3 A preference relation is rational if for all m 2 and there exists no x1 ; x2 ; :::; xm 2 X such
that x1 x2 ::: xm x1 .
As the next proposition shows, De…nition 8 is exactly what is needed to avoid the pathological cases in
Example 1. For the preference relation in that example, points 1 and 2 violate the de…nition of rationality.2
Example 1 shows that a preference relation which is not rational is problematic because it does not induce a
most preferred alternative. Hence, we lose the correspondence between preferences and choices (see Section
4.1): an individual with irrational preferences (or a group of individuals with irrational social preferences)
is unable to make a de…nitive choice! The following proposition formalizes the link between stable choices
and rational preferences:
1 The careful reader will notice that this point is the same as Case 4 in Section 4.1.
2 Point 1 is a violation of rationality for the following reason: a b a violates De…nition 8 for m = 2, x1 = b, x2 = c in .
Point 2 is a violation of rationality for the following reason: a b c a violates De…nition 8 for m = 3, x1 = a, x2 = b,
and x3 = c.

9
Proposition 1 A preference relation is rational if and only if for all subset S of X, there exists a stable
choice in S according to (see De…nition 2).

Note that Proposition 1 can be used to check whether a preference is rational or not. If you are given
a preference relation, it is hard to check whether it is rational using De…nition 3.3 However, one can check
instead whether it has a most preferred alternative for any subset S of X using Remark 2. The following
exercise illustrates this method.

Exercise 2 Let X = fa; b; cg, and suppose that is such that

a c, b a, and c b. (4)

1. Draw a 3*3 table to represent . Is completely speci…ed by the conditions in (4)?

2. Is rational? What is the best alternative in X according to this preference relation?
3. What are the most preferred alternatives in fa; bg according to this preference relation?

Solution.

1. The conditions in (4) imply that 0 1

a b c
B a : ? 1 C
M =B
@ b
C;
1 : ? A
c ? 1 :
where the symbol ? correspond to the pairwise comparisons which are not speci…ed by the conditions
in (4). So is not completely speci…ed by (4).
2. Even though is not completely speci…ed by the conditions in (4), we can say that is not rational.
To see the violation, take m = 3, x1 = a, x2 = c, x3 = b in De…nition 3: we have a c b a.
There is no most preferred alternative according to this preference relation: a is dominated by b, b
is dominated by c, and c is dominated by a. One can also answer this question using Remark 2: all
columns of the matrix M have a 1, so there is no most preferred alternative in X: However, it is
important to note that one cannot answer that question by using directly Proposition 1. The reason is
that Proposition 1 only implies the following: if is not rational, then there exists some subset S of
X such that there is no most preferred alternative in S according to . But we do not know a priori
whether this is the case for S = X:
3. Given the information we have, we cannot tell whether there is a best alternative in fa; bg: we know
that a is dominated by b, so a cannot be a most preferred alternative in fa; bg, but we do not know
whether b is a most preferred alternative in fa; bg because we do not know whether a b or a b.
The conditions in (4) do not provide this information.

The following exercise shows that in some cases, there can be more than one most preferred alternative:

Exercise 3 Let X = fa; b; cg, and let be the preference relation which corresponds to the following matrix:
0 1
a b c
B a : 1 0 C
M =B @ b 0
C:
: 0 A
c 0 1 :

What is the most preferred alternative in X according to ?

3 To
check that, one would have to consider all possible sequences x1 ; :::; xm for any m 2, and check whether the implication
in De…nition 3 is satis…ed. For a given set of alternatives, there are lots of such possible sequences.

10
Solution. Alternative a is a most preferred alternative, because no other alternative is preferred to a. To
see that, just look at the column of a, you will see that there are only 0s (see Remark 2). Likewise, there is
only 0s on the column of alternative c, so alternative c is also a most preferred alternative. Alternative b is
the least preferred alternative.
To make sure that there exists a unique most preferred alternative, one can impose the following require-
ment:

De…nition 4 A preference relation is resolute if for any two distinct alternatives x; y 2 X, x y or

y x.

Hence, a preference relation is resolute if any two distinct alternatives can be strictly ranked. In other
words, is resolute if it is never "indi¤erent" between any two distinct alternatives (see case 3 in Section
4.1). If M denotes its matrix representation (see Remark 1), then is resolute if and only if there is no
nondiagonal entry of M such that this entry contains a 0 and the symmetric entry also contains a 0:

Proposition 2 A preference relation is resolute if and only for any subset S of X, admits at most one
most preferred alternative in S.

Exercise 4 Determine whether the preference relations de…ned in Exercises 1, 2, 3, and Example 1 are
resolute.

Solution. One can simply check that each nondiagonal entry of M is either 1 or its symmetric is 1. In
the case of Exercises 1 and Example 1, it is true. In the case of Exercise 3, it is not true, because the entries
(c; a) and (a; c) are both 0. In the case of Exercise 2, some nondiagonal entries are not speci…ed. But the
preference relation is nevertheless resolute because any nondiagonal entry is either 1 or its symmetric is 1.

Throughout this course, we will assume that all individuals have rational and resolute preferences:

Assumption 1 For all i 2 N , we assume that i is rational and resolute (see Notation 2 and De…nitions
3 and 4).

The assumption that individual have rational preferences is relatively mild. As shown in Proposition
1, it amounts to assume that individuals know what is best for them, and can make choices accordingly.
We assume that individuals are rational because our main concern in this chapter is whether one can make
rational and democratic choices for a group of rational individuals. The main question in political economy
is not whether individuals are dysfunctional, but whether the political system (here, interpreted as the social
preferences) is dysfunctional.
The assumption that individuals have resolute preferences is only for simplicity. As the following remark
shows, preferences which are rational and resolute can be represented in a very compact and convenient
form:

Remark 3 If a preference relation is rational and resolute, there is a unique way to order the alternatives
in X from the most preferred to the least preferred: let x1 denotes the most preferred alternative in X for ,
let x2 denote the most preferred alternative in X n fx1 g for , let x3 denote the most preferred alternative
in X n fx1 ; x2 g for ,..... , and let xjXj be the least preferred alternative in X for ,4 then we have that

x1 x2 ::: xjXj :

This ranking completely characterizes the preferences relation .

Representing a preference relation by ordering the alternatives from the most preferred to the least
preferred (as explained in Remark 3) is clearly shorter than representing it via a matrix, as in Remark 1.
To rank alternatives from the most preferred to the least preferred, one simply needs to specify jXj 1
comparisons. To …ll out the matrix M , one needs to specify jXj (jXj 1) comparisons. However, it is very
important to note that this representation is possible only when preferences are rational and resolute.
4 Observe that from Propositions 1 and 2, at each step of this procedure, there is a unique most preferred alternative, so the
sequence x1 ; x2 ; :::; xjXj exists and is unique.

11
Remark 4 (Important!!!) Since we assume throughout that individual preferences are rational and res-
olute, we will typically use the ranking suggested by Remark 3 to specify a particular individual preference
relation. However, as we will see latter on, social preferences are not always rational or resolute. So if an ex-
ercise asks you to specify the social preferences in a particular context, you CANNOT in general assume that
these preferences are rational or resolute. For this reason, you MUST specify the social preferences between
ALL possible pairs of alternatives, and the most convenient way is via the matrix introduced in Remark 1.

4.3 Preferences and utility functions

In economics, individual preferences are often derived from utility functions. In this chapter, we will use
mostly preferences relations, but in Chapter III, we will consider speci…c political economy environment in
which individual preferences will be speci…ed via utility functions. This section explains the relationship
between the two concepts of preference relation and utility function.
De…nition 5 A utility function U is a function from X to R. The preference relation U derived from a
utility function U is de…ned as follows: for all x; y 2 X,
x U y if U (x) > U (y) .
For any preference relation , we say that a utility function U represents the preference relation if is
equal to U .
Exercise 5 Find two utility functions that represents the preference relation in exercise 1.
Solution. One utility function is given by U (a) = 9 and U (b) = 4 and U (c) = 7. Another possibility
is U 0 (a) = 4=9 and U 0 (b) = 2 and U 0 (c) = 0. One can easily check that the preference relations U and
U 0 derived from these two utility functions coincide with the preference relation de…ned in Exercise 1.
An important property of utility functions is that the preference relation derived from a utility function
is always rational:
Proposition 3 If is represented by a utility function U , then is rational.

4.4 More exercises on preference relations

Exercise 6 What is the most preferred alternative for the preference relation of Exercise 1? Answer that
question for any subset S of X. Do the same exercise for the preference relation of Exercise 3.
Exercise 7 Is the preference relation de…ned in exercise 1 and 3 rational?
Exercise 8 Suppose that a preference relation is represented by some utility function U . Under which
condition on U is resolute?
Solution. Resoluteness requires than for any x 6= y, x y or y x. If is represented by some utility
function U , this means that U (x) > U (y) or U (y) > U (x), or equivalently that U (x) 6= U (y). Hence,
resoluteness means that no two alternatives generate the same utility level.
Exercise 9 (Harder) Find a preference relation that cannot be represented by a utility function (See De…n-
ition 5).
Solution. Let be the preference relation represented by the following table
0 1
a b c
B a : 1 0 C
M =@ B C;
b 0 : 1 A
c 1 0 :
and let U be a utility function that represents . One can readily check by comparing a and b that necessarily,
U (a) > U (b). Likewise, by comparing b and c, we see that U (b) > U (c). And …nally, by comparing c and
a, we see that U (c) > U (a). Adding up the three inequalities, we get that
U (a) + U (b) + U (c) > U (a) + U (b) + U (c) ;
a contradiction. So cannot be represented by a utility function.

12
5 Preference aggregation
De…nition 6 A preference pro…le is a collection of preference relations, one for each individual in N . It is
denoted ( 1 ; ::; N ).

A preference pro…le simply lists the preferences of each individual in the society. One can view a preference
pro…le as a speci…c situation for the social planner. For instance, if N = 3 and X = fa; b; cg, from Remark
4, the following conditions characterize the preference pro…le ( 1 ; 2 ; 3 ):
8
< a 1 b 1 c,
a 2 c 2 b, (5)
:
c 3 a 3 b:
Our main goal in this chapter is to aggregate a pro…le of individual preferences into a social preference
relation. The corresponding mapping is called a preference aggregation rule.

De…nition 7 A preferences aggregation rule (henceforth, PAR) f is a function whose argument is a pref-
erence pro…le ( 1 ; ::; N ) and whose image f ( 1 ; ::; N ) is a preference relation. The preference relation
f ( 1 ; ::; N ) is interpreted as the social preferences for society N when the preferences of its members are
( 1 ; ::; N ) and they are aggregated using the PAR f .

Hence, a PAR can be viewed as a transparent mechanism which maps the possibly con‡icting interests
of the citizens, as captured by the preference pro…le ( 1 ; ::; N ), into a preference for the society as a whole,
as captured by the social preferences f ( 1 ; ::; N ).

Notation 3 In the sequel, when the PAR f and the preference pro…le ( 1 ; ::; N ) is clear from the context,
I will simply denote the social preference relation f ( 1 ; ::; N ) by S , or blabla where blabla is the name
of the PAR f .

To make things more concrete, let us de…ne a few PARs, and see how they aggregate individual preferences
in practice.

Example 2 Examples of PAR:

The Simple Majority Rule (MR): for any two alternatives x and y in X, x is socially preferred
to y according to the MR if a strict majority of individuals in N prefers x to y. Formally:
N +1 5
x MR y if jfi 2 N : x i ygj :
2
The MR is probably the most common way to aggregate preferences in democracies. It embodies the
democratic principle that a decision that makes a majority of people better o¤ should be implemented.
Put di¤ erently, the majority should get its way. This rule, or variants of it, is used in popular refer-
enda, but also in legislatures to pass laws and amendments. Likewise, in most elections involving two
candidates, votes are aggregated by simple majority rule.6 Note that in the case of legislatures, N is
the set of lawmakers, not the citizens themselves.
The 2/3 Majority Rule (2/3MR): for any two alternatives x and y in X, x is socially preferred
to y according to the 2/3MR if at least 2/3 of the individuals in N prefer x to y. Formally:
2N
x 2=3M R y if jfi 2 N : x i ygj :
3
This kind of quali…ed majority rules (also called supermajority rules) are often used in legislatures to
approve constitutional amendments, whereas nonorganic laws are typically approved by simple majority
rule. Such rules are also used to pass standard laws in some countries. For instance, in the US Senate,
most laws must be approved by 60% of the legislators to be enacted.
6 In some countries, votes are aggregated in multiple steps, and thus the procedure is not exactly equivalent to simple majority

rule. For instance, in the U.S., votes for the presidential election are aggregated at the state level using simple majority rule,
and then at the national level using a population weighted simple majority rule among states.

13
 The Unanimity Rule (UR): for any two alternatives x and y in X, x is socially preferred to y
according to the UR if all individuals in N prefer x to y. Formally:
x UR y if x i y for all i 2 N .
This rule can be viewed as a special case of a quali…ed majority rule in which the quali…ed majority is
the entire set N . The UR is the rule most commonly used in international treaties and organizations
such as the World Trade Organization, the Kyoto protocol, the E.U. In these cases, the individuals in
N are the representatives of the members countries, not the citizens themselves.
To illustrate how these PAR work, let us determine how they aggregate individual preferences when the
preference pro…le is given by (5). As explained in Remark 4, we have to determine the social preferences
between ANY possible pairs of alternatives for the MR, the 2/3MR, and the UR.
between a and b: all individuals prefer a over b. Therefore, a MR b, a 2=3M R b, a UR b, and
b M R a, b 2=3M R a, b U R a.
between a and c: two individuals prefer a over c while one prefers c over a. Therefore, a MR c,
a 2=3M R c, a U R c, and c M R a, c 2=3M R a, c U R a.
between b and c: two individuals prefer c over b while one prefers b over c. Therefore, c MR b,
c 2=3M R b, c U R b, and b M R c, b 2=3M R c, b U R c.
between any alternative x and itself: one can readily check that for all of the above PARs, x is never
socially preferred to x (which is reassuring!) because no individual prefer x to x (since individuals have
rational preferences).
An alternative way to represent these social preferences is to exhibit the corresponding matrix:
0 1 0 1
MR a b c MR a b c
B a : 1 1 C B a : 1 0 C
M MR = B @ b
C ; M 2=3M R = M M R ; and M U R = B C
0 : 0 A @ b 0 : 0 A
c 0 1 : c 0 0 :
There exists a wide variety of other PARs, some of them being more exotic or questionable than MR,
2/3MR, or UR. In the example below, I de…ne a few such PARs. They will be used to illustrate the properties
that a reasonable PAR should or should not satisfy.
Example 3 More examples of PAR:
Dictatorship of individual i (denoted D (i)): under the dictatorship of individual i, the social
preferences are exactly the preferences of individual i. Formally, for all x; y 2 X,
x D(i) y if x i y:
Hence, the social preferences depend only on the individual who is the dictator, the preferences of the
other individuals are not taken into account. It corresponds to an extreme concentration of power.
Unanimity rule with an arbitrary default (abbreviated URAD). We de…ne this rule only in
the case of two alternatives X = fa; bg. The unanimity rule with default a, U RAD (a), is de…ned as
follows. For all preferences pro…le ( i )i2N , a U RAD(a) a; b U RAD(a) b: Moreover, if for all i 2 N;
b i a; then b U RAD(a) a and a U RAD(a) b: Otherwise, a U RAD(a) b and b U RAD(a) a: In words,
if all individuals unanimously prefer b to a, then b is socially preferred to a. Otherwise the default
alternative a is socially preferred to b: Thus, this PAR does not treat alternative a and b symmetrically.
It is biased in favor of alternative a because a is deemed superior as soon as at least one individual
prefers it to b: Note that this bias is a bias of the PAR in the sense that it is not based on individual
preferences. The unanimity rule with default b, U RAD (b), is de…ned analogously.
This rule is used for instance in jury trials. In some legal systems, the defendant is innocent unless
all jurors agree that he is guilty. In this case, N is the set of jurors, X = fguilty; innocentg ; and the
default is innocent.

14
 The 1/3 Majority Rule (1/3MR): for any two alternatives x and y in X, x is socially preferred
to y according to the 1/3MR if at least 1/3 of the individuals in N prefer x to y. Formally:
N
x 1=3M R y if jfi 2 N : x i ygj :
3
Exercise 10 Suppose X = fa; b; cg, N = 3 and consider the following preference pro…le:
8
< a 1 b 1 c,
c 2 b 2 a,
:
a 3 b 3 c:
Determine the social preferences for that preference pro…le using MR, UR, D(2), and 1/3MR.
Solution. MR: Observe that individuals 1 and 3 have the same preferences, and 2 individuals out of 3
is a majority. Therefore, the majority preferences will be exactly the preferences of individual 1 and 3, and
are thus given by 0 1
MR a b c
B a : 1 1 C
M MR = B @ b
C:
0 : 1 A
c 0 0 :
2/3MR: When there are only three individuals, a 2/3 majority requires only two individuals. So with
three individuals, 2/3MR coincides with MR. However, note that this is not true in general when there are
more than three individuals!
UR: With the UR, to get x M R y for some x; y 2 X, we need all individuals to prefer x over y.
However, one can readily check that at the preference pro…le of this exercise, for any two alternatives, there
are two individuals who have opposite strict preferences. To see this, it su¢ ces to notice that individuals 1
and 2 rank the alternatives in opposite order. So in that case, the social preferences according to UR are
indi¤erent between any two alternatives. In terms of matrix representation, this means that
0 1
UR a b c
B a : 0 0 C
M UR = B @ b
C:
0 : 0 A
c 0 0 :
So UR is unable to discriminate between any two alternatives.
D(2): That case is easy: the preferences D(2) are simply given by 2, so
0 1
D(2) a b c
B a : 0 0 C
M D(2) = M 2 = B @
C:
b 1 : 0 A
c 1 1 :
1/3MR: When there are only three individuals, a 1/3 majority requires only one individual. So
0 1
1=3M R a b c
B a : 1 1 C
M 1=3M R = B@
C
b 1 : 1 A
c 1 1 :

6 Properties of preference aggregation rules

As the previous section illustrates, there exists a wide variety of PARs, some of which appear satisfactory
(see Example 2) whereas others have less appealing properties (see Example 3). Rather than using our
intuition to decide which PAR are satisfactory or not, our goal in this section is to formulate properties that
can discriminate between satisfactory and unsatisfactory PAR. Ideally, these properties should be su¢ ciently
demanding so as to rule out the questionable PAR in Example 3, but not too demanding so that there exists
some PAR that satis…es all of them.

15
6.1 When does a PAR lead to unambiguous collective decisions?
The following properties guarantee that a PAR leads to collective choices that are coherent and unambiguous.
De…nition 8 A PAR f is
rational if for all rational preference pro…les ( 1 ; ::; N ), the corresponding social preference relation
f ( 1 ; ::; N ) is rational (see De…nition 3).
resolute if for all resolute preference pro…les ( 1 ; ::; N ), the corresponding social preference relation
f ( 1 ; ::; N ) is resolute (see De…nition 4).
Hence, a PAR is rational if the social preferences inherit the rationality of the individual preferences
(recall that we have assumed that individual preferences are rational). From Proposition 1, this requirement
is necessary if we want the PAR to generate some stable collective decision. Hence, an irrational PAR can be
viewed as a form of “political failure” because if that PAR is used to decide whether the current collective
decision should stay in place or whether it should be replaced by another alternative, then this PAR will not
be able to covnerge on a collective decision that will never be replaced by some alternative.
Likewise, a PAR is resolute if the social preferences inherit the resoluteness of the individual preferences
(recall that we have assumed that individual preferences are resolute). Resoluteness is a desirable property
because from Proposition 2, if the social preferences are not resolute, there may be more than one stable
choice, in which case the PAR does not lead to an unambiguous collective choice: it is silent on which of the
stable alternatives should become the …nal collective choice. If you think of a PAR as a rule used to select
candidate in a popular election, being unable to pick a unique winner is problematic: how should ties be
resolved?
It is important to note that the requirements of rationality and resoluteness are quite di¤erent. Rationality
guarantees that there exists a stable collective decision, whereas resoluteness guarantees that there is only
one. Intuitively, having no stable collective decision is more problematic than having more than one, because
if there is none, the PAR does not lead to any coherent decision at all, whereas if there exists many, the PAR
can lead to a stable decision, the only problem being that which collective decision is eventually implemented
may involve some arbitrariness. Thus, rationality is an essential property of a PAR, while resoluteness is
only a desirable property.
Exercise 11 Suppose N = 3 and X = fa; bg. Determine whether MR, UR, URAD, D(i), and 1/3MR are
rational and/or resolute.
Solution. Resoluteness: Consider …rst the 1/3MR. It turns out that 1/3MR is resolute. To show that, we
need to show that for any two arbitrary distinct alternatives x; y 2 X, we have x 1=3M R y or y 1=3M R x:
Suppose that x 1=3M R y: This means that less than a third of the individuals prefer x to y: Since individual
preferences are resolute, those who do not prefer x to y must prefer y to x: Since they are at least 2/3 of the
population, this implies that y 1=3M R x: So 1/3MR is always resolute.
The UR is not resolute because whenever two individuals have opposite preferences between a and b (e.g.,
a 1 b and b 2 a), then a U R b and b U R a, so the UR preferences are not resolute.
Dictatorship is resolute because if all individuals have resolute preferences, the dictator also has resolute
preferences, and thus the dictatorship preferences are resolute.
Consider now the MR, and suppose that there is an odd number of individuals. Since individual preferences
are resolute, for any pair of alternatives x; y 2 X; any individual i prefers x to y or she prefers y to x: So
with an odd number of individuals, there cannot be exactly the same number of individuals who prefer x
to y and who prefer y to x: Therefore, there must be a strict majority of individuals who prefer x to y; in
which case x M R y; or a strict majority who prefer y to x; in which case y M R x:
When there is an even number of voters, the MR is not resolute. To see why, suppose N = 2k for some
integer k, and consider the following preference pro…le: k individuals prefer a to b whereas the other k
individuals prefer b to a. Then
N + 1;
jfi 2 N : a i bgj = k and k < so a MR b;
2
N +1
jfi 2 N : b i agj = k and k < ; so b MR a;
2

16
which shows the MR is not resolute.
URAD is resolute. This can be seen from the de…nition. Suppose for concreteness that the default is a.
Either all individuals prefer b to a, in which case b S a; or there is no such consensus and a S b. In both
cases, the social preferences are resolute.
Rationality: In the case of two alternatives X = fa; bg, there is only one possible cycle for a preference
relation: a b a: Thus, to determine whether the PARs of this exercise are rational, we only need to
determine whether one can …nd a preferences pro…le ( 1 ; :::; N ) over fa; bg such that the social preferences
for that preferences pro…le are a S b and b S a:
Consider …rst D(i). If a D(i) b; then by de…nition of D(i), this means that a i b: Since i is rational, this
means that b i a; and therefore that b D(i) a: Thus, there is no cycle, and D(i) is rational. Another way
to see that is that by construction, the social preferences according to D(i) are the preferences of individual
i, which are rational by assumption.
Consider next the MR. If a M R b; then by de…nition of the MR, a strict majority M of individuals
prefer a to b: Since these M individuals are rational, they do not prefer b to a: As a result, there cannot be a
majority of individuals who prefer b to a; so we cannot have both a M R b and b M R a at the same time,
which proves that MR is rational.
Consider then the UR. If a U M R b; then by de…nition of the UR, this means that for all individuals
i 2 N , a i b. Since these individuals have rational preferences, there is no individual i such that b i a.
Therefore, b U R a. So UR is rational. The reasoning for URAD is analogous.
Consider …nally the 1/3MR. It turns out that the 1/3MR is not rational. To see this, note that we can
always divide the population N into two groups M and P each of which has at least 1/3 of the population.
If all individuals in M prefer a to b and all individuals in P prefer b to a; then by de…nition of the 1/3MR,
a 1=3M R b and b 1=3M R a.

6.2 Democratic properties of a PAR

Even though the properties of rationality and resoluteness are important, the previous exercise shows that
they are satis…ed by PARs which are clearly not satisfactory from a democratic point of view, such as D(i),
URAD, or the 1/3MR. Thus, we need to introduce additional properties that guarantee that a PAR is
su¢ ciently “democratic”.

