social choice
DESCRIPTION
Social Theory introduction up to Arrow's and Gibbard–Satterthwaite's Theorems. Have fun.TRANSCRIPT
Social Choice Theory
Enrico Franchi [email protected]
Group Decisions } Agents are required to choose among a set of outcomes
} Agents can choose one outcome in Ω } Agents can express a preference of outcomes } Let the set of preference orderings of outcomes } We also write to express that agent i prefers ω1
to ω2
{ }1 2, ,ω ωΩ = …
( )Π Ω
ω1 i ω 2
2 Enrico Franchi ([email protected])
Social Welfare } A social welfare function takes the voter preferences and
produces a social preference order: or in the slightly simplified form:
} We write to express that the first outcome ranked above the second in the social outcome
( ) ( ): Nf Π Ω →Π Ω
( ): Nf Π Ω →Ω
ω1 * ω 2
3 Enrico Franchi ([email protected])
Plurality } Simplest voting procedure: used to select a single
outcome (candidate) } Everyone submits his preference order, we count how
many times each candidate was ranked first } Easy to implement and to understand } If the outcomes are just 2, it is called simple majority voting } If they are more than two, problems arise
4 Enrico Franchi ([email protected])
Voting in the UK } Three main parties:
} Labour Party (left-wing) } Liberal Democrats (center-
left) } Conservative Party (right-
wing)
} Left-wing voter: } Center voter: } Right-wing voter: } Tactical Voting } Strategic Manipulation
Labour Party 44%
Liberal Democ
rats 12%
Conservative Party 44%
Voters
ω L ω D ωC
ω D ω L ωC
ωC ω D ω L
5 Enrico Franchi ([email protected])
Condorcet’s Paradox } Consider this election:
} No matters the outcome we choose: two thirds of the electors will be unhappy
Ω = ω1,ω 2 ,ω3{ } Ag = 1,2,3{ }
ω1 1 ω 2 1 ω3
ω3 2 ω1 2 ω 2
ω 2 3 ω3 3 ω1
6 Enrico Franchi ([email protected])
Sequential Majority } Series of pair-wise elections, the winner will go on to the
next election } An agenda is the strategy we choose to order the
elections (linear, binary tree) } An outcome is a possible winner if there is some agenda
which would make that outcome the overall winner } An outcome is a Condorcet winner if it is the overall
winner for every possible agenda } Can we choose the agenda to choose a winner?
7 Enrico Franchi ([email protected])
Borda Count and Slater Ranking } Borda Count
} We have K outcomes } Each time an outcome is in the j-th position for some agent, we
increment its counter by K-j } We order the outcomes according to their counter } Good for single candidates
} Slater ranking } Tries to be as close to the majority graph as possible } Unfortunately, is NP-hard
8 Enrico Franchi ([email protected])
Properties } Pareto condition: if every agent ranks ωi above ωj,
then } Plurality, Borda
} Condorcet winner: if an outcome is a Condorcet winner, then it should be ranked first } Sequential majority elections
} Independence of Irrelevant Alternatives (IIA): social ranking of two outcomes should only be affected by the way that they are ranked in their preference orders } Almost no protocol satisfies IIA
ω i * ω j
9 Enrico Franchi ([email protected])
Properties } Dictatorship: a social welfare function f is a dictatorship
if for some voter j we have that:
} Unrestricted Domain: for any set of individual voter preferences, the social welfare function should yield a unique and complete ranking of societal choices. } E.g., not random, always answers, does not “loop”
( )1, , N jf ω ω ω… =
10 Enrico Franchi ([email protected])
Arrow’s Theorem
} There is no voting procedure for elections with more than two outcomes that satisfies } Non-dictatorship } Unrestricted Domain } Pareto } Independence of Irrelevant Alternatives
11 Enrico Franchi ([email protected])
Gibbard-Satterthwaite’s Theorem } Sometimes voters “lie” in order to obtain a better
outcome } Is it possible to devise a voting procedure that is not
subject to such manipulation?
} Manipulation (i prefers ωi):
} The only procedure that cannot be manipulated and satisfies the Pareto condition is dictatorship
f ω1,…,ω i '…,ω n( ) i f ω1,…,ω i…,ω n( )
12 Enrico Franchi ([email protected])
Complexity and Manipulation
Enrico Franchi ([email protected]) 13
} Even if all procedures can be manipulated, can we devise procedures which are hard to manipulate?
} Hard means “difficult to compute” in an algorithmic sense, e.g., NP-Hard procedures
} These procedures are easy (polynomial) to compute? } Second-order Copeland may be “difficult” to manipulate
} In theory it is NP-Hard } However, it is only a worst case complexity
References
Enrico Franchi ([email protected]) 14
1. Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations; Yoav Shoham and Kevin Leyton-Brown; Cambridge Press
2. Game Theory: Analysis of Conflict; Roger B. Myerson; Harvard Press
3. An Introduction to Multi-Agent Systems; Michael Wooldridge; Wiley Press