Information Revelation and Pandering in Elections Navin Kartik Francesco Squintani

Pandering and Elections

Information Revelation and Pandering in Elections

Navin Kartik Francesco Squintani Katrin Tinn

April 2014

Kartik, Squintani, Tinn

Motivation

I A key question for representative democracy is whether citizens are sufficiently informed when they vote.

I There is much empirical evidence of voter ignorance (e.g. Campbell et al. 1960)

I But politicians have much better access to information than voters

I Competition provides incentives to gather and transmit information to voters

I There is evidence that voters learn or refine their beliefs/preferences during campaigns (Schuman and Presser 1981, Zaller 1992, Fishkin

1997)

Pandering and Elections Kartik, Squintani, Tinn

So, does electoral competition among office-motivated candidates succeed in making voters informed?

Motivation

I A key question for representative democracy is whether citizens are sufficiently informed when they vote.

I There is much empirical evidence of voter ignorance (e.g. Campbell et al. 1960)

I But politicians have much better access to information than voters

I Competition provides incentives to gather and transmit information to voters

I There is evidence that voters learn or refine their beliefs/preferences during campaigns (Schuman and Presser 1981, Zaller 1992, Fishkin

1997)

So, does electoral competition among office-motivated candidates succeed in making voters informed?


Motivation

I Two opposing informal views on this question

• Competition promotes information revelation and efficiency

The arguments made for the voter’s being uninformed implicitly assume that the major cost of information falls on the voter. However, there are returns to an informed political entrepreneur from providing the information to the voters, winning office , and gaining the direct and indirect rewards of holding office. [Donald Wittman, 1989]

• Competition induces pandering to the electorate’s prior , precluding information revelation and efficiency

I Which one is correct?


Preview of our Downsian Model

I Two office-motivated politicians simultaneously commit to platforms

I The (median) voter updates beliefs and votes for the platform closest to expected state


I Unlike previous work, a richer space of platforms and information

• Continuum policy space; baseline Normal-Normal model

I This allows possibility of both pandering to the prior and for overreaction to information

Preview of our Downsian Model

I Two office-motivated politicians simultaneously commit to platforms

I The (median) voter updates beliefs and votes for the platform closest to expected state

I Unlike previous work, a richer space of platforms and information

• Continuum policy space; baseline Normal-Normal model

I This allows possibility of both pandering to the prior and for overreaction to information


Preview of Results in our Downsian Model

Most general results (minimal assumptions on information structure):

1.

Elections cannot efficiently aggregate both politicians’ information

2.

In any equilibrium, voter welfare is effectively determined by only one politician’s information

•

• not necessarily using this information efficiently even though there are two information sources


Hence, we have found a sharp welfare bound

Note: This welfare bound applies even with cheap talk , for all plausible equilibria



1.


2.


•







1.


2.


•






For a class of information structures (Conjugate-Exponential family):

4.

Incentive to distort platforms by overreacting / anti-pandering

• due to a simple statistical property of Bayesian updating

5.

Optimal pandering would be good for electorate

• better than not only any equilibrium but also better than “unbiased” strategies


Overall, both initial views are incomplete/incorrect


For a class of information structures (Conjugate-Exponential family):

4.

Incentive to distort platforms by overreacting / anti-pandering

• due to a simple statistical property of Bayesian updating

5.

Optimal pandering would be good for electorate

• better than not only any equilibrium but also better than “unbiased” strategies

Overall, both initial views are incomplete/incorrect


The Commitment Assumption

I We are interested in whether candidates’ information is revealed in a canonical/realistic model of elections (Downsian model).


I We maintain the Downsian assumption of commitment to platforms.

For many unmodeled reasons, commitment is a realistic assumption.

I If we do not assume commitment, and electoral promises can be revised, there is an equilibrium with fully revealing communication strategies.

But this equilibrium is implausible.

A candidate can secure victory by not revealing her signal.

I Formally, the fully revealing equilibrium is not neologism-proof .

We prove that the only neologism-proof equilibrium strategies are completely uninformative































Message of the Paper

I The question of how informed voters are is fundamental to assess representative democracy.