De…nition 9 A PAR is

anonymous if it treats all individuals symmetrically. Formally, for any preference pro…le ( 1 ; ::; N ),
if we change ( 1 ; ::; N ) by permuting the preferences of two or more individuals, then the social
preferences should remain the same.
neutral if it treats all alternatives symmetrically. Formally, for any preferences pro…le ( 1 ; ::; N ) and
any two alternatives x; y 2 X; if ( 01 ; ::; 0N ) di¤ ers from ( 1 ; ::; N ) only in that the positions of x
and y have been permuted in the preferences of all individuals, and if S and 0S denote the social
preferences for the preference pro…les ( 1 ; ::; N ) and ( 01 ; ::; 0N ) ; respectively, then 0S must di¤ er
from S only in that the role of x and y have been permuted. If we represent the social preferences in
their matrix forms, this means that the matrix of 0S must be equal to the matrix of S in which we
have permuted the lines and the columns of x and y:

Anonymity means that all individuals have the same in‡uence on the social preferences. To be more
concrete, suppose we aggregate the preferences of Pierre, Paul, and Jacques. If, say, Paul reports the
preferences of Pierre, Pierre reports the preferences of Jacques, and Jacques reports the preferences of
Pierre, the social preferences should not change. The name of each individual does not matter.
Neutrality means that the PAR treats all alternatives impartially. They are ranked solely on the basis of
individual preferences. In the case of two alternatives, this property means the following: if we reverse the
preferences of all individuals (that is, those who prefer a to b now prefer b to a and vise versa), the social
preferences should also be reversed. In other words, the name of the alternatives should not matter.
If the collective choice under consideration is a popular referendum (and X are the two options on which
voters have to make a choice), anonymity means that all citizens have the right to vote and all votes have

17
the same weight, and neutrality means that none of the two alternatives on the ballot is given a special
treatment.
The following exercise illustrates the properties in De…nition 9.
Exercise 12 Suppose X = fa; bg : Determine whether MR, UR, D(1), and URAD (see Examples 2 and 3)
are neutral and/or anonymous.
Solution. Neutrality: An easy way to check whether neutrality is satis…ed by a PAR is to check whether
the de…nition of the PAR treats two di¤erent pairs of alternatives di¤erently. For instance, from the de…nition
of MR (see Example 2), when comparing two alternatives x and y, the MR counts the votes between x and
y exactly in the same way irrespective of whether (x; y) = (a; b) or (x; y) = (b; a). Hence, the MR is neutral.
For the same reason, the 1/3MR and the UR are neutral. D(i) is also neutral because any two pairs of
alternatives are compared in the same way: according to the dictator’s preferences.
On the contrary, URAD is not neutral. It is easy to see from its de…nition that the alternative that is the
default is treated di¤erently from the other alternative. To see this formally, suppose f is URAD with default
a: Then if the preference pro…le is a 1 b and b 2 a; the social preferences are such that a S b: Suppose
now that we switch the role of a and b in the individual preferences. The preference pro…le becomes b 1 a
and a 2 b: For this preference pro…le, the social preferences according to URAD with default a are such
that a S b: So the role of a and b are not reversed in the social preferences, a violation of neutrality.
Anonymity: A PAR is anonymous if it treats any two individuals in a symmetric way. For instance,
from the de…nition in Example 2, we see that the MR gives the same weight to any two individuals: no
individual is given a special treatment, the MR simply counts the number of individuals who prefer one
alternative against another. Hence, the MR is anonymous.
For the same reason, the 1/3MR is anonymous, the UR is anonymous (the notion of unanimity is anonymous),
and the URAD is also anonymous (this PAR treats alternatives di¤erently, but it treats all individuals
equally).
However, dictatorship is obviously not anonymous: the dictator has all the decision power while the other
individuals are ignored. To see this formally, suppose f is D (1) : Then if the preference pro…le is a 1 b
and b 2 a; the social preferences are such that a D(1) b: Suppose now that we switch the role of 1 and 2
in the individual preferences. The preference pro…le becomes b 1 a and a 2 b; and the social preferences
according to D (1) are such that a D(1) b: So by changing the role of 1 and 2 in the preferences pro…le, we
have changed social preferences, a violation of anonymity.
Let us summarize what we have learned from Exercises 11 and 12:
The unanimity rule is rational, anonymous, and neutral, but not resolute (and the same is true of
2/3MR, and of any supermajority rule more generally). The violation of resoluteness captures the fact
that requiring unanimity to discriminate between two alternatives is too demanding. Most of the time,
the UR will be unable to decide whether one alternative is socially superior to another (because in a
large and heterogeneous population, for any two alternatives x and y, there will always be someone
who prefers x and another one who prefers y).
The unanimity rule with an arbitrary default is rational, resolute, and anonymous, but not neutral.
Thus, URAD breaks the numerous ties that occur under UR by giving a preferential treatment to a
particular alternative. Thus, it solves the problem of resoluteness of the UR, but it does at the cost of
violating neutrality. In practice, in institutions that take decision by unanimity, the default choice is
simply the status quo that in place at the time of the collective decision. So a preferential treatment
is given to the current status quo, and the violation of neutrality takes the form of status quo inertia.
The dictatorship is rational, resolute, and neutral, but not anonymous. Thus, unlike to UR and URAD,
dictatorship is able to satisfy resoluteness and neutrality at the same time, but it does so by violating
anonymity. The violation of anonymity captures the fact that dictatorship takes only into account the
preferences of the dictators and ignores the other individuals.
The 1/3 majority rule is anonymous, neutral, and resolute but not rational. The 1/3MR violates
rationality because it requires less than a majority of approval to decide whether one alternative is
socially superior to another. Relying on the approval of less than a majorityis not enough to guarantee
the existence of stable collective decisions.

18
 In contrast to the above PARs, the majority rule is rational, resolute, anonymous, and neutral.

Thus, among the PARs we have seen so far, the majority rule is the only rule that satis…es all the desirable
properties we have introduced in De…nitions 8 and 9. Thus, these properties are demanding enough to rule
out undesirable PARs such as UR, URAD, D(i), or 1/3MR, but not too demanding in the sense that there
is at least one PAR that satisfy all of them. This …nding raises two questions. First, are there other PARs
on top of MR that satisfy all of these properties? We address that question in the next section. Second, are
all of the properties in De…nitions 8 and 9 necessary to get rid of unacceptable PARs?
The answer to the latter question is yes. To see why, observe that the UR satis…es all these properties
except resoluteness, URAD satis…es all these propertie except neutrality, dictatorship satis…es all these
propertie except anonymity, and 1/3 majority rule satisfy all of them except rationality. So these four
requirements are necessary to rule out unsatisfactory PARs such that UR, URAD, dictatorship, or minority
rules.

6.3 More exercises on PARs

Exercise 13 Suppose X = f 1; 0; 1g and N = 3. Determine, for the MR, the UR, D(1) and for the 1/3MR,
the social preferences for the preference pro…le derived from the following utility functions: for all x 2 X,
2
U1 (x) = (x + 1) ;
2
U2 (x) = (x 0:1) ;
2
U3 (x) = (x 2) :

Solution. The preference pro…le derived from this pro…le of utility function is
8
< 1 1 0 1 1;
0 2 1 2 1;
:
1 3 0 3 1:

The corresponding social preferences for the MR, the UR, D(1) and for the 1/3MR are
0 1 0 1
MR 1 0 1 1=3M R 1 0 1
B 1 : 0 0 C B 1 : 1 1 C
M MR = B @ 0
C , M 1=3M R = B C
1 : 1 A @ 0 1 : 1 A
1 1 0 : 1 1 1 :
0 1 0 1
UR 1 0 1 D(1) 1 0 1
B 1 : 0 0 C B 1 : 1 1 C
M UR = B @ 0
C , M D(1) = B C
0 : 0 A @ 0 0 : 1 A
1 0 0 : 1 0 0 :

Exercise 14 Check whether the 2/3MR satis…es anonymity, neutrality, and resoluteness. In each case,
provide a short argument or a counter example.

Solution. The 2/3MR satis…es anonymity because in the de…nition of 2/3MR, any two individuals are
treated in the same way (only the number of people who prefer x over y matters, not the identity of these
people).
It is neutral because any two alternatives are treated in the same way in the de…nition (the 2/3 majority
requirement is the same for any pair of alternatives x and y).
With three voters, 2/3MR is equivalent to the MR, so it is resolute. However, with more than three
voters, it is not resolute. To see this consider a preference pro…le in which half of the individuals (or N2+1
is N is odd) prefer x to y and the other half prefer y to x: In that case, x 2=3M R y and y 2=3M R x, so
2/3MR is not resolute.

19
Exercise 15 (This exercise is important!) N is composed of two individuals, X is composed of two al-
ternatives fa; bg : Consider the PAR de…ned by: a S b if and only if a 1 b; and b S a if and only if
b 2 a: Of the four properties in May’s theorem, which one are satis…ed and which one are violated. Provide
a justi…cation for the property(ies) that is (are) violated.
Solution. Neutrality, anonymity, resoluteness and rationality are all violated.
Justi…cation for violation of rationality: consider the preference pro…le
a 1 b
(6)
b 2 a
Social preferences are a S b S a:
Justi…cation for violation of anonymity: consider the preference pro…le (6), and swap individual 1 and 2; we
obtain
b 1a
; (7)
a 2b
we obtain a S b and b S a; which di¤er from social preferences for the preference pro…le (6).
Justi…cation for violation of neutrality: consider the preference pro…le (6), and swap alternatives a and b;
we obtain (7). The social preferences in the …rst case are given by a S b S a and in the second case, they
are given by a S b and b S a; so the role of a and b has not been swapped in the social preferences.
Justi…cation for violation of resoluteness: the social preferences for the preference pro…le (7) are such that
a S b and b S a; which violates resoluteness.
Exercise 16 N is a …nite set of individuals, and X is a …nite set of alternatives from which the group of
individuals N has to choose.
1. Consider the following preference aggregation rule: for all distinct x; y 2 X; x S y: Of the four
properties in May’s theorem, which one are satis…ed and which one are violated. Provide a justi…cation
for the property(ies) that is (are) violated.
2. Answer the same question for the following preference aggregation rule: for all distinct x; y 2 X;
x S y:
Solution.
1. Only rationality is violated. Justi…cation for violation of rationality: For any two distinct alternatives
x; y 2 X; by de…nition of the PAR, x S y S x; a cycle.
2. Only resoluteness is violated. Justi…cation for violation of resoluteness (3pts): for any two distinct
alternatives x; y 2 X; by de…nition of the PAR, x y and y x:

Exercise 17 Let f be a PAR for the set of individuals N = f1; 2; 3g over the set of alternatives X = fa; b; cg.
You do not know what that PAR f is. In the following questions, I will give you some partial information
about that PAR, and I will ask to you to guess, given that information, what PAR f could or could not be.
A) Suppose that f is such that for the preference pro…le ( 1 ; 2 ; 3) given by
8
< c 1b 1a
b 2a 2c ; (8)
:
a 3b 3c
the social preferences f ( 1; 2; 3) (denoted simply by S) are given by the following matrix:
0 1
S a b c
B a 0 0 1 C
B C: (9)
@ b 1 0 1 A
c 0 0 0
Given the information you have about f , answer the following questions:

20
 – Can f be dictatorial? If yes, who could be the dictator?
– Can f be the majority rule?
– Can f be the unanimity rule?
– Can f be something else?
0 0 0
B) Suppose further that for the preference pro…le ( 1; 2; 3) given by
8 0 0
< c 1 b 1 a
0 0
b 2 a 2 c ; (10)
: 0 0
c 3 b 3 a

the social preferences 0S (computed using the same preference aggregation rule f as in part A) are
given by the following matrix: 0 0 1
S a b c
B a 0 0 0 C
B C
@ b 1 0 0 A: (11)
c 1 1 0
Given the information you have about f (from part A AND from part B), answer the same questions
as in part A.
C) Given the information you have about f (from part A AND from part B), can you tell whether f is
rational?

Solution.

A) Dictatorship: if f is dictatorial with i as a dictator, by de…nition, S and i must be the same, which
means that for all x; y 2 X, we have x S y whenever x i y and x S y whenever x i y. In this
case, we see that S and 2 are the same, so f could be the dictatorship of player 2. Individual 1
cannot be the dictator, because b 1 c but b S c. Likewise, individual 3 cannot be the dictator,
because b 3 a but b S a.
MR: one can readily check that given the preference pro…le in (8), the majority preferences M R are
the same as the preferences S given in (9). So f could be the MR.
UR: player 1 prefer c to b, so not all individuals prefer b to c. Therefore, if f was the UR, we should
have b S c. So f cannot be the UR.
To sum up, f could be the dictatorship of player 2, or the majority rule. Notice that f could also be
something completely di¤erent. For instance f could be the following PAR, which we will denote f 0 :
f 0 is the dictatorship of player 2 when the preference pro…le is given by (8), and it is the dictatorship
of player 3 for all other preference pro…les. That PAR would be consistent with the information given
in question A, but it is di¤erent from MR, and from the dictatorship of player 2.
B) We already know from question A that f cannot be the dictatorship of player 1 or 3, and it cannot be
the UR. The remaining question is: given the additional information that is given in question B, can
f still be dictatorship of player 2, or the majority rule?
Dictatorship of player 2: The social preferences are di¤erent from the preferences of player 2 (for
instance, c 02 b but c 0S b) so f cannot be the dictatorship of player 2.
MR: one can readily check that given the preference pro…le given in (10), the majority preferences
0 0
M R are the same the preferences S given in (11). So f could be the majority rule.
Notice again that f could also be something else. For instance f could be the PAR f 0 de…ned in the
answer to question A. One can easily check that PAR would be consistent with the information given
in question A and question B, but it is di¤erent from MR.
C) Given all the information we have, one cannot tell whether f is rational. One can easily check that the
two social preferences given in (9) and in (11) are rational, but for f to be rational, it must be that
the social preferences are rational FOR ALL pro…le of rational individual preferences, not only for the
preference pro…le given in (8) and (10), but we do not know the social preferences for other preferences

21
 pro…le, so we cannot tell whether f is rational. For instance, if f is the MR, it would not be rational
(because of the Condorcet paradox), but if f is the PAR f 0 de…ned in the answer to question A, it is
rational (I leave that to you as an exercise).

7 What is special about the majority rule? May’s theorem

Throughout this section, we restrict attention to collective choice problems in which individuals have to
choose between only two alternatives, i.e., X = fa; bg : We further assume that the number of individuals
N is odd. The assumption that N is odd is for technical convenience (similar results can be derived for an
even number of individuals). The crucial assumption is the restriction to two alternatives. This assumption
is clearly restrictive, but it corresponds to important real-world situations. In particular, in most popular
referenda, there are only two alternatives to choose from, and citizens can cast their vote only for one of
these two alternatives. Likewise, general elections with two candidates are quite common as well.

Theorem 1 (May, 1952) When X = fa; bg and N is odd, the majority rule is rational, resolute, anony-
mous, and neutral. Moreover, it is the only PAR that satis…es these four properties together with an additional
minor technical property.7

Proof. (this proof is given only for your information, you do not need to memorize it) We already know
from Exercise 12 that MR satis…es these properties. Let us now show that this is the only one. Suppose
that a PAR f satis…es these 4 properties together with positive responsiveness. We now prove that f must
be the MR. Step 1 (Sketch): Since f is anonymous, the social preferences according to f can only depend
on the number of individuals such that a i b (since individual preferences are resolute, this number pins
down the number of individuals such that b i a). Since f is positively responsive, it must be that a S b
whenever the number of individuals such that a i b is greater or equal than some threshold Ta ; and that
b S a whenever this number is smaller or equal than a certain threshold Tb , where Ta ; 2 f0; :::; N + 1g :
Step 2: Suppose Ta individuals prefer a over b (and thus that N Ta individuals prefer b over a). By
de…nition of Ta , we have a S b. Since f is rational, we cannot have a S b and b S a at the same time
(see Section 4.2), which means that Ta > Tb .
Step 3: Since f is neutral, Ta = N Tb , so from step 2, Ta > N=2.
Step 4: Suppose N2+1 individuals prefer a over b (and thus that N 2 1 individuals prefer b over a). Since
f is resolute,either a S b, so N2+1 Ta or b S a, so N 2 1 Tb . Since Ta = N Tb (Step 3), in the latter
N +1
case, we also have that Ta 2 :
N +1
Step 5: from step 3 and 4, N=2 < Ta 2 . Since Ta is an integer and N is odd, this means that
N +1
T = 2 . So the PAR is the MR.
The …rst part of May’s theorem states that the MR is optimal both from a decision making perspective—
as a rational and resolute PAR, it always leads to a unique collective decision— and from a democratic
perspective— anonimity implies that the MR aggregates individual preferences in a fair way, and neutrality
implies that it treats all alternatives in an impartial way. The second part of May’s theorem states that the
MR is the only PAR that satis…es these properties. Thus, May’s theorem captures what is special about the
majority rule, and why so widely used in democratic decision processes.

8 What is wrong with majority rule? The Condorcet paradox

Theorem 1 states that the MR is the only PAR that is optimal both from a decision making perspective— in
the sense that it always leads to a unique collective decision— and from a democratic perspective— in the
7 This property is called “positive responsiveness”. I decided not to mention it in these notes because it does not play

a fundamental role, and it confuses many students. It basically guarantees that switching the preferences of an individual
between a and b cannot switch the social preferences in the reverse direction. That is, it cannot be the case that by changing
the preferences of an individual from a i b to b i a we change the social preferences from b S a to a S b: All the PAR we
have seen so far satisfy this very innocuous requirement.

22
sense that it treats all individuals and all alternatives in a fair way. Why don’t we decide everything via
majority rule then? As the next exercise illustrates, an important shortcoming of the MR is that for more
than two alternatives, it is still anonymous, neutral, and resolute (it is easy to extend the justi…cations in
Exercise 12 to more than two alternatives), but it is not rational anymore.
Exercise 18 (The Condorcet paradox) Consider the following preference pro…le for N = f1; 2; 3g and
X = fa; b; cg: 8
< a 1 b 1 c;
b 2 c 2 a;
:
c 3 a 3 b:
What are the social preferences derived by the MR from this preference pro…le? For that preference pro…le,
are the social preferences rational (see de…nition 3)? What is the alternative that is majority preferred by
this group of individual?
Solution. For three individuals, any group of 2 individuals or more constitutes a majority. Individuals 1
and 3 prefer a to b, so a M R b. Individuals 1 and 2 prefer b to c, so b M R c. Individuals 2 and 3 prefer c
to a, so c M R a. Summing up, we have shown that
a MR b, b MR c, and c MR a.
Therefore, the social preferences are not rational. As a result, there is no socially most preferred alternative:
a is socially dominated by c, c is socially dominated by b, and b is socially dominated by a.
Exercise 18 is the famous Condorcet paradox. It shows that even if all individuals have rational pref-
erences, the majority preferences might not be rational. Using De…nition 8, the Condorcet paradox can be
restated as follows:
Theorem 2 When jXj > 2, the MR is not rational.
As argued in Section 6.1, this is a serious problem if we want to use majority rule as a guide for collective
decision making. Hence, the Condorcet paradox can be viewed as a form of political failure.
Exercise 19 (This exercise is important!) Suppose that the set of alternatives is X = fa; b; c; dg, the set of
individuals is N = f1; 2; 3g, and the preference pro…le is
8
< b 1 a 1 c 1 d;
d 2 c 2 b 2 a;
:
c 3 a 3 d 3 b:
1. Suppose that we let individuals choose some x 2 X via majority rule. What are the social preferences?
Are they rational? Given the social preferences, what prediction can you make about the collective
decision(s)?
2. Answer question 1 for the following preference pro…le:
8
< b 1a 1c 1 d;
d 2c 2b 2 a;
:
a 3d 3b 3 c:

Solution.
1. The social preferences are 0 1
MR a b c d
B a : 0 0 1 C
B C
B b 1 : 0 0 C
B C:
@ c 1 1 : 1 A
d 0 1 0 :
They are not rational because a M R d M R b M R a: Alternative c is the only socially most preferred
alternative, so it is the most likely outcome of the collective decision process.

23
 2. The social preferences are 0 1
MR a b c d
B a : 0 1 1 C
B C
B b 1 : 1 0 C
B C:
@ c 0 0 : 0 A
d 0 1 1 :
They are not rational because a M R d M R b M R a: There is no socially most preferred alternative,
but we can see that c is the only socially least preferred alternatives, so c is unlikely to be the outcome
of the collective choice problem, but the other three alternatives are equally likely.

Exercise 20 Describe in at most two sentences the Condorcet paradox.

Solution. The Condorcet paradox asserts that when there are more than two alternatives, majority
preferences can be irrational even though individual preferences are perfectly rational. In particular, there
might not exists a stable choice according to the MR.

Exercise 21 Show that for N = 4 and X = fa; b; c; dg, the 2/3MR is not rational (hint: You have to …nd
a preference pro…le ( 1 ; ::; N ) such that the social preferences f ( 1 ; ::; N ) according to 2/3MR are not
rational. One possibility is to extend the pro…le in Exercise 18 to 4 individuals and 4 alternatives.)

Solution. For N = 4, x is socially preferred to y according to the 2/3MR if 3 or more individuals prefer x
to y. Consider the following preference pro…le
8
>
> a 1 b 1 c 1 d,
<
b 2 c 2 d 2 a,
>
> c 3 d 3 a 3 b,
:
d 4 a 4 b 4 c:

One can readily check that at this preference pro…le, a 2=3M R b 2=3M R c 2=3M R d 2=3M R a, as needed.

9 The tension between rationality and democracy: a generaliza-

tion
May’s theorem shows that with two alternatives, the majority rule is arguably the most decisive and de-
mocratic way to aggregate preferences. Unfortunately, the Condorcet paradox tempers our enthusiasm for
the majority rule: when individuals have to choose between more than two alternatives, the majority pref-
erences can be irrational. As a result, the MR can lead to irrational collective decisions. In this section, we
investigate whether there exists reasonable PARs which are not subject to the Condorcet paradox.

Exercise 22 Show that for any X; dictatorship and UR are rational PARs.

Proof. By assumption, the preferences of any indivudal i are rational. Therefore, the social preferences
according to the dictatorship of i are rational.
To show that UR is rational, let m 2, let ( 1 ; :::; N ) be a preference pro…le, and let U R denote the
corresponding social preferences under unanimity rule. Suppose by contradiction that U R are not rational.
Then there exists x1 ; x2 ; :::; xm 2 X such that x1 U R x2 U R ::: U R xm U R x1 . By de…nition of UR,
this means that for all individuals i 2 N , x1 i x2 i ::: i xm i x1 . This is impossible since individual
preferences are rational, a contradiction.
Thus, dictatorship and the unanimity rule are always rational irrespective of the number of alternatives.
However, we know from Section 7 that these PARs are not as virtuous as the MR, either in terns of decision-
making capabilities or in terms of democratic properties. Nevertheless, it is instructive to try and understand
better what is wrong with these two PARs.

24
 Suppose that an individual prefers x to y, and all others prefer y to x. If that individual is the dictator,
then she will impose her preferences on the rest of society. With the unanimity rule, an analogous problem
arises. That individual is able to prevent y to be socially preferred to x; even if that is the opinion of an
overwhelming majority of individuals. In other words, dictatorship gives the power to one individual to
dictate in which direction the social preferences should go, whereas UR gives the power to any individual
to unilaterally prevent the social preferences to go in a particular direction, which is arguably at odd with
the democratic ideal. If x is the prevailing status quo at the time of the collective decision, neither the
dictatorship nor the UR will not approve a change of policy from x to y.
However, the UR and the dictatorships are not the only alternatives to the MR. There exists PARs which
are not as democratic as the MR but stil more democratic than the dictatorship of the UR, for instance,
the 2/3MR and all other quali…ed majority rules. A natural question is then whether one can …nd a PAR
that is “reasonably democratic” and rational at the same time. The following de…nition formalizes what we
mean by “reasonably democratic”.

De…nition 10 For a given PAR f; an individual i is said to have veto power at some preference pro…le
( i )i2N if there exists x; y 2 X such that x i y; for all j 6= i; y j x; and y S x.
A PAR f is minimally democratic if for all preference pro…les, no individual has a veto power.

In words, a PAR is said to be minimally democratic if no single individual can ever prevent the social
preferences to follow the unanimous will of all other individuals. From the preceding discussion, it should
be clear that this requirement is violated by the dictatorships and the UR. It is worth noting that this
requirement is very mild: De…nition 10 only requires that no single individual should prevail over the rest of
society, but it allows for groups of size greater than 1 to have a veto power. The following exercise formalizes
the latter claim.

Exercise 23 Suppose N 3: For all M 2 N2+1 ; :::; N , consider the PAR called supermajority rule with
majority threshold M , which we de…ne as follows: for all x; y 2 X; x S y if at least M individuals prefer
x to y: Note that MR corresponds to M = N2+1 whereas UR corresponds to M = N: For what value of M is
that PAR minimally democratic?

Solution. Let fM denote the PAR de…ned in the exercise. We already know that for M = N; fM
coincides with the UR, which is not minimally democratic. Let us show that when M < N; fM is minimally
democratic. Consider a pro…le of preferences such that for some x; y 2 X and some i 2 N; x i y and for
all j 6= i; y j x: Since M < N; by de…nition of the PAR, y fM x: From De…nition 10, this means that i
does not have veto power, so fM is minimally democratic.
Thus, the question that we ask in this section can be reformulated as follows. Is there a PAR that is
rational and minimally democratic? Unfortunately, the next theorem shows that (when there is a large
number of alternatives) such a PAR does not exist. Hence, even the mildest democratic requirement is
incompatible with the requirement of collective rationality!

Theorem 3 (Brown, 1975) Suppose there is at least as many alternatives as the number of individuals
(jXj N ). Then there is no PAR that is rational and minimally democratic.

