Institute of Governmental

Studies
(University of California, Berkeley)
Year  Paper WP-

Political Motivations
Steven Callander
Northwestern University

This paper is posted at the eScholarship Repository, University of California.

http://repositories.cdlib.org/igs/WP2007-4
Copyright 2007
c by the author.
 Political Motivations∗
Steven Callander†
Northwestern University
October 24, 2006

Abstract
Are politicians motivated by policy outcomes or by the perks of office? The
answer to this question is central to understanding the behavior of office holders
and the policies they produce. Despite the question’s importance, however, the
existing literature is not well suited to provide an answer. To shed light on the
issue of political motivations I exploit a basic fact: to hold office, one must first
win election. By characterizing the types of candidates that succeed in elections,
I am able to predict the types that hold office and the policies that are produced.
Toward this end, I develop a simple model of two candidate electoral competition in
which candidates may be either office or policy motivated. In a second departure
from standard formulations, the model incorporates both campaign and post-
election behavior of candidates. In this environment I find that office motivated
candidates are favored in electoral competition, but that their advantage is limited
by the electoral mechanism itself, and policy motivated candidates win a significant
fraction of elections. More importantly, I show that the competitive interaction
among candidates of different motivations affects the incentives of all candidates
— both office and policy motivated — and that this competition affects policy
outcomes.

∗
First draft: September 2003. I thank Tim Feddersen, Walter Mebane, Sophie Bade, Dan Bernhardt,
Mattias Polborn, and several seminar audiences for helpful comments.
†
Managerial Economics and Decision Sciences, Kellogg School of Management, Northwestern Uni-
versity, Evanston, IL 60208; [email protected].
1 Introduction
What are the motivations of those who hold public office? Are they inspired by policy
outcomes, or do they merely seek the perquisites of office? These questions are of more
than philosophical interest. In modern indirect democracies the holders of public office
are awarded broad powers to determine both the direction and efficacy of government
policy. It is important to know, therefore, whether elected leaders view public office as
a means to an end or as an end in itself.
In this paper I attempt to shed light on the basic question of political motivations.
The method of investigation is simple and builds upon a basic fact: to hold public office,
one must first win election. By studying which types of candidates succeed in elections,
we may begin to understand which types of candidates hold office and the policies that
result. To formalize this approach, I develop a model of electoral competition with
heterogeneously motivated candidates and allow them to compete against each other. I
suppose that, in addition to ideological differences, candidates vary in how much they
value policy outcomes over the perks of office; candidates are, therefore, in the language
of the literature, either policy or office motivated. Each candidate’s type is drawn from
a known distribution and is private information.
In this setting I find that office motivated candidates are favored in elections (as
intuition would suggest), but to a lesser extent than previously thought (see Calvert
1985) as their advantage is constrained by the electoral mechanism itself. Indeed, I
find that policy motivated candidates win a significant fraction of elections. Moreover,
and more importantly, I show that the competitive interaction among candidates of
different motivations affect the incentives of all candidates, and that this competition
affects policy outcomes.
The issue of candidate motivations is not new and has been prominent in the study
of elections. Yet, despite this emphasis, the extant literature is not suited to answer
the fundamental questions of interest. Previous research has allowed for different mo-
tivations, with one stream of literature following Hotelling (1929) and assuming office
motivated candidates, while another stream assumes that candidates are policy moti-
vated, as proposed by Wittman (1977, 1983) and Calvert (1985). It is the case in both
literature streams, however, that within individual models candidates are assumed to
be homogeneous in their motivations.1 Thus, in these models the type of candidate
that wins office is exogenously imposed rather than, as is the case here, endogenously
determined through electoral competition.
This characteristic of the literature is critiqued disapprovingly by Persson and Tabellini
(2000) — “. . . it is not satisfactory to leap back and forth between such different assump-
tions” (p. 484) — and they list its resolution as an outstanding challenge of political
economics.2 The model introduced here begins to address the challenge of Persson and
1
A more detailed discussion of the related literature is delayed momentarily.
2
Persson and Tabellini (2000) employ different terminology to that used here, referring to policy
and office motivated candidates instead as partisan and opportunistic candidates, respectively.

1
Tabellini by constructing an integrated model of candidate motivations and showing
which types are likely to hold office. Moreover, I show that in this model the sum is
greater than the parts as behavior in the integrated model is not a convex combination
of the extreme cases (of all office or all policy motivated candidates), but rather different
altogether and instead a product of the competitive interactions among candidates of
different motivations.
Driving the model is a novel conception of policy development. The model is inspired
by two observations: that election campaigns matter but they are not all that matter
and policy is only finalized after the election. To capture this richness, I model policy
outcomes as determined by both the choice of a policy instrument and the quality of
implementation, where candidates can commit to policy positions before the election
but the quality of implementation depends on the efforts of election winners after they
have been installed in office.
In this richer environment campaign promises play dual roles: as commitments to
particular policy instruments and as signals about how a candidate will behave once in
office. This duality induces a signaling game and driving this signaling game is a simple
fact: policy motivated candidates care more about policy outcomes and, consequently,
exert more effort if elected to office.3 Thus, ceteris paribus, voters prefer to elect a policy
motivated candidate rather than one that is office motivated. This endogenous voter
preference creates an incentive for candidates to separate and imitate: policy motivated
candidates wish to separate from office motivated candidates and signal their type to
voters, and office motivated candidates seek to imitate policy motivated candidates and
hide their type from voters. This game of cat-and-mouse is iterated continuously by
strategic candidates such that equilibrium behavior differs significantly from that when
motivations are homogeneous.
The expanded view of policy formation that is developed requires an expanded view
of voter learning and behavior. It is no longer the case necessarily that voters support
the candidate closest to them in the ideological space as they must weigh ideology
against the quality of implementation. This balancing act is cognitively more demanding
than in standard models, yet it is closer to real voting behavior where “character”
(more formally, valence) issues are an important determinant of vote choice (Stokes
1963). In contrast to existing models (Ansolabehere and Snyder 2000; Groseclose 2001;
Aragones and Palfrey 2002, 2005) the valence advantage appears endogenously here,
demonstrating explicitly how valence can be generated by policy considerations.
This voting behavior is of interest in its own right and for what it implies for the
interpretation of Black’s (1958) median voter theorem. The theorem continues to hold in
this environment as it is still very much the preferences of the median voter that drives
outcomes. However, it is no longer the case that the most centrist candidate necessarily
wins election, which implies in turn that policy convergence need not obtain.4 In addition
3
One may here interpret effort broadly as that exerted over a lifetime to build up competence.
Corruption is another possible interpretation, with policy motivated candidates engaging in less.
4
This possibility arises in existing models of valence competition although it requires asymmetry

2
to exposing a subtlety in applying the median voter theorem in incomplete information
environments, this finding goes some way to providing an informational explanation as
to why a “race toward the center” is not always observed in politics.
The focus of this paper is on electoral competition, and I take as given the demo-
graphics of the candidate pool. This choice leaves unanswered the complementary ques-
tion of what motivates individuals to transform from citizens into candidates. Although
the answer to this question is largely empirical (and dependant upon non-political out-
side options), the analysis provided here is an essential step in understanding political
motivations. The equilibria of the game I study provide calculations that citizens must
make in deciding whether to enter politics, and demonstrates the political outcomes
that will result once these decisions have been made. In that sense, a contribution of
the paper is to show that policy motivated candidates can win elections, and therefore
have a reason to enter the political fray, dispelling the notion that “A good politician is
quite as unthinkable as an honest burglar” (H.L. Mencken; 1880-1956).

