Finding the Best Voting Protocol I

Why should we consider so many different voting protocols?

What is the best protocol? Can we mathematically prove this?

Robert Bredereck Computational Social Choice

7
Finding the Best Voting Protocol I
Why should we consider so many different voting protocols?
What is the best protocol? Can we mathematically prove this?

Answer: No, but one can prove the “converse”.

Robert Bredereck Computational Social Choice

7
Finding the Best Voting Protocol I
Why should we consider so many different voting protocols?
What is the best protocol? Can we mathematically prove this?

Answer: No, but one can prove the “converse”.

Kenneth Joseph
Arrow, 1921–
Theorem (Arrow’s Impossibility Theorem)
There is no preference-based voting system that satisfies the Nobel Memorial
Prize in
properties unrestricted domain, non-dictatorship, Pareto efficiency, Economic
and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950] Sciences 1972

Robert Bredereck Computational Social Choice

7
Finding the Best Voting Protocol I
Why should we consider so many different voting protocols?
What is the best protocol? Can we mathematically prove this?

Answer: No, but one can prove the “converse”.

Kenneth Joseph
Arrow, 1921–
Theorem (Arrow’s Impossibility Theorem)
There is no preference-based voting system that satisfies the Nobel Memorial
Prize in
properties unrestricted domain, non-dictatorship, Pareto efficiency, Economic
and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950] Sciences 1972
Comments
proper subsets can be satisfied
∃ alternative formulations with different minimal sets of properties
=⇒ There is no “best voting protocol”!
Wait, who said these properties are desirable?
Robert Bredereck Computational Social Choice
7
Finding the Best Voting Protocol I
Why should we consider so many different voting protocols?
What is the best protocol? Can we mathematically prove this?

Answer: No, but one can prove the “converse”.

Kenneth Joseph
Arrow, 1921–
Theorem (Arrow’s Impossibility Theorem)
There is no preference-based voting system that satisfies the Nobel Memorial
Prize in
properties unrestricted domain, non-dictatorship, Pareto efficiency, Economic
and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950] Sciences 1972
Comments
proper subsets can be satisfied
∃ alternative formulations with different minimal sets of properties
=⇒ There is no “best voting protocol”!

Robert Bredereck
But wait, are these four properties really desirable? Computational Social Choice
7
Finding the Best Voting Protocol II

Are these properties desirable?

Unrestricted domain: all preferences are allowed to occur.

Robert Bredereck Computational Social Choice

8
Finding the Best Voting Protocol II

Are these properties desirable?

Unrestricted domain: all preferences are allowed to occur.

Non-dictatorship: there is no single individual that decides on the winner.

Robert Bredereck Computational Social Choice

8
Finding the Best Voting Protocol II

Are these properties desirable?

Unrestricted domain: all preferences are allowed to occur.

Non-dictatorship: there is no single individual that decides on the winner.

Pareto efficiency: if there is any pair of candidates {a, b} such that every voter
prefers a over b, then the system also prefers a over b.

Robert Bredereck Computational Social Choice

8
Finding the Best Voting Protocol II

Are these properties desirable?

Unrestricted domain: all preferences are allowed to occur.

Non-dictatorship: there is no single individual that decides on the winner.

Pareto efficiency: if there is any pair of candidates {a, b} such that every voter
prefers a over b, then the system also prefers a over b.

Independence of irrelevant alternatives: if the system prefers a over b, then, after

changing the votes while keeping the relative orderings of a and b, the
system still prefers a over b.

Robert Bredereck Computational Social Choice

8
Independence of Irrelevant Alternatives
Morgenbesser, ordering dessert, is told by the waitress that he can choose between
apple pie and blueberry pie.

Bon-AppetEats on DeviantArt

Robert Bredereck Computational Social Choice

9
Independence of Irrelevant Alternatives
Morgenbesser, ordering dessert, is told by the waitress that he can choose between
apple pie and blueberry pie.