Proof. Suppose by contradiction that f is minimally democratic and rational . To obtain a contradiction,
we need to exhibit a preference pro…le which generates a cycle in the social preferences (see De…nition 3).
Since jXj N , there exists N distinct alternatives x1 ; x2 ; ::; xN 2 X. Consider the following (rational)
preference pro…le on fx1 ; x2 ; ::; xN g (we do not need to specify the individual preferences over the other
alternatives for the argument below)
8
>
> x1 1 x2 1 x3 1 ... 1 xN 2 1 xN 1 1 xN ,
>
>
>
> x2 2 x3 2 x4 2 ::: 2 xN 1 2 xN 2 x1 ,
<
x3 3 x4 3 x5 3 ::: 3 xN 3 x1 3 x2 ,
>
> :::
>
>
>
> x N 1 N 1 x N N 1 x1 N 1 ::: N 1 xN 4 N 1 xN 3 N 1 xN 2 ,
:
xN N x1 N x2 N ::: N xN 3 N xN 2 N xN 1 .

25
Observe that all individuals but 1 prefer xN to x1 . Since f is minimally democratic, we must have xN S x1 .
Likewise, all individuals but 2 prefer x1 over x2 . Since f is minimally democratic and rational, we must have
x1 S x2 . With a similar reasoning, we must have that x1 S x2 S :::xN S x1 , as needed.

Exercise 24 Determine which individuals have a veto power in the following PAR: MR, 2/3MR, UR, D(1)
(see De…nition 10, and Examples 2 and 3).

Solution. With the MR, no individual has a veto power: if some individual j prefers y over x but everybody
else prefer x over y, the majority prefers x over y, so x M R y.
With the UR, every individual has a veto power: if some individual j prefers y over x but everybody else
prefer x over y, there is no unanimity between x and y, so x U R y.
With 2/3MR, no individual has a veto power: if some individual j prefers y over x and everybody else
prefers x over y, then there is a 2/3 majority in favor of x, so x 2=3M R y.
With D(1), the social preferences are exactly the preferences of the dictator, so the dictator has a veto
power, and the other individuals do not.

Exercise 25 (Harder, optional) Show that with three individuals and three alternatives, there is no PAR
that is rational resolute, neutral, and anonymous. (Hint: suppose that f satis…es all of the above properties,
consider the preference pro…le of Exercise 18, and observe that this preference pro…le is the same if one
permutes the role of the individuals and the alternatives as follows: (a; b; c) ! (b; c; a) and (1; 2; 3) ! (2; 3; 1) ;
use then anonymity and neutrality to show that at this preference pro…le, necessarily, for all x; y 2 X; x y).

Solution. Consider the preference pro…le of Exercise 18. A neutral, anonymous and rational PAR should
be socially indi¤erent between a, b, and c ((in the sense that for all x; y 2 X, x S y; so for all x; y 2 X;
x S y). The reader can check that by permuting the role of the alternatives as follows: (a; b; c) ! (b; c; a)
and then permuting the role of the individuals as follows: (1; 2; 3) ! (2; 3; 1), we obtain the exact same
preference pro…le. To see that, let ( 1 ; 2 ; 3 ) be the preference pro…le in Exercise 18, let ( 01 ; 02 ; 03 ) be the
preference pro…le obtained after permuting the role of the alternatives in ( 1 ; 2 ; 3 ), and let ( 001 ; 002 ; 003 )
be the preference pro…le obtained after permuting the role of the alternatives in ( 01 ; 02 ; 03 ), then we obtain:
8 8 8
< a 1 b 1 c; < b 01 c 01 a; < a 001 b 001 c;
0 0
b 2 c 2 a; , c 2 a 2 b; , and b 002 c 002 a; ;
: : 0 0 :
c 3 a 3 b: a 3 b 3 c: c 003 a 003 b:

so we see that ( 1 ; 2 ; 3 ) = ( 001 ; 002 ; 003 ).

Let us now show that the social preferences corresponding to these preferences pro…les must all be the
same. To see this, recall that ( 001 ; 002 ; 003 ) is obtained from ( 01 ; 02 ; 03 ) by permuting the individuals in
( 1 ; 2 ; 3 ) as follows: (1; 2; 3) ! (2; 3; 1). So by anonymity, the social preferences 00S at the preference
pro…le ( 001 ; 002 ; 003 ) should be the same as the social preferences 0S at the preference pro…le ( 01 ; 02 ; 03 )
by permuting the alternatives in the same way. Moreover, we know that ( 001 ; 002 ; 003 ) = ( 1 ; 2 ; 3 ), so 00S
must be equal to S .
Let us further show that 00S is obtained from S by permuting the role of the alternatives as follows:
(a; b; c) ! (b; c; a) : To see this, recall that ( 01 ; 02 ; 03 ) is obtained from ( 1 ; 2 ; 3 ) by permuting the
alternatives in ( 1 ; 2 ; 3 ) as indicated above. So by neutrality, the social preferences 0S at the preference
pro…le ( 01 ; 02 ; 03 ) should be obtained from the social preferences S at the preference pro…le ( 1 ; 2 ; 3 )
by permuting the alternatives in the same way.
Finally, we will show that for any two distinct alternatives x; y 2 X, necessarily, x y. Suppose by
contradiction that a S b, for instance. From what precedes, 00S is obtained from S by permuting the
role of the alternatives as follows: (a; b; c) ! (b; c; a) : Therefore, a S b implies that b 00S c. But since
00 00
S = S , we must have b S c. Since S is obtained from S by permuting the role of the alternatives as
follows: (a; b; c) ! (b; c; a) ; b S c implies in turn that c 00S a, but since 00S = S , we must have c S a.
Hence, we have shown that if a S b, necessarily, b S c and c S a. This cycle is in contradiction with the
requirement of rationality of the PAR. So the initial hypothesis a S b must be false, so a S b. One can
show with the same reasoning that for any two distinct alternatives x; y 2 X, rationality implies that x y.
This means in turn that the PAR violates resoluteness.
Since it is rational, the PAR must be indi¤erent between all alternatives at that preference pro…le.

26
10 The main take-away from this chapter so far
If we combine May’s theorem, The Condorcet Paradox, and Brown’s theorem (which is basically a simpler
version of Arrow’s famous impossibility theorem), we obtain that individual preferences can be aggregated
in a rational and democratic way only when there is a very small number of alternatives. These results can
be viewed as a form of "political failure". They show that even if we abstract away from he problem of
delegating decisions to uninformed, incompetent, or corrupted representatives, even if we assume that all
individuals perfectly understand the collective decision under consideration and know what is good for them,
and even if we assume that these individuals report their preferences truthfully, it is nevertheless impossible
to make collective decisions in a coherent and democratic way.

11 Some famous preference aggregation rules

11.1 Electoral rules
In this section, we use the tools introduced in this Chapter to study two important and widely used electoral
rules, namely the plurality rule (or …rst-pass-the-post), and the run-o¤ rule. These rules, of variants of them
are often used in presidential elections (e.g., U.S., France). Variants of the plurality rule are often used for
presidential elections (e.g., in the US).
Under plurality rule, the winner of the election is the candidate that receives the most votes (which
may or may not be a majority of votes). This electoral rule does not immediately de…ne a PAR, because
an electoral rule aggregates individual votes whereas a PAR aggregates individual preferences. However,
as a …rst approximation, we can interpret a vote by individual i 2 N for candidate x 2 X in terms of i’s
preferences as follows: the most preferred candidate of voter i is x: In a sense, we are implicitly assuming
that voters vote “truthfully”— if time permits, we will discuss what happens when we allow voters to vote
strategically, instead of sincerely. Under the truthful voting assumption, the PR can be de…ned as a PAR as
follows.

De…nition 11 (Plurality Rule) For any two candidates x; y 2 X; x is socially preferred to y according to
the PR if x is the most preferred candidate of more voters than y: Formally, for any x 2 X; the plurality
score of x; denoted by p (x) ; is the number of individuals whose most preferred candidate is x. Then for all
x; y 2 X; x P R y if p (x) > p (y) :

The following exercise studies the most important properties of PR.

Exercise 26

1. Suppose …rst that there are two candidates, X = fa; bg ; and N is odd (as in May’s theorem). Is PR
Anonymous? Neutral? Resolute? Rational? How can you reconcile those results with May’s theorem
(see Section 7)?
2. Suppose now that we have three candidates and seven voters, i.e., X = fL; M; Rg, N = 7; and the
following preference pro…le:
8
< for i = 1; 2; 3; L i M i R; (leftist voters)
for i = 4; 5; M i R i L; (centrist voters)
:
for i = 6; 7; R i M i L (rightist voters)

Derive the social preferences according to PR.

3. Are the PR and the MR equivalent?
4. Is PR rational when X contains more than 2 alternatives?

Solution. The solution to this exercise will be given in class.

The previous exercise shows that according to the properties we have seen in Section 7, then PR is as
virtuous as MR when comparing 2 alternatives. Moreover, unlike the MR, the PR satis…es all these properties

27
even when there is more than 2 alternatives.8 Why don’t we use the PR in all collective choice problems
then? The following exercise address that question and test the limits of PR by looking at environments of
particular relevance in which PR leads to questionable collective decisions.

Exercise 27 Consider the following two collective choice problems.

1. N individuals have to choose how to share a pie (say, how to spend the public budget). For simplicity,
we consider only the following divisions of the pie: the pie can be allocated entirely to individual 1;
entirely to individual 2; ... entirely to individual N; or it can be shared equally across all individuals.
Thus, we consider N + 1 possible alternatives. Each individual cares only about his share of the pie,
so the most preferred allocation of each individual i is to have all the pie for herself, his second most
preferred allocation is to have the pie shared equally, and these two allocations are preferred to all other
allocations (with the PR, we do no need to specify how individual i ranks these other allocations).

(a) What are the social preferences according to PR and according to MR?
(b) Does PR rank alternatives in a satisfactory way in this environment? Explain intuitively why.
(c) Consider the following two alternatives: individual i gets all the pie, and equal sharing of the
pie. What are the individual and the social preferences (according to PR) between these two
alternatives?
(d) Reformulate the …nding of the previous question using the notions introduced in Section 9.
(e) Explain how one can use the results of Exercise 26 together with the results in Section 9 to derive
the same conclusion as in the previous question.

2. The following collective choice problem is inspired by the 2000 US presidential elections: voters have
to elect a president among three candidates X = fBush; Gore; N aderg (ranked rom most conservative
to most liberal). Assume for the sake of this exercise that voters’ preferences are as follows:9
8
< for 49% of the voters, Bush i Gore i N ader; (conservative voters)
for 48% of the voters, Gore i N ader i Bush; (moderate liberal voters) :
:
for 3% of the voters, N ader i Gore i Bush; (liberal voters)

(a) Who is the socially most preferred candidate according to PR? According to MR? Does this pref-
erence pro…le generate a Condorcet paradox?
(b) Explain intuitively why the PR and the MR lead to di¤ erent outcomes.

Solution. The solution to this exercise will be given in class.

To avoid the shortcomings of the PR highlighted in Exercise 27, some presidential democracies (e.g.,
France) use an alternative electoral rule called “run-o¤”. In the …rst round, the two candidates with the
most votes are retained. In the second round, the voters choose among these two candidates according to
the MR. As for PR, this electoral rule does not immediately de…ne a PAR since it aggregates votes instead
of individual preferences, but under the assumption that voters vote truthfully, we can model the run-o¤
electoral rule as a PAR as follows: the social preferences compares any two candidates x and y according to
their plurality score p (x) and p (y), as with the PR, except when x and y are the two candidates with the
greatest plurality score. The run-o¤ rule compares these two candidates according to the majority preferences.

Exercise 28
8 In Exercise 26, we showed that PR is always rational. We only showed neutrality and anonymity for 2 alternatives, but

it is straightforward to see that these properties are satis…ed for any set of alternatives X: PR is not always resolute, but it
is arguably almost resolute in the sense that the PR leads to social indi¤erences between two alternatives x and y only when
the exact same number of individuals have x and y as most preferred alternatives. If one thinks about popular elections, this
con…guration is very unlikely to occur.
9 The particular preference pro…le postulated is not meant to be an accurate description of the actual preferences of voters

for that particular election. It is only meant to capture a particular con…guration that had noticable implications in terms of
the PAR analyzed in this exercise.

28
 1. Explain intuitively how the run-o¤ rule could remedy the shortcomings of the PR highlighted in Exercise
27.
2. Consider the environment inspired by the 2000 US elections, as explained in Exercise 26. What is
the outcome of the each round according to the run-o¤ rule? Explain intuitively whether this result
con…rms your answer to the previous question.
3. Consider now the collective choice problem of the division of a pie described in Exercise 26.

(a) Explain why the run-o¤ rule is not well de…ned in this case.
(b) To solve the problem identi…ed in the previous question, suppose that each individual is in fact a
group of individuals, that each group is a minority, and no two groups contain the same number
of individuals, and that in the second round, those who are indi¤ erent between the two options do
not participate (so we rank the two options according to the MR among those who participate).
What are the social preferences according to the run-o¤ rule? According to PR?
(c) Explain intuitively why the run-o¤ rule solves or not the problem of the PR identi…ed in Exercise
26 in this case.

4. Consider …nally the following environment, inspired from the 2002 French presidential elections: a
moderate left-wing candidate, Jospin, a moderate right-wing candidate, Chirac, a populist candidate,
Le Pen, and a dozen of other candidates. The most important feature of these candidates is that
they were mostly left-wing and were thus mostly taking votes away from Jospin. To simplify matters,
we assume that there is only one other candidate, and we will call her Laguiller (she was a self-
proclaimed Trotskyist candidate, and she was actually the candidate who got the most votes after the
three aforementioned candidates). So X = fLaguiller; Jospin; Chirac; LeP eng (from most leftist to
most leftist). Assume for the sake of this exercise that voters’ preference pro…le was as follows:10
8
>
> for 24% of the voters, Laguiller i Jospin i Chirac i LeP en; (leftist/Trotskyist voters)
<
for 21% of the voters, Jospin i Laguiller i Chirac i LeP en; (moderate leftist voters)
:
>
> for 30% of the voters, Chirac i Jospin i Laguiller i LeP en; (moderate rightist voters)
:
for 25% of the voters, LeP en i Jospin i Laguiller i Chirac, (anti-establishment/angry voters)
(a) What is the outcome in each round according to the run-o¤ rule?
(b) What is the socially most preferred candidate according to the PR? According to the MR?
(c) Explain in words why the MR, PR, and run-o¤ rules lead to di¤ erent collective decisions.

Solution. The solution to this exercise will be given in class.

11.2 The Borda rule and the Condorcet rule

In this section, we use the tools introduced in this Chapter to study two PAR of particular historical
importance: the Borda rule (henceforth BR) and the Condorcet rule (henceforth CR).
Let us …rst de…ne the Borda rule. For any preference pro…le ( i )i2N , the Borda score bi of an alternative
x for an individual i 2 N is given by
8
>
> bi (x) = jXj if x is the best alternative for the preference relation i ,
<
bi (x) = jXj 1 if x is the second best alternative for the preference relation i ,
>
> :::
:
bi (x) = 1 if x is the worst alternative for the preference relation i .11
The Borda rule aggregates individual preferences as follows: for all x; y 2 X,
X X
x BR y if bi (x) > bi (y) .
i i
1 0 Theparticular preference pro…le postulated is not meant to be an accurate description of the actual preferences of voters
for that particular election. It is only meant to capture a particular con…guration that had noticable implications in terms of
the PAR analyzed in this exercise.

29
Observe that the Borda score of an individual i is actually a utility function for that individual: ri is a
utility function which represents the preference relation i , (see De…nition 5). P The BR then sums these
utility functions across all individuals, and thus create a “social welfare function” i ri (x) (you might have
encountered this concept in a public economics course). This social welfare function is nothing else than a
utility function which represents the social preferences at the preference pro…le ( i )i2N . Hence, the Borda
rule aggregate individual preferences simply by maximizing the sum of their utilities.
Exercise 29
1. Compute the social preferences according to the BR in the case in which X = fa; b; cg ; N = 10; and 6
individuals prefer a to b to c whereas the other 4 individuals prefer b to c to a:
2. Is the BR anonymous? Neutral? Rational?
3. In the case of two alternatives and an odd number of individuals, is the BR resolute? What can you
conclude on the BR in the case of two alternatives?
4. Suppose jXj N . Is the BR minimally democratic? (hint: you can use a theorem from the lecture
notes)
5. Consider an election with three candidates X = fa; b; cg ; 10 voters with the same preferences as in
question 1.
(a) Suppose that the election is decided via the BR. Who is the winner? What if the election is decided
via the MR?
(b) Suppose candidate c decides not to run. Derive the voters’preferences over fa; bg ; and determine
who wins the election according to BR.
(c) Compare your answer to questions 5a and 5b and comment.
Solution. The solution to this exercise will be given in class.
In the previous exercises on the plurality rule, the run-o¤ rule and the Borda rule, we saw that the PR and
run-o¤ are rational, so they are not subject to the same Condorcet paradox as the majority rule. However,
we have also seen that in various environments (sharing a pie, US 2000 elections, French 2002 elections),
these PARs lead to a collective decision that is arguably not satisfactory. In all such environments, we saw
that there is no Condorcet paradox, that is, the majority preferences are rational, and the MR leads to a
collective decision that is arguably more satisfactory than the PR or the run-o¤ rule. Thus, the plurality
rule, the run-o¤ rule, and te Borda rule solve the Condorcet paradox, but they depart from majority rule
even in environments in which majoritarian preferences are not irrational. In light of these examples, a
natural question is whether one can …nd a PAR that is always rational and that coincides with the majority
preferences whenever the majority preferences are rational. The following de…nition formalizes the latter
requirement:
De…nition 12 A PAR f is Condorcet consistent whenever an alternative x is socially most preferred ac-
cording to the MR, it is also most preferred according to f:
Exercise 30 When jXj = 2; are the Plurality rule, the Run-O¤ rule, and the Borda rule Condorcet consis-
tent? What about when jXj = 3?
Finding a PAR that is rational and Condorcet consistent was the main motivation of the marquis de
Condorcet when he devised the Condorcet rule (henceforth CR). Let us now de…ne this rule. For all alter-
native x; we de…ne the Condorcet score c (x) of an alternative x as the number of alternatives y 2 X such
that x M R y: Then for all x; y 2 X,
x CR y if c (x) > c (y) .
Exercise 31
1. Is the CR anonymous? Neutral? Rational? In the case of two alternatives and an odd number of
individuals, is the CR resolute?
2. Is the CR Condorcet consistent?

30
Chapter III
Does majority rule take good collective
decisions?
12 Introduction
In Chapter II, we discussed the normative properties that a preference aggregation rule should satisfy
(anonymity, neutrality, etc...) and investigated which of these properties are satis…ed by various PARs.
This normative approach takes the perspective of a constitution designer who has to decide which voting
rule should be used to aggregate the preferences of the members of a society.
In this chapter, we adopt a more descriptive approach. As in Chapter II, we abstract away from the issue
of political delegation, and consider a world in which collective decisions are taken directly by the citizens
via a voting rule, and the citizens vote sincerely (i.e., their report their true preferences to the aggregation
rule). We depart from Chapter II in that we start from the empirical observation that the majority rule is
the preference aggregation rule most commonly used in modern democracies, and investigate whether this
rule always leads to a satisfactory collective decision. In other words, we focus on the outcome of the voting
rule, rather than on the voting rule itself.
In Chapter II, we showed that the majority rule can lead to an important political failure, namely the
Condorcet paradox. When such a paradox occurs, the majority rule does not know what it wants in the
sense that it cannot produce a stable collective decision. In most of this chapter, we abstract away from this
potential political failure by focusing on some speci…c, but economically relevant, environments in which the
Condorcet paradox never occurs. In those environment, majority preferences are rational and always lead
a stable collective choice. Our main goal is to investigate whether in those environments the majority rule
leads to a satisfactory collective decision. This question is of interest because in most of the environment we
have seen so far (see, e.g., Exercises 26, 27, and 28), when majority preferences are rational, the majority
preferred outcome was an arguably satisfactory collective decision. In this chapter, we will show that in
some cases, even when the majority knows leads to a stable collective decision, this decision might be unfair
or ine¢ cient.

13 Majority rule equilibrium

Before discussing whether majority rule leads to optimal decisions, we must …rst …gure out what collective
decision is likely to come out of a political process governed by majority rule. Throughout this chapter, we
assume that the outcome of a majoritarian political process is given by the following de…nition.

De…nition 13 An alternative x 2 X is a majority rule equilibrium, or a Condorcet winner, at the prefer-

ence pro…le ( i )i2N if for all x 2 X, x M R x .

In words, a Condorcet winner is an alternative x such that no other alternative is majority preferred to
x . Thus, a Condorcet winner is simply a socially most preferred alternative/stable choice when the PAR is
the majority rule (see De…nition 2). When a Condorcet winner exists on a given collective choice problem,
it is reasonable to expect that any open political process that is governed by majority rule will implement
the Condorcet winner, irrespective of the institutional details of this political process (e.g., proposal power,
veto power, deliberations, etc...). To see this, observe that once a Condorcet winner x is implemented, for
any alternative x; we have x M R x ; which means that a proposal to repeal x and implement x instead
will not be approved by a majority of individuals, so x is likely to stay in place. Conversely, suppose that
the collective decision in place x is not the Condorcet winner x . Since x M R x; a majority of voters are
better o¤ under x than under x, so if the political system is open— open in the sense that no individual has
the power to veto new proposals— it is quite likely that someone within that majority will propose x (via
an amendment in the legislative bargaining, via an electoral campaign, or via a referendum), and x will be
approved by a majority against x: Thus, a Condorcet winner is likely to stay in place, and any policy that is

31
not a Condorcet winner is likely to be replaced. That is why the Condorcet winner is also called a majority
voting equilibrium.
It is nevertheless important to remember that the concept of Condorcet winner is just a prediction about
the outcome of a majoritarian decision process, and since politics is not a hard science, there is no guarantee
that this prediction will actually materialize in all possible democratic environments.

Remark 5 It is important to keep in mind that the concept of Condorcet winner is just a prediction about
the outcome of a majoritarian decision process, and since politics is not a hard science, there is no guarantee
that this prediction will actually materialize in all possible democratic environments. There are other possible
predictions (which we will discuss later on if time permits) which are not based on purely on the majority
preferences, but are arguably reasonable, depending on the …ne details of the political decision process. So
rather than making claims of the sort "whenever it exists, the Condorcet winner is the outcome of any
majoritarian decision process ", it is more appropriate to claim that "a reasonable prediction for the outcome
of a majoritarian decision process is the Condorcet winner, if it exists".

Exercise 32

1. Determine whether there exists a Condorcet winner, and what it is for the following preference pro…le:
8
< a 1 b 1 c,
a 2 c 3 b,
:
c 2 a 3 b:

What alternative would be chosen by the group if collective decisions are taken by majority rule?
2. Same question for 8
< a 1 b 1 c;
b 2 c 2 a;
:
c 2 a 3 b:

Solution.

1. Majority preferences are such that b M R a, c M R a, and of course, a M R a, so a is a Condorcet

winner. Since a M R b, b is not a Condorcet winner, and since a M R c, c is not a Condorcet winner
either. So a is the only Condorcet winner.
Since a is a Condorcet winner, and since it is the only one, a is a reasonable prediction for the collective
decision if collective decisions are taken by majority rule.
2. The careful reader will recognize the preference pro…le used in the Condorcet paradox example (see
Exercise 18). In that case, a is not a Condorcet winner because c M R a, b is not a Condorcet winner
because a M R b, and neither is c because b M R c. In that case, it is impossible for us to make a
prediction that is based solely on the fact that the decision process uses the MR.

Hence, the Condorcet winner is a useful prediction tool, but only in environments in which a Condorcet
winner exists. However, as we will see in the following sections, there are several interesting political and
economic environments in which a Condorcet winner always exists.

14 Unanimous majorities
The simplest case in which a Condorcet winner exists is when a majority of the population agree on what
the ideal policy is.

Proposition 4 If the pro…le of preferences ( i )i2N is such that a majority of individuals have the same
most preferred alternative x, then x is a Condorcet winner.

