14.

770: Introduction to Political Economy

Lectures 1 and 2: Collective Choice and Voting

Daron Acemoglu

MIT

September 6 and 11, 2018

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 1 / 64
 Collective Choice and Static Voting Models Introduction

Introduction

Much of economics takes preferences, technology and institutions

(market structure, laws, regulations, policies) as given.
Thus institutions matter in the same way as preferences do.
But in general, in the background
Institutions are not just about laws, but more generally about how
collective choices are made and con‡icting preferences (and
di¤erential information) are aggregated.
Political economy is about systematically investigating this type of
aggregation and resolution of con‡icts.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 2 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Collective Choices: Introduction

One happy settlement of the questions of political economy would be

to show that there is a “natural” and perhaps even “e¢ cient” way of
resolving con‡icts.
For example, one could imagine a set of aggregation rules that would
take the con‡icting preferences of the members of a group and arrive
to a mutually agreeable collective decision.
Unfortunately, as we will next see this is not possible.
This is the essence of Arrow’s Impossibility Theorem, which has two
key implications:
1 Con‡ict will not have an easy solution.
2 Details of how individuals with con‡icting preferences interact will
matter greatly.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 3 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Basics

Abstract economy consisting of a …nite set of individuals H, with the

number of individuals denoted by H.
Individual i 2 H has a utility function

u (xi , Y (x, p ) , p j αi ) .

Here xi is his action, with a set of feasible actions denoted by Xi ;

p denotes the vector of political choices (for example, institutions,
policies or other collective choices), with the menu of policies denoted
by P ;
Y (x, p ) is a vector of general equilibrium variables, such as prices or
externalities that result from all agents’actions as well as policies,
and x is the vector of the xi ’s.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 4 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Example Preferences

For example, we could have that given aggregates and policies,

individual objective functions are strictly quasi-concave so that each
agent has a unique optimal action

xi (p, Y (x, p ) , αi ) = arg max u (xi , Y (x, p ) , p j αi ) .

x 2X i

Then, indirect utility function

U (p; αi )

The preferred policy or the (political) bliss point of individual i

p (αi ) = arg max U (p; αi ).

p 2P

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 5 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Preferences More Generally

Individual individual i weakly prefers p to p 0 ,

p i p0

and if he has a strict preference,

p i p0.

Assume: completeness, re‡exivity and transitivity (so that z i z 0 and

z 0 i z 00 implies z i z 00 ).

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 6 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Collective Preferences?

Does there exist welfare function U S (p ) that ranks policies for the
society.
Let us simplify the discussion by assuming that the set of feasible
policies is, P RK
Let < be the set of all weak orders on P , that is, < contains
information of the form p1 i p2 i p3 and so on, and imposes the
requirement of transitivity on these individual preferences.
An individual ordering Ri is an element of <, that is, Ri 2 <.
Since our society consists of H individuals, ρ = (R1 , ..., RH ) 2 <H is
a preference pro…le.
Also ρjP 0 = R1 jP 0 , ..., RH jP 0 is the society’s preference pro…le when
alternatives are restricted to some subset P 0 of P .

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 7 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Restrictions on Collective Preferences I

Let = be the set of all re‡exive and complete binary relations on P

(but notice not necessarily transitive).
A social ordering R S 2 = is therefore a re‡exive and complete binary
relation over all the policy choices in P :

φ : <H ! =.

We have already imposed “unrestricted domain,” since no restriction

on preference pro…les.
A social ordering is weakly Paretian if

p i p 0 for all i 2 H =) p S
p0.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 8 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Restrictions on Collective Preferences II

Given ρ, a subset D of H is decisive between p, p 0 2 P , if

p i p 0 for all i 2 D and p i0 p 0 for some i 0 2 D =) p S

p0

If D 0 H is decisive between p, p 0 2 P for all preference pro…les

ρ 2 <H , then it is dictatorial between p, p 0 2 P .
D H is decisive if it is decisive between any p, p 0 2 P
D 0 H is dictatorial if it is dictatorial between any p, p 0 2 P .
If D 0 H is dictatorial and a singleton, then its unique element is a
dictator.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 9 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Restrictions on Collective Preferences III

A social ordering satis…es independence from irrelevant alternatives, if

for any ρ and ρ0 2 <H and any p, p 0 2 P ,

ρjfp,p 0 g = ρ0jfp,p 0 g =) φ (ρ)jfp,p 0 g = φ ρ0 jfp,p 0 g

.