I We prove that electoral competition by office-motivated candidates has poor efficiency properties.

I Principled, benevolent politicians may be needed to achieve informational efficiency.

I Independent and unbiased media are necessary to achieve informational efficiency.



















Plan

Introduction

Literature

Downsian Elections

Cheap Talk

Normative Analysis


Introduction

Literature

Downsian Elections

Cheap Talk

Normative Analysis


Literature

I Pandering to influence voters’ belief about policy-state match:

Heidues & Lagerlof (2003)

• binary policy / state / signal model

• related: Schultz (1996), Laslier & Van der Straten (2004),

Loertscher (2010), Gratton (2010)


I “Career concerns” about perceived ability or preferences

• Canes-Wrone et al (2001), Maskin & Tirole (2004), ....

• Anti-pandering in Prendergast & Stole (1996), Levy (2004)

I Group polarization

• Sobel (2006), Glazer and Sunstein (2009)

I Responsible Parties

• Bernhardt, Duggan and Squintani (2009)

Literature














Literature














Literature














Introduction

Literature

Downsian Elections

Cheap Talk

Normative Analysis


Baseline Downsian Model

I A (median) voter has preferences represented by

U ( y , θ ) = − ( y − θ )

2

, where y ∈

R is the policy and θ ∈

R is an unknown state

• can allow for a continuum of ideological voters

I Two purely office-motivated candidates, A and B : u i

= 1

{ i is elected }

I θ ∼ N (0 , 1 /α )

I Each candidate i receives a signal θ i

= θ + ε i

, where ε i

∼ N (0 , 1 /β )

• can allow for β

A

= β

B



Timing:

1.

Nature draws θ and then θ

A

, θ

B

(conditionally iid)

2.

Each candidate i privately observes θ i

3.

A and B simultaneously commit to platforms y

A and y

B

4.

Voter elects a candidate i , and i implements y i


Study Perfect Bayesian equilibria where

I Candidates use pure strategies, y i

( θ i

) (not essential)

I Voter’s strategy, p ( y

A

, y

B

) , randomizes 50-50 when indifferent


Timing:

1.

Nature draws θ and then θ

A

, θ

B

(conditionally iid)

2.

Each candidate i privately observes θ i

3.

A and B simultaneously commit to platforms y

A and y

B

4.

Voter elects a candidate i , and i implements y i

Study Perfect Bayesian equilibria where

I Candidates use pure strategies, y i

( θ i

) (not essential)

I Voter’s strategy, p ( y

A

, y

B

) , randomizes 50-50 when indifferent


Some Definitions

I A strategy y i

( θ i

)

• is informative if it is not constant

• is fully revealing if it is one-to-one

• is unbiased if y i

( θ i

) =

E

[ θ | θ i

]

• has pandering if it is informative and

1.

θ i

> 0 ⇒ y i

( θ ) ∈ [0 ,

E

[ θ | θ i

])

2.

θ i

< 0 ⇒ y i

( θ ) ∈ (

E

[ θ | θ i

] , 0]

• has overreaction if

1.

θ i

> 0 ⇒ y i

( θ ) >

E

[ θ | θ i

]

2.

θ i

< 0 ⇒ y i

( θ ) <

E

[ θ | θ i

]


I Because of latitude in off-path beliefs, any pair of constant strategies is an equilibrium

I Our main interest is in informative equilibria

Some Definitions

I A strategy y i

( θ i

)

• is informative if it is not constant

• is fully revealing if it is one-to-one

• is unbiased if y i

( θ i

) =

E

[ θ | θ i

]

• has pandering if it is informative and

1.

θ i

> 0 ⇒ y i

( θ ) ∈ [0 ,

E

[ θ | θ i

])

2.

θ i

< 0 ⇒ y i

( θ ) ∈ (

E

[ θ | θ i

] , 0]

• has overreaction if

1.

θ i

> 0 ⇒ y i

( θ ) >

E

[ θ | θ i

]

2.

θ i

< 0 ⇒ y i

( θ ) <

E

[ θ | θ i

]

I Because of latitude in off-path beliefs, any pair of constant strategies is an equilibrium

I Our main interest is in informative equilibria


Unbiased Strategies

Proposition.