32
Proof. (you are not requested to memorize this proof) Let y 2 X be an arbitrary alternative, and let
M N be the set of individuals whose most preferred alternative is x. For all i 2 M , y i x, so the set of
individuals j such that y j x is included into N nM . Since M is a majority, N nM cannot be a majority. By
de…nition of the majority rule, this implies that y M R x. Since this is true for all y 2 X, x is a Condorcet
winner.
A unanimous majority is rather the exception than the norm, but it is not so uncommon. For instance,
in popular elections with su¢ ciently few candidates, it can happen that a majority of voters agree on who
the best candidate is. It also happens in legislative bargaining (i.e., when legislators vote), when a party has
a majority in the legislature, and all legislators from the majoritarian party agree on a given topic (or when
the party can impose su¢ cient discipline among its members so as to guarantee that they all follow the line
imposed by the party).
One can check for instance that in the case of Exercise 32, the …rst pro…le of preferences

14.1 Application 1: the tyranny of the majority

Consider the following environment: N = 5, and the set of alternative X includes all the di¤erent ways to
spend a given public budget (normalized to 1) among 3 possible types of public goods a, b, and c (they can
also be publicly provided private goods) Formally,

X = fga ; gb ; gc : ga 0; gb 0; gc 0 and ga + gb + gc 1g ;

where ga , gb , and gc are the level of public spending on some goods a, b, and c, respectively. The preferences
of the individuals are given by the following utility functions:
8
< U1 (ga ; gb ; gc ) = U2 (ga ; gb ; gc ) = ga ;
U3 (ga ; gb ; gc ) = gb ;
:
U4 (ga ; gb ; gc ) = U5 (ga ; gb ; gc ) = gc ;

Thus, individuals 1 and 2 care only about public good a, individual 3 cares only about public good b and
individuals 4 and 5 care only about public good c. For concreteness, you can think of ga , gb , gc as the level
of spending on mosques, synagogues, and churches, with individual 1 and 2 being Muslim, individual 3 being
Jew, and individuals 4 and 5 being Christians. According to this interpretation, the preferences of each
individual are admittedly quite sectarian and narrow minded, but these extreme preferences are assumed
solely for the sake of illustration.
It turns out that in the public good environment, there is no Condorcet winner. No majority of voters
can agree on a division of the pie. For example, if the status quo is to spend 50% of the budget on public
good a and 50% on public good b, voters 4 and 5 can propose to voter 3 to spend 60% on b and 40% on c;
which would be approved by voters 3, 4, and 5. But individuals 1 and 2, who have been left out of the deal,
can then propose to spend 50% on a and 50% on c. Voters 1, 2, 4, and 5 would bene…t from this budget
reallocation, relative to the previous proposal. But then, voter 3 can propose to spend 60% on a and 40% on
c; which would be approved by voters 1; 2; and 3 relative to the previous proposal, and the cycle continues
inde…nitely...
This stylized example illustrates a more general result in voting theory: when the space of alternatives
is multidimensional (which is very often the case when the collective decision is about how to redistribute a
given amount of money),12 majority preferences are very likely to be irrational.
Now suppose that individual 4 has the same preferences as individual 1 and 2 (that is, she converts to
Islam). In that case, (ga ; gb ; gc ) = (1; 0; 0) is the most preferred alternative of individuals 1; 2; and 4: From
Proposition 4, this alternative is a Condorcet winner. Thus, 1; 2 and 4 form a unanimous majority that can
use majority rule to control the public budget and spend it entirely on their most preferred good, ignoring
the preferences of individuals 3 and 5.
This stylized example illustrates a fundamental hindsight in political sciences: when a majority of indi-
viduals have su¢ ciently homogeneous preferences and interests over some …scal policies, this group can use
1 2 In this case, the space of alternative is in fact 2 dimensional: ga and gb can take any value in the set
fga ; gb : ga 0; gb 0 and ga + gb 1g, the level of spending on remaining public good gc must be equal to 1 ga gb
(there is no reason to leave a surplus at the end of the division).

33
the majority rule to spend all the public funds on their most preferred goods, at the expense of the minority.
If left unimpeded, the majority rule can even completely expropriate the minority. For instance, in the above
example, the public funds are exogenously …xed at 1: But the majority group could increase the amount of
available public funds by implementing a con…scatory tax on the minority group. This abuse of majoritarian
power is one of the most important example of political failure. It is called the tyranny of the majority.
The tyranny of the majority occurs in collective choice problems in which a public good or a tax can be
targeted to speci…c regions/ethnic groups/religious groups, and one of the group contains a majority of the
voters. However, this phenomenon is not con…ned to …scal issues. The tyranny of the majority can also lead
a majority to deny basic civil rights to a minority. For instance, heterosexuals can use the majority rule to
make homosexuality illegal. Likewise, the most populous ethnic group can use the majority rule to prohibit
the use of the language of the minority.
Majoritarian abuses of power are an inherent danger of democracy because the founding principle of
democracy is that the will of the majority should prevail. There is an long debate in political sciences as to
whether the majority should always get its way (as in a pure democracy) or should be constrained (as in
a republic). The tyranny of the majority can be avoided, or at least mitigated, by constitutional laws such
as bills of rights. Such constitutional laws de…ne a set of inalienable rights that no policy can violate, even
if supported by a majority of voters. For instance, in the previous example, if ga , gb , and gc are level of
spending in a local public good in regions a; b; and c; respectively, one can avoid the tyranny of the majority
by imposing the same level of public spending per capita across regions. Thus, ruling out discriminatory
and divisive policies at the constitutional level can prevent majoritarian groups to use the political process
to treat unfairly minority groups.

14.2 Application 2: intense minority versus apathetic majority

Suppose that on some issue (say gun control, or gay marriage), a majority favors a certain policy (e.g.,
background check for gun owners, opposition to gay marriage) but does not care much, while the minority
care deeply about the alternative policy (no gun control, legalization of gay marriage). Political equality
requires that all votes be weighted equally, so the will of the apathetic majority should prevail. But in those
cases, the one-person-one-vote principle does not seem very appealing. If the preferences of the minority
are su¢ ciently more intense than that of the majority on this issue, the socially optimal policy might very
well be that favored by the minority. The recent example of the 2013 popular referendum on gay marriage
in Croatia, or the 2012 referendum on banning Minarets in Switzerland o¤er possible examples of intense
minority versus apathetic majority.

14.3 Application 3: voting on nondistortionary redistribution

Redistribution represents the single largest public spending category in most modern democracies. The main
tool for redistribution is income taxation. In this subsection, we investigate what degree of redistribution
would emerge if the tax scheme was chosen directly by the voters via majority rule. For simplicity, we focus
on a purely redistributive tax scheme. That is, the purpose of the taxation is not to …nance some public
program, but to redistribute income within the population. Since we are more interested in the politics
than the economics of redistribution, we further assume that income taxation is not distortionary: it does
not a¤ect the pre-tax income of individuals, and the money raised via taxation is fully redistributed: some
individuals pay negative taxes (they will receive positive transfer from the tax scheme) whereas others pay
positive taxes.

14.3.1 A basic income …nanced by a ‡at income tax

For simplicity, we consider only two …scal instruments: a ‡at income tax and a universal basic income. A ‡at
income tax means that each citizen i 2 N pays a …xed share s 2 [0; 1] of her pre-tax income wi : A universal
basic income means that thePmoney raised by taxation is divided equally among all citizens, irrespective
of their salaries. So if W i2N wi denote the aggregate pre-tax income, then the revenue raised by the
W
‡at income tax is sW; so the basic income received by each citizen is b = sw; where w N is the average

34
income. Thus, the after-tax income of each individual i is

wiAT = (1 s) wi + sw: (12)

Note that s can be interpreted as the degree of redistribution: a share s of total income is redistributed
equally among all citizens, and the remaining share 1 s is not redistributed. So s = 0 corresponds to no
redistribution— each individual keeps her entire income: for all i 2 N; wiAT = wi — whereas s = 1 corresponds
to complete redistribution— all individuals have the same after-tax income: for all i 2 N; wiAT = w.13

14.3.2 Voters’induced preferences over the degree of redistribution

To make a prediction on the outcome of the collective choice problem, we must …rst de…ne the choice
variables— that is, what are the citizens voting on exactly— and then derive voters’ preferences over these
choice variables. These preferences are often called the induced preferences, because they are induced by the
particular collective choice problem under consideration.
In this case, what the voters must choose is the ‡at tax rate s 2 [0; 1], because as argued in the previous
section, once s is …xed, the level of the basic income b is …xed via the budget constraint of the government.
Thus, we are facing a one-dimensional collective choice problem.
To derive voters’ induced preferences over s; we must …rst specify their preferences over the di¤erent
goods and outcomes that are a¤ected by the variable of the collective choice problem. In this simplistic
environment, we do not model formally what consumers purchase with their after-tax income. The choice
variable s a¤ects voters only indirectly via their after-tax income. It is reasonable to assume that when
comparing two tax policy, a given voter will prefer the policy that leaves her with the greater their after-tax
income. So we assume that the utility function of a voter i is simply her after-tax income wiAT : Therefore,
from (12), her induced preferences are given by

Vi (s) = wi + s (w wi ) ;

which shows that the utility of an individual i is increasing in s if wi < w, decreasing in s if wi > w, and
constant in s if wi = w. Hence, individuals with a smaller (greater) than average income always want more
(less) redistribution.
The intuition for this result can be explained through the notion of tax-price. The notion of tax-price
extends the notion of price to publicly provided goods.

De…nition 14 The tax price of a publicly provided good g for an individual i is the extra tax that i has to
pay when the level of public provision of g increases by one unit.

In our case, the publicly provided good is the basic income. So the tax-price of the basic income for
individual i is the extra tax paid by i if we increase the basic income by one unit. Since the basic income b
is entirely …nanced by the ‡at income tax, as argued previously, we have that b = sw; so the tax paid by i
is ti = swi = bwi =w: Therefore, increasing b by one unit increases ti by wi =w unit. That is, the tax price of
the basic income for individual i is wi =w:
The …rst thing to notice is that this tax-price is increasing in wi ; because richer individuals pay a
greater share of the total cost of the basic income. As a result, the basic income appears cheaper to poorer
individuals, which is why poorer individuals want more basic income (and thus more redistribution) than
richer individuals. More precisely, for a poorer than average individual i, the tax price wi =w is smaller than
1; so the tax price of an additional unit of basic income is smaller than the bene…t (which is 1; because
one unit of basic income result in one unit of after-tax income). Therefore, when comparing two levels of
redistribution, i will always prefer the one that yields a greater basic income. That is, i always prefer more
redistribution, and vice versa when i is richer than average.
1 3 The constraint that s 2 [0; 1] guarantees that no individual has a negative after-tax income. To see this, simply take w ! 0
i
and wi ! +1 in (12).

35
14.3.3 The Condorcet winner
From what precedes, the most preferred degree of redistribution of an individual i is s = 1 if wi < w; and
s = 0 if wi > w: Thus Proposition 4 implies that if a majority of individuals have an income smaller (greater)
than the average income, the Condorcet winner is sCW = 1 (sCW = 0). This result can be stated in terms
of median income. The median income is the income level wm such at least half of the population has a
weakly smaller income and at least half of the population has a weakly greater income. Therefore, if wm < w
(wm > w), then the income of majority of voters is weakly smaller (greater) than wm ; and therefore smaller
(greater) than w; which proves the following.

Proposition 5 Suppose redistribution is achieved via a basic income …nanced by a nondistortionary ‡at
income tax. Then if we let citizens vote on the degree of redistribution, the Condorcet winner is full redistri-
bution (i.e., all incomes are fully taxed and equally redistributed) if the median income is smaller than the
average income, and it is no redistribution (i.e., income are not taxed at all) when the median income is
greater than the average income.

In almost all countries, the average income is greater than the median income, so Proposition 5 suggests
that if we let citizens decide directly the degree of redistribution via majority rule, the most likely outcome
is complete redistribution. Of course, such a prediction should be taken with a grain of salt. No democratic
country has ever implemented such an extreme degree of redistribution. Nevertheless, this stylized model
highlight an important point (this is the main take away from this simple model, remember it!): How much
a given individual bene…ts from or contribute to income redistribution depends on her relative position in the
income distribution: relatively poor citizens bene…t from redistribution whereas relatively rich citizens con-
tributes more than they receive. So if we let self-interested citizens choose directly the degree of redistribution
via the majority rule, the outcome will depend on the pre-tax distribution of income in the society, an in
particular on how the median income compares with the mean income.

14.3.4 Discussion
Given the result of the previous section, one might ask why the poor do not expropriate the rich in all
democracies. This question has received a lot of attention in the political economy literature, and several
answers have been proposed, which involve political and economic e¤ects that have been ignored in our
stylized model. These mechanisms are beyond the scope of this course, so I will only brie‡y summarize some
arguments below.
Our model makes a number of simplifying assumptions: (i) the tax scheme is decided directly by the
voters, (ii) the gross income of each voter is una¤ected by taxation, i.e., incentives to generate income are
not taken into account (iii) voters care only about their after-tax income. There are several reasons why
these assumptions are too simplistic.
Regarding assumption (i), income taxation is typically not decided directly by the voters via majority
voting. The tax code in a country is typically the result of a complex bargaining process. Some political
scientists have argued that this process is more sensitive to the preferences of richer constituents. Lobbying
by special interests is likely to be an important driver of this bargaining process. When lobbying plays a role,
the principle underlying the decision process on the tax policy moves away from the one-citizen-one-vote
principle towards one-dollar-one-vote principle. Some empirical evidence has been found along these lines.14
As for assumption (ii), complete redistribution does not give any incentive for workers and entrepreneurs
to earn an income in the …rst place. Moreover, in a globalized world, such a taxation scheme would make
rich people and successful entrepreneurs ‡ee to less redistributive countries. Hence, even if the redistributive
scheme is dictated by the preferences of the relatively poor voters, the latter might realize that a less extreme
level of redistribution can create more income and attract more economic activity in her economy, and can
thus increase the tax base from which they can redistribute. Exercise 37 below considers the more realistic
case in which redistributive taxation generates distortions and ine¢ ciencies.15 Finally, regarding assumption
1 4 See, for instance, "Economic Inequality and Political Representation" by Larry Bartels, available on the web, and the

references therein.
1 5 The seminal paper on voting, redistribution, and workers incentives is the following: Meltzer and Richards. 1981. A rational

theory of the size of government. Journal of Political Economy 89 (5): 914-927.

36
(iii) ; voters do not care only about their after-tax income. They care about other values. For instance,
they might care about merit. A voter might consider that hard workers and risk takers deserve a greater
income. If such an ideology is prevalent in a society, even poor voters may not want to equalize income
across individuals.

14.4 More exercises on unanimous majorities

The following exercise considers an economy in which the goods and services provided by the public sector
are …xed, and the money needed to fund these goods and services can be raised by two types of taxes, a
labor tax and a land tax. Thus, unlike in the redistribution problem analyzed in Section 12, the collective
choice problem under consideration is not how much tax the government should raise, but what tax base it
should use to …nance its (exogenously given) …scal needs.
Exercise 33 (Land versus Labor Taxation) Consider an economy with three goods: a private good (say
food), labor, and land. We …x the price of the private good at 1 by convention, the price of labor (the wage)
is denoted w, and land is assumed to be non tradable in that model, for simplicity.16
The public sector can levy two types of taxes:

– A labor tax. This tax is levied on every unit of labor exchanged in the economy. It is paid by
the buyer of the unit of labor (i.e., the …rm that hires the worker). The labor tax rate is denoted
tw 2 [0; 1]. So for each unit of labor it hires, a …rm pays w + tw .
– A land tax. This tax is levied on every unit of land owned in that economy. It is paid by the
landowner. The land tax rate is denoted tl 2 [0; 1]. So for each unit of land he owns, a citizen
pays tl .

Each citizen i 2 N is endowed with 1 unit of labour, and an amount li 0 of land.

Each unit of land can produce 1 unit of private good.
Firms use labor to produce the private good. They are competitive (that is, they are price takers), and
they all have the same technology: they can produce one unit of private good for each unit of labor they
hire.
Individuals care only about the amount of private good they consume, so the utility of each individual
is simply the amount of private good she consumes. As a result, individuals inelastically supply all of
their labor, they use all of their after-tax labor income to purchase the private good, and they consume
all of the after-tax production of the land they own.
Given the information above, answer the following questions.
1. Show that at the competitive equilibrium, the equilibrium wage is w = 1 tw .
If we take into account the land tax, how much units of private good does each unit of land generate
for its owner?
2. Determine how much an individual i pays in taxes, and how much private good she consumes in
equilibrium, as a function of the amount of land li she owns, the labor tax rate tw , and the land tax
rate tl .
3. Determine the total public revenue as a function
P of the level of taxes tw and tl , the number of tax
payers N , and the total amount of land = i2N li .
4. Suppose that the government must raise a certain amount of money T to provide some essential public
goods for the population (e.g., a sewage system, …re protection...). So the government must choose a
pair of tax rates (tw ; tl ) such that the total tax revenue is T . Given this constraint, what is the most
preferred tax policy (tw ; tl ) of an individual i as a function of her land ownership li , the total amount
of land , the population N , and the …scal need T (you may assume for simplicity that N > T and
> T )?
1 6 One can easily check that at any rate, at any competitive equilibrium, no one would like to sell or buy land.

37
 5. What tax policy (tw ; tl ) is likely to emerge if the tax rates tw and tl are decided directly by the citizens
via majority rule?
6. Suppose now that the private good is interpreted as money, and land is interpreted as capital (capital
generates money just as land generates food). How should you interpret your answer to question 5?

Solution. This exercise will be solved in class.

15 The Median Voter Theorem

In this section, we introduce the class of single-peaked preferences. This class of preference relations is of
particular interest because if all individuals have single-peaked preferences, a Condorcet winner always exist,
and it can be easily characterized.

15.1 Single-peaked preferences

In this section, X is a subset (…nite or in…nite) of R. So single-peaked preferences must be de…ned on a
one-dimensional policy such as a tax rate, or the level of provision of a public good.

De…nition 15 A preference relation on X is single-peaked if it there exists x 2 X such that for all
y; z 2 X,
if x < y < z, then x y z,
(13)
if x > y > z, then x y z:
A preference pro…le ( i )i2N is single-peaked if for all i 2 I, i is single-peaked. In this case, the unique
“peak” of each individual i 2 N is denote by xi :

In words, the preferences of an individual i 2 N are single-peaked if that individual has a unique most
preferred alternative xi , and the farther an alternative x is from xi , either to the left or to the right of xi ,
the less preferred is x according to i . The alternative xi is usually called the peak, or the ideal policy of
individual i.17

Remark 6 The term “single-peaked” comes from the fact that if a preference relation is represented by a
utility function U (see Section 4.3), then is single-peaked whenever has a unique local maximum. To see
this, observe that the …rst line of (13) means that U must be decreasing to right of x , and the second line of
(13) means that U must be increasing to left of x . This alternative de…nition of single-peakedness is usually
more practical to use than De…nition 15, whenever preferences are de…ned via utility functions. When U is
twice di¤ erentiable and X is an interval, the above condition on U is satis…ed in particular when U 00 (x) < 0
for all x 2 X:

Exercise 34 Suppose X = Z, N = f1; 2; 3; 4; 5g and for all i 2 N , i is derived from the utility function:
2
for all x 2 X, Ui (x) = (x xi ) ;

where x1 = 9, x2 = 2, x3 = x4 = 1, x5 = 2. Show that the preference pro…le derived from these utility
functions is single-peaked, and determine the peak for each individual.

Solution. For individual i = 1, @U @x =

1
2 (x + 9) ; so @U
@x > 0 for x <
1
9 and @U@x < 0 for x >
1
9. This
means that U1 is increasing to the left of 9 and decreasing to the right of 9: From Remark 6, this implies
that 1 is single-peaked with a peak at 9. A similar reasoning shows that the preferences of individuals
2; 3; 4 and 5 are also single-peaked, with a peak at 2, 1; 1; and 2, respectively. Alternatively, it su¢ ces to
2
notice that @(@x)
Ui (x)
2 = 2 < 0: From Remark 6, this implies that preferences are single-peaked, although this
method does not allow us to determine the peak of each preferences directly.
1 7 Single-peaked preferences are also called unimodal preferences, and xi is called the mode.

38
15.2 The Median Voter Theorem
Why do we care about single-peaked preferences? As the next theorem shows, existence of a Condorcet
winner is guaranteed when preferences are single-peaked. For simplicity, we state this result in the case of
an odd number of individual, and we will maintain that assumption throughout this Chapter.

De…nition 16 Let ( i )i2N be a pro…le of single-peaked preferences, and let (xi )i2N be the corresponding
pro…les of the voters’ peaks. Then there is a unique median peak xm de…ned by

N N
jfi 2 N : xi < xm gj and jfi 2 N : xi > xm gj :
2 2
The voter(s) whose peak is xm is (are) called the median voter(s).

The median voter is simply the individual whose peak xm is such that no more than half of the individuals
have a peak to the left of xm and no more than half of the voters have a peak to the right of xm . Observe
that since we have assumed that there is an odd number of voters, there is a unique median peak.18 There
might be more than one median voter, but all the median voters must all have the same peak (see Exercise
35 below).

Theorem 4 (the Median Voter Theorem) Let ( i )i2N be a pro…le of single-peaked preferences. Then
there exists a unique Condorcet winner. It is given by the median peak (see De…nition 16).

Proof. To show that xm is the unique Condorcet winner, we will show that it is majority preferred to any
alternative x < xm and to any alternative x > xm . Let x 2 X be such that x < xm . For all i 2 N such that
xi = xm , by de…nition of xi , xm is individual i’s most preferred alternative, so xm i x. For all i 2 N such
that xi > xm , x < xm < xi , and since preferences are single-peaked, xi xm x. Hence, we have shown
that for all i 2 N such that xi xm , we have xm i x. By de…nition of the median voter m, the set of such
individuals form a majority. So by de…nition of the majority rule, xm M R x; so x cannot be a Condorcet
winner. A symmetric reasoning shows that for all x > xm , for all i 2 N such that xi xm ; xm i x, and
since the set of such voters form a majority, xm M R x; so x cannot be a Condorcet winner. Since we have
shown that for all x 6= xm , xm M R x, xm is a Condorcet winner.
The Median Voter Theorem (henceforth abbreviated MVT) states two results. First, when preferences are
single-peaked, a Condorcet winner always exists. Second, the Condorcet winner can be easily characterized:
it is the most preferred policy of the median voter.
Hence, when preferences are single-peaked, the majority rule gives a sort of dictatorial power to the
median voter, because it always select her most preferred alternative. It is not exactly a dictatorial power
though, because the median voter cannot implement any policy she wants. To see that, observe that if her
preferences change, she may no longer be the median voter, in which case the Condorcet winner will not
be her preferred policy anymore. Thus, the median voter is only a “positional dictator”: she is a dictator
only as long as her peak remains at the median position in the distribution of peaks within the society (see
Exercise 35 below).

15.3 Application 1: who is the median voter?

The following exercise illustrates how the identity of the median voter can change as individual preferences
change.

Exercise 35 (Mandatory and important exercise!) Consider the preference pro…le given in Exercise
34.

1. What is the Condorcet winner?

2. Suppose now that we change the preferences of individual 5 as follows: x5 = 0. How does that a¤ ect
the Condorcet winner?
1 8 That
is the only reason why we assume that there is an odd number of voters. If the number of voters is even, there might
be two median peaks.

39
 3. Let us consider again the preference pro…le given in Exercise 34. Which individual can change the
location of the Condorcet winner by unilaterally changing the location of her peak?

Solution.

1. The peak of individual 3; i.e., x3 = 1; satis…es the de…nition of the median peak (see De…nition 16),
because
fi 2 N : xi < 1g = f1; 2g so jfi 2 N : xi < 1gj = 2 25 ;
fi 2 N : xi > 1g = f5g so jfi 2 N : xi > 1gj = 1 25 :
Theorem 4 implies then that the Condorcet winner is xCW = 1. Individual 3 and 4 are both median
voters, but they have the same peak, so the Condorcet winner is unique. One can easily check that no
other peak satis…es the de…nition of the median peak.
2. Suppose now that x5 = 0. Then one can easily check that x5 = 0 is the median peak, because
5
fi 2 N : xi < x5 g = f1; 2g so jfi 2 N : xi < x3 gj = 2 2;
5
fi 2 N : xi > x5 g = f3; 4g so jfi 2 N : xi > x3 gj = 2 2:

So the unique median voter is now individual 5; and the Condorcet winner is xCW = 0. Hence, the
main qualitative take away from this exercise is the following: when individual preferences change, the
identity of the median voter may change, so the median voter is not really a dictator because she can
impose her most preferred policy only as long as she remains the median voter.
3. Consider for instance individual 1. By moving her peak to any location x1 in ( 1; 1] ; she remains
on the same side of the the median peak 1, so she does not a¤ect the location of the median peak,
so it does not a¤ect the Condorcet winner according to the MVT. Suppose now that she change the
location of her peak to some x1 in [1; +1) ; then she moves it to the other side of the median peak.
However, one can easily check that since there are two individuals whose peak is median, this change
does not a¤ect the identity of the median peak either, and thus does not a¤ect the Condorcet winner.
The same argument implies that moving the peak of individual 2 cannot change the location of the
Condorcet winner either.
Consider now the peak of individual 3. If we move her peak to some x3 > 1; the unique median
becomes that of individual 4, but it remains at the same position. However, if we move her peak to
some x3 < 1; then the median peak becomes max fx2 ; x3 g (to see this, just consider the two cases
x3 2 and x3 2 ( 1; 2)), so the Condorcet winner becomes xCW = max fx2 ; x3 g : Since in the
original preferences pro…le, x3 = x4 , moving the peak of individual 4 only can a¤ect the Condorcet
winner exactly in the same way as with individual 3.
From the previous question, moving the peak of individual 5 can also a¤ect the position of the Condorcet
winner. So the only individuals whose peak can a¤ect the Condorcet winner are individuals 3, 4, and
5.