This axiom states that if two preference pro…les have the same choice
over two policy alternatives, the social orderings that derive from
these two preference pro…les must also have identical choices over
these two policy alternatives, regardless of how these two preference
pro…les di¤er for “irrelevant” alternatives.
While this condition (axiom) at …rst appears plausible, it is in fact a
reasonably strong one. In particular, it rules out any kind of
interpersonal “cardinal” comparisons— that is, it excludes information
on how strongly an individual prefers one outcome versus another.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 10 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Arrow’s Impossibility Theorem

Theorem
( Arrow’s (Im)Possibility Theorem) If a social ordering, φ, is transitive,
weakly Paretian and satis…es independence from irrelevant alternatives,
then it is dictatorial.

An immediate implication of this theorem is that any set of minimal

decisive individuals D within the society H must either be a
singleton, that is, D = fi g, so that we have a dictatorial social
ordering, or we have to live with intransitivities.
Also implicitly, political power must matter. If we wish transitivity,
political power must be allocated to one individual or a set of
individuals with the same preferences.
How do we proceed?! Restrict preferences or restrict institutions.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 11 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Proof of Arrow’s Impossibility Theorem I

Suppose to obtain a contradiction that there exists a non-dictatorial

and weakly Paretian social ordering, φ, satisfying independence from
irrelevant alternatives. Contradiction in two steps.
Step 1: Let a set J H be strongly decisive between p1 , p2 2 P if
for any preference pro…le ρ 2 <H with p1 i p2 for all i 2 J and
p2 j p1 for all j 2 HnJ , p1 S p2 (H itself is strongly decisive
since φ is weakly Paretian).
We …rst prove that if J is strongly decisive between p1 , p2 2 P , then
J is dictatorial (and hence decisive for all p, p 0 2 P and for all
preference pro…les ρ 2 <H ).
To prove this, consider the restriction of an arbitrary preference pro…le
ρ 2 <H to ρjfp1 ,p2 ,p3 g and suppose that we also have p1 i p3 for all
i 2 J.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 12 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Proof of Arrow’s Impossibility Theorem II

Next consider an alternative pro…le ρ0jfp1 ,p2 ,p3 g , such that

p1 i0 p2 i0 p3 for all i 2 J and p2 i0 p1 and p2 i0 p3 for all
i 2 HnJ .
Since J is strongly decisive between p1 and p2 , p1 0S p2 .
Moreover, since φ is weakly Paretian, we also have p2 0S p3 , and
thus p1 0S p2 0S p3 .
Notice that ρ0jfp1 ,p2 ,p3 g did not specify the preferences of individuals
i 2 HnJ between p1 and p3 , but we have established p1 0S p3 for
ρ0jfp1 ,p2 ,p3 g .
We can then invoke independence from irrelevant alternatives and
conclude that the same holds for ρjfp1 ,p2 ,p3 g , i.e., p1 S p3 .
But then, since the preference pro…les and p3 are arbitrary, it must be
the case that J is dictatorial between p1 and p3 .

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 13 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Proof of Arrow’s Impossibility Theorem III

Next repeat the same argument for ρjfp1 ,p2 ,p4 g and ρ0jfp1 ,p2 ,p4 g , except
that now p4 i p2 and p4 i0 p1 i0 p2 for i 2 J , while p2 j0 p1 and
p4 j0 p1 for all j 2 HnJ .
Then, the same chain of reasoning, using the facts that J is strongly
decisive, p1 0S p2 , φ is weakly Paretian and satis…es independence
from irrelevant alternatives, implies that J is dictatorial between p4
and p2 (that is, p4 S p2 for any preference pro…le ρ 2 <H ).
Now once again using independence from irrelevant alternatives and
also transitivity, for any preference pro…le ρ 2 <H , p4 i p3 for all
i 2 J.
Since p3 , p4 2 P were arbitrary, this completes the proof that J is
dictatorial (i.e., dictatorial for all p, p 0 2 P ).

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 14 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Proof of Arrow’s Impossibility Theorem IV

Step 2: Given the result in Step 1, if we prove that some individual

h 2 H is strongly decisive for some p1 , p2 2 P , we will have
established that it is a dictator and thus φ is dictatorial. Let Dab be
the strongly decisive set between pa and pb .
Such a set always exists for any pa , pb 2 P , since H itself is a
strongly decisive set. Let D be the minimal strongly decisive set
(meaning the strongly decisive set with the fewest members).
This is also well-de…ned, since there is only a …nite number of
individuals in H.
Moreover, without loss of generality, suppose that D = D12 (i.e., let
the strongly decisive set between p1 and p2 be the minimal strongly
decisive set).
If D a singleton, then Step 1 applies and implies that φ is dictatorial,
completing the proof.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 15 / 64
 Collective Choice and Static Voting Models Arrow’s Impossibility Theorem