1.1 Related Literature

Intriguingly, the central premise of the current model (to allow office and policy moti-
vated candidates to compete against each other) is raised informally by Calvert (1985)
who notes “there is no reason to expect policy motivation to be “selected out” among
candidates.” (p. 73). In an environment of complete policy commitment (and, signif-
icantly, candidate homogeneity), Calvert concludes however, that “. . . candidate mo-
tivations don’t affect the conclusions of the spatial model.” (p. 73) This conclusion
contrasts with the one presented here, although the reason for such is easy to see. In
assuming full commitment, Calvert allows no role for post-election behavior, precluding
the signaling that drives the current model. My results show that Calvert’s conjec-
ture does not extend to the more realistic world in which candidate motivations are
heterogeneous and private information.
A growing literature in political economy allows candidates to differ on characteristics
other than ideology, although the focus of these models is not on candidate motivations
and electoral competition. An exception is Roemer (1999) who models heterogeneity
within political parties and explores how this affects the ability of parties to compete in
a model with policy commitment. A second exception is Aragones and Palfrey (2005)
who study a general model of candidate type, allowing for heterogeneity in valence,
ideology and motivations (and purify the mixed equilibrium of Aragones and Palfrey
(2002)). They assume, however, full commitment over policy at the electoral stage and
the type of the winning candidate is not consequential (and signaling concerns do not
arise).
The model is also related to the small literature on signaling models of two candidate
across candidates (Ansolabehere and Snyder 2000, Groseclose 2001, Aragones and Palfrey 2002). The
subtlety exposed here is that in incomplete information environments the same property holds with
even ex-ante symmetric candidates and without an exogenously imposed valence advantage.

3
elections that began with Banks (1990). These models, in various capacities, offer
explanations of policy divergence, although the focus of each differs from the current
paper. Banks (1990) develops a model of costly lying in elections, although he imposes
these costs exogenously.5 Schultz (1996) studies a model with a state of the world that
is known by both candidates but unobserved by voters, and shows that if candidate
preferences are sufficiently close to those of the median voter then this information can
be transmitted through campaign platforms. Prat (2002a, 2002b) incorporates lobby
groups into the model and studies the information transmission provided by campaign
contributions.6
Models of single elections with policy commitment were introduced by Hotelling
(1929) and have since then provided a popular modeling apparatus (Wittman 1983;
Calvert 1985; Schultz 1996; Ansolabehere and Snyder 2000; Aragones and Palfrey 2002;
etc.). The dual restrictions of a single election and commitment are interrelated and it
is often suggested that reputational concerns in repeated play provide a foundation for
policy commitment. However, to the extent that the credibility of campaign commit-
ments depends on verifiability, it seems reasonable to assume that commitment is not
possible on all dimensions of policy, and that effort levels are more difficult to verify than
ideological promises. The model offered here captures this distinction in its most simple
and extreme form, showing how all dimensions of policy can play a role in elections even
if the ability to commit is only partial.7
An advantage of this approach is that it integrates into a single model both pre- and
post-election behavior. This capability speaks to a second unresolved issue in political
economy of how and when election campaigns matter. As is the case for candidate
motivations, the literature on this question has proceeded along two parallel streams.
The literature following Hotelling (1929) assumes not only that elections are important
but that they are all-important as pre-election commitments fully determine policy
outcomes. The opposing literature, originating with Barro (1973) and Ferejohn (1986),
interprets literally the incompleteness of the electoral contract and treats all campaign
statements as cheap talk, focusing instead on the behavior of agents in office. This state
of affairs (perhaps not surprisingly) is also critiqued by Person and Tabellini (2000):

“It is thus somewhat schizophrenic to study either extreme: where promises

have no meaning or where they are all that matter. To bridge the two models
is an important challenge.” (page 483)
5
Banks (1990) also shares the trait (and is one of the few papers to do so) that the link between
campaign promises and policy outcomes is partially relaxed; see also Bernhardt and Ingberman (1985).
6
In work done after this paper was circulated, Kartik and McAfee (2006) develop some related
results. Candidates in their model are not concerned with policy, yet one type is assumed exogenously
to possess a valence advantage (what they label ‘character’). To induce a signaling game, they assume
that the advantaged candidate is less flexible in policy offerings and, in fact, is entirely nonstrategic. In
contrast, candidates in my model are fully strategic and differ only in their motivation. Moreover, the
voter preference for one type of candidate emerges endogenously rather than being imposed exogenously.
7
This formulation also captures the reality that politicians do not choose outcomes, but rather they
choose policies which must be implemented.

4
 The model introduced here offers one way to bridge this gap. By assuming partial
commitment I allow for meaningful election campaigns while respecting the importance
of post-election policy making. More importantly, this approach shows how promises
can mean more than they appear on the surface, connecting issues of commitment to
other policy dimensions. More generally, this connection only arises when motivations
are heterogeneous, showing how the dual challenges of Persson and Tabellini are in fact
closely connected.

2 The Model
There are n citizens, where n is odd, and two candidates, denoted by A and B. A single
election is to be held and determined by majority rule. The timing of the election game
is depicted in Table 1.
The policy space is the real line, R. Prior to the election each candidate must
commit to a policy position, where xj ∈ R is the policy committed to by candidate
j. The candidate who wins the election, denoted w, receives a benefit of W and also
chooses a level of effort, ew ∈ [0, 1], for w = A, B. Effort is costly, with cost given
by γ (ew )2 . The effort exerted by the office-holder may be interpreted as determining
the quality of implementation of a policy choice, where the quality is assumed to equal
(1 − e). The policy outcome, then, is composed of both a policy instrument and the
quality of implementation, and the outcome space is P = R× [0, 1]. A key feature of
the model is that voters observe only x = (xA , xB ) prior to casting their votes.
Preferences are as follows. All actors — voters and candidates — derive utility from the
policy outcome (xw , ew ), the policy chosen and implemented by the winning candidate.
Voters and candidates prefer more effort to less (effort is a public good), although they
differ on ideological preferences. The ideal points for all actors, therefore, are on the
x-axis. Denote voter ideal points by pi for each voter i, with the median of citizens’ pi
values at pm = 0. The candidates are located symmetrically about the median voter
with ideal points −α and α for candidates A and B, respectively, where α ≥ 0.
Voters and candidates also differ on the utility weighting assigned to policy. Each
actor is endowed by nature with a type t, where t ∈ {h, l}, and the prior probability
of being type l is q ∈ [0, 1]. It is assumed that type is private information and that
β h > β l ≥ 0. Voter utility is as follows:
£ ¤
Ui (xw , ew |t) = −β t (pi − xw )2 + (1 − ew )2 ,

where i ∈ {1, 2, ..., n}. Thus, voter preferences do not depend on the identity or type
of the winning candidate, other than through the policy outcome. Also note that the
separability of utility across dimensions aids in the exposition but is not necessary for
the results.8
8
In some instances conservatives may prefer incompetent liberal office holders (and vice-versa). Even
if this is true (and it is not on issues such as defense and corruption), it cannot reverse the preferences

5
 Stage 1: Nature draws motivations for candidates and voters.

Stage 2: Candidates commit to policy positions.

Stage 3: Voters observe policy positions and cast votes.

Stage 4: Winner decided; the incumbent implements policy commitment

and exerts effort.

Stage 5: Policy outcome is determined and payoffs received.

Table 1: Sequence of Play

Candidates have identical preferences to voters, although they may additionally be

elected to office, receiving the perks of office and paying for effort to implement their
chosen policy. The utility of candidate A is given by:
£ ¤ £ ¤
UA (xw , ew |t) = −β t (−α − xw )2 + (1 − ew )2 + ΓA (w) W − γ (ew )2 ,

where ½
1 if w = A
ΓA (w) = .
0 otherwise
An actor’s type affects the disutility felt from different policy outcomes. For voters
this is inconsequential to behavior (although, if voting were costly, type may influence
the turnout decision). For candidates type determines how much they value policy
outcomes over the perks of office. For obvious reasons, therefore, I refer to candidates
of type h as “policy motivated” and those of type l as “office motivated.” Note that
the difference between policy and office motivated candidates is only one of degree (as
the traditional restriction for office motivated candidates of β l = 0 is permitted but not
imposed).9
A candidate’s type determines the effort exerted should he be elected to office. Vot-
ers do not observe type and have only the information contained in a policy commitment
(and the candidate’s identity) to draw inferences about type. I make the standard as-
sumption that all voters draw identical inferences (as they possess the same information);
of any voter. Instead the preferences of partisan voters would be strengthened, and electoral outcomes
would still hinge on the median voter, whose preference is more likely to be symmetric across candidates.
9
One may generalize this framework and suppose that voters and candidates assign different weights
to the different components of policy outcomes, and that these values are imperfectly correlated (that
is, a voter’s concern for policy is only loosely correlated with that for effort). The assumption of equal
weights is without loss of generality and adjustments to the normalization can produce any weighting
desired. On the other hand, assuming that the weights are correlated is necessary for the results,
although perfect correlation is not. It is sufficient to assume that the weights are positively correlated
to some degree so that voters can draw inferences about expected effort from the policy commitments.