15.4 Application 2: voting over the location of a public facility

Exercise 36 (Mandatory and important exercise!) Consider a village which consists of a single road
along which leave 3 inhabitants. Their location can be parametrized by a real number (which indicates where
they live along that road). Individual 1, 2, and 3 are located at x1 = 1, x2 = 0, and x3 = 1, respectively.
They want to build a well, and have to decide where to build it. If xw denote the location of the well, then
2
the welfare of an individual i who lives in location xi is given by Ui (xw ) = (xi xw ) .

1. If the inhabitants decide the location of the well by majority rule among the villagers, what is the likely
outcome of the vote?
2. Suppose that individual 3 moves to a new location, x3 = 4. How does that change your answer to
question 1?

40
 3. Suppose now that the location of the well is decided by a benevolent mayor whose goal is to maximize
the sum of the welfare of the inhabitants of the village. How would your answer to questions 1 and 2
change ? Comment.

Solution.

1. Observe …rst that the preferences of all individuals are single-peaked. To see this, note that the
@ 2 Ui
collective decision problem is choosing xw , so it is one-dimensional, and for all i 2 N , (@x 2 = 2,
w)
which is negative. The peak of individual 1, 2, and 3 are respectively 1, 0, and 1 (their most preferred
location for the well). The median peak is 0, so from the MVT, the unique Condorcet winner is xw = 0.
The Condorcet winner is a reasonable prediction when collective decisions are taken by majority rule.
2. The median peak is still 0, so the likely outcome of the majoritarian decision process is still xw = 0.
3. The policy that maximizes the sum of individual’s preferences in case A) is given by
2 2 2
max ( 1 xw ) (0 xw ) (1 xw ) :
xw

the …rst order condition of that maximization program is

2( 1 xw ) + 2 (0 xw ) + 2 (1 xw ) = 0:

The solution to the above equation is xw = 0:

The policy that maximizes the sum of individual’s preferences in case A) is given by
2 2 2
max ( 1 xw ) (0 xw ) (4 xw ) :
xw

the …rst order condition of that maximization program is

2( 1 xw ) + 2 (0 xw ) + 2 (4 xw ) = 0:

The solution to the above equation is xw = 1.

So we see that a benevolent social planner would take into account the fact that individual 3 has
moved father from the well, and would relocate the well accordingly, whereas the majoritarian decision
process would not. The reason is that the outcome of the vote is driven solely by the preferences of the
median voter. Hence, the main qualitative take away from this exercise is the following: when decisions
are taken by majority rule, the interests of individuals with nonmedian preferences is not taken into
account in the sense that the majoritarian outcome does not depend on how extreme these nonmedian
individuals are. In contrast, a social planner would tilt the decision in favor of the individuals that
have more extreme (and thus more intense) preferences. Thus, this exercise illustrates an another
potential source of political failure: the majority rule is insensitive to the intensity of the preferences
of nonmedian individuals.

15.5 Application 3: distortionary redistribution and the majority rule

We have already analyzed a voting model of redistribution in Section 14.3. A somewhat extreme implication
of that model is that the most preferred tax policy of any given individual is either complete redistribution
or no redistribution at all. As argued earlier, this unrealistic result was driven by our assumption that the
proportional income tax is not distortionary (that is, the gross income of all individuals were una¤ected by
taxation). In reality, income taxes reduce the incentives to work, and thus to earn an income in the …rst
place. In the following exercise, we take the distortions induced by income taxes into account, and investigate
how they a¤ect the majority preferred degree of redistribution.

41
Exercise 37 (Distortionary redistribution) Consider the same …scal framework as in Section 14.3: all
individuals pay a share s 2 [0; 1] of their gross income, and the revenue raised by this proportional tax
is then redistributed equally among all individuals under the form of a basic income b. Using the same
notations as in Section 14.3, the tax paid by each individual is swi ; so the aggregate tax payment is sW:
However, instead of assuming that this sum of tax payment turns into the same quantity of tax revenue for
the government, we assume that a fraction f of sW is lost due to the ine¢ ciencies of taxation. This loss
of f sW is meant to capture the deadweight loss due to distortion induced by the income tax, but it can also
capture the administrative costs or corruption generated by the tax collection. Economists often use the term
“cost of public fund” to refer to the ine¢ ciencies generated by the payment and collection of taxes. Clearly,
the more redistributive the income tax is, the less incentives it gives to work, so the more distortions and
ine¢ ciencies it generates. Accordingly, we assume that the fraction f of the tax payments that is lost is
increasing in the tax rate s. More precisely, f = 2 s; where the parameter captures the magnitude of that
e¤ ect (the fraction 12 is there only to simplify the algebra). Hence, if the tax rate is s, the total taxes paid by
the citizens is sW , but the revenue raised by the government is only

sW f sW = sW s2 W:
2
As in the model in Section 14.3, the tax revenue are then redistributed equally among all individuals, so the
level of the basic income is b = N1 sW 2
2s W :

1. Describe how the total revenue from the income tax varies with the tax rate s. Draw a graph and
comment.
2. Determine the after-tax income of an individual as a function of the parameters of the problem.
3. Suppose individuals care only about their after tax income (therefore, one can interpret their after-tax
income as their utility function). What can you say about the preferences of an individual i with pre-tax
income wi over the tax rate s? What is her most preferred tax rate si (remember that s 2 [0; 1])?
4. If the tax rate s is determined directly by the tax payers via majority rule, what is the likely outcome
of the vote as a function of the distribution of income and of ?
5. Suppose > 1. Describe how your answer to the previous question varies as a function of the degree
of income inequality and of the e¢ ciency of taxation .

Solution. This exercise will be solved in class.

16 Public good provision and the majority rule

One of the most important question in political economy is what determines the size of public spending.
In most modern democracies, the two single biggest public spending categories are redistributive transfers
(monetary transfers such as universal income or pensions, or in kind transfers via the public provision of
private goods such as health care, or higher education) and the provision of public goods (national defense,
primary education, transportation infrastructure, police...). We have already studied majority preferences
over the degree of redistribution in Sections 14.3 and 15.4. In this section, we let citizens vote over the
level of provision of a public good under di¤erent institutional assumptions, and investigate how e¢ cient the
majoritarian decision is.

16.1 A simple public good model

Consider an economy with N individuals and two goods, a private good andPa public good. All individuals
are initially endowed with a quantity ! i > 0 of the private good, and i2N ! i denotes the aggregate
endowment. The public good is produced from the private good, and we normalize units so that 1 unit of
private good produces 1 unit of public good.
An allocation of resources speci…es a pro…le of private good consumption (ci )i2N and a level of public
good consumption g— by de…nition of a public good, g is the same for all individuals. To simplify the

42
analysis, we do not require private consumption levels to be positive: for all i 2 N , ci can be negative if
needed. Therefore, an allocation is feasible
Pif (i) the level of public good g is positive, and (ii) the sum of
the aggregate private good consumption i2N ci and of the amount of private good used to produce the
public good g is no greater than the initial endowment of the economy . Thus, an allocation (ci )i2N ; g
is feasible if
P g 0; (14)
i2N ci + g :
Therefore, the choice of a feasible allocation can be decomposed into two subchoices: …rst, how much private
good should be allocated to the production of the public good, that is, what g should we choose in [0; ]?
Second, how should we allocate the remaining quantity of private goods g among the individuals,
P that is,
what pro…le of private consumption (ci )i2N should we choose, given the resource constraint i2N ci g:
To fully specify this collective choice problem, it remains to specify the pro…le of preferences of the
individuals. The preferences of each individual i over the feasible allocations depend only on her level of
consumption of the private good ci and of the public good g, and are derived from the following utility
function:
Ui (ci ; g) = i ln (g) + ci ; (15)
where i > 0 is a parameter, and ln is the natural logarithm function.19 The preferences derived from the
utility function (15) have the following properties:

The term ci corresponds to the utility derived from private good consumption. For simplicity, prefer-
ences are quasi-linear in the private good, i.e., ci appears linearly in (15). This implies that there is
no income e¤ect: the demand for public good of a given individual is independent of her wealth.
The term i ln (g) corresponds to the utility derived from public good consumption. The marginal
2
willingness to pay for the public good20 is positive ( @[ i@g
ln(g)]
> 0) but decreasing ( @ [(@g)
i ln(g)]
2 < 0). This
means that individuals prefer to consume more public good, but the amount of private consumption
they are willing to forego to consume an additional unit of public good decreases as the level of public
good increases. Roughly speaking, there is a satiation e¤ect for the public good.
The preferences of any individual i are characterized by the parameter i : the greater i; the more
individual i cares about public good consumption relative to private good consumption.

16.2 The socially optimal allocation: the Samuelson rule

As a benchmark, suppose …rst that the economy is governed by a benevolent and omniscient planner whose
goal is to maximize the sum of all individual’s welfare. That is, she implements the allocation that solves
the following program: X
max Ui (ci ; g) (16)
(ci )i2N 2RN ;g 0
P i2N
s.t. i2N ci +g ;

where Ui (ci ; g) is given by (15).21 To simplify that program, observe

P…rst that at the social optimum, no
resources is wasted, which means that the technological constraint i2N ci + g holds with equality.
1 9 The @ ln(g) 1
only thing you are supposed to know about that function is that it is strictly increasing, concave, that @g
= g
,
2
@ ln(g)
and that (@g)2 = g12 . You also supposed to know how to draw the function g ! ln (g) on a graph.
2 0 The marginal willingness to pay for the public good is how much private consumption individual i is willing to forgo to

consume an additional unit of public good. In that case, the extra utility she gains by consuming an additional unit of public
good is i ln (g + 1) i ln (g), which is approximately equal
Pto i =g.
2 1 Note that the social welfare function W (c )
i i2N ; g i2N Ui (ci ; g) de…nes a preference aggregation rule as introduced
in Chapter II: for any two feasible allocations (ci )i2N ; g and c0i i2N
; g0 ,

(ci )i2N ; g S c0i i2N

; g0 if W (ci )i2N ; g > W c0i i2N
; g0 :

This PAR is similar in spirit to the Borda rule (see Section 11.2) in the sense that it is based on a utilitarian social welfare
function.

43
This allows usPto express the aggregate level of consumption of the private good as a function of the level of
public good, i2N ci = g, and thus to simplify (16) as follows:
" ! #
X X
max i ln (g) + ci
(ci )i2N 2RN ;g 0
P i2N i2N
=
s.t. i2N ci +g ;
" ! #
X
, max i ln (g) + g : (17)
g 0
i2N

We see from (16) that the private good allocation (ci )i2N has disappeared from the maximization program of
the social planner. In mathematical terms, we have transformed a maximization program with N +1 variables
(ci )i2N ; g and two constraints into a simpler program with one variable g and only one constraint g 0.
In economic terms, this means that our utilitarian planner cares only about the quantity of private good that
is allocated to the production of the public good, i.e., g; but but she is indi¤erent as to how
P the remaining
quantity of private good g should be allocated among the individuals.22 The term i2N
= i ln (g) in
(17) captures the social bene…ts of public good provision, whereas the term g captures its social costs
in terms of foregone private good consumption.
Proposition 6 At any socially optimal allocation, i.e., at any solution of (16), the level of public good
provision g O is X
gO = i: (18)
i2N

Proof. As shown in the text above, the optimal level of public good provision g O solves (17). One can easily
check that
P P
@ i2N i ln (g) + g i2N i
= 1;
@g g
P P
@2 i2N i ln (g) + g i2N i
2 = < 0;
(@g) g2
so the maximand in (17) is concave, and its maximum is given by the …rst order condition
P !
@ i ln (g) + g X 1
i2N
=0, i = 1:
@g g
i2N

Note that we have ignored the constraint P g 0, but one can see that the solution to the above …rst order
condition is indeed strictly positive, since i2N i > 0.
Equation (18) has a natural interpretation. To see this, note that @U i @Ui
@ci = @g = g ; so g is the marginal
i i

rate of substitution for individual i between the public and the private good. Moreover, the marginal rate of
transformation between these two goods is 1,P because 1 unit of private good can produce 1 unit of public good.
Finally, note that (18) can be rewritten as i2N gi = 1, which then states that the sum across individuals
of the marginal rate of substitution between the public and the private good is equal to the corresponding
marginal rate of transformation. This equality is known as the Samuelson condition. It generalizes the
standard optimality condition to the case of a non rival good. Since g is non rival, an additional unit of
public good increases the welfare of all individuals by gi . As a result, this e¤ect is summed across all
individuals.

16.3 Public good provision by voluntary contributions: the free market alloca-
tion
Let us now consider an economy without social planner in which individuals decide themselves in a decentral-
ized and voluntary way how much public good should be provided. Speci…cally, we assume that the public
2 2 This
simpli…cation comes from the fact that a unit of private good consumption increases the social welfare function in (15)
by one unit irrespective of which individual consumes it.

44
good is produced by a benevolent organization which has access to the same technology as the benevolent
planner (that is, she can transform every unit of private good she controls into one unit of public good) but
this organization does not control the allocation of resources. Instead, each individual can decide freely how
much of her initial endowment to consume as private good, and how much to give to the benevolent organi-
zation to produce the public good. These individual decisions can be viewed as voluntary contributions to
the public good. It is important to note at teh outset that the fact that the organization is benevolent does
not matter for our result, it would be equivalent to assume that the public good is produced and sold by a
large number of perfectly competitive …rms. In any case, the total level of public good is equal to the sum
of all individual contributions. The key di¤erence with the social planner scenario is that the public good is
…nanced in a decentralized fashion by the voluntary contributions of individuals.

16.3.1 Individual contributions

If gi denotes the voluntary contribution of individual i— i.e., the quantity of private good that she decides
to use to purchase some public good— then once gi is …xed, her level of consumption of the private good is
ci = ! gi . Therefore, the only decision
P that individual i faces is choosing gi 0: The level of public good
g consumed by all individual is g = j2N gj .
The privately optimal contribution of an individual i is given by the solution of:
0 1 2 0 1 3
X X
max Ui @! gi ; gj A = max 4 i ln @ gj A + ! gi 5 : (19)
gi 0 gi 0
j2N j2N

Note that when solving (19), individual i chooses her own contribution gi , but she treats the others’ con-
tribution (gj )j6=i as exogenous parameters. So the solution to (19) is a function of the others’contribution
(gj )j6=i :

Proposition 7 Given the others’ contributions (gj )j6=i ; the contribution giP O that is privately optimal for
i— i.e., the solution to (19)— is 8 9
< X =
giP O = max 0; i gj : (20)
: ;
j6=i

P
Proof. To solve (19), observe that the maximand i ln j2N gj +! gi is strictly concave and the unique
constraint gi 0 is linear, so it is a convex program. Therefore, if gi denotes the solution to the program
(19) without the constraint gi 0; the solution to (19) is gi whenever gi 0; and it is 0 otherwise. In other
words, the solution is max f0; gi g. The F.O.C. of the unconstrained program is
h P i
@ i ln j2N gj + ! gi
= 0;
@gi
P
which yields gi = i j6=i gj , so the privately optimal contribution for i is (20).
The term i in the right hand-side P of (20) captures the intrinsic willingness of individual i to pay for the
public good, whereas the term j6=i gj captures the fact that i’s willingness
P to contribute for the public
good is crowded out by the others’contributions. This is because gi and j6=i gj are perfect substitutes Pin the
production of the public good. In particular, when the others’contribution is su¢ ciently large, i j6=i gj
is negative, so giP O = 0; which means that i prefers to free-ride on the contribution of the others instead of
contributing to the collective e¤ort.

16.3.2 The free-riding equilibrium

The privately optimal contribution of each individual i; as given by (20), does not immediately characterize
the equilibrium of this economy, because the others’contributions (gj )j6=i are unknown at this stage. The
private contribution gj of any other individual j 6= i is also given by the same condition as (20), which itself

45
depend on the contributions of the other individuals including i. Hence, the equilibrium contribution of each
individual cannot be derived in isolation. In game theoretic lingo, (20) de…nes the best response gi of each
individual i to the strategy (gj )j6=i of the other individuals, and an equilibrium is a pro…le of contribution
(gi )i2N such that each i 2 N; gi is a best response for i to gj j6=i . The solution to that system is given by
the following proposition.

Proposition 8 In equilibrium, only the individual(s) with the highest i contributes to the public good,
and all the other individuals free-ride. That is, in any equilibrium giF R i2N ; giF R > 0 if and only if
i = maxj2N j : Moreover, X
gF R giF R = max i : (21)
i2N
i2N

In particular, if only one individual, say individual 1; has the greatest i; then the unique equilibrium is given
by g1F R = 1 and for all i 6= 1; g1F R = 0:

Proof. An equilibrium of this economy is a pro…le of voluntary contributions (gi )i2N such that each gi solves
(19), and thus (20), given the equilibrium contributions (gj )j6=i made by the other individuals. Observe that
P
(20) implies that for all i 2 N; gi i P j6=i gj ; and gi = 0 whenever the latter inequality is strict.
Equivalently, (20) implies that for all i 2 N; j2N gj i ; and
P gi = 0 whenever the latter inequality is
strict. Since this inequality
P holds for all i 2 N; it implies that j2N gj maxj2N j ; so for all i 2 N such
that i < maxj2N j ; j2N gj > i , and from what precedes, gi = 0: This proves that only the individual(s)
with the greatest i P make positive contribution. P
Let us now showP j2N gj = maxj2N j : We have already proved that j2N gj maxj2N j : Suppose P by
contradiction that j2N gj > P max j2N j . Then from what precedes, no individual contribute, so j2N j =
g
0; which is impossible since j2N gj maxj2N j > 0:
P
If we compare Equations (21) and (18), since maxi2N i < i2N i , we conclude that g F R < g O :Thus,
providing the public good via decentralized and voluntary contributions leads to an ine¢ ciently low level
of provision, as compared to the social optimum. Moreover,when N is large (as is the case for a national
public good) and individuals
P have similar preferences (that is, they all have similar i ), maxi2N i is very
small compared to i2N i , so g F R is very small compared to g O . Thus, the ine¢ ciency of the “free-riding”
equilibrium— which, again, can be viewed as the free market equilibrium— increases rapidly as the number
of individuals increases.

16.3.3 The main take-away from the voluntary contribution model

This model shows that an economy without a coercive government in which the public good is …nanced by
the decentralized and voluntary contributions of the individuals performs very poorly, as compared to the
social optimum. The public good is dramatically underprovided. This severe market failure naturally calls
for a public intervention.
The intuition for the failure of the system of decentralized and voluntary contribution is the following.
When considering a voluntary contribution to the public good, each individual realizes that she bears the
full costs of her contribution, but since the public good bene…ts all N individuals in the economy (because
it is non rival and non excludable), she gets only a fraction 1=N of the social bene…t of her contribution.23
2 3 More precisely, when an individual makes a private contribution g to the public good, the total level of public good increases
P P i
from j6=i gj to gi + j6=i gj , so the impact of her contribution on her own utility is simply
0 0 1 0 11
X X
i @ln @gi + gj A ln @ gj AA ;
j6=i j6=i

while the impact of her contribution on social welfare is

0 1 0 0 1 0 11
X X X
@ k
A @ln @gi + gj A ln @ gj AA :
k2N j6=i j6=i

P i
Therefore, she internalizes a share of the social bene…t of her contribution. When all individuals have the same i,
k2N k
that share is indeed 1=N .

46
Thus, she internalize the full cost but internalize only 1=N of the social bene…ts of her contribution. For this
reason, private incentives to contribute are weak, especially when N is large. As a result, most individuals
prefer to consume their endowment of private good and let individuals who care the most about the public
good provide it. This type of behavior is known as the free riding problem.
It is important to note however that the ine¢ ciency of the voluntary contribution allocation does not
come from the fact that some individuals enjoy the public good without making any contribution. Whether
the cost of provision of the public good should be shared evenly across all individuals or concentrated on
a single individual is a fairness issue that does not a¤ect the level of utilitarian welfare as we de…ned it in
Section 16.2. As argued in Section 16.2, the level of utilitarian welfare does not depend on how a given
amount of aggregate consumption of private good is shared among individuals, so for a given level of public
good provision g, the level of utilitarian welfare does not depend on how the cost of providing g is shared
among individuals. The ine¢ ciency of the voluntary contribution model comes from the fact that individuals
do not internalize all the social bene…ts of their contributions. As a result, their incentives to contribute to
the public good are too weak.

16.4 Public good provision and the majority rule: the case of head tax …nancing
One of the most important tenet of political economy is that one cannot simply assume that decision makers
(politicians, civil servants, voters...) are benevolent and omniscient. To solve the “market failure” that
occurs when the public good is provided via a decentralized system of voluntary contributions, the “political
economy solution”is not to delegate the allocation of all resources to a hypothetical benevolent social planner
as we did in Section 16.2. Political economists prefer to rely on institutions and collective decision procedures
that lead to outcomes that are socially satisfactory even if the decision makers within these institutions and
procedures are neither omniscient nor benevolent. In this section, we analyze one such collective decision
procedure in which the public good is …nanced by a mandatory and uniform tax and the level of provision
is decided by majority rule.24

16.4.1 The head tax model

In the “head-tax model”, we assume that the public good is …nanced by a uniform tax imposed on on all
citizens— that is, all citizens are forced to make the same contribution— and citizens decide collectively the
level of tax and of public good provision via majority rule. The uniform tax is also often called a head tax,
or a uniform lump-sum tax. What outcome is likely to emerge from such a collective decision procedure?
The variables of this collective choice problem are the level of public good g and the level of the head tax
needed to …nance the public good. Note that the tax revenue is N , so the tax level needed to provide a
quantity g of public good must be such that g = N . Hence, for a given g; the head-tax must be = g=N;
so the collective choice problem is de facto one-dimensional: what level of public good g should be provided?

16.4.2 Voters’demand for the public good

To determine the majoritarian equilibrium, we must …rst determine the induced preferences of each voter on
the collective decision variable g (the notion of induced preferences is de…ned in Section 14.3). For a given
g 0; the tax paid by each individual i is = Ng ; so she consumes ! i Ng units of private good and g
units of public good. Therefore, her induced preferences over g are derived from the following induced utility
function:
g g
Vi (g) = Ui ! i ; g = i ln (g) + ! i : (22)
N N
Under majority rule, no single individual can unilaterally impose her most preferred collective choice. Nev-
eretheless, it is instructive to determine for each individual i what level of public good provision gi she would
pick if she was a dictator. This ideal policy gi solves the following maximization program:
h gi
max Vi (g) , max i ln (g) + ! i :
g 0 g 0 N
2 4 In the case of a perfectly competitive market without externalities, public goods, and information imperfection, a system of

free markets o¤ers an example of such a decision process: market participants are only required to know their own preferences,
observe public prices, and make consumption decisions that maximize their own preferences. The market achieves an e¢ cient
outcome nevertheless.

47
The solution to that maximization program is given by the …rst order condition @[ i ln(g)+! @g
i g=N ]
= 0,
provided that the constraint g 0 is not binding (we will check that it is indeed not the case afterwards).
Simple calculus shows that @V@g
i (g)
= gi N1 , so the …rst-order condition yields gi = N i . Observe that since
i > 0, gi > 0 as we assumed in the …rst place.
One can interpret gi as the demand for public good by individual i when the public good is …nanced by
head taxes. Note that each additional unit of public good increases the tax payments of individual i by 1=N:
Thus, the public good price for voter i under this tax system is 1=N: This price her tax-price for the public
good (the notion of tax-price is de…ned in Section 14.3).
Note that gi is increasing in i , which means that individuals which a greater willingness to pay for the
public good demand a higher level of public good, which is intuitive. Note also that gi is increasing in N .
The reason for that result is that in a more numerous society, the cost of the public good can be spread
on more people. Put di¤erently, the tax-price 1=N faced by voters is decreasing in N; so the public good
becomes cheaper as the population grows.

16.4.3 The Condorcet winner

Since individuals disagree over the ideal level of public good gi , the collectice choice problem is not straight-
forward, and it is not clear what level of public good would emerge under majority rule. To help us …gure
that out, we can use the Median Voter Theorem. To make sure we can use it, we must …rst check whether
the induced preferences over the level of public good, as given by (22), are single-peaked (see De…nition 15).
As argued in Section 15, these preferences are single-peaked if Vi (g) is concave. Di¤erentiating (22), we
obtain
@Vi (g) 1 1 @ 2 Vi (g) 1
= i and 2 = 2
< 0;
@g g N (@g) g
so Vi (g) is concave.
Let g HT denote the Condorcet winner when the public good is …nanced by head taxes. The MVT tells
us that g HT exists, and that g HT is the median of the “peaks” (gi )iN of all individuals. Since gi = N i ; if
m denote the median of the distribution ( i )i2N ; we have that

g HT = median (N i) =N m: (23)

If we compare the collective choice under majority rule and head taxes (23) with the equilibrium allocation
(21) under voluntary contributions, we see that if N m > maxi2N i , then g HT > g F R : This means that the
collective choice with majority rule and head taxes mitigates the underprovision of the public good relative to
the decentralized system of voluntary contributions, and vice versa if N m < maxi2N i . Intuitively, in the
case of national public goods, N is very large, so the case N m > max is the most likely. This is obviously
true when all citizens have similar preferences, in which case, m and max are of similar magnitude.
This discussion suggests that in this case, the “political solution” which consists in …nancing the public
good via a uniform tax, and letting the citizens decide by majority rule the level of public good provision, is
socially better than the “market solution” which consist in letting consumers purchase the public good on
a voluntary basis. Observe that consistently with our objective, this political solution does not assume that
voters are benevolent or omniscient. It only assumes that each citizen knows his own preferences, and votes
for the alternative that is best for her.
An interesting question is how the collective choice under majority rule and head taxes compares with
the social optimum. By comparing (23) to the Samuelson optimal condition (21), we see that whether this
political solution leads
P
to too much or too little public good provision (i.e. whether g HT 7 g O ) depends
on whether m 7 i2N N
i
. Speci…cally, if the median of ( i )i2N is greater than the mean of ( i )i2N ,
majority rule together with head taxes leads to too much public good and vice versa. Hence, the e¢ ciency of
the majority rule equilibrium depend on the symmetry and heterogeneity of the distribution P of preferences.
i
When the distribution of preferences is su¢ ciently symmetric or homogeneous, then m ' i2N N , and the
Condorcet winner is approximately socially optimal.