Proof of Arrow’s Impossibility Theorem V

Thus suppose that D 6= fi g. Then, by unrestricted domain, the
following preference pro…le (restricted to fp1 , p2 , p3 g) is feasible
for i 2 D p1 i p2 i p3
for j 2 Dn fi g p3 j p1 j p2
for k 2
/D p2 k p3 k p1 .
By hypothesis D is strongly decisive between p1 and p2 . Thus
p1 S p2 .
Next if p3 S p2 , then given the preference pro…le here, Dn fi g
would be strongly decisive between p2 and p3 , and this would
contradict that D is the minimal strongly decisive set.
Thus p2 %S p3 . Combined with p1 S p2 , this implies p1 S p3 . But
given the preference pro…le here, this implies that fi g is strongly
decisive, yielding another contradiction.
Therefore, the minimal strongly decisive set must be a singleton fh g
for some h 2 H. Then, from Step 1, fh g is a dictator and φ is
dictatorial,
Daron completingPolitical
Acemoglu (MIT) the proof.
Economy Lectures 1 and 2 September 6 and 11, 2018 16 / 64
 Collective Choice and Static Voting Models Voting and the Condorcet Paradox

Voting

Could be voting help?

No because Arrow’s Theorem already covers voting.
But voting may impose additional “institutional structure”
But in fact the Condorcet paradox has many of the same features as
Arrow’s Theorem.
This was anticipated by the great and indomitable Marquis de
Condorcet.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 17 / 64
 Collective Choice and Static Voting Models Voting and the Condorcet Paradox

The Condorcet Paradox

Imagine a society consisting of three individuals, 1, 2, and 3, three

choices and preferences

1 a c b
2 b a c
3 c b a

Assume “open agenda direct democracy” system.

A1. Direct democracy. The citizens themselves make the policy choices
via majoritarian voting.
A2. Sincere voting. Individuals vote “truthfully” rather than strategically.
A3. Open agenda. Citizens vote over pairs of policy alternatives, such
that the winning policy in one round is posed against a new alternative in
the next round and the set of alternatives includes all feasible policies.
Implication: cycling
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 18 / 64
 Collective Choice and Static Voting Models Voting and the Condorcet Paradox

The Condorcet Winner

We can avoid the Condorcet paradox when there is a Condorcet

winner.

De…nition
A Condorcet winner is a policy p that beats any other feasible policy in
a pairwise vote.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 19 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Single-Peaked Preferences
De…nition
Consider a …nite set of P R and let p (αi ) 2 P be individual i’s unique
bliss point over P . Then, the policy preferences of citizen i are single
peaked i¤:

For all p 00 , p 0
2 P , such that p 00 < p 0 p (αi ) or p 00 > p 0 p ( αi ),
we have U (p 00 ; αi ) < U (p 0 ; αi ).

Essentially strict quasi-concavity of U

Median voter: rank all individuals according to their bliss points, the
p (αi )’s. Suppose that H odd. Then, the median voter is the
individual who has exactly (H 1) /2 bliss points to his left and
(H 1) /2 bliss points to his right.
Let us denote this individual by αm , and his bliss point (ideal policy)
is denoted by pm .
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 20 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Median Voter Theorem

Theorem
(The Median Voter Theorem) Suppose that H is an odd number, that
A1 and A2 hold and that all voters have single-peaked policy preferences
over a given ordering of policy alternatives, P . Then, a Condorcet winner
always exists and coincides with the median-ranked bliss point, pm .
Moreover, pm is the unique equilibrium policy (stable point) under the
open agenda majoritarian rule, that is, under A1-A3.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 21 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Proof of the Median Voter Theorem

The proof is by a “separation argument”.

Order the individuals according to their bliss points p (αi ), and label
the median-ranked bliss point by pm .
By the assumption that H is an odd number, pm is uniquely de…ned
(though αm may not be uniquely de…ned).
Suppose that there is a vote between pm and some other policy
p 00 < pm .
By de…nition of single-peaked preferences, for every individual with
pm < p (αi ), we have U (pm ; αi ) > U (p 00 ; αi ).
By A2, these individuals will vote sincerely and thus, in favor of pm .
The coalition voting for supporting pm thus constitutes a majority.
The argument for the case where p 00 > pm is identical.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 22 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Median Voter Theorem: Discussion

Odd number of individuals to shorten the statement of the theorem

and the proof. It is straightforward to generalize the theorem and its
proof to the case in which H is an even number.
More important: sincere voting
Alternative: Strategic voting.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 23 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Strategic Voting
A20 . Strategic voting. De…ne a vote function of individual i in a pairwise
contest between p 0 and p 00 by vi (p 0 , p 00 ) 2 fp 0 , p 00 g. Let a voting
H
(counting) rule in a society with H citizens be V :fp 0 , p 00 g ! fp 0 , p 00 g for
any p 0 , p 00 2 P .
Let V (vi (p 0 , p 00 ) , v i (p 0 , p 00 )) be the policy outcome from voting rule V
applied to the pairwise contest fp 0 , p 00 g, when the remaining individuals
cast their votes according to the vector v i (p 0 , p 00 ), and when individual i
votes vi (p 0 , p 00 ).
Strategic voting means that
vi p 0 , p 00 2 arg max U V ṽi p 0 , p 00 , v i p 0 , p 00 ; αi .
ṽi (p 0 ,p 00 )