There is no equilibrium in which both candidates use unbiased strategies.


Reason: would be optimal for voter to elect more extreme candidate, and hence a profitable deviation is to overreact.

I each candidate’s estimate is based on one signal

I the voter’s estimate is based on two signals (b/c full revelation)

I so the voter’s estimate puts less weight on the prior and hence is more extreme than the average of the candidates’ estimates

I hence the voter would elect the more extreme candidate

Unbiased Strategies

Proposition.








Unbiased Strategies

Proposition.








Deviation from Unbiased Strategies

Proof.

I Given y i

( θ i

) =

E

[ θ | θ i

], the mid-point of the platforms is m ( θ

A

, θ

B

) =

E

[ θ | θ

A

] +

E

[ θ | θ

B

]

2

=

β ( θ

A

+ θ

B

)

.

2 α + 2 β

I On the other hand,

E

[ θ | θ

A

, θ

B

] =

β ( θ

A

+ θ

B

)

α +2 β

.


I Same sign but latter has larger magnitude, hence if | θ i

| > | θ

− i

| then

| y i

( θ i

) −

E

[ θ | θ

A

, θ

B

] | < | y

− i

( θ

− i

) −

E

[ θ | θ

A

, θ

B

] | , the candidate with the more extreme platform wins.

I So, a profitable deviation is to overreact (not to pander).

Deviation from Unbiased Strategies

Proof.

I Given y i

( θ i

) =

E

[ θ | θ i

], the mid-point of the platforms is m ( θ

A

, θ

B

) =

E

[ θ | θ

A

] +

E

[ θ | θ

B

]

2

=

β ( θ

A

+ θ

B

)

.

2 α + 2 β

I On the other hand,

E

[ θ | θ

A

, θ

B

] =

β ( θ

A

+ θ

B

)

α +2 β

.

I Same sign but latter has larger magnitude, hence if | θ i

| > | θ

− i

| then

| y i

( θ i

) −

E

[ θ | θ

A

, θ

B

] | < | y

− i

( θ

− i

) −

E

[ θ | θ

A

, θ

B

] | , the candidate with the more extreme platform wins.

I So, a profitable deviation is to overreact (not to pander).


Linear Updating

Quadratic prefs ⇒ voter selects platform closest to expected state

Normal-Normal structure implies linear updating:

E

[ θ | θ

A

, θ

B

] =

α

2 β

+ 2 β

θ

A

+ θ

B

2


More generally, expectation is linear when distribution of θ i

| θ is in the exponential family and the prior is a conjugate prior

I includes continuous and discrete distributions: normal, exponential, gamma, beta, chi-squared, binomial, Dirichlet, Bernoulli, Poisson...

I most results extend to this class; main welfare result is even more general

Linear Updating

Quadratic prefs ⇒ voter selects platform closest to expected state

Normal-Normal structure implies linear updating:

E

[ θ | θ

A

, θ

B

] =

α

2 β

+ 2 β

θ

A

+ θ

B

2

More generally, expectation is linear when distribution of θ i

| θ is in the exponential family and the prior is a conjugate prior

I includes continuous and discrete distributions: normal, exponential, gamma, beta, chi-squared, binomial, Dirichlet, Bernoulli, Poisson...

I most results extend to this class; main welfare result is even more general


Symmetric FRE

Despite the incentive for overreaction, can information be revealed?


Proposition.

There is a symmetric fully revealing equilibrium. It has overreaction: y ( θ i

) =

E

[ θ | θ i

, θ

− i

= θ i

] =

2 β

α + 2 β

θ i

.

The voter chooses each candidate with prob.

1

2 for all platform pairs.

Symmetric FRE

Despite the incentive for overreaction, can information be revealed?

Proposition.

There is a symmetric fully revealing equilibrium. It has overreaction: y ( θ i

) =

E

[ θ | θ i

, θ

− i

= θ i

] =

2 β

α + 2 β

θ i

.

The voter chooses each candidate with prob.

1

2 for all platform pairs.