48
16.4.4 The main take-away from the head tax model
The analysis in the previous sections delivers an important result: a public intervention that consists in
deciding the level of public good provision via the majority rule and …nancing it via head taxes can lead
to an almost socially optimal level of public good, whereas public good provision via competitive markets
and voluntary contributions leads to a severe underprovision problem. The fact that majority rule together
with uniform, mandatory taxation can solve the free-rider problem is a powerful justi…cation for a public
intervention. In particular, when N ! 1 (which is is the case for pure national public goods), the market
failure (i.e., the di¤erence between the market equilibrium g F R and the social optimum g O ) is much greater
than the political failure (the di¤erence between the political equilibrium g HT and the social optimum g O ).
It is important to understand why uniform and mandatory taxation together with majority rule can
solve the free-riding problem. Recall the intuition behind the free-riding problem: when making a voluntary
contribution, an individual bears the full social cost of her contribution, but receives only a fraction 1=N
of the social bene…t (each voter care only about her own consumption of the public good, so she does
not internalize the bene…t of public good consumption by others). As a result, individuals’ incentives to
contribute are very weak, and they contribute too little in equilibrium. On the contrary, when the public
good is …nanced by a uniform tax, a given individual still receives only 1=N of the social bene…t of the public
good provision (she still cares only about her own consumption), but she now bears only a fraction 1=N
of the social cost, since the cost is shared equally among all individuals! In that case, the fact that voters
internalize only 1=N of the bene…t of public good provision is not a problem anymore, because this bias is
exactly compensated by the fact that they internalize the same fraction of the costs.

16.4.5 Discussion
Consistently with our political economy perspective, to establish the "almost social optimality" of the polit-
ical equilibrium, we did not assume that the public intervention was under the control of an omniscient and
benevolent planner. Neither did we assume that voters were omniscient and benevolent (they do not inter-
nalize the costs and bene…ts of other individuals). However, for this "political solution" to the free-riding
problem to work, we have made a certain number of questionable assumptions. More precisely:

The government must have the authority to raise taxes. That is, we need a coercive and legitimate
government with an e¢ cient tax administration, an assumption that is not satis…ed in some developing
countries.
The government must respect the preferences of the majority. Roughly speaking, we need the govern-
ment to be accountable to the population, and a majority of voters must be able to oust it if they are
dissatis…ed with its policy.

The government must use the revenue solely to …nance the public good. In other words, we need a
transparent government that cannot divert money for its own bene…t.
The individuals must have su¢ ciently homogeneous preferences for the public good. In other words,
this solution works only for public goods that a¤ect all individuals in a similar way, not for local public
goods (e.g., building a bridge in a remote district), or for public goods that are targeted to a subgroup
of the population (religious facilities).
All individuals must be able to a¤ord the head tax. This condition is not met in real-world democracies,
in which income can vary greatly across taxpayers.

Any departure from these assumptions will exacerbate the political failure. Quantifying the extra political
distortion generated by any such departures is beyond our scope. In the next section, we relax the last one.
In modern democracies, the tax paid by each individual is linked to her income/wealth level in a way that
guarantees that every individual is able to a¤ord the tax. We will see that such progressive tax systems
generate additional political distortions.

49
16.5 Public good provision and the majority rule: the case of a proportional
tax
In this section, we analyze whether the results derived in Section 16.4 in favor of a public intervention depends
on the way the public good is …nanced. In most democracies, taxes are not uniform. Richer individuals
pay more taxes than poorer individuals. The morale imperative of equality and compassion dictates that
individuals who are more fortunate— be it because they are more productive, better educated, or born from
richer parents— should help their less fortunate pears by making greater contribution to the public good.
The point of that section is not to discuss this rather uncontroversial feature of most tax systems. Instead,
our goal is to investigate how the progressivity of the tax scheme used to …nance the public good a¤ects the
popular demand for the public good, and the majoritarian equilibrium.

16.5.1 The proportional tax model

To investigate the impact of …nancing the public good with a progressive tax, we assume that the tax paid
by any individual is proportional to her initial endowment ! i .25 For simplicity, we assume that individuals
share the same preferences: for all i 2 N; i = 1.
Proportional taxation is arguably a very mild form of redistribution: in most democracies, richer indi-
viduals not only pay more taxes in absolute terms, but they also typically pay a larger share of their income
or wealth in taxes— absent any tax loopholes! In fact the poorest individuals often pay no income taxes, or
negative taxes. However, this simple tax structure will allow us to make a …rst step in understanding the
impact of redistribution on voters’preferences over public good provision.
Let us denote by the tax rate, so that the tax paid by each individual i is ! i . We assume for simplicity
that this tax is not distortionary, so total tax revenue is . Therefore, the feasibility constraint implies that
the tax rate needed to provide a quantity g of public good must solve g = , so = g . Hence, the level
of public good g uniquely pins down the tax rate that each individual must face, so the collective choice
problem is once again one-dimensional: what level of public good g should be provided?

16.5.2 Voters’demand for the public good

To determine the majoritarian equilibrium, we must …rst determine the induced preferences of each voter on
the collective decision variable g (the notion of induced preferences is de…ned in Section 14.3). For a given
level of public good g, the tax paid by each individual i is ! i = g ! i ; so she consumes ! i 1 g units
of private good and g units of public good. Therefore, her induced preferences over g are derived from her
utility function as follows:
!i !i
Vi (g) = Ui ! i g; g = ln (g) + ! i g:

If individual i could choose unilaterally the level of public good provision, she would choose the level gi that
maximizes Vi (g). The solution to the F.O.C. @V @g = 0 is gi = ! i , or equivalently,
i

!
gi = N ;
!i

where ! = N is the average wealth in the economy. As in the case of head taxes, gi can be interpreted as
her demand for the public good. Note that each additional unit of public good increases the tax payments
of individual i by !i : So !i can be interpreted as her tax-price for the public good.
We see that gi is decreasing in ! i : poorer individuals want more public good. The intuition for that
result is straightforward: since all individuals have the same intrinsic preferences (they all have the same
preferences parameter i ), poorer people demand a greater quantity of the public good, because it is cheaper
for them than for richer individuals: their-tax price for the public good is lower than for richer individuals.
A large quantity of public good allows them to indirectly extract some wealth from the wealthier individuals
via their greater public good contribution.
2 5 We can introduce money in this economy, and normalize the price of the private good to 1. In that case, ! can be
i
interpreted as individual i’s income/wealth. The tax is then a proportional income tax.

50
16.5.3 The Condorcet winner
Let us …rst check whether we can use the median voter theorem. Note that
@Vi (g) 1 !i @ 2 Vi (g) 1
= and 2 = < 0;
@g g (@g) g2

which shows that Vi (g) is concave. Therefore, voters have single-peaked preferences over g; so we can use
the MVT.
From our previous discussion, the “peak” gi = N !!i of each individual i is decreasing in her wealth ! i .
So the median peak (see De…nition 16) is given by the peak of the individual with the median wealth, which
we denote ! m . From Theorem 4, under proportional taxation, the unique Condorcet winner g P T is the
median peak:
!
gP T = N : (24)
!m
The socially optimal level of public good is still given by the Samuelson condition (18), and since we have
assumed that i = 1 for all i 2 N , it is
g O = N: (25)
If we compare the Condorcet winner under proportional taxation (24) to the social optimum (25), we obtain

a result very similar to the one we have derived in Section 16.4 with head taxes: if ! m < !, g P T > g O ; so the
public intervention leads to too much public good, whereas if ! m > !, g P T > g O ; so the public intervention
leads to too little public good.
Hence, whether there is over or underprovision of the public good in the voting equilibrium depends on
whether the median wealth ! m is smaller or greater than the average wealth !. This comparison depends in
turn on on the asymmetry (or skewness) of the wealth distribution. In the case of head taxes with preferences
heterogeneity analyzed in Section16.4, it was not clear whether the median i was typically greater or smaller
than the average i , because the preference parameter i is hard to map to quanti…able observables. On
the contrary, it is quite clear that income and wealth vary widely among citizens of a given country. And we
know the typical shape of income distribution in most countries: it is highly skewed to the right. That is,
the mean income ! is signi…cantly greater than the median income ! m , because the mean income is “pulled
to the right” by the very rich. This means that the public good will be typically be overprovided! Hence,
when the public good is …nanced via a redistributive tax system, the direction of the political failure is clear:
there is too much public good, so the political failure goes in opposite direction to the market failure.
It is instructive to compare public good provision under proportional taxes and under head taxes. Under
head taxes, when all i are equal to 1 (as we have assumed in this section), the majority rule equilibrium is
g U T = N , while under proportional taxation, it is g P T = N !!m . So with uniform taxation, the public good
is optimally provided, while under proportional taxation, it is systematically overprovided (in the realistic
case ! m < !). It is important to recall the intuition for that result: under head taxes, the tax-price faced by
the median voter is N1 , whereas under proportional taxes, the tax-price faced by the median voter is N1 !!m ;
which is typically smaller than N1 since ! m is typically smaller than !. So under proportional taxation, the
median voter pays a smaller share of the public good, and the public good appears cheaper to her, so she
demands more of it.

16.5.4 The main take away from the proportional tax model
The results derived in the proportional tax model are fundamental in political economy: when public goods
are …nanced via redistributive taxation, poor voters have an incentive to vote for an ine¢ ciently high level of
public good provision, because they pay a smaller fraction of the public good relative to rich voters, so they
use public good provision to redistribute money from the rich to themselves. This form of redistribution is
ine¢ cient, because it is achieved by an ine¢ ciently high level of public good provision, rather than via a
direct transfer from the rich to the poor.
In the real world, this distortion could be even larger than the model suggests because in modern democ-
racies, taxes are more progressive than a simple proportional tax. This means that rich people pay more
taxes not only because they are taxed on a greater income, but also because face a higher tax rate (their pay

51
a larger share of their income in taxes). With the notation of our model, is not the same for all individuals:
is greater for richer individuals. For instance, in the U.S., about 50% of the households do not pay any
income tax because their income is too low. Obviously, in such an environment, the incentives of a majority
of voters to vote for an ine¢ ciently high level of public goods and services are quite strong.26
It is very important to note here that the problem is not that poorer individuals pay less taxes. This is
precisely what redistribution is about, and most economists and political scholars agree that some degree of
redistribution is desirable. The problem highlighted in this section is that when the public good is …nanced by
a redistributive tax, the degree of redistribution distort voters’demand for the public good, and this distortion
induces ine¢ ciently high demand for the public good. Pushing this reasoning further, ideally, we should vote
separately on a redistributive tax scheme, and on public good provision …nanced via head taxes. The …rst
vote would achieve the desired degree of redistribution, and the second vote would allow citizens to express
their willingness to pay for the public good, independently of their desire for redistribution.
Thus an important lesson from the public good model is that economic institutions can interact with po-
litical institutions, because economic institutions (the tax structure in this case) can a¤ect voters’incentives
(that is, they a¤ect their induced preferences), and thus the political equilibrium. In this case, we have seen
that more redistributive …scal institutions tend to lead to an excessively large public sector.

17 Voting over multiple policy dimensions

To understand where we are going now, let me summarize what we have learned in Chapter III so far.
The goal of the previous sections was to make predictions on the majoritarian outcome in di¤erent settings.
We argued that when a Condorcet winner exists, the Condorcet winner is a realistic prediction for the
majoritarian outcome. We introduced the median voter theorem, which states that under some conditions,
a Condorcet winner exists. These conditions are basically twofold: the space of alternative must be one-
dimensional, and the preferences of the voters must be single-peaked. We then applied this theorem to various
economic environments. We saw that the assumption of single‘peaked preferences was satis…ed under rather
mild conditions.
However, the main limitation of these applications is that we do not allow the citizens to vote on multiple
issues simultaneously. In most voting situations (e.g., general elections, or legislative bargaining), several
policy dimensions are considered at the same time. So the key question that we are interested in this section
is: can we extend the median voter theorem to more than one policy dimension? To investigate this question,
we consider a multidimensional set of alternatives, but we still assume that individuals have preferences which
are single-peaked on each dimension.

17.1 The multidimensional Euclidean model

Let us …rst consider the simple case of two dimensions with Euclidean preferences (the term Euclidean
preferences will become clear in a moment). Formally, the space of alternatives is X = R2 , so an alternative
x can be decomposed into two coordinates, which we denote x1 and x2 . One can think of x1 as the level of
public spending on public good 1, say primary education, and of x2 as the level of public spending on public
good 2, say transportation infrastructure. In that interpretation, x1 and x2 would have to be positive.
The preferences of an individual i 2 N are given by the following utility function: for all x 2 R2 ,
i 2 i 2 i 2
Ui (x) = x = x1 1 x2 2 ; (26)
i
where the vector 2 R2 is the “peak” of individual
q i. The term kx yk refers to the Euclidean distance
2 2
between any two points x; y 2 X, that is kx yk = jx1 y1 j + jx2 y2 j . In this section, we do not derive
the preferences of voters on policies from their primitives preferences on outcomes as we did for instance in
Section 16. Instead, we specify them directly because the details of the preferences will not matter for the
main message of this section. Hence, one can view (26) as an induced utility function on policies x1 and x2 .
As you can see from this utility function,
2 6 It should be noted that even though these individuals do not pay income taxes, they contribute in a substantial way to

the public budget through other taxes such as payroll taxes and sales taxes, though it is not clear whether they take these tax
contributions into accont when voting.

52
 i i
Ui = 0 and Ui (x) < 0 for all x 6=
i
Ui (x) becomes more and more negative as x gets farther away from

Hence, i is individual i’s most preferred alternative in X, and her preferences are such that for any
two policy bundles x = (x1 ; x2 ) and y = (y1 ; y2 ), x is preferred to y if and only if x is closer— in the sense
of the Euclidean distance— than y to her ideal point i . This is why such preferences are called Euclidean
preferences.

17.2 Voting issue by issue: An illustrative example

i 2
Remark 7 From (26), if we …x x2 ; then Ui (x) = x1 1 + cte; where the term cte does not depend on
i
x1 ; so Ui is concave in x1 with a maximum at 1 : This implies that if we …x x2 , the preferences of individual
i over x1 are single-peaked (in the sense of De…nition 15) with a peak at i1 . Likewise, if we …x x1 , then
i 2
Ui (x) = x2 2 + cte; where the term cte does not depend on x2 : This means that if we …x x2 , the
preferences of individual i over x2 are single-peaked with a peak at i2 . Hence, this type of preference is the
natural multidimensional extension of the class of single-peaked preferences that we have introduced in the
one-dimensional case in De…nition 15. To make things more concrete, in this environment, i1 and i2 can
be interpreted as the demand of individual i for public good 1 and 2; respectively.

In this environment, the pro…le of preferences ( i )i2N is completely characterized by the pro…le of ideal
points i i2N . A natural question is whether the Median Voter Theorem extends to 2 dimensions, that is,
whether the single-peakedness of preferences in two dimensions imply the existence of a Condorcet winner.
It turns out that the answer to that question is almost always no. To see the kind of di¢ culty one can get
into when voting on multiple dimensions, consider the following pro…le of ideal points: N = 3 and
8 1
< = (0; 2)
2
= (1; 0) :
: 3
= (2; 1)

To make some sense of the above pro…le of preferences, here is a little story that is consistent with the above
preference pro…le. Recall that public good 1 is education and public good 2 is investment on transportation
infrastructure (roads). Voter 1 is a salesman who has no children, so he does not care about education, but
he spends a lot of his working hours on roads, so he cares a lot about road quality and would like 2 units of
spending roads. Voter 2 has three children and no car, so he does not care about roads, but he cares about
primary education. However, since he is relatively rich, he understands that too much public spending on
education will increase disproportionately her tax contribution, so he wants only 1 "unit’ of spending on
education. Voter 3 is a truck driver with children, so he cares a lot about roads and about education, and
he is less concerned about high taxes than voter 1 and 2 because his income is relatively low, so he pays
a small fraction of the public goods. Thus, ideally, he would like 1 unit of spending on education and two
units of spending on roads.

17.2.1 Voting in each public good separately

Suppose that we …x the coordinate of the second dimension x2 , and consider the preferences of the voters
over the …rst dimension x1 . From Remark 7, all individuals have single-peaked preferences over x1 , and their
peaks are 11 = 0, 21 = 1, and 31 = 2. So on the …rst dimension, voter 2 is the median voter. This implies
that if we …x x2 , and if we let individuals vote on x1 , the median voter theorem applies (because we are
considering a single dimension at a time), and the Condorcet winner is xCW1 = 1.
Let us now …x the coordinate of the …rst dimension x1 and consider the preferences of the voters over the
second dimension x2 . From Remark 7, all individuals have single-peaked preferences over x2 , and their peaks
are 12 = 2, 22 = 0, and 32 = 1.27 So on the second dimension, voter 3 is the median voter. This implies
2 7 To see that, observe the utility function of individual i is given by
i 2
Ui (x) = x2 2 + Cte;

53
that if we …x x1 , and if we let individuals vote on x2 , the median voter theorem applies (again, because we
are considering a single dimension at a time), and the Condorcet winner is xCW 2 = 1.
From this reasoning, it seems that a natural candidate for the Condorcet winner is the alternative
xCW = (1; 1): the level of public spending should be 1 unit on each type of public good. However, it turns
out that this alternative is not a Condorcet winner. For instance, one can easily see from a graph that
the point x0 = (3=2; 1=2) is closer to the ideal points of individual 2 and 3, so it is majority preferred to
x = (1; 1). In fact, as we shall how in that class, in that example, there is no Condorcet winner! So the
median voter theorem does not hold in that environment.

17.2.2 Voting on the size and allocation of the public budget

To see why decomposing the space of alternatives into two dimensions and letting people vote on each
dimension is problematic, consider the following alternative voting mechanism. One could …rst vote on the
total spending on both public goods, that is, x1 + x2 , and on how much more (or less) should be spent on
public good 1, that is, x1 x2 . This voting mechanism has a natural interpretation in that it let voters
vote on the size of the public budget (or equivalently, how much tax should be raised), and then one how
to allocate that budget across public goods. Along those lines, one can interpret i1 + i2 as a measure of
how much voter i wants the public sector to spend on both public goods together (that is, the dimension
x1 + x2 ), or equivalently, how this voter trade o¤ more public good versus less taxes, while i1 i
2 can be
i i
viewed as a measure of how much voter i prioritize public good 1 versus public good 2 if 1 2 > 0 or vice
versa if i1 i
2 < 0, and thus whether she would like to spend more on public good 1 (that is, x1 x2 > 0)
or on public good 2 (that is, x1 x2 < 0).

Remark 8 If we rede…ne the variables of the collective choice problem as follows:

x1 +x2
x
^1 = 2 ;
x1 x2
x
^2 = 2 ;

then one can express the welfare Ui (x) of each voter i, as given by (26), as a function of the variables (^
x1 ; x
^2 )
as follows:
i 2 i 2
Ui (x) = U^i (^
x) 2 x^1 ^ 2 x^2 ^ 1 ; 2

i
where the new ideal points ^ are given by
(
^i =
i i
1+ 2
1 2 ;
^i =
i i

2
1
2
2
:

Hence, we see from the above formula for U ^ for that if we …x x^2 , the preferences of individual i on x
^1 are
i
single peaked with a peak at ^1 , and if we …x x
^1 , the preferences of individual i on x
^2 are single peaked with
i
a peak at ^ .
2

Proof. One can express x as a function of x

^ as follows:
i i i
x1 = x
^1 + x
^2 , x2 = x
^1 x
^2 , i
= ^1 + ^2 , and i
= ^1 ^i :
1 2 2
2
@
where the term Cte does not depend on x2 . Therefore, @x2
Ui (x) = 2 < 0, so the utility function of individual i over the …rst
2
i
dimension x2 is indeed single-peaked with a peak at 2.

54
This allows us to rewrite (26) as a function of x
^ as follows:
2 2
Ui (x) = (^
x1 + x
^2 ) ^i + ^i (^
x1 x
^2 ) ^i ^i
1 2 1 2
2 2
= x
^1 ^i + x
^2 ^i x
^1 ^i x
^2 ^i
1 2 1 2
2 2
= x
^1 ^i 2 x
^1 ^i x
^2 ^i x
^2 ^i
1 1 2 2
2 2 2
x
^1 ^i +2 x
^1 ^i x
^2 ^i x
^2 ^i
1 1 2 2
2 2
= 2 x
^1 ^i 2 x
^2 ^i :
1 2

The distribution of preferences with that new interpretation of the policy space is given by
8 1
< ^ = (1; 1)
>
^2 = (1=2; 1=2) :
>
: ^3
= (3=2; 1=2)

So the median voter on x ^1 is 1, with a peak at x

^1 = 1, and the median voters on x
^2 are voters 2 and 3, with
a peak at x
^2 = 1=2. So when voting along these dimensions, the outcome of the vote would be x ^ = (1; 1=2),
or equivalently, in the old system of policy coordinates, x = (3=2; 1=2). So the outcome with this voting
mechanism di¤ers from the outcome with the previous voting mechanism!

17.2.3 Agenda power

This last result is quite disturbing, because there is no obvious reason why one of the two voting mechanism
is better than the other. Both are based on majority rule, they di¤er only in that they use a di¤erent system
of coordinates on the policy space. Put di¤erently, these two voting mechanisms di¤er only in that they
phrase the collective choice problem di¤ erently to the voters. The …rst mechanism asks: what level of each
public good do you want? The second mechanism asks: how much do you want to spend on public goods,
and how do you want to allocate that budget across public goods? This suggests that the person who is in
charge of designing the voting mechanism or the referendum can have some in‡uence on the outcome. In
voting theory, this phenomenon is called agenda power.
The fact that the outcome of the vote depend on the way the question is phrased to the voters is driven
by the fact that in the above example, the median voter was not the same in all policy dimensions. When
voting on x1 , voter 2 was the median voter; when voting on x2 , voter 3 was the median voter; when voting
on x^1 , voter 1 was the median voter. This disturbing feature of the preferences pro…le was driven by the
fact that in that example, there was no Condorcet winner. Hence, a natural question is whether the above
example was a special case.

17.3 When does a multidimensional Condorcet winner exists?

Let us …rst point out that there are also examples of pro…les of ideal points i i2N such that one individual
is the median voter in all directions. A trivial example is a pro…le of ideal points such that everybody has
the same ideal point. That ideal point would obviously be the Condorcet winner, because it is preferred to
any other alternative by all voters. Consider the following, less trivial example: N = 5 and
8 1
>
> = (1; 1)
>
> 2
< = (2; 1)
3
= (1; 2) :
>
>
> 4 = (0; 1)
>
: 5
= (1; 0)

55
One can easily see by drawing a graph that this pro…le of ideal points is symmetric around the point (1; 1),
and voter 1 is the median in all directions. In that case, it turns out that the point (1; 1) is the unique
Condorcet winner. The following de…nitions explains formally what is needed for an ideal point to be a
median in all directions:

De…nition 17 Given a pro…le of ideal points ( i )i2N , a median line L in R2 is a straight line such that less
than a majority of the voters are (strictly) one side of L, and less than a majority of voters are (strictly) on
the other side of L.

De…nition 18 Given a pro…le of ideal points ( i )i2N , we say that a voter m 2 N is a median voter in all
directions if every median line L in R2 goes through m .

For instance, in the above example with N = 3, if we consider the vertical line that goes through
4
= (0; 1), four voters are on the right side of the line, which is a majority so that line is not a median line.
On the contrary, if we draw the vertical line that goes through 1 = (1; 1), we see that only voter 4 is on the
left of the line, and only voter 2 is on the right of the line, so that line is a median line. Of course, to check
that voter 1 is the median voter in all direction, we would have to check that this is true for all lines that
go through 1 , not only the vertical one. However, one can easily see that because the picture is symmetric
around 1 , it must be true.
The following theorem formalizes the previous discussion.

Theorem 5 (Plott, 1973) If X = R2 , and if for all i 2 N , the preferences of individual i are given by the
utility function (26) for some pro…le of ideal points ( i )i2N 2 RN , then there exists a Condorcet winner if
and only if one voter is median in all directions. When this is the case, the Condorcet winner is the ideal
point of that median voter.