So we are now essentially looking for a Nash equilibrium or Bayesian

Nash equilibrium with strategic players.
A weakly-dominant strategy for individual i is a strategy that gives
weakly higher payo¤ to individual i than any of his other strategies
regardless of the strategy pro…le of other players.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 24 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Median Voter Theorem with Strategic Voting

Theorem
(The Median Voter Theorem With Strategic Voting) Suppose that H
is an odd number, that A1 and A20 hold and that all voters have
single-peaked policy preferences over a given ordering of policy
alternatives, P . Then, sincere voting is a weakly-dominant strategy for
each player and there exists a unique weakly-dominant equilibrium, which
features the median-ranked bliss point, pm , as the Condorcet winner.

Notice no more “open agenda”. Why not?

Why emphasis on weakly-dominant strategies?
Here we have an example where strategic voting doesn’t matter.
What about in plurality elections with more than two candidates?

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 25 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Proof of the Median Voter Theorem with Strategic Voting

The vote counting rule (the political system) in this case is

majoritarian, denoted by V M .
Consider two policies p 0 , p 00 2 P and …x an individual i 2 H.
Assume without loss of any generality that U (p 0 ; αi ) U (p 00 ; αi ).
Suppose …rst that for any vi 2 fp 0 , p 00 g, V M (vi , v i (p 0 , p 00 )) = p 0 or
V M (vi , v i (p 0 , p 00 )) = p 00 , that is, individual i is not pivotal.
This implies that vi (p 0 , p 00 ) = p 0 is a best response for individual i.
Suppose next that individual i is pivotal, that is,
V M (vi (p 0 , p 00 ) , v i (p 0 , p 00 )) = p 0 if vi (p 0 , p 00 ) = p 0 and
V M (vi (p 0 , p 00 ) , v i (p 0 , p 00 )) = p 00 otherwise. In this case, the action
vi (p 0 , p 00 ) = p 0 is clearly a best response for i.
Since this argument applies for each i 2 H, it establishes that voting
sincerely is a weakly-dominant strategy and the conclusion of the
theorem follows.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 26 / 64
 Collective Choice and Static Voting Models Single-Peaked Preferences and the Median Voter Theorem

Strategic Voting in Sequential Elections

Sincere voting no longer optimal in dynamic situations.
1 a b c
2 b c a
3 c b a
These preferences are clearly single peaked (e.g., alphabetical order).
Consider the following dynamic voting set up: …rst, there is a vote
between a and b. Then, the winner goes against c, and the winner of
this contest is the social choice.
Sincere voting: in the …rst round players 2 and 3 vote for b, and in the
second round, 1 and 2 vote for b, which becomes the social choice.
However, when players 1 and 2 are playing sincerely, in the …rst round
player 3 can deviate and vote for a (even though she prefers b), then
a will advance to the second round and would lose to c.
Consequently, the social choice will coincide with the bliss point of
player 3. What happens if all players are voting strategically?
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 27 / 64
 Collective Choice and Static Voting Models Party Competition and the Downsian Policy Convergence Theorem

Moving Away from Direct Democracy

Towards representative democracy, with parties.

Two parties that can announce and commit to policies.
Rent Q > 0 from coming to power and no ideological bias.
Thus the maximization problem of the two parties are

Party A : max P(pA , pB )Q

pA
Party B : max (1 P(pA , pB )) Q
pB

P(pA , pB ) is the probability that party A comes to power when the

two parties’platforms are pA and pB respectively.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 28 / 64
 Collective Choice and Static Voting Models Party Competition and the Downsian Policy Convergence Theorem

Party Competition

Let the bliss point of the median voter be pm .

When the median voter theorem applies, we have

P ( pA , pB = pm ) = 0, P(pA = pm , pB ) = 1, and
P ( pA = pm , pB = pm ) 2 [0, 1] .

A4. Randomization:

P(pA = pm , pB = pm ) = 1/2.

Why?

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 29 / 64
 Collective Choice and Static Voting Models Party Competition and the Downsian Policy Convergence Theorem

Downsian Policy Convergence Theorem

Theorem
(Downsian Policy Convergence Theorem) Suppose that there are two
parties that …rst announce a policy platform and commit to it and a set of
voters H that vote for one of the two parties. Assume that A4 holds and
that all voters have single-peaked policy preferences over a given ordering
of policy alternatives, and denote the median-ranked bliss point by pm .
Then, both parties will choose pm as their policy platform.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 30 / 64
 Collective Choice and Static Voting Models Party Competition and the Downsian Policy Convergence Theorem

Proof of the Downsian Policy Convergence Theorem

The proof is by contradiction.