6
denote by πj (xj ) the posterior belief of voters that candidate j is of type l given policy
commitment xj , and set π (x) = {πA , π B } for x = {xA , xB }. I assume voters do not use
weakly dominated strategies, and therefore vote for the candidate that maximizes the
expectation of utility (and voting is costless).10
A significant — and deliberate — feature of the model is that a candidate’s motivation
is the only information that is privately held (in particular, therefore, voters know
candidate ideal points). This focuses attention on political motivations, ensuring that
the strategic obfuscations of candidates are driven by their motivations and not their
ideology, a topic that has been studied elsewhere (Banks 1990; Schultz 1996). In an
appendix I relax this assumption and show that the model and the results extend easily
in this direction.
The above restrictions imply that voter pm (the voter with the median ideal point)
is also pivotal in determining majority preference. A strict preference held by m for
one candidate implies that a strict majority of voters hold the same preference. If,
on the other hand, m is indifferent then all other voters are either split equally over
the two candidates or are all indifferent also. To avoid unnecessary complication, I
assume that an indifferent voter votes for candidate A if her ideal point is to the left of
m’s (i.e., pi < 0) and for candidate B if to the right (i.e., pi > 0). Focus can then be
restricted to the preference of voter m. For the pair of policy commitments, x, denote by
r (x) the probability that the median voter votes for candidate A. If the median voter
is indifferent, assume that r (x) ∈ (0, 1) such that both candidates win with positive
probability.
The election environment just described is formally a signaling game, although with
several non-standard attributes. The election game has dual senders (the candidates)
and multiple receivers (the voters), and one of the senders (the winner) takes an action
after the receivers. Moreover, the receivers’ action is collective and limited to a binary
choice of one of the two senders. Although different to most signaling models, these
features are necessary ingredients of a meaningful model of elections. Two more signifi-
cant features that are different from other models (including Banks 1990) are that the
voters intrinsically care about the signal that is sent by the candidates (the policy com-
mitment) and that the candidates care about what happens should they lose (i.e., the
policy to be implemented by their opponent). In this sense, candidates here can truly
be said to be policy motivated. This dependence creates the possibility for free-riding
by candidates who may prefer to lose the election and have the other³candidate ´ exert
2
costly effort; to avoid this possibility I impose the constraint11 W ≥ γ β β+γ h
.
h
An equilibrium satisfies the above conditions and involves a set of (possibly mixed)
strategies for candidates and voters that optimize utility at every history based on their
types and the beliefs π (.), where beliefs are derived by Bayes’ rule when possible (a
10
To ensure meaningful voting it is required that β l > 0.
11
This constraint is sufficient but not necessary for the results to follow. It ensures that the di-
rect benefit of winning office exceeds the cost of effort for candidates of all types and for all policy
commitments.

7
Perfect Bayesian Equilibrium). For analytic simplicity I restrict attention to equilibria
such that A locates to the left of B.12 To avoid technicalities, I also consider only
equilibria in which both candidates win election with positive probability.
The set of signaling equilibria is large and to provide sharp insight a refinement is
required. For the following results to hold it is required only that, according to the
standard notion, belief following a deviation be shifted toward the type that is “most
likely” to deviate. Unfortunately, however, the simplest refinements that capture this
notion have little technical bite in the policy environment (e.g., the intuitive criterion,
divinity, D1; Cho and Kreps 1987, Banks and Sobel 1987). Elections differ from standard
signaling games both in contextual details and what these details imply for signaling
costs. In standard games the type most preferred by the receiver can send signals at
lower cost. Here the opposite is true: all candidates prefer to commit to their ideal point
and it is “cheaper” for office motivated candidates to move away from this ideal point.
A further difficulty is that the cost of the signals themselves (the policy commitments)
depend on the reaction of voters and so beliefs can always be found to render any
deviation profitable.
One strategy here would be to adjust the definition of divinity or D1 to suit the
electoral environment. As this would require excessive technical detail, I instead follow
Banks (1990) and apply the stronger requirement of universal divinity, due to Banks
and Sobel (1987). This requirement is far stronger than is necessary for the results
presented here, and I point out the excess at the appropriate times (see, in particular,
the discussion surrounding Lemma 5). A technical statement of universal divinity is
provided in the appendix.
Before proceeding, some language on strategies is in order. A strategy for a candi-
date j is said to be pooling if j commits to the same policy position whether office or
policy motivated (or mixes over the same distribution). Similarly, a strategy is said to be
separating if voters precisely learn the candidate’s type for all actions, and a strategy is
semi-separating if it is neither pooling nor separating (alternatively this may be thought
of as semi-pooling). Equilibria are referred to as pooling, separating, or semi-separating
if both candidates A and B use those strategies. Although asymmetric strategies are
permitted, all equilibria are symmetric and so the type of an equilibrium will follow
immediately from the equilibrium strategy of any one candidate. Note that these defin-
itions apply to individual candidates and how their behavior depends on their type; in
particular, whether candidate A is pooling is independent of the choices of candidate B.
If both candidates pool and locate at the same policy position (e.g., the median voter’s
ideal point) then strategies are said to be convergent, otherwise they are divergent.
12
This applies the natural restriction that in equilibrium candidates do not locate on the far side of
the median voter from their ideal point, formalizing the notion that partisans implement policies more
favorable to themselves than does the opposition.

8
3 Results
3.1 A Simple Property of Equilibrium
The central problem for voters is to infer a candidate’s type so as to predict his level of
effort should he be installed in office. The first part of this process may be confounded by
strategic behavior. The second part — the connection between type and effort level — is,
however, clear and described in Lemma 1. The additive separability of preferences (and
subgame perfection) implies that the equilibrium effort level for each type of candidate
is independent of his policy commitment. This allows for the following simplification of
equilibrium behavior.13
βt
Lemma 1 In equilibrium the effort level of the elected candidate is β t +γ
for all t.

The lemma establishes that the equilibrium effort level is increasing in the strength
of policy preference. Simply because they intrinsically care more about policy outcomes,
policy motivated candidates implement policies more effectively than do office motivated
candidates. Critically, this induces in voters, ceteris paribus, an endogenous preference
for policy motivated candidates.
Although simple, the property in Lemma 1 drives the electoral signaling game and
is a central determinant of policy outcomes. This property is important as it provides
the connection between post- and pre-election behavior of the candidates. Important
to note is that this voter preference derives solely from candidate motivations and does
not require the assumption of additional heterogeneity.

3.2 Benchmark Results

To show most clearly how the dual modeling innovations of the model interact, I consider
here the two benchmark cases in which only one of the innovations is included. In both
benchmark cases, full policy convergence is the unique outcome.
Homogeneous candidate motivations. Candidate homogeneity implies, by Lemma 1,
that every candidate will behave identically if elected to office. As such, the voters’
inference problem becomes inconsequential as although voters care about the imple-
mentation of policy, the quality is now independent of who wins the election (and thus
the signaling game has no bite). The only dimension of electoral competition then is pol-
icy, and standard results for this environment guarantee that platforms must converge
in equilibrium.14
13
The statement of results within the body of the paper are in verbal form; the appendix contains
formal statements of results as well as all proofs.
14
Technically, multiple equilibria exist as convergence can be supported by multiple voter reaction
functions r. Throughout the paper, claims of uniqueness are with respect to the policy outcomes that
are possible.

9
Lemma 2 If q ∈ {0, 1} a unique equilibrium exists in which both candidates A and B
locate at the median voter’s ideal point.