Bon-AppetEats on DeviantArt

He orders the apple pie.

Robert Bredereck Computational Social Choice

9
Independence of Irrelevant Alternatives
Morgenbesser, ordering dessert, is told by the waitress that he can choose between
apple pie and blueberry pie.

Bon-AppetEats on DeviantArt

He orders the apple pie. Shortly thereafter, the waitress comes back and says that
cherry pie is also an option.

Robert Bredereck Computational Social Choice

9
Independence of Irrelevant Alternatives
Morgenbesser, ordering dessert, is told by the waitress that he can choose between
apple pie and blueberry pie.

Bon-AppetEats on DeviantArt

He orders the apple pie. Shortly thereafter, the waitress comes back and says that
cherry pie is also an option.
Morgenbesser says: “In that case I’ll have the blueberry pie”.
Robert Bredereck Computational Social Choice
9
Proving Arrow’s Impossibility Theorem I
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]
(IIA)

Robert Bredereck Computational Social Choice

10
Proving Arrow’s Impossibility Theorem I
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]
General Setting
Let C be a finite set of candidates and V be a finite set of n voters.
Let �i be a weak order over C representing the preferences of voter i. We write
x �i y if and only if ¬(y �i x).
Binary relation over C:
* trasitive: for all x, y and z: x y and y z implies x z
* connex: for all x and y: x y or y x
* reflexive: for all x: x x

Robert Bredereck Computational Social Choice

10
Proving Arrow’s Impossibility Theorem I
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]
General Setting
Let C be a finite set of candidates and V be a finite set of n voters.
Let �i be a weak order over C representing the preferences of voter i. We write
x �i y if and only if ¬(y �i x).
Let P denote the set of all possible weak orders over C .
A social preference function �S : P n → P assigns to each preference profile a societal
preference order. (Omit the profile if clear.)

Robert Bredereck Computational Social Choice

10
Proving Arrow’s Impossibility Theorem I
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]
General Setting
Let C be a finite set of candidates and V be a finite set of n voters.
Let �i be a weak order over C representing the preferences of voter i. We write
x �i y if and only if ¬(y �i x).
Let P denote the set of all possible weak orders over C .
A social preference function �S : P n → P assigns to each preference profile a societal
preference order. (Omit the profile if clear.)
Decisive Voters (Intuitively, voters with more power than others)
For a social preference function �S , a voter v is decisive for x over y �= x if x �v y
Robert Bredereck implies x �S y . Computational Social Choice
10
Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

Robert Bredereck Computational Social Choice

11
Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .

Robert Bredereck Computational Social Choice

11
Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈

Robert Bredereck Computational Social Choice

11
 Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈

Robert Bredereck Computational Social Choice

11
 Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈

Robert Bredereck Computational Social Choice

11
 Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈

Robert Bredereck Computational Social Choice

11
 Proving Arrow’s Impossibility Theorem II
Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
6. Voter v ∗ is decisive for a over b and b over a.

Robert Bredereck Computational Social Choice

11
 Homework: 1) Which properties are not so desirable?
2) Which were (surprisingly?) strong in the proof?

Proving Arrow’s Impossibility Theorem II

Theorem (Arrow’s Impossibility Theorem - General Form)
There is no social preference function satisfying unrestricted domain, non-dictatorship,
Pareto efficiency, and independence of irrelevant alternatives. [Arrow, J. Polit. Econ. 1950]

High-Level Approach
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
6. Voter v ∗ is decisive for a over b and b over a.
7. For all c, c � ∈
/ {a, b}, voter v ∗ is decisive for c over c � .
Robert Bredereck Computational Social Choice
11
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .

Robert Bredereck Computational Social Choice

12
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .

Robert Bredereck Computational Social Choice

12
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .
Consider the profile Pb�a where all voters strictly prefer b � a � . . . .

Robert Bredereck Computational Social Choice

12
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .
Consider the profile Pb�a where all voters strictly prefer b � a � . . . .
Due to Pareto efficiency, we have a �S (Pa�b ) b and b �S (Pb�a ) a.