Suppose not, then there is a pro…table deviation for one of the parties.
For example, if p A > p B > pm , one of the parties can announce pm
and win the election for sure.
When p A 6= pm and p B = pm , party A can also announce pm and
increase its chance of winning to 1/2.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 31 / 64
 Collective Choice and Static Voting Models Party Competition and the Downsian Policy Convergence Theorem

Downsian Policy Convergence Theorem: Discussion

What happens without Assumption A4?

Why is this theorem important?
A natural generalization of this theorem would be to consider three or
more parties. What happens with three parties?

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 32 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Multidimensional Policies?

Unfortunately, single-peakedness does not work would

multidimensional policies.
But political economy is interesting with multidimensional policies.
Generalizations, e.g., intermediate preferences.
But not widely applicable.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 33 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Single Crossing

More useful:

De…nition
Consider an ordered policy space P and also order voters according to
their αi ’s. Then, the preferences of voters satisfy the single-crossing
property over the policy space P when the following statement is true:

if p > p 0 and αi 0 > αi , or if p < p 0 and αi 0 < αi , then

U (p; αi ) > U (p 0 ; αi ) implies that U (p; αi 0 ) > U (p 0 ; αi 0 ).

Notice that while single peakedness is a property of preferences only,

the single-crossing property refers to a set of preferences over a given
policy space P . It is therefore a joint property of preferences and
choices.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 34 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Single Crossing versus Single Peakedness

Single-crossing property is does not imply single-peaked preferences.

1 a b c
2 a c b
3 c b a

These preferences are not single peaked. But they satisfy single
crossing
The natural ordering is a > b > c:

α = 2: c b =) α = 3: c b
a c a c
α = 2: =) α = 1: .
a b a b

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 35 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Single Crossing versus Single Peakedness (continued)

The following preferences are single peaked with the natural order
a > b > c > d:
1 a b c d
2 b c d a
3 c b a d
For them to satisfy single crossing, we need to adopt the same order
over policies (given 1’s preferences) and the order 3 > 2 > 1 over
individuals.
But then the fact that d 2 a should imply that d 3 a, which is not
the case. (It is easy to verify that if one chooses the order 2 > 3 > 1
over individuals, one would obtain a similar contradiction as c 3 b,
but b 2 c).
This shows that single peakedness does not ensure single crossing.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 36 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Median Voter Theorem with Single Crossing

Theorem
(Extended Median Voter Theorem) Suppose that A1 and A2 hold and
that the preferences of voters satisfy the single-crossing property. Then, a
Condorcet winner always exists and coincides with the bliss point of the
median voter (voter αm ).

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 37 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Proof

The proof works with exactly the same separation argument as in the
proof of Theorem 4.
Consider the median voter with αm , and bliss policy pm .
Consider an alternative policy p 0 > pm . Naturally,
U ( pm ; α m ) > U ( p 0 ; α m ) .
Then, by the single crossing property, for all αi > αm ,
U ( pm ; α i ) > U ( p 0 ; α i ) .
Since αm is the median, this implies that there is a majority in favor
of pm .
The same argument for p 0 < pm completes the proof

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 38 / 64
 Collective Choice and Static Voting Models Beyond Single-Peaked Preferences

Extended Policy Convergence

Theorem
(Extended Downsian Policy Convergence) Suppose that there are two
parties that …rst announce a policy platform and commit to it and a set of
voters that vote for one of the two parties. Assume that A4 holds and that
all voters have preferences that satisfy the single-crossing property and
denote the median-ranked bliss point by pm . Then, both parties will
choose pm as their policy.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 39 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation I

Consider situation with two parties competing to come to power.
Suppose that agents have the following preferences
u i c i , x i = c i + h (x i )
where c i and x i denote individual consumption and leisure, and h ( )
is a well-behaved concave utility function.
There are only two policy instruments, linear tax on earnings τ on
lump-sum transfers T 0 (and this is important).
The budget constraint of each agent is
ci (1 τ )l i + T ,
The real wage is exogenous and normalized to 1.
Individual productivity di¤ers, such that the individuals have di¤erent
amounts of “e¤ective time” available. That is, individuals are
subject to the “time constraint”
αi xi + li ,
i
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 40 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation II

Assume that αi is distributed in the population with mean α and
median αm .
Since individual preferences are linear in consumption, optimal labor
supply satis…es
l i = L( τ ) + ( αi α ),
1
where L(τ ) α (h0 ) (1 τ ) is decreasing in τ by the concavity
of h ( ).
To derive this, note that from quasi-linear preferences, the …rst order
condition of each individual is
(1 τ ) = h 0 (x i ).
Inverting this, writing x i = αi l i , adding and subtracting α, and
1
de…ning L(τ ) α (h0 ) (1 τ ) we obtain the desired expression.
A higher tax rate on labor income distorts the labor-leisure choice and
induces the consumer to work less. This will be the cost of
redistributive taxation in this model.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 41 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation III

Let l denote average labor supply. Since the average of αi is α, we

have l = L(τ ). The government budget constraint can therefore be
written:
T τl τL(τ ).