Convergence occurs in equilibrium regardless of whether candidates are office or

policy motivated. Thus, despite different motivations (and disutility from converging
on policy), electoral competition among homogeneous candidates induces policy choices
that are independent of those motivations.
A constant quality of policy implementation. Extending the model to include hetero-
geneity in motivations does not weaken the incentive to converge if the quality of policy
implementation is fixed. Lemma 3 shows that again complete convergence of policy
must arise in equilibrium. Heterogeneity in motivations is, therefore, not sufficient in
isolation to affect equilibrium behavior.

Lemma 3 If P = R then, for all q, there exists a unique equilibrium in which both
candidates A and B locate at the median voter’s ideal point, regardless of type.

With the implementation of policy at a fixed level, the winning candidate’s type does
not influence the policy outcome and competition again is only over policy instruments;
convergence must then result. Clearly, this logic does not depend on fixing the effort
level to 1 (such that P = R), and holds for any constant effort level.15

3.3 Heterogeneous Motivations: Policy Divergence

The inclusion of heterogeneity in candidate motivations and the quality of implemen-
tation change the nature of electoral competition by inducing a signaling game at the
election stage. The following lemma reflects basic properties of signaling games. These
properties hold in any equilibrium and do not require the universal divinity refinement
of beliefs.

Lemma 4 In every equilibrium:

(i) Office motivated candidates win with weakly higher probability.
(ii) Policy motivated candidates locate weakly closer to their ideal point.

The monotonicity results of Lemma 4 reflect the fact that movement towards the
median voter is at greater cost to policy motivated candidates than to office motivated
candidates. As office motivated candidates are less concerned about policy, they are
more prepared to trade policy outcomes for an increased chance of winning office.
The simple intuition of Lemma 4, when combined with the requirements of universal
divinity, leads to an important feature of voter beliefs. Because policy motivated candi-
dates are less willing to trade policy for the chance of winning office, it follows that they
15
Calvert (1985) and Wittman (1983) establish divergence results in the single dimension framework
of Lemma 3 (with all policy motivated candidates; q = 0) by assuming candidates are uncertain about
the location of the median voter’s ideal point.

10
benefit more from moving towards their ideal point (ceteris paribus). Consequently,
upon observing a deviation from a pooling strategy, voters believe the deviator to be
policy motivated if the deviation is towards the ideal point. This leads to the following
powerful result.

Lemma 5 For q ∈ (0, 1) the following holds:

(i) a pooling equilibrium at the median voter’s ideal point does not exist if α > 0.
(ii) if a pooling equilibrium exists then candidate A locates at −α and candidate B locates
at α.

Driving Lemma 5 is that a candidate’s motivation provides a critical connection

between policy commitments and effort levels, the two components of policy outcomes.
This connection allows voters to infer valuable information about outcomes from policy
commitments, in addition to the commitment itself.
To gain intuition for the lemma, suppose both candidates pool at the median voter
and α > 0. In this case the median voter receives her most preferred policy position
irrespective of who wins, but the quality of implementation is uncertain as she receives
no information about candidate type. If candidate B deviates towards his ideal point,
inducing voters to believe he is policy motivated, the median voter faces a choice between
different alternatives: her ideal policy with uncertain quality from candidate A, or a less
attractive policy position from candidate B but with an expectation of high effort (a
more effective implementation of the policy). As this logic applies for deviations of any
size (as long as the deviation is towards the ideal point), the median voter prefers the
divergent candidate if the deviation is sufficiently small. Consequently, the benchmark
pooling equilibrium breaks down.
More generally, this logic (and lemma) shows that the standard intuition about the
median voter theorem doesn’t extend to incomplete information environments. Driving
the lemma is the fact that the median voter need not necessarily support the closest
candidate in the policy space. Consequently, although the theorem still holds as it is
the preferences of the median voter driving outcomes, the power of the median voter
does not necessarily induce convergence.
The lemma also shows clearly the role played by universal divinity. For the lemma
to go through it is required that following a deviation beliefs shift in favor of policy
motivated candidates enough to ensure the median voter prefers the deviator. This can
hold for small changes in beliefs and is possible even if beliefs vary continuously (in the
deviation). The strength of universal divinity exceeds these requirements as it shifts all
belief on to the possibility that the candidate is policy motivated.
The intuition provided above does not depend on the candidates pooling at the
median and generalizes to any pooling strategy that isn’t at a candidate’s ideal point
(as a small deviation towards the ideal point must be available). When the candidates
pool at their ideal points such a deviation is not possible, and an equilibrium may be
supported (part ii of the lemma). By the symmetry of the model, therefore, pooling
equilibria must be symmetric.

11
 Of most interest here is that Lemma 5 immediately implies, for any degree of candi-
date heterogeneity, that the benchmark results for electoral competition no longer apply
and policy divergence must occur. Moreover, equilibrium behavior in the general model
is not a convex combination of the benchmark cases (which both predict policy conver-
gence in this environment), but rather behavior is different altogether. If a candidate
is pooling across types then there must be significant divergence and behavior unwinds
all the way to the ideal point. The only other alternative is that candidates don’t pool,
instead separating in some capacity which leads to signaling behavior and voter learning
(and divergence), effects that also don’t arise in the benchmark model.
The breakdown of the benchmark results not only highlights the effect of candidate
heterogeneity in a general model, but goes some way to explaining empirical phenomena
as policy divergence is a commonly observed regularity in a variety of political systems
(Budge, Robertsen, and Hearl 1987). In the following section I characterize precisely
the nature of policy divergence and voter learning in equilibrium.

3.4 Equilibrium Behavior

The preceding discussion of pooling strategies considered only deviations by candidates
toward their ideal points. For deviations away from ideal points the logic of Lemma 5 is
reversed: it is more costly for policy motivated candidates to deviate and voters believe
the deviator to be office motivated (expecting a low level of effort).
Suppose then that candidates are each pooling at their ideal point and one candi-
date, say candidate A, deviates towards the median voter. If the deviation is small then
candidate A is worse off as voters will favor candidate B (who is still playing the equi-
librium strategy). But what if the deviation is large? The loss from revealing his type
is the same as for a small deviation, but the attractiveness to the median voter of the
policy offered is increasing in the amount of convergence. For large enough deviations,
therefore, the favor of the median voter can be secured and the election won. This logic
requires that a sufficiently large deviation “towards” the median voter is possible, which
in turn requires that the candidate ideal points (and the potential pooling equilibrium)
are sufficiently extreme. The definition of α1 provides the critical value.

Definition 1 α1 satisfies:
µ ¶2 µ ¶2
βl α21 βh
1− = + 1− .
βl + γ (1 − q) βh + γ

The value α1 defines the policy position such that the median voter is indifferent
over a policy of α1 with beliefs equal to her priors, and a policy of 0 from an office
motivated candidate; that is, one policy is at the median voter’s ideal point with the
expectation of low effort, and the second is a policy position at a candidate’s ideal point
with expectations equal to prior beliefs. Figure 1 shows this graphically via the median
voter’s indifference curve. The following result then obtains.

12
 6
Effort ³ ´
βl
1− β l +γ
³ ´ ³ ´
βl βh
q 1− β l +γ + (1 − q) 1 − β h +γ

³ ´
βh
1− β h +γ

-
−α2 −α1 0 α1 α2
Policy

Figure 1: Median Voter’s Indifference Curve

Theorem 1 Suppose q ∈ (0, 1). For all α ∈ [0, α1 ] the unique equilibrium is pooling:
candidate A locates at −α and candidate B locates at α, irrespective of their types. A
pooling equilibrium does not exist if α > α1 .
The breakdown of pooling equilibria for α > α1 follows from the fact that office
motivated types willingly separate from policy motivated types to win the support of
voters. This behavior is unusual as a general property of signaling models is that, if
types are revealed, then the type less preferred by the receiver (in this case an office
motivated type) receives a less desirable reaction. This general property does not hold
here as the voters intrinsically care about the signal that is sent, and this induces office
motivated candidates, under certain circumstances, to willingly separate from their more
preferred counterparts.
At the critical threshold of α1 the nature of electoral competition changes. Rather
than candidates pooling in their policy announcements to hide their types from voters,
opportunistic candidates now willingly reveal their types and pander to the preferences
of the median voter. As a consequence of these actions, voters learn valuable informa-
tion about expected future performance from campaign platforms. The remaining two
results of this section characterize equilibrium behavior in the domain where pooling
equilibria don’t exist, describing the amount of separation by candidates and the de-
gree of information conveyed to voters. Toward this end, a second critical value of α is
important.
Definition 2 α2 satisfies:
µ ¶2 µ ¶2
βl βh
1− = α22 + 1− .
βl + γ βh + γ
The value α2 defines the policy position such that the median voter is indifferent
over a policy of α2 from a policy motivated candidate, and a policy of 0 from an office
motivated candidate; that is, a policy at the median voter’s ideal point with the expec-
tation of low effort, and a second policy at α2 with the expectation of high effort. The
value of α2 depends on the difference β h − β l (and γ), and is independent of q.

