1 Median voter & c.

1.1 Arrow Theorem.

How can I build a social ordering on a set of alternatives starting from individual
orderings on the same set?
(Ordering: complete and transitive preferences on outcomes). Pure aggre-
gation issue.
Extreme informational ass: individual preferences are ordinal and not com-
parable (i.e. some types of voting mechanisms (e.g. allowing no indications of
intensity of preferences).. but not only).
Let a; b A set of choices (at least three elements).
i = 1; :::n; individuals (nite, n _ 3):
Each i has a preference ordering on A (complete, transitive).
a %
i
b i U
i
(a) _ U
i
(b); where a ~
i
b if a %
i
b and b
i
a; a s b if a %
i
b
and b _
i
a
a %
i
b = aR
i
b
R = (R
1
; R
2
:::R
n
) prole of preferences.:
W = f(R) social order
Arrows conditions
1. f(R) must be an ordering (complete, transitive)
P: if aR
i
b all i; then aRb;
U: f(R) dened for any R;
I (independence of irrelevant alternatives) Let R; R
0
and suppose af(R)b;
and i : aR
i
b = i : aR
0
i
b : Then af(R
0
)b:
D f(R) is dictatorial if there exists an i such that aR
i
b implies af(R)b all
a; b A .
P,U,I implies welfarism (Sen, 1970). Social ordering depends only on infor-
mation on the utilility of individuals.
1
Theorem " If f(R) satises 1,P,U,I then f(R) is dictatorial"
Proof (Salanie, 2000 or Mas-Colell & c.)
1.Prove by U,I, 1 that if a group of agents is decisive on a couple of alterna-
tives is decisive upon all of them;
2. Prove by U,I, 1 that if a group is decisive it can always be split in
subgroups, until the decisive set is a singleton.
3. Note by P that the entire set of agents is decisive.
Illustration.
1.2 Voting and Condorcets Paradox
Let n = 1; 2; 3; and A = a; b; c : Suppose preferences are
1: a ~ c ~ b;
2: b ~ a ~ c;
3: c ~ b ~ a;
Assume:
A.1 Direct democracy; citizens directly make the policy choices.
A.2 Sincere voting: each citizen votes for the alternative that gives her the
highest utility.
A.3 Open Agenda: vote over pairs of alternatives and vote over all pairs of
alternatives until you nd a winner.
A Condorcet winner is an alternative that beats any other feasible alternative
in pairwise voting.
Remark: (Condorcets paradox) There is no Condorcet winner with the
previous preferences ( b beats a; c beats b, a beats c). Cycle, intransitive social
prefrences, no ordering. Agenda manipulation (closed agenda) advantageous,
incentives for strategic voting.
Given Arrow, something has to be given in order to make progress.
Voting mechanisms with intensity of preferences (e.g. Bordas count) solves
problems of building a "social ordering" (by denition) but more prone to strate-
gic voting.
Abandon U (restricting preferences).
2
1.3 Single-peaked preferences and beyond
Suppose a; b; c are scalar (unidimensional issue) and that there is some natural
metric on A (weight, size, lenght, time..), so that a > b > c: Then 1s preferences
"strange" (two peaks). Suppose 1: a ~ b ~ c (single-peak). Then, under A.1,
A.2,A.3 , Condorcet winner is b, the preferred alternative of "middle" voter.
More generally:
(PT notation)
Let W(q;
i
) be the political preferences of individual i as dened on (uni-
dimensional) policy q, where
i
captures dierences in preferences across indi-
viduals. Let q(
i
) = arg max W(q;
i
):
(Can be derived by primitives; let U(c
i
; p; q; ;
i
) be is economic prefer-
ences on private consumption, policies q; prices p; where p may depend on q:
Each i max U(c
i
; p; q; ;
i
) by choice of c
i
s.t. H((c
i
; p; q; ;
i
) _ 0: Assuming
U(:) quasi-concave and H(:) convex, single optimal choice c
i
= c(p(q); q;
i
),
substituting W(q;
i
) = U(c
i
; p; q; ;
i
) ).