The most important statement of that theorem is that no Condorcet winner exists when no individual
is a median in all directions. The intuition for that result is that there exists a median voter in all direction
only for very special pro…le of ideal points. In fact, one can make that statement a little bit more precise. If
the ideal points of the voters are generated by any "reasonable" random process (for concreteness, think of
a random process as throwing N darts on the blackboard, and the ideal points of the voters are the point
at which the darts hit the board), then the probability that a voter is median in all direction is 0.28 Hence,
with more than one dimension, a Condorcet winner exists only under very exceptional circumstances! In
other words, when the collective decision problem has more than one dimension, there will almost always be
a Condorcet paradox: the majority rule cycles inde…nitely and there is no majority preferred alternative.
The previous gives you an easy way to check whether there exists a Condorcet winner for a given
distribution of ideal points ( i )i2N . If you can …nd three median lines that do not intersect in the same
point, then from De…nition 18, there is no voter who is median in all directions, and from Plott’s theorem,
there is no Condorcet winner.29 I will illustrate in class how to use that powerful technic.

17.4 The Chaos Theorem

This results will be explained graphically in class.

17.5 More exercises

Exercise 38 Consider a two dimensional space of alternative X = R2 , and …ve voters N = f1; 2; 3; 4; 5g
whose preferences are derived from the following utility function: for all i 2 N , for all x 2 R2 ,
i 2 i 2
Ui (x) = x1 1 x2 2 :

In each of the following questions, justify your answer, with a graph, with math, and/or with words.
2 8 By N
reasonnable, I mean any random process which admits a probability density function over R2 .
2 9 You
need at least three median lines, because two lines always cross in a single point. You do not need more than three
median lines to show that a Condocet winner does not exist, because if there is a set of lines that do not all intersect in the
same point, necessarily, three of them must not intersect at the same point.

56
 i
1. Suppose that the ideal point of each of the …ve voters are:
1 2 3 4 5
= (2; 0) , = (2; 1) , = (1; 2) , = (2; 1) , and = (3; 2) :

(a) Suppose that voters …rst vote on x1 and then on x2 by majority rule. What is the outcome of the
vote?
(b) Suppose that voters …rst vote on x2 and then on x1 by majority rule. What is the outcome of the
vote?
x1 +x2 x1 x2
(c) Suppose that voters …rst vote on 2 and then on 2 by majority rule. What is the outcome
of the vote?
5
2. Suppose now that we change the ideal point of voter 5 as follows: = (1; 0). Is there a Condorcet
winner on X?
5
3. Suppose now that we change the ideal point of voter 5 as follows: = (4; 3). Is there a Condorcet
winner on X?

Exercise 39 (Voting separately on each public program versus voting on the total budget) In this
exercise, we consider an economy with a set N if individuals and three goods: one private good and two public
goods. The welfare of each individual i 2 N depends her level of consumption of the private good, denoted
ci , the level of consumption of the public good 1, denoted g1 , and the level of consumption of the public good
2, denoted g2 . Her preferences over ci , g1 , and g2 are derived from the following utility function:

Ui (ci ; g1 ; g2 ) = i ln (g1 ) + (1 i ) ln (g2 ) + ci ;

where i 2 [0; 1] is a parameter that determines the preferences of individual i for public good 1 relative
to public good 2. Individuals di¤ er not only in their public good preference parameter i , but also with in
their income ! i . The two public goods are publicly provided, and are …nanced by a proportional income tax:
an individual i with income ! i pays a tax s! i . We assume that the proportional tax does not generate any
ine¢ ciency or deadweight loss, so the total tax revenue is s ; where denotes the aggregate income. The
provision of one additional unit of either public good requires 1 unit of private good. So for a given tax rate
s, the levels of public good consumption (g1 ; g2 ) that are feasible are

g1 0, g2 0,
g1 + g2 s :
Notice that once s, g1 , and g2 are …xed, the allocation of goods across all individuals in this economy is
completely determined: each individual i consumes an amount g1 of public good 1, an amount g2 of public
good 2, and an amount ! i (1 s) of private good. So the collective choice problem is a three dimensional
one. As usual, to reduce the dimensionality of the problem, we can assume without loss of generality that
the government budget constraint binds. Since g1 + g2 = s , (g1 ; g2 ) pins down s, so we have in fact a two
dimensional collective choice problem.
Throughout this exercise, we will assume for simplicity that N = 3 and that the preferences of these three
individuals are given by

( 1 ; !1 ) = (0:2; 1) , ( 2 ; !2 ) = (0:5; 3) , ( 2 ; !2 ) = (0:8; 4) (27)

1. Derive the level of welfare of each voter i as a function of (g1 ; g2 ) only, and let us denote that induced
utility function Vi (g1 ; g2 ). Show that if we …x g2 , Vi (g1 ; g2 ) is single peaked in g1 and determine its
peak g1i; . Provide the general formula for g1i; for an arbitrary N; and then the value of g1i; for each
voter i 2 f1; 2; 3g given the preferences assumed in (27).
2. Suppose that the level of provision of the public good 2 is …xed at some exogenous level g2 , and that g1
is determined directly by the tax payers via majority rule. What is a likely outcome of the vote on g1
as a function of the distribution of income, preferences, and of g2 ?
3. Answer questions 1 and 2 for g2 , …xing g1 . Is there a voter that is median on both dimensions? Can
you change the preference of one voter so as to change your answer to that question?

57
 4. Suppose now that instead of voting separately on each public good, voters vote separately on the tax rate
s 0, and on the share of the budget that is allocated to public good 1, that is, sg1 2 [0; 1]. The vote on
s can be interpreted as a vote on the level of taxation, or equivalently on the level of public spending
(because s determines g1 + g2 via the government budget constraint g1 + g2 = s ), and the vote on sg1
can be interpreted as a vote on how to allocate the level of public spending across public goods (because
the share of the budget sg1 spent on public good 1 determines the share of the budget sg2 spent on public
good 2 via the budget constraint: sg2 = 1 sg1 ).
Formally, if we denote r sg1 the share of public spending allocated to the public good 1, we can express
the “old” variables of the collective choice problem (g1 ; g2 ) as a function of the new variables of the
collective choice problem (r; s) as follows:

g1 = rs
g2 = (1 r) s

Given the above identities, show that the level of welfare of each voter i as a function of (s; r) can be
written as follows: (hint: you can simplify this expression a great deal by using the fact that ln (ab) =
ln (a) + ln (b))

Wi (r; s) = i ln (r) + (1 i ) ln (1 r) + ln (s) + ln ( ) + ! i s! i :

5. Show that if we …x r, Wi (r; s) is single peaked in s and determine its peak si . Provide the general
formula for si; for an arbitrary N; and then the value of si; for each voter i 2 f1; 2; 3g given the
preferences assumed in (27).
6. Suppose that r is …xed at some exogenous level, and that s is determined directly by the tax payers via
majority rule. What is a likely outcome of the vote on s as a function of the distribution of wealth,
preferences, and of r?
7. Answer questions 5 and 6 for r, …xing the tax rate s.
8. Using your answers to the previous questions, answer the following question: it is equivalent to vote
on g1 and then on g2 , and to vote on s and then on r? Can you change the preference of one voter so
as to change your answer to that question?
9. Suppose now that all public decisions are taken by a single individual i 2 N (that public decision maker
is subject to the same tax as the other individuals). Would her decision be di¤ erent if he chooses …rst
g1 and then g2 , or if she chooses …rst s and then r? Comment.

Solution. This exercise will be solved in class.

58
Chapter IV
Electoral competition
18 Introduction
In Chapter III, we have investigated what kind of economic outcomes majority rule would lead to into
various economic environments. The crucial assumption that underpins the analysis in Chapter III is that
whenever a Condorcet winner exists, the political decision procedure would lead to that outcome. The basic
justi…cation for that assumption is that if a Condorcet winner exists, then as soon as that alternative is
proposed, by de…nition, it would beat the current status quo by majority rule. It would thus become the
new status quo. Moreover, by de…nition of the Condorcet winner, no other alternative would beat it by
majority rule.
However, this justi…cation does not explain why the Condorcet winner would be proposed in the …rst
place if it is not already the current status quo. This is a reasonable assumption whenever policies are
proposed directly by the citizens (because some voters will prefer the Condorcet winner to the status quo).
However, in modern democracies, the impetus for policy change does not typically come directly from the
voters. Most decisions are taken by political representatives, not by the citizens themselves. of course, these
representatives are elected by the voters, but this means that voters choose representatives, not policies
directly. This distinction is important because during elections, voters typically have the choice between a
small number of candidates, and there is no guarantee that one of them will propose a policy that is close
to what voters want.
That is where electoral competition comes into the picture. For the supply of private goods and services,
economic competition is the driving force that is supposed to induce …rms proposes the goods and services
that the consumers want at a reasonable price. When it comes the supply of collective decisions and policies,
the force that is supposed to induce politicians to propose the right policies is electoral competition, that is,
the competition for votes.
In this section, we will investigate to what extent electoral competition drives politicians to propose
the policies that the majority prefers. To do so, we need to understand how the incentives faced by the
candidates and their expectation about voters’behavior shape the political equilibrium. To do so, we will
analyze a strategic model of electoral competition in which the main driver of candidates behavior is the
competition for winning o¢ ce and to implement the policy they prefer. That is what the Downs model and
its many variants are about.
In a sense, the approach in Chapters II and III was focusing exclusively on the demand for public policies.
We will now look at the supply of public policies by political candidates.

19 The standard electoral competition model: the Downs model

19.1 The assumptions of the Downs model
In this course, for the sake of clarity, we will analyze a simpli…ed version of the Downs model. The simpli…-
cation is that we will restrict attention to the case of three alternatives. This simpli…ed version of the model
is enough to convey the main intuition of the original model.

The alternatives:
The set of alternative (also called the policy space) is denoted X as usual, and X = fl; m; rg.
The candidates:

– Two candidates denoted A and B

– Each candidate A and B proposes a policy platform denoted pA 2 X and pB 2 X, respectively.
– The full commitment assumption: the elected candidate has to implement his policy platform.
In other words, the winner cannot revise her policy platform once elected. She cannot lie during

59
 the election period. Therefore, if w 2 fA; Bg denotes the winner, and pw her policy platform, the
policy that will be implemented after the election is pw .
– O¢ ce motivated candidates: candidates care only about being elected, they do not care what
policy platform they propose. More formally, candidates are assumed to choose the policy platform
that maximize the probability of being elected.
The voters:
– The set of voters is denoted by N as usual.
– For simplicity, voters do not abstain.
– Each voter has (rational and resolute) preferences over the alternatives. Voters do not care about
who is elected, they only care about what policy (i.e., what alternative) the elected candidate
implements once in o¢ ce. Therefore, for each voter, it is optimal to vote for the candidate
that proposes the policy she prefers.30 In the case of a tie (a voter is indi¤erent between the
policies proposed by the two candidates), we assume that voters vote for each candidate with
equal probability.
– The preference pro…le: for simplicity, N is composed of three kind of voters: leftist, moderate,
and rightist voters.
The preferences of leftist voters L are given by:
l L m L r:
The preferences of moderate voters M are given by:
m M l and m M r:
Whether moderate voters prefer r to l or vice versa does not play an important role. To …x
ideas, we will assume that they strictly prefer l to r.
The preferences of rightist voters R are given by:
r R m R l:
– We assume that neither the leftists nor the rightists form a majority of the population. Therefore,
the coalition of the moderates and the rightists form a majority, and so does the coalition of the
moderates and the leftists.
The electoral rule: the winner of the election is the candidate that get the most votes (plurality
rule). In case of a tie, we assume that each candidate is elected with equal probability.
The timing of the game:
1) Candidate propose their policy platform pA and pB simultaneously
2) Voters observe the pro…le of policy platforms (pA ; pB ) and vote accordingly
3) The winner is determined according to the electoral rule and implements her policy platform.
Remark 9 Note under our assumptions, the majority preferences over the alternatives are rational (there
is no Condorcet paradox), and are as follows:
m MR l MR r:
In particular, alternative m is the Condorcet winner. As we will see, whether the Condorcet winner is m,
r, or l does not play an important role in the Downs model. What matters is that there exists a Condorcet
winner. The case in which there is no Condorcet cycle is discussed in Section 22.
3 0 When there are only two candidates, there is no reason for a voter not to vote for the candidate that propose her most

preferred policy. Let me state this more formally (i.e., in game theoretic lingo). Suppose that candidates have already proposed
their policy platform pA and pB , and consider the voting game between voters once pA and pB are …xed. For each voter, among
these two policy platforms, there is one that she prefers (two is she is indi¤erent between pA and pB ). Voting for the candidate
that proposes that preferred policy is the unique (weakly) dominant strategy for that voter. This not necessarily true with
more than 2 candidates. It might make sense to vote for a second best candidate who is more likely to win the election than
the …rst best candidate.

60
19.2 Electoral and policy outcomes
For reasons that will be clearer in the next section, to analyze the electoral competition between the two
candidates, we …rst need to determine who wins the election in every possible scenario, that is, for every
possible policy proposals (pA ; pB ) 2 X 2 by the two candidates. This information is conveniently summarized
in a matrix called the matrix of electoral outcomes M e :
De…nition 19 The matrix of electoral outcome is the matrix that give the election probability of each can-
didate ( A ; B ) for every possible pro…le of policy platforms of the two candidates (pA ; pB ).
By convention, the lines of M e correspond to the policy platform of candidate A and the columns of M e
correspond to the strategy of candidate B. Each cell of the matrix M e contains two numbers: the …rst is
the probability A that A wins the election, the second is the probability B that B wins the election.
For instance, if candidate A proposes the policy platform pA = l and candidate B proposes the policy
platform pB = m, given the assumption we have made about voters’behavior, leftist voters will vote for A,
while moderate and rightist voters will vote for B, so candidate B will win with probability 1 and candidate
A is elected with probability 0. By assumption, the coalition of moderates and rightist form a majority.
e
Therefore, if Ml;m denotes the entry of the matrix M e that corresponds to the line l and the column m,
e
Ml;m = (0; 1).
Consider now the case in which candidate A and B proposes the same policy platform, say, pA = pB =
l. Given the assumption we have made about voters’ behavior, all voters will vote randomly with equal
probability for each candidate, so each candidate will receive half of the vote share, and each candidate will
e
be elected with probability 1=2. Therefore, Ml;m = (1=2; 1=2).
Following the same logic, for all possible policy platform pro…les (pA ; pB ) 2 X 2 , given the assumed
preference pro…le, we get the following electoral outcome matrix:
0 1
pA npB l m r
B l (0:5; 0:5) (0; 1) (1; 0) C
Me = B@ m
C (28)
(1; 0) (0:5; 0:5) (1; 0) A
r (0; 1) (0; 1) (0:5; 0:5)
In some cases, we will be interested not in the identity of the winner of the election, but in the policy
that the winner of the election will implement (remember that under our assumption of full commitment,
the winning candidate w 2 fA; Bg is forced to implement the policy pw he has proposed during the election).
This information is conveniently summarized in the matrix of policy outcomes M p .
De…nition 20 The matrix of policy outcome is the matrix that gives the policy implemented by the winner
of the election for every possible pro…le of policy platforms of the two candidates (pA ; pB ).
For instance, as argued earlier, if candidate A proposes the policy platform pA = l and candidate B
proposes the policy platform pB = m, given the assumption we have made about voters’behavior, candidate
B wins with probability 1, so under the full commitment assumption, policy m is implemented. Therefore,
p
Ml;m = m. Following the same logic, for all possible policy platform pro…les (pA ; pB ) 2 X 2 , we get the
following policy outcome matrix:
0 1
pA npB l m r
B l l m l C
Mp = B @ m
C: (29)
m m m A
r l m r
Remark 10 Observe that as claimed in the introduction, contrary to Parts II and III, the set of policies on
which voters can vote is not exogenously given (voters do not vote directly on the set of alternatives X), but
are provided by the political candidates. The main focus of our analysis will be to investigate what policies
are proposed by the political candidates.
Exercise 40 When deriving the matrices of electoral outcome and policy outcome, we have assumed that
the moderate voters prefer policy l to policy r. Derive these two matrices if we make the reverse assumption
that moderate voters prefer policy r to policy l.

61
19.3 Predictions of the Downs model
The main value of the Downs model is to make predictions on what policy platform the candidates will
propose. Note …rst that making such a prediction is not easy, because even if we assume for simplicity that
candidates only care about the probability of being elected (the o¢ ce motivated candidate assumption), one
can easily see from the electoral outcome matrix M e in (28) that the probability that a candidate is elected
does not depend only on her policy platform, but also on the policy platform of the other candidate. For
instance, if candidate A proposes pA = l and candidate B propose pB = r, then we can see from M e (just
read the top right entry) that A is elected with probability 1, but if B proposes pB = m instead of pB = r,
then A is elected with probability 0. Hence, if A thinks that B will propose pB = r, then pA = l is a sensible
strategy for candidate A. However, if candidate B proposes pB = m instead, then pA = l is not anymore
a sensible strategy for candidate A. Therefore, each candidate must react to the policy proposed by the
other candidate, which means that each candidate must form a conjecture about the strategy that the other
candidate will choose.
In economics as in political science, the tool that it used to analyze this type of complex strategic
interaction is game theory. The central prediction tool of game theory is the notion of Nash equilibrium. It
basically states that the likely outcome of a strategic interaction of the sort described above is that all players
correctly forecast the strategy played by other players, and all players choose the strategy that maximize
their payo¤ given the belief they have about the others’strategies. In the case of electoral competition, the
concept of Nash equilibrium translates as follows:

De…nition 21 An electoral equilibrium with o¢ ce motivated candidates is a pair of policy platforms (peA ; peB ) 2
X 2 such that taking the platform of the other candidate as given, no candidate can change her own policy
platform and increase strictly her probability of being elected:

if candidate B keeps her policy platform at peB , candidate A cannot increase her probability of winning
by proposing policy platform pA 2 X instead of peA .
if candidate A keeps her policy platform at peA , candidate B cannot increase her probability of winning
by proposing policy platform pB 2 X instead of peB .

Hence, an electoral equilibrium is a Nash equilibrium of the electoral competition game where the payo¤
of the candidate is given by the electoral outcome matrix in (28).

Remark 11 In terms of electoral outcome matrix M e , an electoral equilibrium is a cell of M e such that the
…rst coordinate cannot be increased by going up or down (no pro…table deviation for candidate A), and the
second coordinate cannot be increased by going to the left or to the right (no pro…table deviation for candidate
B).

For instance, let us check whether (peA ; peB ) = (l; r) is an equilibrium. As discussed in the previous section,
in that case, one can easily read from the electoral outcome matrix in (28) that candidate A wins the election
with probability 1. But candidate B can increase its winning probability by proposing pB = m instead of
pB = r. Therefore, (peA ; peB ) = (l; r) is not an electoral equilibrium.
Let us now check whether (peA ; peB ) = (m; m) is an equilibrium. At that strategy pro…le, one can easily
read from the electoral outcome matrix in (28) that both candidate wins with probability 1=2. Candidate A
cannot increase its chance of being elected, because by proposing pA = l or pB = r, he looses the election for
sure. Therefore, candidate A has no incentives to deviate from the strategy pA = m. By symmetry, neither
does candidate B. Hence, (peA ; peB ) = (m; m) is an electoral equilibrium.
A careful inspection of that matrix shows that (peA ; peB ) = (m; m) is the only electoral equilibrium. Hence,
the prediction of the Downs model are the following:

Proposition 9 In the canonical Downs model, there is a unique electoral equilibrium. In that equilibrium,
both candidates proposes the Condorcet winner, and each candidate is elected with probability 50%.

Remark 12 (for those of you who have had an introduction to game theory) It turns out that for
the this canonical version of the Downs model, the strategy pro…le (peA ; peB ) = (m; m) is more than a simple

62
Nash equilibrium. The strategy peA = m is a weakly dominant strategy for player A: one can easily check that
for any possible policy proposal pB of candidate B, the policy proposal pA = m gives a probability of winning
for candidate A which is at least as large as any other possible policy proposal p0A . Hence, the strategy pA = m
is optimal for player A irrespective of her conjecture about what candidate B will propose (in other words, it
is always a best response). Moreover, it is the only weakly dominant strategy for player A. Likewise, pB = m
is the only weakly dominant strategy for candidate B. Therefore, the strategy pro…le (peA ; peB ) = (m; m) is
the unique pro…le of weakly dominant strategy.

There are two main qualitative predictions of the Downs model of political competition are the following:

1. Political competition forces candidates to propose to voters what the majority wants (that is, the
Condorcet winner). That conclusion has a positive ‡avor which is similar to the well-known result in
industrial organization: in perfectly competitive industries, the sellers’surplus is competed away, and
…rms are obliged to sell what consumers want at a price equal to marginal cost.
2. In equilibrium, candidates propose identical platforms. That result is less clearly in favor of electoral
competition. On the one hand, both candidates propose what voters want, but on the other hand,
voters do not have much of a choice.

The …rst prediction of the Downs model seems realistic. In most elections, parties try hard to seduce the
median voters on each issue (e.g., it tries to seduce the middle class on …scal issues). The second prediction,
if taken literally, seems a bit unrealistic. Even though it is a well documented that candidates’orientations
tend to converge as the general election approaches, candidates almost never propose the same policies.
Candidates typically try to distinguish themselves from their rival. The result that electoral competition
tend to force candidates to converge nevertheless captures an important aspect of the nature of electoral
competition: the competition for votes is a very brutal form of competition in which the winner takes all. By
that I mean that whichever candidate succeed in proposing the policy that attract the most votes wins the
whole prize, even if he did not convince all voters (whether a candidate wins 1% or 49% if the votes does not
matter, he will not spend any minute in o¢ ce). In contrast, in economic competition, a …rm which attracts
a minority of customers can still make positive pro…t. This winner-takes-all feature forces candidates to
compete …ercely for the median voter, and leads them to converge to the same location on the policy space.
This model raises the following question: how robust is this prediction that electoral competition leads
candidates’proposal to converge? In other words, what makes candidates diverge in real elections, and how
can we introduce these ingredients in the Downs model to make it more realistic? The possible ingredients
that we have left outside the Downs model and that could potentially explain policy divergence in real
elections are the following:

1. Policy motivation: Candidates might not only care about being elected, they might also be interested
in what policy they implement. That could potentially drive them away from the median, if they think
that the ideal policy of the median is not the right policy for the country.
2. Electoral uncertainty: In real elections, no one knows exactly what the majority preferences are.
3. Irrational majority preferences: we have implicitly assumed that majority preferences over alter-
natives are rational. As we have extensively discussed in class, that might not be the case, even with
only three alternatives.
4. More than two candidates: What happens when there are more than two candidates?
5. Commitment: In real elections, candidates make electoral promises but can renege once they are
in o¢ ce. Allowing for non binding campaign promises might change the electoral incentives of the
candidates, and the voting behavior of citizens.
6. Reputation: Candidates are typically associated with a certain ideology (which might very well be
di¤erent from their own, true beliefs). Most candidates are known to the public before the beginning
of the campaign, they have taken clear positions on public debates in the past (by conviction or by
political calculus). Because of that, it is not without risk to simply change their policy orientation

63
 during a campaign to target the median voter (which varies from one election to the next), for they
would be perceived as opportunistic, unpredictable, and unreliable (“‡ip-‡op” candidates). Those
reputational constraints might explain why we never observe two candidates proposing the exact same
policy in a given election.
7. Political parties: Candidates are typically chosen from a party, and they need the support of their
party to win the election. To be credible and to receive the support of the party, a candidate has to
be consistent with the ideology of the party, and the o¢ cial positions it has taken in the past. Those
constraints might prevent a candidate from proposing the policy favored by the median voter.

In the following sections, we will introduce some of these ingredients in the basic Downs model and
analyze how each of these ingredient a¤ects the prediction of the model.

20 Policy motivated candidates

Suppose now that candidates are like voters: they care only about the policy that is implemented, indepen-
dently of who is elected. This assumption is typically referred to as "policy motivated candidates" in the
literature. This assumption might seem a bit unrealistic at …rst sight, because most politicians care, at least
to some extent, about being elected. Hence, the reality is probably in between the two extreme assumptions
of "o¢ ce motivated candidates" and "policy motivated candidates", but analyzing what happens in the two
extreme can give us some insight on what happens in between these two extremes.
To …x ideas, suppose that one candidate, say candidate A, has the preferences of a leftist voters, while the
other candidate, candidate B, has the preferences of a rightist voters. Since the rules of the game have not
changed (only the payo¤ of the candidate has changed as compared to the original Downs model), the policy
outcome matrix M p given in (29) is unchanged. With policy motivated candidates, an electoral equilibrium
is a pair of policy platforms (peA ; peB ) 2 X 2 such that taking the platform of the other candidate as given,
no candidate can change her own policy platform and secure a policy that is better according to her own
preferences:

De…nition 22 An electoral equilibrium with policy motivated candidates is a pair of policy platform (peA ; peB ) 2
X 2 such that:

For all pA 2 X, the policy outcome that results from the pair of policy platform (pA ; peB ) is not preferred
by candidate A to the policy outcome that results from the pair of policy platform (peA ; peB ).
For all pB 2 X, the policy outcome that results from the pair of policy platform (peA ; pB ) is not preferred
by candidate B to the policy outcome that results from the pair of policy platform (peA ; pB ).