13
 An obvious property of all equilibria is that candidates do not choose policy posi-
tions more extreme than their ideal point (as moving towards their ideal point induces
voters to believe they are policy motivated and is more attractive to themselves and the
median voter). An immediate implication of this property for α < α2 is that neither
candidate can play a separating strategy: if types separate completely, then the median
voter must strictly prefer the policy motivated type, thereby violating Lemma 4. Thus,
for candidate preferences in the domain (α1 , α2 ) candidates must be semi-separating.
Theorem 2 describes the equilibrium.

Theorem 2 Suppose q ∈ (0, 1). For all α ∈ (α1 , α2 ) the unique equilibrium is semi-
separating: policy motivated candidates locate at their ideal point and office motivated
candidates mix over their ideal point and the ideal point of the median voter. The median
voter is indifferent over all equilibrium announcements.

A semi-separating strategy requires one type of candidate to play a mixed strategy

and for there to be some imitation. By the same logic for pooling equilibria (Lemma 5),
if the semi-pooling is anywhere other than at the candidate’s ideal point then a small
deviation towards the ideal point is profitable. Thus, it must be the office motivated type
that mixes and the semi-pool must be at the ideal point. When an office motivated type
doesn’t pool at the ideal point, his type is fully revealed to the voters. His reputation,
therefore, cannot be damaged further and so he aggressively competes for voter favor
on the remaining policy dimension. His incentive to converge is unencumbered and
unabated (as in the benchmark results) and he mixes over his ideal point and zero. This
logic applies to both candidates equally and the equilibrium is symmetric.
What is perhaps less obvious is that the office motivated candidates must mix in a
precise ratio to leave the median voter indifferent over all possible announcements. This
property is required in equilibrium to ensure in turn that office motivated candidates
are indifferent over their possible actions and prepared to mix. To see why this is
necessary, consider the incentives of candidate B if he is policy motivated and if voters
strictly prefer a policy commitment of zero over a commitment to α. In this situation,
candidate B loses whenever candidate A commits to zero and ties if A commits to −α,
in which case he wins a fraction of the time and the expected policy outcome is zero
but with variance. By deviating to an announcement of zero himself, candidate B then
strictly improves the policy outcome (as he is risk averse) and wins the election with
higher probability, and so is better off. If, on the other hand, the median voter prefers
an announcement at α then office-motivated candidates would not be prepared to mix.
In equilibrium, therefore, the median voter must be indifferent.
The median voter is free to mix over all policy commitments offered in equilibrium
(as she is indifferent). She does not mix equally if β l > 0. If she did so then an
office motivated candidate would strictly prefer to announce his ideal point rather than
zero as it offers a more attractive policy outcome for the same probability of victory.
It is only if β l = 0, and a candidate is unconcerned with policy, that in equilibrium
a non-centrist announcement wins with the same frequency as does a commitment to

14
the median voter’s ideal point. For β l > 0, therefore, centrist policy commitments are
rewarded with an increased chance of victory.
For candidate policy preferences more extreme than α2 it is not possible for there to
be any degree of imitation among candidates and simultaneously maintain the indiffer-
ence of the median voter (as, recall, any imitation must be at a candidate’s ideal point).
For α ≥ α2 equilibrium strategies must be separating; they are described in Theorem 3.

Theorem 3 Suppose q ∈ (0, 1). For all α ≥ α2 the unique equilibrium is separating:
policy motivated candidates A and B locate at −α2 and α2 , respectively, and all office
motivated candidates locate at the ideal point of the median voter; the median voter is
indifferent over all equilibrium announcements.

As is the case in Theorem 2, once office motivated types separate from policy moti-
vated candidates their type is fully revealed and they converge completely to the median
voter’s ideal point. The policy motivated types are similarly liberated and seek to con-
verge, but they only are willing to do so while their reputation for high effort is kept
intact. With office types all the way at 0, policy motivated types can move inward from
their ideal point while retaining their reputation, but only until they reach the threshold
±α2 , the point at which the median voter is indifferent over the candidates. Further
convergence would induce voters to believe the deviator to be office motivated. Thus,
α2 provides a bound on policy positions for all equilibria.
In contrast to the previous equilibria, voters in separating equilibria are able to
precisely determine a candidate’s type for all equilibrium announcements, and there
remains no residual uncertainty about the quality of implementation to be expected.
This result, in combination with the equilibria of Theorems 1 and 2, leads to the sur-
prising conclusion that the more different are the preferences of the candidates and the
median voter (the larger is α), the more information is conveyed in equilibrium. This
finding contrasts with Schultz (1996) and results from other settings that assume het-
erogeneous preferences, such as legislative models of information transmission (Gilligan
and Krehbiel 1987) and the cheap-talk literature generally (Crawford and Sobel 1982).

3.5 Properties of Equilibrium

The equilibria described in Theorems 1-3 allow insight into the nature of electoral com-
petition when information is incomplete and the policy space rich, providing unique
predictions about behavior and policy outcomes. In particular, and most germane, the
equilibria provide answers to the foundational questions posed in the introduction.
(i) What are the motivations of candidates that win elections? In the equilibria with
some separation (Theorems 2 & 3) it is true that more centrist policy commitments
provide an electoral advantage, and that office motivated candidates benefit from this.
However, this property not only doesn’t hold in all equilibria (see Theorem 1), but even
when it does hold, the advantage is not absolute and candidates with relatively extreme
policy commitments are still able to win elections. Indeed, the electoral advantage exists

15
only if office motivated candidates are also motivated to some degree by policy outcomes
(i.e., β l 6= 0).
It follows, to answer question (i), that office motivated candidates do not always win
elections and overwhelm policy motivated candidates, and in many situations perform
no better in terms of winning elections. The greater willingness of office motivated
candidates to trade policy for office is restrained by the electoral mechanism itself, and
policy motivated candidates win a significant fraction of elections.
(ii) How does the competitive electoral process shape the policies that are produced?
Equilibrium policy outcomes are very different when the motivations of candidates are
heterogeneous. Not only do candidates diverge on policy, but they frequently diverge all
the way to their ideal policy positions, thereby providing an informational explanation
as to why candidates may truthfully campaign according to their own policy preferences
even when policy commitments are possible.16 In answer to the question, therefore,
the interaction of candidates of different motivations has a significant impact on policy
outcomes.
For any set of parameter values, the effect of equilibrium behavior on the distribution
of candidate platforms and policy outcomes can be traced out. Figure 2 depicts the
distribution of policy commitments for values of q when β l = 0, β h = γ, and α = 12 .
The first three panels, (a)-(c), depict the distribution of campaign promises by both
candidates, with these commitments spread between the candidates’ ideal points and
the ideal point of the median voter. As the median voter mixes equally for β l = 0,
these figures also represent the distribution of policy outcomes. Panel (d) shows more
specifically how the probability that a candidate locates at zero varies as q varies between
zero and one. √ √
For√ the parameters of Figure 2 the critical values of α are α1 = 23 1 − q and
α2 = 23 , where α2 is independent of q, and α1 is decreasing in q. This implies that for
α of sufficient size (α ≥ α2 ) the equilibria are separating for all q whereas, in contrast,
for more moderate α (α < α2 ) equilibria are pooling for some values of q and semi-
separating for others, as depicted in Figure 2; notably, a fully separating equilibrium is
not possible for α < α2 .
The distributions of policy arising in equilibrium can be compared to empirical ob-
servations. Poole and Rosenthal (1997) provide evidence consistent with Figure 2 and
the finding of bimodal policy distributions. In particular, Panel (b) most closely reflects
practice in that, in addition to groupings of candidates at non-centrist locations, there
is a smattering of policy positions across more centrist locations.17
The results also relate to the empirical literature on the forces that drive political
behavior (Levitt 1996). As the median voter theorem does not imply policy convergence
16
This provides a contrast to the citizen-candidate literature (Osborne and Slivinski 1996; Besley and
Coate 1997), in which such positions follow from an inability to commit.
17
Although modeling elections from a different perspective, a similar clustering of policy positions is
found in Reed (1994), Duggan (2000), Bernhardt, Dubey, and Hughson (2004), and Banks and Duggan
(2006).