Robert Bredereck Computational Social Choice

12
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .
Consider the profile Pb�a where all voters strictly prefer b � a � . . . .
Due to Pareto efficiency, we have a �S (Pa�b ) b and b �S (Pb�a ) a.
Identifying the specific voter v ∗
Transform profile Pa�b into profile Pb�a by successively switching a and b. Let v ∗ be
the first voter where �S changes from a �S b to something else.

Robert Bredereck Computational Social Choice

12
 Proving Arrow’s Impossibility Theorem - Step 1a
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .
Consider the profile Pb�a where all voters strictly prefer b � a � . . . .
Due to Pareto efficiency, we have a �S (Pa�b ) b and b �S (Pb�a ) a.
Identifying the specific voter v ∗
Transform profile Pa�b into profile Pb�a by successively switching a and b. Let v ∗ be
the first voter where �S changes from a �S b to something else.
voter 1: b � a � . . .
..
.
Profile P1 : voter v ∗ − 1: b � a � . . . a �S (P1 ) b
∗
voter v : a � b � . . .
..
.
Robert Bredereck
voter n: a � b � . . . Computational Social Choice
12
 Proving Arrow’s Impossibility Theorem - Step 1b
1. Fix a specific voter v ∗ “influencing” some pair {a, b} ⊆ C .
Consider the profile Pa�b where all voters strictly prefer a � b � . . . .
Consider the profile Pb�a where all voters strictly prefer b � a � . . . .
Due to Pareto efficiency, we have a �S (Pa�b ) b and b �S (Pb�a ) a.
Identifying the specific voter v ∗
Transform profile Pa�b into profile Pb�a by successively switching a and b. Let v ∗ be
the first voter where �S changes from a �S b to something else.
voter 1: b � a � . . .
..
.
Profile P2 : voter v ∗ − 1: b � a � . . . b �S (P2 ) a
∗
voter v : b � a � . . .
..
.
Robert Bredereck
voter n: a � b � . . . Computational Social Choice
13
 Proving Arrow’s Impossibility Theorem - Step 2a
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈

Robert Bredereck Computational Social Choice

14
 Proving Arrow’s Impossibility Theorem - Step 2a
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P3 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: a � b � c � . . .

Robert Bredereck Computational Social Choice

14
 Proving Arrow’s Impossibility Theorem - Step 2a
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P3 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: a � b � c � . . .
By independence of irrelevant alternatives (IIA) and P1 : a �S b.

Robert Bredereck Computational Social Choice

14
 Proving Arrow’s Impossibility Theorem - Step 2a
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P3 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: a � b � c � . . .
By independence of irrelevant alternatives (IIA) and P1 : a �S b.
By Pareto-efficiency: b �S c and a �S c (transitivity).

Robert Bredereck Computational Social Choice

14
 Proving Arrow’s Impossibility Theorem - Step 2b
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: {b, c} � a � . . .
..
.
Profile P4 : voter v ∗ − 1: {b, c} � a � . . .
voter v ∗ : b � a � c � . . .
..
.
voter n: a � {b, c} � . . .

Robert Bredereck Computational Social Choice

15
 Proving Arrow’s Impossibility Theorem - Step 2b
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: {b, c} � a � . . .
..
.
Profile P4 : voter v ∗ − 1: {b, c} � a � . . .
voter v ∗ : b � a � c � . . .
..
.
voter n: a � {b, c} � . . .
a and b are ranked as in P2 : b �S a (again IIA)

Robert Bredereck Computational Social Choice

15
 Proving Arrow’s Impossibility Theorem - Step 2b
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: {b, c} � a � . . .
..
.
Profile P4 : voter v ∗ − 1: {b, c} � a � . . .
voter v ∗ : b � a � c � . . .
..
.
voter n: a � {b, c} � . . .
a and b are ranked as in P2 : b �S a
a and c are ranked as in P3 : a �S c =⇒ b �S c (trasitivity)