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 42 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation IV

Let U (τ; αi ) be utility for αi from tax τ with T determined as
residual. By straightforward substitution into the individual utility
function, we can express the policy preferences of individual i as

U (τ; αi ) L( τ ) + h ( α L(τ )) + (1 τ )(αi α ). (1)

Are the preferences represented by (1) single-peaked?

The answer depends on the shape of the average labor supply
function L(τ ). By putting enough structure on dysfunction, we could
ensure that U (τ; αi ) is strictly concave or quasi concave, thus
satisfying single-peakedness. However, this function could be
su¢ ciently convex that U (τ; αi ) could have multiple peaks (multiple
local maxima). As a result, preferences may not be single peaked.
But it is straightforward to verify that (1) satis…es the single-crossing
property.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 43 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation V

Therefore, we can apply MVT, and party competition gives
τ m = arg max U (τ; αm )
τ
Hence, we have
L0 ( τ m ) 1 h0 (α L(τ m )) ( αm α) 0 (2)
with complementary slackness.
If the mean is greater than the median, as we should have for a
skewed distribution of income, it must be the case that αm α < 0
(that is median productivity must be less than mean productivity).
This implies that τ m > 0— otherwise, (2) would be satis…ed for a
negative tax rate, and we would be at a corner solution with zero
taxes (unless negative tax rates, i.e., subsidies, were allowed).
Now imagine a change in the distribution of α such that the di¤erence
between the mean and the median widens. From the above …rst-order
condition, this’ll imply that the equilibrium tax rate τ m increases.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 44 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation VI

This is the foundation of the general presumption that greater

inequality (which is generally, but not always, associated with a
widening gap between the mean and the median) will lead to greater
taxation to ensure greater redistribution away from the mean towards
the median.
Notice also that greater inequality in this model leads to greater
“ine¢ ciency” of policy.
Why is this? The reason is only weakly related to the logic of
redistribution, but more to the technical assumptions that have been
made.
In order to obtain single-peaked preferences, we had to restrict policy
to a single dimensional object, the linear tax rate.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 45 / 64
 Collective Choice and Static Voting Models Application: Redistributive Taxation

Application: Redistributive Taxation VII

Moreover, is this “ine¢ ciency” the same as Pareto suboptimality?

Imagine, instead, that di¤erent taxes can be applied to di¤erent
people. Then, redistribution does not necessitate distortionary
taxation. But in this case, preferences will clearly be
non-single-peaked— agent i particularly dislikes policies that tax him a
lot, and likes policies that tax agents j and k a lot, where as agent j
likes policies that tax i and k a lot, etc.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 46 / 64
 Collective Choice and Static Voting Models Inequality and Redistribution

Inequality and Redistribution

One interpretation of the previous result is that greater inequality

should lead to greater redistribution.
Despite these claims in the literature, however, there is no such
unambiguous prediction.
More importantly, there is no empirical evidence that greater
inequality leads to more distribution.
In fact, why many highly unequal societies do not adopt more
redistributive policies will be one of the teams we will investigate
when we come to understanding the nature of institutions

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 47 / 64
 Collective Choice and Static Voting Models Inequality and Redistribution

Inequality and Redistribution in the MVT Models

Consider the previous model with the mean greater than the median
Consider the following redistribution: take money from everybody
below the median and redistribute to everybody above the median in
a way that leaves all voters’ranking the same.
This is a mean preserving spread and thus increases inequality.
But the median has become richer relative to demean and thus there
is less redistribution.
The gap between the mean in the median generally has little to do
with inequality (except for the log normal distribution).

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 48 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Understanding Nonexistence

Game theoretically, the Condorcet paradox is not about “cycling” ,

but nonexistence of pure strategy equilibria.
Example: three (groups of) voters, i = 1, 2, 3 of equal size with
strictly increasing preferences

U (p ) = u p i ,

where p = p 1 , p 2 , p 3 , with ∑3i =1 p i = 1.