Symmetric FRE

Proof.

Only need to show that the voter’s is indifferent between the two candidates regardless of the platforms they propose.


Since y ( · ) is fully revealing and onto, each candidate i ’s platform choice y i is equivalent to reporting a signal θ i

.

Using mean-variance decomposition of utility, it suffices that for any θ

A and θ

B

,

( y ( θ

A

) −

E

[ θ | θ

A

, θ

B

])

2

= ( y ( θ

B

) −

E

[ θ | θ

A

, θ

B

])

2

, or

2 β

α + 2 β

θ

A

−

2 β

α + 2 β

θ

A

+ θ

B

2 which is true for any θ

A

, θ

B

.

2

=

2 β

α + 2 β

θ

B

2 β

−

α + 2 β

θ

A

+ θ

B

2

2

,

Symmetric FRE

Proof.



.



A and θ

B

,

( y ( θ

A

) −

E

[ θ | θ

A

, θ

B

])

2

= ( y ( θ

B

) −

E

[ θ | θ

A

, θ

B

])

2

, or

2 β

α + 2 β

θ

A

−

2 β

α + 2 β

θ

A

+ θ

B


A

, θ

B

.

2

=

2 β

α + 2 β

θ

B

2 β

−

α + 2 β

θ

A

+ θ

B

2

2

,

Symmetric FRE

Proof.



.


A and θ

B

,

( y ( θ

A

) −

E

[ θ | θ

A

, θ

B

])

2

= ( y ( θ

B

) −

E

[ θ | θ

A

, θ

B

])

2

, or

2 β

α + 2 β

θ

A

−

2 β

α + 2 β

θ

A

+ θ

B


A

, θ

B

.

2

=

2 β

α + 2 β

θ

B

2 β

−

α + 2 β

θ

A

+ θ

B

2

2

,


A Fundamental Welfare Result

While the previous eqm is fully revealing, there is policy distortion from overreaction

I This is because platforms are commitments

But perhaps this is the best eqm, in terms of ex-ante voter welfare?



I For any candidates’ strategy profile, σ , let v i

( σ ) be the voter’s welfare from electing candidate i no matter the platform pair.

I If σ is an equilibrium profile, voter’s welfare must be at least

V ( σ ) ≡ max { v

A

( σ ) , v

B

( σ ) } .

I Political competition is beneficial only if welfare is strictly larger than V ( σ ).


Theorem.

In any equilibrium with candidates’ strategies σ , voter welfare is V ( σ ) .

⇒ in any equilibrium, it is as if welfare only depends on one candidate’s policy function!

⇒ sharp upper bound on eqm voter welfare





V ( σ ) ≡ max { v

A

( σ ) , v

B

( σ ) } .


Theorem.









V ( σ ) ≡ max { v

A

( σ ) , v

B

( σ ) } .


Theorem.





The Ex-Post Lemma

Lemma

Fix an informative equilibrium ( y

A

( · ) , y

B

( · ) , p ( · )) and let µ be the induced ex-ante probability measure over platform pairs. There is p

∗ ∈ { 0 ,

1

2

, 1 } such that for µ -a.e.

( y

A

, y

B

) , p ( y

A

, y

B

) ≡ Pr( elect A | y

A

, y

B

) = p

∗

.


I This Lemma implies the previous Proposition because:

• Uninformative equilibria yield weakly lower welfare than equilibria in which one candidate always wins

• If p = 1 / 2, then voter welfare is the same as if either candidate were always elected

The Ex-Post Lemma

Lemma

Fix an informative equilibrium ( y

A

( · ) , y

B

( · ) , p ( · )) and let µ be the induced ex-ante probability measure over platform pairs. There is p

∗ ∈ { 0 ,

1

2

, 1 } such that for µ -a.e.

( y

A

, y

B

) , p ( y

A

, y

B

) ≡ Pr( elect A | y

A

, y

B

) = p

∗

.