16
 6 6
Prob. Prob.
1- 1-

- -
−α1 −α 0 α α1 −α−α1 0 α1 α
Policy Policy
¡ 2¢ 7
(a) q ∈ 0, 3 : pooling equilibrium (bimodal) (b) q = 9
: semi-separating equilibrium (bimodal)

6 6
Prob. Prob.
1- 1-
¢¢
¢
¢
¢
¢
¢
¢
¢
¢
- ¢ -
−α −α1 0 α1 α 0 2 1 q
Policy 3
8
(c) q = 9 : semi-separating equilibrium (unimodal) (d) Prob. that a candidate locates at 0

1
Figure 2: Distribution of Policy Commitments: β l = 0, β h = γ, α = 2
.

17
in incomplete information environments, it is no longer clear that politicians who deviate
from the ideological preferences of voters are pursuing their own interests rather than
pandering to voters (as such behavior is interpreted in Levitt 1996). Rather, the results
presented here show that a “race to the center” may not be observed, despite the desire
of candidates to offer voters the most attractive policy they possibly can.
Equilibrium behavior also gives rise to a nonmonotonic relationship between a candi-
date’s policy commitment and his ideal point. If the ideal point is taken as a parameter,
the position of policy motivated candidates is strictly increasing in their ideal point until
they hit the bound of α2 , where it remains for more extreme candidates. In contrast,
for office motivated candidates the policy choice is not monotonic as the switch from a
pooling to semi-separating equilibrium induces the office motivated candidate to jump
to the center. Supporting this relationship, Blomberg and Harrington (2000) provide
evidence from the U.S. Congress that members at more extreme positions are viewed as
less flexible and more ideological.

Voter Learning and Behavior

Voter learning is important both for the instances when it occurs, and because the
threat of voter learning casts a shadow over candidate behavior. Voters do not learn
any valuable information in the pooling equilibria, yet behavior is very different because
of the possibility that information could be conveyed. Implied by this property is that
political campaigns can matter, and they can matter in ways that are not directly
measurable by what voters learn during the campaign.
As described above, voters learn more information about candidates type the more
extreme are candidate ideal points. However, also increasing in ideal points is the
divergence of policy commitments by policy motivated candidates. This creates a trade-
off for voters and begs the question: Does the information about type revealed by
electoral competition compensate for the inefficiency of policy divergence? As perhaps
is clear by now, the answer is no. In fact, when α ≥ α1 the median voter obtains the
same payoff as she would when q = 1 and all candidates are office motivated. In these
situations the policy cost incurred by candidates attempting to signal their type exactly
offsets the improvement in quality that flows from an increase in the fraction of policy
motivated types to the pool of candidates.
For more moderate policy preferences, the inclusion of policy motivated candidates
improves voter welfare (as divergence is limited by α) yet voters cannot extract the full
surplus. Indeed, the election mechanism leads to the same outcome as would the process
of selecting a candidate at random and allowing him complete freedom to determine all
components of the policy outcome (both effort and the policy instrument). Thus, the
ability for candidates to commit to policy does not provide any benefit to voters. An
interesting implication of this discussion is that, because elections are costly, democracy
itself may be inefficient, with voters preferring to select a candidate randomly from the
available pool rather than engage in the electoral process.18
18
This conclusion is dependent on policy motivation being bounded below by β l . If the election

18
 With respect to voter behavior, this discussion suggests that voters are better off
focusing on the issues alone as if they do then candidate convergence obtains.19 Conse-
quently, myopic, or ill-informed, voting may be welfare improving and a rational response
to an environment with incomplete information. That is, voters who ignore some dimen-
sions of policy may make themselves better off if such ignorance can affect the policies
offered by candidates. This incentive may go some way to explaining why voters may
be attracted to a candidate who proclaims that the election should only be “about the
issues.”
The ability of voters to learn about candidate type is also important for the efficiency
of policy outcomes. In pooling equilibria voters do not learn at all about candidate type
and the electoral mechanism does not shape the quality of policy implementation (in
the sense that selecting a candidate randomly would achieve the same result). In semi-
separating and separating equilibria voters learn about type, but this is traded off against
policy in such a way that the quality of implementation is lowered as a result of the
election (as office motivated types win more frequently if β l > 0).
This property suggest an equilibrium explanation for excessive government ineffi-
ciency. Moreover, a testable empirical prediction can be deduced. As office motivated
candidates converge whereas policy motivated candidates do not, a relationship between
policy and efficiency is produced: more centrist politicians (or governments) implement
policies less efficiently than do those with non-centrist policy positions. This topic has
not been investigated empirically (to the best of my knowledge), although it is key to
understanding the trade-off between policy and efficiency in democracies.

3.6 Comparative Statics: Approaching Homogeneity

In this section I examine equilibrium behavior as the model approaches the benchmark
environment of homogeneous candidate motivations. A complication is that there are
three ways by which the benchmark can be approached: q → 1, q → 0, and β h − β l → 0.
Intriguingly, I find that behavior varies depending on how the limit is approached, and,
consequently, convergence to the benchmark results does not always obtain.20
Corollary 1 analyzes the situations where one type of candidate is increasingly com-
mon in the candidate pool. The policy positions of candidates converge to the median
voter’s ideal point if policy motivated candidates disappear and only office motivated
types remain, but significant divergence persists in the reverse case. Let Pj (x) be the
equilibrium probability that candidate j commits to policy x. Recall that α1 is a function
mechanism were abolished then this bound may not persist.
19
These conclusions rely on the assumption that the location of the median voter is known with
certainty. If uncertainty exists then a degree of policy divergence may improve welfare. An interesting
open question is then whether electoral dynamics can induce the “right” amount of divergence to
optimize voter welfare.
20
Calvert (1985) interprets his study of policy motivated candidates as a robustness check on the
standard office motivated model. The current investigation can be interpreted in a similar way with
respect to the standard homogeneous motivations model.

19
of q, whereas α2 depends only on the difference β h − β l .

Corollary 1 (i) Office motivated candidates remain: if q → 1 then α1 → 0, and

PA (0) = PB (0) → 1.
(ii) Policy motivated candidates remain: if q → 0 then α1 → α2 , and PA (−α̂) → 1 and
PB (α̂) → 1, where α̂ = min [α, α2 ].

As the limit is approached a signaling game persists, and it is only at the limit that
voter inference is moot. Candidate incentives to imitate and pool remain, therefore, and
equilibrium behavior is driven by the relative costs and benefits of revealing one’s type to
the voters. If candidates are pooling then as policy motivated types disappear (q → 1)
voter expectations are for low effort. If a candidate separates and leads voters to believe
he is office motivated, the change to voter utility is small and the deviation is profitable
if the change in policy position is not too small. Thus, for pooling or semi-separating
equilibria to be maintained the pool must increasingly approach the median voter’s ideal
point. In contrast, if it is office motivated candidates that disappear (q → 0) then voter
expectations are for high effort, and separating from a pool is not profitable unless the
deviation is large. Consequently, non-negligible divergence can persist as the limit is
approached, even if the chance a candidate is office motivated is vanishingly small.
The benchmark environment can also be reached by allowing differences in policy
motivations to vanish. In this case, regardless of the convergent value of β, equilibrium
behavior must converge. Convergence in this case is effective rather than absolute: a
significant fraction of candidates locate away from zero at α2 , but as the difference in
motivations disappears (β h − β l → 0), the value of α2 itself approaches zero.