Policy preferences are single-peaked if q
00
_ q
0
_ q(
i
) or q
00
_ q
0
_ q(
i
)
then W(q
00
;
i
) _ W(q
0
;
i
)
Theorem "Suppose policy preferences are single peaked. Then under A.1-
A.3 there is a Condorcet winner and it coincides with the preferred policy of
the median ranked voter"
Proof. Separation argument. Order individuals on the basis of q(
i
): Let
q
m
the preferred policy of median voter. Pitch q
m
against an alternative q
0
.
Suppose q
0
> q
m
. Under single peaked preferences, and A.2 , for q
m
vote
median voter and all voters to the left of median voters (plus voters whose bliss
point is closer to q
m
); at least 50%+1 of voters. Similarly for q
0
< q
m
. Hence,
q
m
beats all alternatives. By A.3 it is the winner.
Note: concavity of U(:) does not necessarily imply single-peakness, even with
unidimensional policy.
More general (sucient) conditions by restricting voters heterogeneity (rather
than preferences). Let
i
be unidimensional distributed on some support.
(Gans-Smart) Single-crossing property.
Political preferences satisfy single-crossing if, for q > q
0
and
0
i
> a
i
, or for
q < q
0
and
0
i
< a
i
;
W(q;
i
) _ W(q
0
;
i
) ==W(q;
0
i
) _ W(q
0
;
0
i
):
3
Theorem " If preferences of voters satisfy single-crossing a Condorcet winner
exists and coincides with the bliss point of the voter with median value of a
i
"
Proof. Separation argument. Let q(
m
) be the preferred policy by voter
with median value of a
i
;
m
: By SC, every voter with a
i
_
m
prefers q(
m
) to
any alternative q < q(
m
); thus in pairwise voting, q(
m
) beats any q < q(
m
):
By SC, every voter with a
i
_
m
prefers q(
m
) to any alternative q > q(
m
);
thus in pairwise voting, q(
m
) beats any q > q(
m
):
SC assumption on preferences and policies; SP only on preferences. Basically,
SC reduces heterogeneity of agents to single dimension & imposes monotonicity
from agents type to policy choices. Hence median voter.
SC more general condition than SP (SP implies SC but the converse is not
true).
Example 1:
Let n = 1; 2; 3; and A = a; b; c : Suppose preferences P are
1: a ~ b ~ c;
2: a ~ c ~ b;
3: c ~ b ~ a;
with natural ordering a > b > c: P does not satisfy sp; 2s pref, two peaks.
P satises SC: Order of individuals
=1,2,3. Observe = 2 : c ~ b == = 3 : c ~ b: = 2 : a ~ c &
a ~ b == = 1 : a ~ c & a ~ b: In open agenda pairwise voting, a beats b and
a beats c. a is the Condorcet winner and 2 is the median voter.
Example 2 (see Person & Tabellini, p 24).
1) w
i
= c
i
+ V (x
i
), preference V (:) concave, i = 1; ::n
2) c
i
_ (1 t)l
i
+ g; consumers budget
3) l
i
= 1 x
i

i
; = average,
m
median.
Max 1) subject to 2, 3):
4)l
i
= L(t) (
i
)
5) ng _ t
X
(L(t) (
i
)) == g _ tL(t) == W(t;
i
) = L(t) (1
t)(
i
)) + V (1 L(t) )
W(t;
i
) may not satisfy SP (depends on sign L
tt
(t)), but satises SC.
With multi-dimensional issues, SP or SC much harder to believe (classical
example, to be developed below, expenditure for public education with private
4
market supply too). Median voter generally fails, hence cyclying, hence more
strategic behaviour, hence agenda setting power. But some condition still en-
sures median voter eq. even with muldimensional setting.
Let voters preferences satisfy the intermediate property if their indirect
utility function can be written as:
W(q;
i
) = J(q) + K(
i
)H(q)
where K(
i
) monotonic in
i
; J(q); H(q), common among voters,q is now
a vector of policy choices.
Theorem If voters have intermediate preferences, a Condorcet winner exists
and it is given by q(
m
) .
Proof. See PT, p.26.
Very restrictive assumption however, very rarely used, in contrast with SP
or SC.