I This Lemma implies the previous Proposition because:

• Uninformative equilibria yield weakly lower welfare than equilibria in which one candidate always wins

• If p = 1 / 2, then voter welfare is the same as if either candidate were always elected


Proof Sketch of Ex-post Lemma

I Fix any voter strategy p ( y

A

, y

B

)

I This induces a complete-information constant-sum game between candidates:

• S

A

= S

B

=

R and u

A

( s

A

, s

B

) = p ( s

A

, s

B

) while u

B

( s

A

, s

B

) = 1 − p ( s

A

, s

B

)

• Let v i

∗ be each player’s minmax value (if it exists)


I Any (Nash) equilibrium of the Bayesian game that candidates are actually playing is a correlated equilibrium of the above complete-information game

I There is a unique correlated equilibrium value



A

, y

B

)


• S

A

= S

B

=

R and u

A

( s

A

, s

B

) = p ( s

A

, s

B

) while u

B

( s

A

, s

B

) = 1 − p ( s

A

, s

B

)

• Let v i







A

, y

B

)


• S

A

= S

B

=

R and u

A

( s

A

, s

B

) = p ( s

A

, s

B

) while u

B

( s

A

, s

B

) = 1 − p ( s

A

, s

B

)

• Let v i





Proof Sketch of Ex-Post Lemma

I Consider any Bayesian Nash equilibrium of electoral competition

We have concluded that each candidate i ’s interim expected probability of winning Π platforms y i i

( y i

; θ i

) is constant in θ played in equilibrium i and for all

I We want to conclude that the ex-post probability of winning p ( y

A

, y

B

) is constant in y

A and y

B

I The conclusion is not immediate because of well known counterexamples such as the matching pennies game

I We shall exploit minimal statistical properties of the electoral game



I Fix an informative equilibrium of the election game (possibly mixed)

I WLOG suppose B is informative; for simplicity, assume only a finite number of on-path platforms for B , say ( y 1

B

, . . . , y m

B

), with m > 1


I By the previous results, for any generic on-path y

A and θ

A

, θ

0

A

, v

∗

A

= Π

A

( y

A

; θ

A

) = Π

A

( y

A

; θ

0

A

) .

I But this implies that for any m types of candidate A ,







Pr y

1

B

.

..

| θ

1

A

Pr y 1

B

| θ m

A

· · · Pr y m

B

.

..

| θ

1

A

· · · Pr ( y m

B

| θ m

A

)











 p y

A

.

..

, y

1

B p ( y

A

, y m

B

)









= 

 v

∗

A

.

..

v

∗

A







.

I Since rows of coefficient matrix change nonlinearly in solution for all generic ( θ 1

A

, . . . , θ m

A

) is a constant p ( y

A

θ j

A

, the only

, · ).




B

, . . . , y m

B

), with m > 1


A and θ

A

, θ

0

A

, v

∗

A

= Π

A

( y

A

; θ

A

) = Π

A

( y

A

; θ

0

A

) .









Pr y

1

B

.

..

| θ

1

A

Pr y 1

B

| θ m

A

· · · Pr y m

B

.

..

| θ

1

A

· · · Pr ( y m

B

| θ m

A

)











 p y

A

.

..

, y

1

B p ( y

A

, y m

B

)









= 

 v

∗

A

.

..

v

∗

A







.


A

, . . . , θ m

A


A

θ j

A

, the only

, · ).




B

, . . . , y m

B

), with m > 1


A and θ

A

, θ

0

A

, v

∗

A

= Π

A

( y

A

; θ

A

) = Π

A

( y

A

; θ

0

A

) .








Pr y

1

B

.

..

| θ

1

A

Pr y 1

B

| θ m

A

· · · Pr y m

B

.

..

| θ

1

A

· · · Pr ( y m

B

| θ m

A

)











 p y

A

.

..

, y

1

B p ( y

A

, y m

B

)









= 

 v

∗

A

.

..

v

∗

A







.



A

, . . . , θ m

A


A

θ j

A

, the only

, · ).




B

, . . . , y m

B

), with m > 1


A and θ

A

, θ

0

A

, v

∗

A

= Π

A

( y

A

; θ

A

) = Π

A

( y

A

; θ

0

A

) .








Pr y

1

B

.