Robert Bredereck Computational Social Choice

15
 Proving Arrow’s Impossibility Theorem - Step 2b
/ {a, b}, voter v ∗ is decisive for b over c.
2. For all c ∈
Consider the following profile:

voter 1: {b, c} � a � . . .
..
.
Profile P4 : voter v ∗ − 1: {b, c} � a � . . .
voter v ∗ : b � a � c � . . .
..
.
voter n: a � {b, c} � . . .
a and b are ranked as in P2 : b �S a
a and c are ranked as in P3 : a �S c =⇒ b �S c
=⇒ by IIA: v ∗ is decisive for b over c
Robert Bredereck Computational Social Choice
15
 Proving Arrow’s Impossibility Theorem - Step 3
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
Consider the following profile:

voter 1: {a, c} � b � . . .
..
.
Profile P5 : voter v ∗ − 1: {a, c} � b � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: {a, c} � b � . . .

Robert Bredereck Computational Social Choice

16
 Proving Arrow’s Impossibility Theorem - Step 3
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
Consider the following profile:

voter 1: {a, c} � b � . . .
..
.
Profile P5 : voter v ∗ − 1: {a, c} � b � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: {a, c} � b � . . .
as v∗ is decisive for b over c: b �S c

Robert Bredereck Computational Social Choice

16
 Proving Arrow’s Impossibility Theorem - Step 3
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
Consider the following profile:

voter 1: {a, c} � b � . . .
..
.
Profile P5 : voter v ∗ − 1: {a, c} � b � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: {a, c} � b � . . .
as v∗is decisive for b over c: b �S c
by Pareto-efficiency: a �S b =⇒ a �S c (by transitivity)

Robert Bredereck Computational Social Choice

16
 Proving Arrow’s Impossibility Theorem - Step 3
/ {a, b}, voter v ∗ is decisive for a over c.
3. For all c ∈
Consider the following profile:

voter 1: {a, c} � b � . . .
..
.
Profile P5 : voter v ∗ − 1: {a, c} � b � . . .
voter v ∗ : a � b � c � . . .
..
.
voter n: {a, c} � b � . . .
as v∗is decisive for b over c: b �S c
by Pareto-efficiency: a �S b =⇒ a �S c
=⇒ by IIA: v ∗ is decisive for a over c
Robert Bredereck Computational Social Choice
16
 Proving Arrow’s Impossibility Theorem - Step 4a
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈

Robert Bredereck Computational Social Choice

17
 Proving Arrow’s Impossibility Theorem - Step 4a
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P6 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: c � a � b � . . .

Robert Bredereck Computational Social Choice

17
 Proving Arrow’s Impossibility Theorem - Step 4a
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P6 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: c � a � b � . . .
ranks a and b as in P1 : a �S b

Robert Bredereck Computational Social Choice

17
 Proving Arrow’s Impossibility Theorem - Step 4a
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � c � a � . . .
..
.
Profile P6 : voter v ∗ − 1: b � c � a � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: c � a � b � . . .
ranks a and b as in P1 : a �S b
by Pareto-efficiency: c �S a =⇒ c �S b (by transitivity)

Robert Bredereck Computational Social Choice

17
 Proving Arrow’s Impossibility Theorem - Step 4b
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � {a, c} � . . .
..
.
Profile P7 : voter v ∗ − 1: b � {a, c} � . . .
voter v ∗ : c � b � a � . . .
..
.
voter n: {a, c} � b � . . .

