Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Power Extending the model Axiomatisation Application Conclusion
The men who weren’t even thereLegislative voting with absentees
László Á. Kóczy1 Miklós P. Pintér2
1IEHAS & Óbuda University, Budapest
2Corvinus University Budapest
Xth Meeting of the Society for Social Choice and Welfare,21–24 July 2010.
Power Extending the model Axiomatisation Application Conclusion
Voting power
Voting situation
Voters cast votes for or againstLegislation rules determine whether notion is passed or not
Examples: National parliaments, shareholder’s meetings, UNSecurity Council, EU Council of Ministers, IMF
Voting power
(Share of) decision probability.
Power Extending the model Axiomatisation Application Conclusion
Voting power 2 – Example
EEC Council ofMinisters, 1958, QMV.
Country weightBelgium 2France 4Germany 4Italy 4Luxemburg 1Netherlands 2Quota 12
β = ( 521 ,521 ,
321 ,
321 ,
521 ,0)
Power index(Normalised) a priori measure of voting power.
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 1. abstention
Yes/no models do not cover all situations→ ternary voting: allowing abstention:
1 may be a No vote – reduces chance for approval2 may be a Yes vote – when veto is not used
Yes Abstain No
Abstain
Yes NoModel 1 (Machover and Felsenthal, 1997) Model 2 (Braham and Steffen, 2002)
Two models3 options: Yes/No/AbstainVoters decide: Abstain/Vote. If latter: Yes/No
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 1. abstention
Yes/no models do not cover all situations→ ternary voting: allowing abstention:
1 may be a No vote – reduces chance for approval2 may be a Yes vote – when veto is not used
Yes Abstain No
Abstain
Yes NoModel 1 (Machover and Felsenthal, 1997) Model 2 (Braham and Steffen, 2002)
Two models3 options: Yes/No/AbstainVoters decide: Abstain/Vote. If latter: Yes/No
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 1. abstention
Yes/no models do not cover all situations→ ternary voting: allowing abstention:
1 may be a No vote – reduces chance for approval2 may be a Yes vote – when veto is not used
Yes Abstain No
Abstain
Yes NoModel 1 (Machover and Felsenthal, 1997) Model 2 (Braham and Steffen, 2002)
Two models3 options: Yes/No/AbstainVoters decide: Abstain/Vote. If latter: Yes/No
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 2.
The voting statistics of theHungarian National Assembly listthose:
1 voting Yes2 voting No3 voting Abstain (F-M abstain)4 not voting (B-S abstain)
5 not present
The numbers do not add upBeing absent is a “non-strategicabstention.”
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 2.
The voting statistics of theHungarian National Assembly listthose:
1 voting Yes2 voting No3 voting Abstain (F-M abstain)4 not voting (B-S abstain)
5 not present
The numbers do not add upBeing absent is a “non-strategicabstention.”
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 2.
The voting statistics of theHungarian National Assembly listthose:
1 voting Yes2 voting No3 voting Abstain (F-M abstain)4 not voting (B-S abstain)5 not present
The numbers do not add upBeing absent is a “non-strategicabstention.”
Power Extending the model Axiomatisation Application Conclusion
The model expanded: 2.
The voting statistics of theHungarian National Assembly listthose:
1 voting Yes2 voting No3 voting Abstain (F-M abstain)4 not voting (B-S abstain)5 not present
The numbers do not add upBeing absent is a “non-strategicabstention.”
Power Extending the model Axiomatisation Application Conclusion
Absenteeism elaborated
Some pointsAbsent = ill, busy elsewhere. Non-strategic decision.Absent voter (party) = abstaining voter (party)In legislative voting a party’s weight is the nr of MP’spresent. Here:absent (individual) voter = smaller weight for party
Weighted voting game [q,w1, . . . ,wn] or [q,w ]With absent voters [q′,w ′1, . . . ,w
′n] or [q′,w ′], where
q′ = q′(w ′1, . . . ,w′n).
E.g. q′ = q, or q′ : q′/∑
w ′i = q/∑
wi .
Power Extending the model Axiomatisation Application Conclusion
Absenteeism continued
Consider game [q,w ] ∈ Γ (Γ = n-player weighted voting games)Induced game [q′,w ′] with probability p(w ′).
A generalised weighted voting game (p,q,w) ∈ Γ̃ is a triple:
a vector of maximum weights wa quota function q : Nn0 −→ N0a probability distribution on all w ′ : 0 ≤ w ′ ≤ w vectors
A power measure κ : Γ −→ Rn
A power measure with absenteeism:
κ̃ : Γ̃ −→ Rn
κ̃i(p,q,w) =∑
∀i: 0≤w ′
Power Extending the model Axiomatisation Application Conclusion
Absenteeism: Properties
Power index=power measure normalised to 1
PropositionThe power index of i is 1 iff i is the sole player.The power index of i is 0 iff i has weight 0.