..

| θ

1

A

Pr y 1

B

| θ m

A

· · · Pr y m

B

.

..

| θ

1

A

· · · Pr ( y m

B

| θ m

A

)











 p y

A

.

..

, y

1

B p ( y

A

, y m

B

)









= 

 v

∗

A

.

..

v

∗

A







.


A

, . . . , θ m

A


A

θ j

A

, the only

, · ).


Introduction

Literature

Downsian Elections

Cheap Talk

Normative Analysis


Elections with Cheap Talk

I Suppose we allow candidates to revise their promises.


I We add an initial cheap talk stage to our Downsian game.

I There exists an equilibrium with perfectly revealing cheap talk.

Both candidates fully reveal their signals, and then commit to

E [ θ | θ

A

, θ

B

] .

The voter is indifferent among the two candidates.

I This equilibrium is implausible.

If a candidate deviates by not revealing her platform, then she will hold better information than the opponent, and secure victory.








E [ θ | θ

A

, θ

B

] .










E [ θ | θ

A

, θ

B

] .











E [ θ | θ

A

, θ

B

] .











E [ θ | θ

A

, θ

B

] .






Neologism-Proofness

I Consider an equilibrium of a game with cheap talk.


I A neologism n is a message not used in the equilibrium cheap talk strategy.

I Given a set of types Θ i

, the neologism n is credible for Θ i

, if

1. each type θ i in Θ i prefers to send the neologism equilibrium message m ( θ i

) , n than her

2. each type θ i not in Θ i prefers to send her equilibrium message m ( θ i

), than the neologism n ,

3. the opponents j = i hold the belief, upon receiving n , that θ i distributed according to their prior beliefs, conditional on Θ i

.

is

I An equilibrium is neologism proof, if it does not allow for any credible neologism.

I (We assume that the equilibrium winning probability in the second stage is ranked as the candidates’ information.)

Neologism-Proofness






, if


) , n than her




.

is



Neologism-Proofness





, if


) , n than her




.

is




Neologism-Proofness





, if


) , n than her




.

is




Neologism-Proofness





, if


) , n than her




.

is




Neologism-Proofness

I Full revelation in the cheap talk stage is not neologism proof.


I All types of either candidate i prefer to send the neologism

“silence”, than revealing his information.

I Upon receiving the neologism “silence”, the other players cannot update their priors.

I Hence, upon adopting the neologism, candidate i is better informed than candidate j , and wins the election.

I Further, all the neologism proof equilibria are completely uninformative.

I Again, the intuition is that, by revealing information in the initial stage, a candidate can only facilitate the opponent’s victory.

Neologism-Proofness









Neologism-Proofness









Introduction

Literature

Downsian Elections

Cheap Talk

Normative Analysis


Optimal Platform Strategies

I Normatively, what are the optimal strategegies in the Downsian game?

I Consider an auxiliary benevolent candidates game:

• Each candidate maximizes the voter’s welfare

• Team problem in sense of Marschak & Radner (1972)

• Pareto-dominant eqm ⇐⇒ solution to normative exercise


Proposition.

In the benevolent candidates game, unbiased strategies are not an equilibrium, because candidates would deviate to pandering.

I Recall: unbiased strategies ⇒ winner has more extreme signal

I This “winner’s curse” ⇒ benevolent moderation of platform when conditioning on winning, i.e. deviate to pandering .

Optimal Platform Strategies

I Normatively, what are the optimal strategegies in the Downsian game?

I Consider an auxiliary benevolent candidates game:

• Each candidate maximizes the voter’s welfare

• Team problem in sense of Marschak & Radner (1972)

• Pareto-dominant eqm ⇐⇒ solution to normative exercise

Proposition.

In the benevolent candidates game, unbiased strategies are not an equilibrium, because candidates would deviate to pandering.

I Recall: unbiased strategies ⇒ winner has more extreme signal

I This “winner’s curse” ⇒ benevolent moderation of platform when conditioning on winning, i.e. deviate to pandering .


Pandering Normatively Desirable

Proposition.

In the benevolent candidates game,

1.