Corollary 2 If (β h − β l ) → 0 then α2 → 0; for all q ∈ (0, 1) there exists an ε > 0 such

that if (β h − β l ) < ε then PA (−α2 ) = 1 − q, PB (α2 ) = 1 − q, and PA (0) = PB (0) = q.

The intuition here is similar to that for q → 1 in Corollary 1. As motivational

differences disappear, voter utility changes little for changes in belief about type, and
deviations from pools (or semi-pools) are profitable unless those pools approach the
median voter’s ideal point.
In summary, the benchmark homogeneous motivations model is robust to some but
not all perturbations. In particular, if there is even an arbitrarily small possibility that
a candidate is of an undesirable type, aggregate behavior will be affected and significant
policy divergence will result. An implication of these results is that the small threat
of a “bad” candidate is more influential on behavior than the small threat of a “good”
candidate.

4 Conclusion
The objective of this paper has been to address a basic question previously overlooked
by the political economy literature: what are the motivations of those elected to public

20
office? In contrast to the existing literature in which this question is resolved by fiat,
I attempt here to address it directly. The method of investigation toward this end
is simple and novel: I take heterogeneity in the candidate pool as a primitive, and
allow candidates of different motivations to compete against each other. This approach
produces a simple model of electoral competition that provides insight into the impact
of candidate motivations on policy outcomes. In an appendix I generalize the model
in two key respects — allowing candidate ideal points that are asymmetric and private
information — and show that the key intuition is robust.
The main finding of the paper is that office motivated candidates do not dominate
elections, despite their greater willingness to trade off policy for the perks of office. Thus,
although a healthy dose of cynicism about the motivations of our elected leaders remains
a wise position, all is not lost and a significant fraction of those who win government
are interested in the policies they produce.
Balancing this positive result is the more concerning finding that policy outcomes
themselves (independent of the quality of implementation) are distorted by the compet-
itive forces within elections. This distortion arises as candidates seek to transmit infor-
mation to voters even though such attempts are often ultimately futile. Consequently,
equilibrium behavior in the model is not a convex combination of the alternative bench-
mark cases, and very different policies are chosen than would otherwise arise, even when
no information is conveyed in the election. Supporting the conclusions presented here
(however concerning) is that equilibrium behavior, both in terms of voter behavior and
the divergence of policy positions, is broadly consistent with stylized facts from empirical
research.
Much that is important to the practice of politics is omitted from the model and the
model itself is but one specification of the policy process. The policy freedoms that real
office holders enjoy extend well beyond effort levels, and the channels of competition in
election campaigns are not restricted to a single policy dimension. These simplifications
have been an intentional feature of the current enterprise, which is best seen as a small
step to connect several hitherto distinct aspects of electoral competition into a single
model, and show how their interdependencies affect policy outcomes. Much remains to
be done to incorporate these insights into broader models of political economy, both
theoretically and empirically.

21
5 Appendix A: Extensions
The model analyzed here is both simple and stylized, intended to expose most clearly the
impact of candidate heterogeneity on policy outcomes. The most prominent restriction
of the model is that the policy preferences of the candidates are symmetric and public
information. In politics the policy preferences of candidates are often endogenous and
possibly asymmetric, the result of either a primary election or strategic determination
by the nominating party. Moreover, there is no compelling reason to think that a
candidate’s policy preferences are known definitively within the party, let alone among
the public at large. In this appendix I relax individually the restrictions that ideal
points are symmetric and public information and show that the main insights of the
model extend to these more general environments.

Asymmetric Candidate Ideal Points

As the model is currently specified, asymmetric candidate ideal points pose a prob-
lem as the equilibria depend on voter indifference. For example, if ideal points are
asymmetric and candidates play the pooling strategies of Theorem 1, one candidate
wins and the other loses with certainty (possibly leading to equilibrium non-existence).
This conclusion is, however, an artefact of the simple information structure and does
not apply if some uncertainty about voter preferences is added to the model. This is
confirmed by Theorem 4, which for brevity considers only pooling equilibria.21

Theorem 4 Suppose the location of the median voter, pm , is a random variable distrib-
uted uniformly on the interval [−δ, δ], where pm is unknown by candidates. Let candidate
ideal points be αA and αB , respectively, such that −α1 < αA ≤ 0, 0 ≤ αB < α1 , and
αA +αB
2
< δ. The unique equilibrium when q ∈ (0, 1) is for all candidate types to pool at
their ideal point; both candidates win election with positive probability.

It is rather natural to assume that candidates are uncertain about voter preferences,
just as voters are uncertain about candidate preferences. The inclusion of this uncer-
tainty “smooths out” candidate payoffs and, although not necessary for the main results,
makes the model more broadly applicable.22

Ideal Points as Private Information

Suppose at Stage 1 of the election game (see Table 1) that, in addition to moti-
vations, candidates also receive a policy preference that is private information. More
specifically, suppose that candidates are either centrists (C) or extremists (E), where
21
Similar issues surround the separating equilibria. Semi-separating equilibria are not so affected:
by mixing in asymmetric ratios, the office motivated candidates can maintain voter indifference and
support an equilibrium.
22
Theorem 4 also suggests that an equilibrium may exist in multiple dimensional policy spaces, even
when a core point does not exist.

22
candidates A and B’s ideal points are 0 if centrists and −α and α, respectively, if ex-
tremists. Let candidates be centrist with probability p and extremist with probability
1 − p. A candidate’s policy preference is assumed for simplicity to be independent of
his motivation. Therefore, a candidate can now be one of four possible types with the
following probabilities:
Pr {C, l} = p.q
Pr {C, h} = p. (1 − q)
Pr {E, l} = (1 − p) .q
Pr {E, h} = (1 − p) . (1 − q) .
The nature of equilibrium in this environment is (surprisingly) similar to the case
considered previously (which corresponds to p = 0). An immediate observation is that
centrists always locate at their ideal point (as it corresponds to the ideal point of the
median voter). Thus, when an extremist office motivated candidate deviates to 0, he no
longer fully reveals his type, but rather he merges into the pool of centrist candidates,
both office and policy motivated. This implies that in equilibrium extremists cannot
pool at their ideal point, as they did in Theorem 1. If they did so then a deviation to
zero would not adversely affect voter beliefs, and the more attractive policy commitment
would win the election. Consequently, equilibria in this generalized environment require
extremists to either semi-separate or separate entirely, although now there will be pool-
ing across policy preferences (office motivated extremists will pool with centrists). These
facts lead to a new critical value of α.
Definition 3 α satisfies:
µ ¶2 µ ¶ µ ¶2
βl p 2 βh
1− = 1−p+ α + 1− .
βl + γ q βh + γ
The value α defines the policy position such that the median voter is indifferent over
a policy of α from a policy motivated candidate, and a policy of zero from either one
of the three types, {0, l} , {0, h}, or {α, l}, with beliefs determined by Bayes’ rule. The
statement of the definition follows from several steps of algebra. Further algebra shows
several additional properties of α: α < α2 , α may be larger or smaller than α1 depending
on parameters, and that α is decreasing in p.
Equilibria in this environment take one of two forms. Candidates with extreme
preferences play a semi-separating strategy in equilibrium only if their policy preferences
are not too extreme.
Theorem 5 Suppose p, q ∈ (0, 1). For all α ∈ [0, α) a unique equilibrium exists that is
symmetric, and for extremists it is semi-separating. The strategy of candidate B is the
following.
⎧
⎨ 0 if of type {C, l} or {C, h} ;
Commit to policy: α if of type {E, h} ;
⎩ mix over α and 0 if of type {E, l} .

23
The median voter is indifferent over all equilibrium announcements.

If instead the preferences of extremists are sufficiently divergent then they fully
separate in equilibrium and policy commitments are bounded by α.

Theorem 6 Suppose p, q ∈ (0, 1). For all α ≥ α a unique equilibrium exists that
is symmetric, and for extremists it is separating. The strategy of candidate B is the
following.
½
0 if of type {C, l}, {C, h}, or {E, l};
Commit to policy:
α if of type {E, h}.

The median voter is indifferent over all equilibrium announcements.