In other words, an electoral equilibrium with policy motivated candidate is a Nash equilibrium of the
game where candidates have preferences over the policy outcome matrix Mp . Concretely, the equilibria can
be found from M p as follows: an equilibrium is a cell of M p such that no other policy outcome in the same
column is strictly better for the leftist preferences (no pro…table deviation for candidate A), and no other
policy outcome on the same line is strictly better for the rightist preferences (no pro…table deviation for
candidate B).
For instance, let us check whether (peA ; peB ) = (l; r) is an equilibrium. As discussed in the previous section,
in that case, one can easily read from the policy outcome matrix in (29) that candidate A wins the election
with probability 1, and that policy l is implemented. But candidate B can change the policy outcome by
proposing pB = m instead of pB = r. In that case, she wins the election, and policy m is implemented.
Since candidate B has rightist preferences, she prefers policy m to policy l. Therefore, (peA ; peB ) = (l; r) is
not an electoral equilibrium.
Let us now check whether (peA ; peB ) = (m; m) is an equilibrium. At that strategy pro…le, one can easily
read from the policy outcome matrix in (29) that candidate A cannot change the policy outcome of the
election, m will always prevail (all the entries in the column pB = m are m). By symmetry, neither can
candidate B. Hence, (peA ; peB ) = (m; m) is an electoral equilibrium.

64
 One can easily check that this game has two equilibria: (peA ; peB ) = (m; m) and (peA ; peB ) = (l; m). Hence,
there exists an additional equilibrium as compared to the case of o¢ ce motivated candidates, but the policy
outcome is the same: for both equilibria, the policy that is eventually implemented is m. Moreover, observe
that in the equilibrium (peA ; peB ) = (l; m), the policy outcome is m and candidate A is not elected. If candidate
A deviates to pA = m (so the policy platform pro…le is (m; m)), the outcome is still m with probability 1, but
candidate A is now elected with probability 50%. Hence, if instead of assuming that candidates are purely
policy motivated, we assume that candidates are also slightly o¢ ce motivated, then equilibrium (l; m) would
disappear, because candidate A could increase her chances of election without a¤ecting the policy outcome.
In that case, only the equilibrium (m; m) would remain, as in the original Downs model.31

Remark 13 There is an alternative way to …nd the electoral equilibria with policy motivated candidates that
some students who have studied game theory might be more comfortable with. Policy motivated candidates
can be modelled via the following payo¤ functions: if x denotes the policy that is implemented at the end of
the election, then the following utility function represent the leftist and rightist preferences of candidate A
and B, respectively: 8 8
< 1 if x = l; < 1 if x = l;
A (x) = 0 if x = m; and B (x) = 0 if x = m;
: :
1 if x = r; 1 if x = r:
Given the policy outcome matrix and the new payo¤ function, the payo¤ matrix of the electoral game with
policy motivated candidates is:
0 1
pA npB l m r
B l (1; 1) (0; 0) (1; 1) C
B C:
@ m (0; 0) (0; 0) (0; 0) A
r (1; 1) (0; 0) ( 1; 1)

One can easily check that the equilibria of that game (that is, the cells such that the …rst coordinate cannot
be increased by going up or down and the second coordinate cannot be increased by going to the left or to the
right) are (peA ; peB ) = (m; m) and (peA ; peB ) = (l; m), as we found with the other method.

21 Electoral uncertainty
In this section, we will assume that candidates do not know exactly the preferences of the voters. More
precisely, the preferences of leftist, moderate, and rightist voters are unchanged, but instead of assuming that
no group of voter has a majority of the votes (and thus that majority preferences are given by m M R l M R r
with probability 1), we assume that a majority of voters are leftist with probability pL , a majority of voters
are rightist with probability pR , and with the remaining probability 1 pL pR , neither leftist nor rightist
form a majority. To …x ideas, in that section, we assume further that pL < 1=2 and pR < 1=2. The analysis
in the case pL 1=2 and pR 1=2 is very similar. We keep the other assumptions we made in section 19
(in particular, we assume that candidates are o¢ ce motivated).
Given these distribution over the voters’preferences, the majority preferences are as follows:
8
< with probability pL , l M R m M R r,
with probability pR , r M R m M R l,
:
with probability 1 pL pR , m M R l M R r.32

Hence, the Condorcet winner is policy l with probability pL , policy r with probability pR , and policy m with
the remaining probability 1 pL pR . Given these majority preferences, the electoral outcome matrix is
3 1 To make that point more formally, assume that candidates have the following preferences over policy platform pro…les: if

two policy platform pro…les (pA ; pB ) and p0A ; p0B result in di¤erent policies x and x0 being implemented, then each candidate
prefers the policy platform pro…le that result in the best implemented policy according to her policy preferences. But if the
two policy platforms (pA ; pB ) and p0A ; p0B result in the same policy x being implemented, each candidate prefers the policy
platform that gives her the greatest probability of being elected. With these preferences over electoral and policy outcomes,
one can easily show that the only equilibrium of the electoral competition game is (pA ; pB ) = (m; m).

65
given by: 0 1
pA npB l m r
B l (0:5; 0:5) (pL ; 1 pL ) (1 pR ; pR ) C
M =B
e
@ m
C:
(1 pL ; pL ) (0:5; 0:5) (1 pR ; pR ) A
r (pR ; 1 pR ) (pR ; 1 pR ) (0:5; 0:5)
Since candidates are o¢ ce motivated, the electoral equilibria can be found directly from the matrix M e as
explained in Remark 11: an electoral equilibrium is a cell of M e such that the …rst coordinate cannot be
increased by going up or down (no pro…table deviation for candidate A), and the second coordinate cannot
be increased by going to the left or to the right (no pro…table deviation for candidate B). One can easily
check that given our assumption that pL 1=2 and pR 1=2, the only equilibrium is (peA ; peB ) = (m; m).
Hence, when there is uncertainty, the electoral equilibrium also entails policy convergence: both candidate
propose the same policy platform.

Remark 14 One can wonder why m is the outcome of the election, rather than l or r, since m is not
anymore the Condorcet winner with probability 1. Under our assumption, both candidates propose m because
m is a “median Condorcet winner” in the following sense: the Condorcet winner is to the left of m with
probability smaller than 1/2 (because the probability that the Condorcet winner is l is pL , which is smaller
than 1=2), and it is to the right of m with probability smaller than 1=2 (because the probability that the
Condorcet winner is l is pR , which is smaller than 1=2). If we assume instead that pL > 1=2, the median
Condorcet winner is l, and the only equilibrium of the electoral game is (peA ; peB ) = (l; l).33 Likewise, if we
assume instead that pR > 1=2, the median Condorcet winner is r, and the only equilibrium of the electoral
game is (peA ; peB ) = (r; r).

Therefore, the main take-away from the extensions of policy motivation and electoral uncertainty in
Sections 20 and 21 is that the main prediction of the Downs model, namely that political competition forces
the candidates to propose what the majority wants, is quite robust. Thus this model captures an important
aspect of electoral competition. Remember that this prediction should not be taken literally, but should be
understood as follows: electoral competition, unlike market place competition, does not foster policy variety,
and it forces candidates to pander to what the majority wants (here, in expected terms).

22 Condorcet cycles
Suppose now that the preferences of the voters of type L are given by l L r L m (instead of l L m L r),
whereas the preferences of the other voters are as in the standard Downs model. Given these preferences,
the majority preferences are such that

r MR m MR l MR r.

Thuse, they are irrational, as in teh Condorcet paradox. Given these majority preferences, the electoral
outcome matrix is given by:
0 1
pA npB l m r
B l (0:5; 0:5) (0; 1) (1; 0) C
Me = B @ m
C:
(1; 0) (0:5; 0:5) (0; 1) A
r (0; 1) (1; 0) (0:5; 0:5)

Since candidates are o¢ ce motivated, the electoral equilibria can be found directly from the matrix M e as
explained in Section 19. And it is quite easy to check that no cell of the matrix M e satisfy the equilibrium
conditions. Hence, there is no electoral equilibria when majority preferences are irrational! For every
3 3 To see why the median Condorcet is l, notice that the probability that the Condorcet winner is to the left of l is 0, because

there is no policy to the left of l, and the probabiltiy that the Condorcet winner is to the right of l (i.e., that the Condorcet
winner is m or l) is 1 pL , which is smaller than 1=2 since pL > 1=2.
To see why peA ; peB = (l; l) is the only electoral equilibrium, one can simply check from the electoral matrix M e that when
pL > 1=2 (and necessarily pR < 1=2), peA ; peB = (l; l) is the only strategy pro…le from which no candidate can improve her
probability of winning.

66
pro…le of policy platform (pA ; pB ), there exists a pro…table deviation for one of the candidate. Hence,
candidate could cycle inde…nitely by best responding to each other platform: if candidate A proposes m,
then candidate B should propose r. This would guarantee her a probability 1 of winning the election. But if
candidate B proposes r, candidate A could revise its campaign announcement and propose l instead. This
would guarantee her a probability 1 of winning the election. But then, candidate B should also revise her
campaign announcement, and propose m instead. One can easily see that this cycle could go on inde…nitely.
The problem with irrational majority preferences is that the majority does not know what it wants. Hence,
the electoral race for vote never converges to an alternative that is most preferred by a majority of voters.
There are several ways to interpret that result. The literal interpretation is that when there are Condorcet
cycles, candidate’s campaign proposal will cycle over and over until the election day. That interpretation
does not seem to square with reality, because candidates typically have policy platform that are relatively
stable through the campaign.
The less literal interpretation is that when there are Condorcet cycles, candidates have an incentive to
constantly reevaluate their campaign promises as they learn about the strategy of their opponent. According
to that interpretation, the last candidate to announce her policy has an advantage, because she can best
respond to the early announcement of her rival.
Another interpretation is that when there are Condorcet cycles, the Downs model is not a good model.
The reason is that there are key aspects of political competition that are missing:

The …rst missing ingredient is time: campaigns are dynamic events, and the timing of the announcement
of policy platform is crucial. The above model seem to suggest that being the last one to announce a
policy platform is bene…cial. However, candidates who announce their campaign proposal at the last
moment will lack visibility, and will not have a lot of time left to convince voters that their policy
platform is a good one.
The second missing ingredient is the reputation and predictability of candidates: candidates who
constantly change their policy platform (‡ip-‡op candidates) are typically perceived by voters as op-
portunistic and unreliable. In particular, candidates who have already taken position in the past on
some key issues will have a hard time convincing voters that their belief has changed. Hence, some
candidate will not be able to announce some policy platforms.
The third missing ingredient is the role of political parties. Candidates are endorsed by parties, and
parties are necessary to candidates because they provide them with crucial resources to win the election
(support of other politicians, contacts to raise campaign funds, advice, credibility). And parties are
composed of men and women with certain ideologies, so the candidate of a given party has to keep in
mind that her policy platform has to conform with the political views of the members of her party.
The reason is simply that even if the candidate of a party is only o¢ ce motivated, the members of her
party will not enjoy the perks from o¢ ce. Therefore, they are only interested in the policy she will
implement once in o¢ ce. In other words, members of a leftist party are not interested in electing a
leftist candidate who will implement a rightist policy.

23 More exercises
Exercise 41 When deriving the electoral equilibrium, we have assumed that the moderate voters prefer policy
l to policy r. Derive the electoral equilibrium in the case in which moderate voters prefer policy r to policy l.

Exercise 42 Consider the standard Downs model with X = fa; b; cg ; N = f1; 2; 3g ; and the following
preferences pro…le for the voters: 8
< a 1 b 1 c;
c 2 b 2 a;
:
c 3 b 3 a;

1. Determine the outcome of the election (who wins and what policy is implemented) for each possible
pairs of policy proposals by the two candidates.
2. Find the electoral equilibria (de…ned as in the Downs model).

67
Solution.

1. The answer to that question can be summarized in two matrices, the matrix M e of electoral outcome
(which gives the probability of winning for each candidate and each pair of policy platform) and the
matrix M p of policy outcome (which gives the implemented policy for each pair of policy platform).
To …ll these matrices, one needs to …rst determine the majority preferences. These preferences coincide
with the preferences of the voters 2 and 3, since they have the same preferences and they form a
majority. Therefore, the majority preferences are rational and are given by

c MR b MR a.

Given these majority preferences, the matrices M e and M p are as follows:

0 1 0 1
pL npH a b c pL npH a b c
B a (0:5; 0:5) (0; 1) (0; 1) C B a a b c C
Me = B@
C ; Mp = B C:
b (1; 0) (0:5; 0:5) (0; 1) A @ b b b c A
c (1; 0) (1; 0) (0:5; 0:5) c c c c

2. An electoral equilibrium is a Nash equilibrium of the electoral competition game where the payo¤ of
the candidate is given by the electoral outcome matrix M e . In other words, an equilibrium is a cell
such that the …rst coordinate cannot be increased by going up or down (no pro…table deviation for
candidate L), and the second coordinate cannot be increased by going to the left or to the right (no
pro…table deviation for candidate H). A careful inspection of that matrix M e shows that the only
equilibrium corresponds to the cell (peL ; peH ) = (c; c).

Exercise 43 Consider an environment with three voters N = f1; 2; 3g, three alternatives X = fa; b; cg, and
the following preference pro…le for the voters:
8
< a 1 b 1 c;
c 2 a 2 b;
:
c 3b 3a

As in the Downs model, we assume that two candidates L and H simultaneously proposes a policy pL 2
fa; b; cg and pH 2 fa; b; cg, voters then vote for the candidate that propose the policy they prefer (if they
are indi¤ erent, they vote for each candidate with probability 1=2), and …nally, the candidate who gets the
most vote is elected (if they receive the same number of votes, they are elected with equal probability). The
objective of candidates is to maximize the probability of being elected.

1.

(a) Determine the outcome of the election and the policy implemented for each for each possible pairs
of policy proposal by the two candidates.
(b) Is (pL ; pH ) = (a; a) an electoral equilibrium? Why?
(c) Is (pL ; pH ) = (b; c) an electoral equilibrium? Why?
(d) Find all the electoral equilibria.
(e) Do you think that the prediction given in your answer to question 1.d is realistic? (no more than
a few lines)

2. Find all the electoral equilibria when the preference pro…le of the voters is given by
8
< c 1 a 1 b;
b 2 c 2 a; :
:
a 3b 3c

Do you think that the prediction of the Downs model in that case is realistic?

68
Solution.

1.

(a) The answer to that question can be summarized in two matrices, the matrix M e of electoral
outcomes (which gives the probability of winning for each candidate and each policy platform
pro…le) and the matrix M p of policy outcomes (which gives the implemented policy for each policy
platform pro…le). To …ll these matrices, one needs to …rst determine the majority preferences over
alternatives. One can readily check that they are rational and satisfy:

c MR a MR b.

Given these majority preferences, the matrices M e and M p are as follows:

0 1 0 1
pL npH a b c pL npH a b c
B a (0:5; 0:5) (1; 0) (0; 1) C B a a a c C
Me = B @
C ; Mp = B
A @
C:
b (0; 1) (0:5; 0:5) (0; 1) b a b c A
c (1; 0) (1; 0) (0:5; 0:5) c c c c

(b) An electoral equilibrium is a Nash equilibrium of the electoral competition game where the payo¤
of the candidate is given by the electoral outcome matrix M e . In other words, an equilibrium
is a cell such that the …rst coordinate cannot be increased by going up or down (no pro…table
deviation for candidate L), and the second coordinate cannot be increased by going to the left or
to the right (no pro…table deviation for candidate H).
The pro…le of policy platform (pL ; pH ) = (a; a) is not an equilibrium because candidate L is not
best responding to pH = a : To see why, note that at (pL ; pH ) = (a; a), L wins with probability
0.5, and if we …x pH = a, L can win with probability 1 by proposing the policy platform pL = c
instead of a. Therefore, there is a pro…table deviation, so (pL ; pH ) = (a; a) is not an equilibrium (it
turns out that H is not best responding either, but it su¢ ces to show that L is no best responding
to show that (pL ; pH ) = (a; a) is not an equilibrium).
(c) Likewise, the pro…le of policy platform (pL ; pH ) = (b; c) is not an equilibrium because candidate
L is not best responding to pH = c: To see why, note that at (pL ; pH ) = (b; c), because candidate
L looses for sure, and by proposing the policy platform pL = c, she could win with probability
0.5. Therefore, (pL ; pH ) = (a; a) is not an equilibrium.
(d) Below, I draw the best response correspondence of each candidate (we draw it in the matrix M e
because candidates are o¢ ce-motivated)
0 1
pL npH a b c
B a (0:5; 0:5) L (1; 0) (0; 1) H C
B C:
@ b (0; 1) H (0:5; 0:5) (0; 1) H A
c L (1; 0) L (1; 0) L (0:5; 0:5) H

The electoral equilibria are the points that are on the best response of both candidates, so the
unique electoral equilibrium in that case is (pL ; pH ) = (c; c) :
(e) Note that c is the Condorcet winner (see question 1). So the Downs model predict two main things.
First, candidates proposes the Condorcet winner, i.e., what the majority wants. This prediction
is rather realistic in that it captures the fact that electoral competition forces candidates to
propose what the majority wants. Candidates do not have a lot of degree of freedom in what they
can propose (as a …rm in a perfectly competitive environment, it does no have a lot of market
power). Second, both candidates propose the same policy (policy convergence). This prediction
is somewhat realistic in that in most single-district elections,34 the candidates will choose its
electoral strategy so as to cater to the majority preferences, but it is too extreme in that in most
real world elections, each candidate tries to di¤erentiate itself from the others in order to convince
3 4 By single-district election, I mean an election in which voters elect only one candidate, e.g., a president.

69
 voters that they are the uniquely best candidate.
N.B.: keep in mind that the text of the question asked to comment the PREDICTIONS of the
model, not the ASSUMPTIONS of the model (e.g., that politicians are only o¢ ce motivated, that
voters believe what politicians propose them, that every voter votes, etc...)

2. With the new preference pro…le, the majority preferences are irrational, they are such that:

a MR b MR c MR a:

Given these majority preferences, the matrices M e and M p are as follows (note that for this question,
the matrix M p is useless, I provide it below just for your information):
0 1 0 1
pL npH a b c pL npH a b c
B a (0:5; 0:5) (1; 0) (0; 1) C B a a a c C
Me = B@
C and M p = B C:
b (0; 1) (0:5; 0:5) (1; 0) A @ b a b b A
c (1; 0) (0; 1) (0:5; 0:5) c c b c

Below, I draw the best response correspondence of each candidate (we draw it in the matrix M e because
candidates are o¢ ce-motivated)
0 1
pL npH a b c
B a (0:5; 0:5) L (1; 0) (0; 1) H C
B C:
@ b (0; 1) H (0:5; 0:5) L (1; 0) A
c L (1; 0) (0; 1) H (0:5; 0:5)
The electoral equilibria are the points that are on the best response of both candidates, so we see that
there does not exists an electoral equilibrium! The reason is that majority preferences are not rational.
Hence, by comparing part 1) and part 2), we see that electoral competition gives the majority what it
wants when the majority knows what it wants, but when majority preferences are irrational, electoral
competition could cycle inde…nitely. The latter prediction is obviously not very realistic, because it
does not map with any known political reality. In the real world, we do not see candidates constantly
revise their policy proposals as a response to their opponent. Instead, they try to look committed to
their initial program so as to appear reliable and predictable. The fact that candidates try to appear
credible and committed to their electoral platform is not taken into account in the Downs model, which
is why it gives such a weird prediction when majority preferences are irrational.

Exercise 44 Suppose that the voters’ preferences are as in the standard Downs model.

1. Find all the electoral equilibria when the candidates are policy-motivated and their policy preferences
are
l A m A r;
r B m B l:

2. Find all the electoral equilibria when the candidates are policy-motivated and their policy preferences
are
l A m A r;
r B l B m:
Why does electoral competition always lead to the majority preferred policy in question 1 but not in 2?
(hint: what are the alternatives that are Pareto dominated for the candidates in both cases)
3. Find all the electoral equilibria when the candidates are policy-motivated and their policy preferences
are
l A r A m;
r B m B l:
Why are the candidates unable to avoid an alternative that is Pareto dominated for them in this case?
(hint: think about the majority preferences)

70
Solution. Given the voters’preferences, the majority preferences are rational, they are such that:

m MR l MR r:

Given these majority preferences, the matrix M p of policy outcomes (which gives the policy implemented
by the winner for each policy platform pro…le) is
0 1
pA npB l m r
B l l m l C
p
M =@ B C
m m m m A
r l m r

1. By de…nition, an electoral equilibrium with policy-motivated candidate is a pro…le of policy platform

(pA ; pB ) such that no candidate can change the implemented policy for a policy she strictly prefers by
changing her platform (taking the policy platform of the other candidate as given). Thus, it is a Nash
equilibrium of the game whose outcome function is given by the matrix M p . Below, we draw the best
response correspondence of each candidate
0 1
pA npB l m r
B l Al AmB Al C
B C
@ m mB AmB mB A
r Al Am rB

The electoral equilibria are the points that are on the best response of both candidates, so the electoral
equilibria in that case are (pA ; pB ) = (m; m) and (pA ; pB ) = (l; m) ; and the policy outcome in both
cases is m, i.e., the Condorcet winner.
2. Below, we draw the best response correspondence of each candidate (note that A’s preferences have
not changed relative to question 1, so there is no need to compute again her best response)
0 1
pA npB l m r
B l AlB Am AlB C
B C
@ m mB AmB mB A
r Al Am rB

The electoral equilibria are the points that are on the best response of both candidates, so the electoral
equilibria in that case are (pA ; pB ) = (m; m) ; (pA ; pB ) = (l; l), and (pA ; pB ) = (l; r) : The policy
outcome for each of these equilibria is m; l; and l, respectively. So we see that electoral competition
does not always lead to what the majority wants. The reason is that the majority preferred alternative
m is Pareto dominated for the candidates by l: That is, both candidates prefer l to m; which was not
the case in question 1 (in the standard Downs model with policy motivated candidates, candidates
disagree on how to rank any two policies). So candidates can "collude" on l, because they both prefer
l to what the majority wants, i.e., m:
3. Below, we draw the best response correspondence of each candidate (note that B’s preferences have
not changed relative to question 1, so there is no need to compute again her best response)
0 1
pA npB l m r
B l Al AmB Al C
B C
@ m mB AmB mB A
r Al Am rB

The electoral equilibria are the points that are on the best response of both candidates, so the electoral
equilibria in that case are (pA ; pB ) = (m; m) and (pA ; pB ) = (l; m) ; and the policy outcome in both
cases is m, i.e., the Condorcet winner, as in question 1. In that case, both candidates prefer r to m;
but they cannot "collude" and impose r because r is the worst alternative according to the majority
preferences, so the only way to impose r is that both candidates propose r; and in that case, candidate
A has an incentive to deviate and propose l; which she and the majority prefer over r:

71
Exercise 45 Suppose that the preferences of the leftist voters are given by l L r L m; whereas the
preferences of the other voters are as in the standard Downs model.

1. Find all the equilibria when candidates are policy-motivated and their policy preferences are

l A m A r;
r B m B l:

2. Find all the equilibria when candidates are policy-motivated and their policy preferences are

l L r L m;
r B l B m:

Solution. Given the voters’preferences, the majority preferences are irrational, they are such that:

l MR r MR m MR l:

Given these majority preferences, the matrix M p of policy outcomes (which gives the implemented policy
for each policy platform pro…le) is
0 1
pA npB l m r
B l l m l C
M =B
p
@ m
C:
m m r A
r l r r

1. Given the candidates’preferences, the best response correspondence of each candidate is given by (we
draw it in the matrix M p because candidates are policy-motivated)
0 1
pA npB l m r
B l Al AmB Al C
B C:
@ m m Am rB A
r Al rB rB

The electoral equilibria are the points that are on the best response of both candidates, so the unique
electoral equilibrium in that case is (pA ; pB ) = (m; m) ; and the policy outcome is m.
2. Given the candidates’preferences, the best response correspondence of each candidate is given by (we
draw it in the matrix M p because candidates are policy-motivated)
0 1
pA npB l m r
B l AlB m AlB C
B C:
@ m m m ArB A
r Al rB rB

The electoral equilibria are the points that are on the best response of both candidates, so the electoral
equilibria in that case are (pA ; pB ) = (l; l) and (pA ; pB ) = (l; r) and the policy outcome is both equilibria
is l. To see the intuition for this result, note that policy m is not anymore the equilibrium outcome as
in 1 because it is the worst policy for both candidates, and since the majority does not know what it
wants (majority preferences are irrational), electoral competition cannot force candidates to propose
the Condorcet winner. So the preferences of the candidates a¤ect the outcome. In that case, they both
prefer l and r to m; so m is not the outcome. The outcome is l because between l and r; the majority
of voters prefers l.

72