Robert Bredereck Computational Social Choice

18
 Proving Arrow’s Impossibility Theorem - Step 4b
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � {a, c} � . . .
..
.
Profile P7 : voter v ∗ − 1: b � {a, c} � . . .
voter v ∗ : c � b � a � . . .
..
.
voter n: {a, c} � b � . . .
ranks a and b as in P2 : b �S a (using IIA)

Robert Bredereck Computational Social Choice

18
 Proving Arrow’s Impossibility Theorem - Step 4b
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � {a, c} � . . .
..
.
Profile P7 : voter v ∗ − 1: b � {a, c} � . . .
voter v ∗ : c � b � a � . . .
..
.
voter n: {a, c} � b � . . .
ranks a and b as in P2 : b �S a (again IIA)
ranks b and c as in P6 : c �S b =⇒ c �S a (+ transitivity)

Robert Bredereck Computational Social Choice

18
 Proving Arrow’s Impossibility Theorem - Step 4b
/ {a, b}, voter v ∗ is decisive for c over a.
4. For all c ∈
Consider the following profile:

voter 1: b � {a, c} � . . .
..
.
Profile P7 : voter v ∗ − 1: b � {a, c} � . . .
voter v ∗ : c � b � a � . . .
..
.
voter n: {a, c} � b � . . .
ranks a and b as in P2 : b �S a
ranks b and c as in P6 : c �S b =⇒ c �S a
=⇒ by IIA: v ∗ is decisive for c over a
Robert Bredereck Computational Social Choice
18
 Proving Arrow’s Impossibility Theorem - Step 5
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈

Robert Bredereck Computational Social Choice

19
 Proving Arrow’s Impossibility Theorem - Step 5
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
Consider the following profile:

voter 1: a � {b, c} � . . .
..
.
Profile P8 : voter v ∗ − 1: a � {b, c} � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: a � {b, c} � . . .

Robert Bredereck Computational Social Choice

19
 Proving Arrow’s Impossibility Theorem - Step 5
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
Consider the following profile:

voter 1: a � {b, c} � . . .
..
.
Profile P8 : voter v ∗ − 1: a � {b, c} � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: a � {b, c} � . . .
as v∗ is decisive for c over a: c �S a

Robert Bredereck Computational Social Choice

19
 Proving Arrow’s Impossibility Theorem - Step 5
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
Consider the following profile:

voter 1: a � {b, c} � . . .
..
.
Profile P8 : voter v ∗ − 1: a � {b, c} � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: a � {b, c} � . . .
as v∗is decisive for c over a: c �S a
by Pareto-efficiency: a �S b =⇒ c �S b (transitivity)

Robert Bredereck Computational Social Choice

19
 Proving Arrow’s Impossibility Theorem - Step 5
/ {a, b}, voter v ∗ is decisive for c over b.
5. For all c ∈
Consider the following profile:

voter 1: a � {b, c} � . . .
..
.
Profile P8 : voter v ∗ − 1: a � {b, c} � . . .
voter v ∗ : c � a � b � . . .
..
.
voter n: a � {b, c} � . . .
as v∗is decisive for c over a: c �S a
by Pareto-efficiency: a �S b =⇒ c �S b
=⇒ by IIA: v ∗ is decisive for c over b
Robert Bredereck Computational Social Choice
19
 Proving Arrow’s Impossibility Theorem - Step 6
6. Voter v∗ is decisive for a over b and b over a.

Robert Bredereck Computational Social Choice

20
 Proving Arrow’s Impossibility Theorem - Step 6
6. Voter v∗
is decisive for a over b and b over a.
Consider the following profile:

voter 1: {a, b} � . . .
..
.
Profile P9 : voter v ∗ − 1: {a, b} � . . .
voter v ∗ : a � c � b � . . .
..
.
voter n: {a, b} � . . .