CorollaryA minority has a positive power.
Claim (proof pending)
For a fixed q, assume that the expected size of the majoritycoalition exceeds the quota. Ceteris paribus the larger theassembly the smaller the minority power
A large assembly is but a voting machine.
Power Extending the model Axiomatisation Application Conclusion
Absenteeism: Properties
Power index=power measure normalised to 1
PropositionThe power index of i is 1 iff i is the sole player.The power index of i is 0 iff i has weight 0.
CorollaryA minority has a positive power.
Claim (proof pending)
For a fixed q, assume that the expected size of the majoritycoalition exceeds the quota. Ceteris paribus the larger theassembly the smaller the minority power
A large assembly is but a voting machine.
Power Extending the model Axiomatisation Application Conclusion
Absenteeism: Properties
Power index=power measure normalised to 1
PropositionThe power index of i is 1 iff i is the sole player.The power index of i is 0 iff i has weight 0.
CorollaryA minority has a positive power.
Claim (proof pending)
For a fixed q, assume that the expected size of the majoritycoalition exceeds the quota. Ceteris paribus the larger theassembly the smaller the minority power
A large assembly is but a voting machine.
Power Extending the model Axiomatisation Application Conclusion
Axiomatisation – Setup
Based on Dubey (IJGT, 1975) and Young (IJGT, 1985).Notation:v ∨ w = max {v ,w} if v ,w ∈ Γ.uT is the unanimity game on T , 1vT indicates T wins in v .
v =∨
T∈Wv
uT =∨
T⊆N1vT uT
Let pv : pv ≥ 0 for all v ∈ Γ and∑
v∈Γ pv = 1.
Generalised voting game
ṽ =∑v∈Γ
pv v .
Power Extending the model Axiomatisation Application Conclusion
Axiomatisation – Axioms
The value κ : Γ̃→ RN satisfies
Efficiency: if ∀ṽ ∈ Γ̃: ṽ(N) =n∑
i=1κi(ṽ),
Symmetry: if ∀ṽ ∈ Γ̃ such that i ∼ṽ j : κi(ṽ) = κj(ṽ),Marginality: if ∀ṽ , w̃ ∈ Γ̃, ∀i ∈ N such that ṽ ′i = w̃ ′i :κi(ṽ) = κi(w̃).
The Shapley value meets Efficiency, Symmetry and Marginality.
Power Extending the model Axiomatisation Application Conclusion
Axiomatisation – Result
Theorem
On the class Γ̃ solution κ satisfies Efficiency, Symmetry andMarginality if and only if κ = φ.
Fact1: ṽ =∑
w∈Γ
∨T⊆N
pw1wT uT =∨
T⊆N
∑w∈Γ
pw1wT uT
1 Write generalised voting games as max of generalisedunanimity games
2 Proof by induction: Divide N into N1 and N2, where in N1players are dummy in some winning coaition
3 Remove coalition, result is weighted voting game, withsame marginality for all other players.
4 Use Marginality and inductive assumption to determinevalue.
5 In N2 veto players get the same value by Symmetry, which,by Efficiency is the Shapley value.
Power Extending the model Axiomatisation Application Conclusion
Axiomatisation – Result
Theorem
On the class Γ̃ solution κ satisfies Efficiency, Symmetry andMarginality if and only if κ = φ.