A policy will be the winner if it gets votes from 2 agents.
Now take a winning policy (p1 , p2 , p3 ) where without any loss of
generality suppose that p1 > 0.
Then the following policy will always beat this winning policy
(p1 2ε, p2 + ε, p3 + ε), proving that there will always be cycling.
Therefore, no pure strategy Nash equilibrium.
Intuition: viewed as a cooperative game, this has an empty core.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 49 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Probablilistic Voting: Main Idea

In the above example, it appears that the discontinuity of best

responses in policies is important in nonexistence.
The main idea of probabilistic voting is to “smooth” best responses in
order to get existence.
Intuitively, there are ideological and non-policy factors, so that a small
advantage due to policies will not sway all voters.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 50 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Probablilistic Voting: Introduction

G distinct groups, with a continuum of voters within each group
having the same economic characteristics and preferences.
Electoral competition between two parties, A and B, that are
“non-ideological” (only care about coming to power; is this
important?).
π gP :fraction of voters in group g voting for party P = A, B, and
λg : share of voters in group g . Then expected vote share of party P is
G
πP = ∑ λg πgP .
g =1

Suppose that individual i in group g has the following preferences:

Ũig (p, P ) = U g (p ) + σ̃gi (P ) (3)
when party P comes to power, where p 2 P RK .
As usual U g (p ) is the indirect utility of agents in group g
σ̃gi (P ) is the non-policy bene…ts for i from party P coming to power.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 51 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Probablilistic Voting I
Let us normalize σ̃gi (A) = 0, so that
Ũig (p, A) = U g (p ), and Ũig (p, B ) = U g (p ) + σ̃gi (4)
In that case, the voting behavior of individual i can be represented as
8 g
< 1 if U g (pA ) U g (pB ) > σ̃i
g
vi (pA , pB ) = 1
if U (pA ) U (pB ) = σ̃gi ,
g g (5)
: 2 g g g
0 if U (pA ) U (pB ) < σ̃i
Suppose that the distribution of non-policy related bene…ts σ̃gi for
individual i in group g is given by a smooth cumulative distribution
function H g de…ned over ( ∞, +∞), with the associated probability
density function hg .
The draws of σ̃gi across individuals are independent.
Consequently, the vote share of party A among members of group g is
π gA = H g (U g (pA ) U g (pB )).
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 52 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Probablilistic Voting II
Supposed to start with that parties maximize their expected vote
share.
In this case, party A sets this policy platform pA to maximize:
G
πA = ∑ λ g H g ( U g ( pA ) U g (pB )). (6)
g =1

Party B faces a symmetric problem and maximizes π B , which is

de…ned similarly. Since π B = 1 π A , party B’s problem is exactly
the same as minimizing π A .
Equilibrium policies determined as the Nash equilibrium of a
(zero-sum) game where both parties make simultaneous policy
announcements to maximize their vote share.
First-order conditions for party A
G
∑ λ g h g ( U g ( pA ) U g (pB ))DU g (pA ) = 0,
g =1
g
where DU (p ) is thePolitical U g 1(and
Daron Acemoglu (MIT) gradient Economyof
Lectures ). 2 September 6 and 11, 2018 53 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Probablilistic Voting Equilibrium

Focus …rst on pure strategy symmetric equilibria. Clearly in this case,
we will have policy convergence with pA = pB = p , and thus
U g ( pA ) = U g ( pB ) .
Consequently, symmetric equilibrium policies, announced by both
parties, must be given by
G
∑ λg hg (0)DU g (p ) = 0. (7)
g =1

Therefore, the probability quoting equilibrium is given as the solution

to the maximization of the following weighted utilitarian social welfare
function:
G
∑ χg λg U g (p ) , (8)
g =1

where χg hg (0) are the weights that di¤erent groups receive in the
social welfare function.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 54 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Weighted Social Welfare Functions

Theorem
(Probabilistic Voting Theorem) Consider a set of policy choices P , let
p 2 P RK be a policy vector and let preferences be given by (4), with
the distribution function of σ̃gi as H g . Then, if a pure strategy symmetric
equilibrium exists, equilibrium policy is given by p that maximizes (8).

Most important: probabilistic voting equilibria are Pareto optimal

(given policy instruments).
Now in fact, looking back, whenever the Median Voter Theorem
applies, the equilibrium is again Pareto optimal.
What does this mean?

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 55 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Existence of Pure Strategy Equilibria

However, the probability voting model is not always used properly.

It is a good model to represent certain political interactions.
But it is not a good model to ensure pure strategy equilibria.
In fact, pure strategy existence requires that the matrices
G
B (0, p ) ∑ λg h g (0)D 2 U g (p )
g =1
G
∂hg (0)
∑ λg ∂x
(DU g (p )) (DU g (p ))T
g =1

is negative semide…nite. (Why?)

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 56 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Existence of Pure Strategy Equilibria I

Since this is di¢ cult to check without knowing what p , the following
“su¢ cient condition” might be useful:

∂hg (x )
B g (x, p ) (DU g (p )) (DU g (p ))T
h g (x )D 2 U g (p ) +
∂x
(9)
is negative de…nite for any x and p, and each g .

Theorem
( Pure Strategy Existence) Suppose that (9) holds. Then in the
probabilistic voting game, a pure strategy equilibrium always exists.