Robert Bredereck Computational Social Choice

20
 Proving Arrow’s Impossibility Theorem - Step 6
6. Voter v∗
is decisive for a over b and b over a.
Consider the following profile:

voter 1: {a, b} � . . .
..
.
Profile P9 : voter v ∗ − 1: {a, b} � . . .
voter v ∗ : a � c � b � . . .
..
.
voter n: {a, b} � . . .
as v ∗ is decisive for a over c and c over b: a �S c �S b =⇒ a �S b (transitivity)

Robert Bredereck Computational Social Choice

20
 Proving Arrow’s Impossibility Theorem - Step 6
6. Voter v∗
is decisive for a over b and b over a.
Consider the following profile:

voter 1: {a, b} � . . .
..
.
Profile P9 : voter v ∗ − 1: {a, b} � . . .
voter v ∗ : a � c � b � . . .
..
.
voter n: {a, b} � . . .
as v ∗ is decisive for a over c and c over b: a �S c �S b =⇒ a �S b
=⇒ by IIA: v ∗ is decisive for a over b

Robert Bredereck Computational Social Choice

20
 Proving Arrow’s Impossibility Theorem - Step 6
6. Voter v∗
is decisive for a over b and b over a.
Consider the following profile:

voter 1: {a, b} � . . .
..
.
Profile P9 : voter v ∗ − 1: {a, b} � . . .
voter v ∗ : a � c � b � . . .
..
.
voter n: {a, b} � . . .
as v ∗ is decisive for a over c and c over b: a �S c �S b =⇒ a �S b
=⇒ by IIA: v ∗ is decisive for a over b
works analogous for showing that v ∗ is decisive for b over a
(swap a and b for v ∗ )
Robert Bredereck Computational Social Choice
20
 Proving Arrow’s Impossibility Theorem - Step 7
7. For all c, c � / {a, b}, voter v ∗ is decisive for c over c � .
∈

Robert Bredereck Computational Social Choice

21
 Proving Arrow’s Impossibility Theorem - Step 7
7. For all c, c �
/ {a, b}, voter v ∗ is decisive for c over c � .
∈
Consider the following profile:

voter 1: {x, y } � {a, b} . . .

..
.
Profile P10 : voter v ∗ − 1: {x, y } � {a, b} � . . .
voter v ∗ : x � a � b � y � . . .
..
.
voter n: {x, y } � {a, b} . . .

Robert Bredereck Computational Social Choice

21
 Proving Arrow’s Impossibility Theorem - Step 7
7. For all c, c �
/ {a, b}, voter v ∗ is decisive for c over c � .
∈
Consider the following profile:

voter 1: {x, y } � {a, b} . . .

..
.
Profile P10 : voter v ∗ − 1: {x, y } � {a, b} � . . .
voter v ∗ : x � a � b � y � . . .
..
.
voter n: {x, y } � {a, b} . . .
as v ∗ is decisive for x over a, for a over b, and for b over y : x �S a �S b �S y =⇒
x �S y (by transitivity)

Robert Bredereck Computational Social Choice

21
 Proving Arrow’s Impossibility Theorem - Step 7
7. For all c, c �
/ {a, b}, voter v ∗ is decisive for c over c � .
∈
Consider the following profile:

voter 1: {x, y } � {a, b} . . .

..
.
Profile P10 : voter v ∗ − 1: {x, y } � {a, b} � . . .
voter v ∗ : x � a � b � y � . . .
..
.
voter n: {x, y } � {a, b} . . .
as v ∗ is decisive for x over a, for a over b, and for b over y : x �S a �S b �S y =⇒
x �S y
=⇒ by IIA: v ∗ is decisive for x over y

Robert Bredereck Computational Social Choice

21
 Proving Arrow’s Impossibility Theorem - Step 7
7. For all c, c �
/ {a, b}, voter v ∗ is decisive for c over c � .
∈
Consider the following profile:

voter 1: {x, y } � {a, b} . . .

..
.
Profile P10 : voter v ∗ − 1: {x, y } � {a, b} � . . .
voter v ∗ : x � a � b � y � . . .
..
.
voter n: {x, y } � {a, b} . . .
as v ∗ is decisive for x over a, for a over b, and for b over y : x �S a �S b �S y =⇒
x �S y
=⇒ by IIA: v ∗ is decisive for x over y
Finally: v ∗ is decisive for all candidate pairs!
Robert Bredereck Computational Social Choice
21