Fact1: ṽ =∑
w∈Γ
∨T⊆N
pw1wT uT =∨
T⊆N
∑w∈Γ
pw1wT uT
1 Write generalised voting games as max of generalisedunanimity games
2 Proof by induction: Divide N into N1 and N2, where in N1players are dummy in some winning coaition
3 Remove coalition, result is weighted voting game, withsame marginality for all other players.
4 Use Marginality and inductive assumption to determinevalue.
5 In N2 veto players get the same value by Symmetry, which,by Efficiency is the Shapley value.
Power Extending the model Axiomatisation Application Conclusion
The National Assembly, Hungary
Simple model: all MPs are present with probability p, nocorrelation. The value of a coalition can be given:
v(S) =wS∑q
(wSi
)pi(1− p)wS−i
where wS =∑
i:Ni∈S wi .In 2009 p = 91.55%
2009 2006 2005 1994party seats seats seats seatsFidesz 139 141 168 20FKGP - - - 26KDNP 22 23 - 22MDF 9 11 9 38MSzP 189 190 177 209SzDSz 18 20 20 70Indep’t 6 1 11 -
Power Extending the model Axiomatisation Application Conclusion
National Assembly – Power of the minority
2009 2005 1994party seats S-S new seats S-S new seats S-S newFidesz 36.3 3.3 11.6 43.5 23.3 31.6 5.2 0 2.3FKGP - - - - - - 6.8 0 2.3KDNP 5.7 3.3 8.6 - - - 5.7 0 2.3MDF 2.4 3.3 2.8 2.3 6.7 4.8 9.9 0 2.3MSzP 49.3 83.3 69.5 45.9 40.0 51.0 54.3 100 88.4SzDSz 4.7 3.3 5.7 5.2 23.3 6.7 18.2 0 2.3Indep’t 1.6 3.3 1.9 2.8 6.7 5.8 - - -
Values closer to size.
Power Extending the model Axiomatisation Application Conclusion
National Assembly – RobustnessPlay with the value of p.
p = seats 1 0.95 0.9 0.85 0.8 0.75 0.7MSzP 49.3 83.3 75.4 66.4 58.6 53.0 50.2 50.0Fidesz 36.3 3.3 7.7 14.2 24.2 38.0 49.0 50.0KDNP 5.7 3.3 7.6 8.5 7.6 3.0 0.2 0.0SzDSz 4.7 3.3 6.9 5.4 6.1 2.9 0.2 0.0MDF 2.4 3.3 1.3 3.5 2.1 1.9 0.2 0.0indep’t 1.6 3.3 1.2 2.1 1.5 1.3 0.2 0.0
Power Extending the model Axiomatisation Application Conclusion
Conclusion
What we haveAbsenteeism leads to generalised voting gamesGood news! Parliamentary democracy 6= periodic electeddictatorship
if your parliament is small enough and yourMPs lack discipline.
The Shapley value is axiomatised for generalised votinggames.
Things to doAxiomatise Banzhaf index/measure (seems difficult)Examples from a smaller parliament (with lazy MPs)More accurate predictions with differences in partydisciplinePartisan voting
Power Extending the model Axiomatisation Application Conclusion
Conclusion
What we haveAbsenteeism leads to generalised voting gamesGood news! Parliamentary democracy 6= periodic electeddictatorship
if your parliament is small enough and yourMPs lack discipline.
The Shapley value is axiomatised for generalised votinggames.
Things to doAxiomatise Banzhaf index/measure (seems difficult)Examples from a smaller parliament (with lazy MPs)More accurate predictions with differences in partydisciplinePartisan voting
Power Extending the model Axiomatisation Application Conclusion
Conclusion
What we haveAbsenteeism leads to generalised voting gamesGood news! Parliamentary democracy 6= periodic electeddictatorship if your parliament is small enough
and yourMPs lack discipline.
The Shapley value is axiomatised for generalised votinggames.
Things to doAxiomatise Banzhaf index/measure (seems difficult)Examples from a smaller parliament (with lazy MPs)More accurate predictions with differences in partydisciplinePartisan voting
Power Extending the model Axiomatisation Application Conclusion
Conclusion
What we haveAbsenteeism leads to generalised voting gamesGood news! Parliamentary democracy 6= periodic electeddictatorship if your parliament is small enough and yourMPs lack discipline.The Shapley value is axiomatised for generalised votinggames.
Things to doAxiomatise Banzhaf index/measure (seems difficult)Examples from a smaller parliament (with lazy MPs)More accurate predictions with differences in partydisciplinePartisan voting
Power Extending the model Axiomatisation Application Conclusion
Conclusion
What we haveAbsenteeism leads to generalised voting gamesGood news! Parliamentary democracy 6= periodic electeddictatorship if your parliament is small enough and yourMPs lack discipline.The Shapley value is axiomatised for generalised votinggames.
Things to doAxiomatise Banzhaf index/measure (seems difficult)Examples from a smaller parliament (with lazy MPs)More accurate predictions with differences in partydisciplinePartisan voting
Power Extending the model Axiomatisation Application Conclusion
SPAM! RePEc and other things
RePEc=Repository of Papers in Economics – the ‘ArXiv foreconomics.’ repec.org 800k+ papersNEP=New Electronic Papers. Mailing lists: 1 email/weekwith 5-25 abstracts. No spams. nep.repec.org
NEP-CDM = NEP-Collective Decision MakingNEP-GTH = NEP-Game THeory
PowerPreliminaries
Extending the modelAbstentionAbsenteeism
AxiomatisationAxiomatisation
ApplicationNational Assembly, Hungary
ConclusionConclusion