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 57 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Existence of Pure Strategy Equilibria II

But (9) is a very restrictive condition. In general satis…ed only if all

the H g ’s uniform.
Thus we have not solved the existence problem at all.
To understand (9), consider the …rst and second order conditions in
the one-dimensional policy case with …rst-order condition
G
∂U g (pA )
∑ h g ( U g ( pA ) U g (pB ))
∂p
=0
g =1

G
∂ 2 U g ( pA )
∑ h g ( U g ( pA ) U g (pB ))
∂p 2
+
g =1
G 2
∂hg (U g (pA ) U g (pB )) ∂U g (pA )
∑ ∂x ∂p
<0
g =1

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 58 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Existence of Pure Strategy Equilibria III

Looking at each group’s utility separately, this requires

∂2 U g (pA )/∂p 2 ∂hg (U g (pA ) U g (pB ))/∂x

2
>
(∂U g (pA )/∂p ) hg (U g (pA ) U g (pB ))

for all g .
At the same time, this point must also be a best response for party B,
so by the same arguments,

∂2 U g (pB )/∂p 2 ∂hg (U g (pA ) U g (pB ))/∂x

> .
(∂U g (pB )/∂p )2 hg (U g (pA ) U g (pB ))

A su¢ cient condition for both of these inequalities to be satis…ed is

j∂hg (x ) /∂x j ∂2 U g (p ) /∂p

sup inf for all g .
x h g (x ) p ∂U g (p ) /∂p

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 59 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Existence of Mixed Strategy Equilibria

Naturally, mixed strategy equilibria are easier to guarantee (for

example, they are immediate from Glicksberg’s Theorem)

Theorem
( Mixed Strategy Existence) In the probabilistic voting game, a mixed
strategy equilibrium always exists.

But do these equilibria have the same features as the canonical

probabilistic voting equilibria?

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 60 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Application: the Power of the Middle Class I

Here is an example showing how with uniform distribution,
probablilistic voting becomes very tractable and useful.
Also, assume now (more realistically) that parties care about
probability of coming to power not vote share.
Key concepts: “swing voters”— who are more responsive to policy.
Three distinct groups, g = R, M, P, representing the rich, the middle
class, and the poor, with preferences
U (p ) = u p i ,
u ( ) is the strictly monotonic utility function common to all groups.
The population share of group g is λg , with ∑3g =1 λg = 1.
The relevant policy vector is again a vector of redistributions
p = p 1 , p 2 , p 3 with ∑3g =1 λg p g = 1.
At the time of the elections, voters base their voting decision both on
the economic policy announcements and on the two parties’ideologies
relative to the realization of their own ideology.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 61 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Application: the Power of the Middle Class II

Voter i in group g prefers party A if

U g (pA ) > U g (pB ) + σig + δ.

Let us assume that this parameter for each group g has group-speci…c
uniform distribution on
1 1
g , g .
2φ 2φ

The parameter δ measures the average (relative) popularity of

candidate B in the population as a whole, and also can be positive or
negative. Assume that it has a uniform distribution on

1 1
, .
2ψ 2ψ

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 62 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Application: the Power of the Middle Class III

The “indi¤erent” voter in group g will be a voter whose ideological
bias, given the candidates’platforms, makes him indi¤erent between
the two parties.
σg = U g (pA ) U g (pB ) δ.
All voters in group g with σig σg prefer party A. Therefore, party
A’s actual vote share is
1
πA = ∑ λg φg σg +
2φg
.
g

Notice that σg depends on the realized value of δ, and thus the vote
share π A is also a random variable.
Party A’s probability of winning is then
" #
3
1 1
PA = Prob π A = + ψ ∑ λg φg [U g (pA ) U g (pB )] ,
δ 2 2 g =1

Party B wins with probability 1 PA .

Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 63 / 64
 Collective Choice and Static Voting Models Probablilistic Voting

Application: the Power of the Middle Class III

Suppose party B has announced the equilibrium policy pB = p .
Then
2 3
λ1 φ1 (u (pA,1 ) u (p1 ))
1 1
PA = Prob π A = + ψ 4 + λ2 φ2 (u (pA,2 ) u (p2 )) 5 ,
δ 2 2
+ λ3 φ3 (u (pA,3 ) u (p3 ))
(10)
Party A will maximize (10) subject to the resource constraint.
The …rst-order conditions are
φ1 u 0 (pA,1 ) = η
φ2 u 0 (pA,2 ) = η
φ3 u 0 (pA,3 ) = η
where η is the Lagrangean multiplier on the resource constraint.
Implication: whichever group has higher φ, thus approximating “a
swing voter group” will have greater in‡uence on policies.
Daron Acemoglu (MIT) Political Economy Lectures 1 and 2 September 6 and 11, 2018 64 / 64