1.4 Side Issues
1) The above assumes sincere voting. Restrictive? No, with SP preferences.
Strategic voting: let v
i
(q
0
; q
00
) be the voting function of voter i between
q
0
; q
00
; v
i
(q
0
; q
00
) q
0
; q
00
:Let V : q
0
; q
00

n
q
0
; q
00
the voting function
that maps votes in policy outcomes (e.g. V
m
the majoritarian vote). Let
V (v
i
(q
0
; q
00
); v
i
(q
0
; q
00
)) be the policy outcome when i votes according to v
i
(q
0
; q
00
)
and all other voters according to v
i
(q
0
; q
00
): Voter i is strategic if
v
i
(q
0
; q
00
) = arg max W(V (e v
i
(q
0
; q
00
); v
i
(q
0
; q
00
);
i
); all e v
i
(q
0
; q
00
)
i.e. each voter chooses her voting strategy so as to max her utility given the
voting strategies of all other agents.
Theorem Suppose A.1 holds, voting rule is V
m
& preferences satisfy SP
over policy space.Then, sincere voting is a weakly dominant strategy for each
voter and there exists a unique weakly dominant equilibrium with q(
m
) as
Condorect winner.
weakly dominant strategy for i: a strategy that gives weakly higher payo
to i than any other is strategy, regardless the voting strategies of other voters.
Proof. Intuition. If voter is not pivotal, how she votes does not aect the
outcome; thus, voting for his preferred alternative is a best responce whatever
the others do. If voter is pivotal, then clearly voting for his preferred alternative
is the best responce. Hence voting sincerely is a weakly dominant strategy for
5
all voters; but if everybody votes sincerely, by SP and V
m
, the median voter is
the Condorcet winner.
2) Why the (simple) majoritarian rule?
Buchanan & Tullock (1962); the optimal decision rule must trade o internal
versus external cost of making decision.
But there is a kink at 50%; below 50% both a and not a could be approved
simultaneously leading to deadlock. Majoritarian rule is the smallest majority
that avoids deadlocks (but among decisive majoritarian rule is the one that max
external costs).
Mays theorem (1952). The Majoritarian rule is the only rule that satises
a number of normative requirements (i.e. anonimity, decisiveness, neutrality,
responsiveness). See Mueller (2007).
Of course, many more electoral rules are possible in theory and are applied
in practise (e.g. supermajority for costitutional amendments). See Cox, 1997.
We will come back to this.
1.5 Conclusive remarks
With multidimensional issues (would seem the normal case) generally electoral
cycles are pervasive. You would need a voter who is median in all dimensions,
that is very umplausible.
Mckelvey (1976) general theorem. With spatial prererences in multidimen-
sional issue, a sequence of pairwise votes connects any starting point with any
point in the Pareto set of agents.
Worse, you would expect lot of strategic behaviour.
Gibbard & Satterthwaite theorem "Assume at least three alternative, relax
informational assumption, keep U and allows agents to falsely state their prefer-
ences. Given U, an ecient social welfare function is implementable in thruthful
strategies i is dicatorial".
see Mas-Colell chapter 23 for an introduction to mechanism design and more
precise statement.
In voting theory, it means (loosely) that for unrestricted preferences there
is no hope to nd a "mechanism" (including majoritarian voting) that gives
individuals incentive to state thruthfully their preferences (voting sincerously).
Unless is a dictatorship, in which case a version of the theorem above applies.
In practice, this means lots of possible equilibria as a result of voting.
But if the above were a description of reality, we should expect to see a lot
of instability in voting results (cycling) and implemented policies. We instead
observe remarkable stability. Basic explanation in literature: most policy de-
cisions are taken in representative democracies and representative democracies
are very dierent from voting in committees. They have mechanisms to select
6
among equilibria (uncertainty in voting behaviour, structure induced equilib-
rium, legislative bargaining, agency model ect.). Furthermore, there may be
reasons to believe that in a representative democracy the single-dimensionality
of policy (i.e. left to right) is not so out to mark and that the assumption of
sincere voting is also not too unrealistic. More to follow.
7