There is a symmetric FRE with pandering: y ( θ i

) =

E

[ θ | θ i

, | θ

− i

| < | θ i

| ] , and voter elects the more extreme candidate.

2.

This is the Pareto-dominant equilibrium subject to i winning whenever | θ i

| > | θ

− i

| .

I We conjecture that caveat in part 2 can be dropped


Extensions

1.

More general information structures

• welfare bound results require minimal assumptions (some correlation)

• (anti-)pandering results apply within conjugate-exponential family


2.

Idelogical motivation

• u i

= − ρ i

( y − θ − b i

) 2 + (1 − ρ i

) 1

{ i wins }

• Theorem: for small ( ρ , b ), any welfare-maximizing equilibrium is almost non-competitive

Details

Extensions

1.

More general information structures

• welfare bound results require minimal assumptions (some correlation)

• (anti-)pandering results apply within conjugate-exponential family

2.

Idelogical motivation

• u i

= − ρ i

( y − θ − b i

) 2 + (1 − ρ i

) 1

{ i wins }

• Theorem: for small ( ρ , b ), any welfare-maximizing equilibrium is almost non-competitive

Details


Conclusion

I The question of how well elections aggregate politicians’ private information is fundamental to assess representative democracy.


I We prove that electoral competition by office-motivated candidates cannot aggregate both candidate’s information.

I Principled, benevolent politicians may be needed to achieve efficiency.

I Independent, unbiased media may be necessary to make voters informed.

I Efficient information revelation does not necessarily fail because of pandering, but also because of over-reaction.

Conclusion







Conclusion







Conclusion







Conclusion







Pandering and Elections

Thank you!

Kartik, Squintani, Tinn

Mixed-Motivated Candidates

Back to Extensions

I A mixed-motivations game is parameterized by ( ρ , b )

• u i

= − ρ i

( y − θ − b i

)

2

+ (1 − ρ i

) 1

{ i wins }

• pure office motivation when ( ρ, b ) = ( 0 , 0 )

I Let Σ

ε

( ρ, b ) be set of ε -competitive equilibria.

I Let U ε ( ρ , b ) := sup

σ ∈ Σ ε ( ρ , b )

ε -competitive equilibria.

U ( σ ) be highest voter welfare among


Theorem.

Assume candidates have mixed motivations. Then,

1.

As ( ρ , b ) → ( 0 , 0 ) , U 0 ( ρ, b ) → U 0 ( 0 , 0 ) .

2.

∀ ε > 0 , ∃ δ > 0 : k ( ρ, b ) − ( 0 , 0 ) k < δ ⇒ U

ε

( ρ, b ) < U

0

( ρ, b ) − δ .

⇒ when close to pure office motivation, any welfare-maximizing eqm must be almost non-competitive.

Mixed-Motivated Candidates

Back to Extensions

I A mixed-motivations game is parameterized by ( ρ , b )

• u i

= − ρ i

( y − θ − b i

)

2

+ (1 − ρ i

) 1

{ i wins }

• pure office motivation when ( ρ, b ) = ( 0 , 0 )

I Let Σ

ε

( ρ, b ) be set of ε -competitive equilibria.

I Let U ε ( ρ , b ) := sup

σ ∈ Σ ε ( ρ , b )

ε -competitive equilibria.

U ( σ ) be highest voter welfare among

Theorem.

Assume candidates have mixed motivations. Then,

1.

As ( ρ , b ) → ( 0 , 0 ) , U 0 ( ρ, b ) → U 0 ( 0 , 0 ) .

2.

∀ ε > 0 , ∃ δ > 0 : k ( ρ, b ) − ( 0 , 0 ) k < δ ⇒ U

ε

( ρ, b ) < U

0

( ρ, b ) − δ .

⇒ when close to pure office motivation, any welfare-maximizing eqm must be almost non-competitive.


Information Revelation and Pandering in Elections Navin Kartik Francesco Squintani

Thank you!

Related documents

Products

Support

Information Revelation and Pandering in Elections Navin Kartik Francesco Squintani

Thank you!

Related documents

Add this document to collection(s)

Add this document to saved

Suggest us how to improve StudyLib