The intuition for these equilibria are the same as for the previous results. In all equi-
libria the centrist candidates pool (therefore, they are centrist in both preference and
action), and voters are always unable to sort between them. In contrast, extremist can-
didates separate to some degree in all equilibria and voters learn information about the
expected quality of implementation. Figure 3 shows equilibrium configurations of policy
commitments as p, the fraction of centrist candidates, varies; the parameters depicted
are the same as in Figure 2 with the additional restriction that q = 14 (corresponding to
the pooling equilibrium of panel (a) in Figure 2). For these parameter values α = α = 12
at the critical value p = 23 ; thus, the semi-separating equilibria of Theorem 5 apply
2
for p <√ 3
, and the separating equilibria of Theorem 6 apply thereafter. √
(Note that
3 3
α → 2 as p → 1, and equilibria are semi-separating for all p if α < 2 ). In panel
(c) the equilibrium is separating and all extremist office motivated candidates deviate
to zero. To maintain voter indifference as p increases, the policy motivated extremists
must converge, although this convergence is bounded away from zero by α.
For simplicity it has been assumed that centrists have ideal points at zero. Equilib-
rium behavior is largely unchanged if instead the preferences of centrists also diverge,
although with minor predictable adjustments. For example, in such a case centrists
pool at their ideal points for low values of q, but, as in Theorem 1, the pooling equi-
librium breaks down for sufficiently large q, and separation arises for both centrists and
extremists.

6 Appendix B: Proofs
Define et = β β+γ
t
for t = h, l. As this is independent of policy commitments, I simplify
t
the strategies of candidates and write them only as the choice of a policy commitment.
A mixed strategy σ j maps candidate j’s type into a probability distribution over policy
positions, where σ j (x|t) represents the probability that candidate j commits to policy
x if of type t. Denote equilibrium strategies by σ ∗ , r∗ , and π ∗ = {π ∗A , π ∗B }. I abuse
notation and use several definitions to also represent their expectation. I generalize

24
 6 6
Prob. Prob.
1- 1-

- -
−α −α 0 α α −α−α 0 α α
Policy Policy
1 1
(a) p = 4
: semi-separating equilibrium (bimodal) (b) p = 2
: semi-separating equilibrium (unimodal)

6 6
Prob. Prob.
1- 1-

- -
−α−α 0 α α 0 2 1 q
Policy 3
3
(c) p = 4
: separating equilibrium (unimodal) (d) Prob. that type {E, l} locates at 0

Figure 3: Distribution of Policy Commitments: β l = 0, β h = γ, α = 12 , q = 1

4
.

25
the function r (.) and allow it to depend on strategies σ. The expected utility for
candidates is written as Uj (σ A , σ B , r|t), and for voters the expected utility from policy
commitment xj is Ui (xj |π) for beliefs π. I allow the support of candidate strategies to be
both countable and uncountable. In all equilibria, however, mixing is over at most two
positions; consequently, to simplify language I refer throughout to probabilities rather
than densities.
The proof of Lemma 1 is trivial and omitted. The proofs of Theorems 5 and 6 follow
similar arguments to preceding results and are also omitted. I begin with a statement
of universal divinity.
Universal divinity: For any out-of-equilibrium announcement x0A , define ρ (x0A |t)
as the set of median voter reaction functions (i.e., r (x0A , .)) that make a type t candidate
A indifferent between deviating and playing his equilibrium strategy. If ρ (x0A |l) ⊂
ρ (x0A |h) then type h benefits from the deviation for a strictly greater set of reactions
by the voters. Thus, he is more likely to deviate than a type l candidate, and universal
divinity requires that voters believe with probability one that the deviator is of type h.
Alternatively, if ρ (x0A |h) ⊂ ρ (x0A |l) then a type l candidate is more likely to deviate,
and voters believe with probability one the deviator is of type l.

Lemma 2: If q ∈ {0, 1} then the unique equilibrium is σ ∗j (0|t) = 1 ∀t, j.

Proof of Lemma 2: For all x and j, π j (x) = q, and the level of effort is inde-
pendent of the winning candidate. Competition is only over policy commitments. The
result then follows from well known arguments; see Calvert (1985).¥

Lemma 3: If P = R then ∀q the unique equilibrium is σ ∗j (0|t) = 1 ∀t, j.

Proof of Lemma 3: Analogous to Lemma 2.¥

Lemma 4: Consider an equilibrium {σ ∗ , r∗ , π ∗ }: If σ ∗A (x|l) > 0 and σ ∗A (y|h) > 0

then r∗ (x, σ ∗B ) ≥ r∗ (y, σ ∗B ) and |−α − x| ≥ |−α − y|. The conditions for candidate B
are analogous.

Proof of Lemma 4: Equilibrium requires for σ ∗B (x|l) > 0 and σ ∗B (z|h) > 0 that:

UB (σ ∗A , x, r∗ |l) ≥ UB (σ ∗A , z, r∗ |l) , and (1)

UB (σ ∗A , x, r∗ |h) ≤ UB (σ ∗A , z, r∗ |h) .

26
Candidate B’s expected utility from committing to x is (for any σ A , r):
∙Z h ¸
2 ¡ ¢i
l 2
UB (σ A , x, r|t) = −β t q σ (y|l) r (y, x) (α − y) + 1 − e ∂y (2)
∙Z h ¸
2 ¡ ¢i
h 2
−β t (1 − q) σ (y|β h ) r (y, x) (α − y) + 1 − e ∂y
∙ ¸
2 ¡ ¢
t 2 γ ¡ t ¢2 W
−β t [1 − r (σ A , x)] (α − x) + 1 − e + e − .
b b
∙ ³ ´ ¸
2
∗ ∗ ∗ b
For β l = 0 this reduces in equilibrium to UB (σ A , x|t) = − [1 − r (σ A , x)] γ b+γ − w ,
and part (i) of the lemma is immediate. So set β l > 0. Substituting Equation 2 into the
equilibrium conditions, dividing by β t and rearranging, produces the conditions:
∙ ¸
∗
¡ ¢
l 2 γ ¡ l ¢2 w
χ ≥ [r (σ ∗A , z)
−r ∗
(σ ∗A , x)]
1−e + e − (3a)
βl βl
∙ ¸
∗ ∗ ∗ ∗
¡ ¢
h 2 γ ¡ h ¢2 w
χ ≤ [r (σ A , z) − r (σ A , x)] 1 − e + e − (3b)
βh βh
¡ h ¢2
for some
h¡ χ that is independent of
i h¡t. Simple algebra and recalling
i W ≥ γ e establishes
¢2 ¡ ¢2 ¢2 ¡ ¢2
that 1 − el + βγ el − βw > 1 − eh + βγ eh − βw . Thus, the inequalities in
l l h h
Equations (3a) and (3b) then require:
r∗ (σ ∗A , z) − r∗ (σ ∗A , x) ≥ 0 ⇒ r∗ (σ ∗A , z) ≥ r∗ (σ ∗A , x) ,
proving part (i) of the lemma.
As oA ≤ oB in equilibrium, UB (σ ∗A , x, r∗ |t) is decreasing in both r (σ ∗A , x) and |α − x|.
Therefore, part (ii) of the lemma follows from part (i) if Equation 1 is to hold.¥
Voter Beliefs: The following lemma describes how beliefs are formed for out-of-
equilibrium policy commitments. Part (i) deals with deviations towards a candidate’s
ideal point, and parts (ii) and (iii) deviations away.
Lemma 6 Suppose the strategy σ satisfies Lemma 4 and previous assumptions. If
σ B (x|h) > 0 and σ B (x0 |h) = σ B (x0 |l) = 0, universal divinity implies the following.
(i) π B (x0 ) = 0 if |α − x0 | < |α − x|.
(ii) Suppose UB (σ A , σ B , r|l) = UB (σ A , x, r|l). Then π B (x0 ) = 1 if |α − x0 | > |α − x|.
(iii) Suppose UB (σ A , σ B , r|l) > UB (σ A , x, r|l). Then π B (x0 ) = 0 if |x0 − x| < ε for
some ε > 0.

Proof of Lemma 6: Set σ B (x|h) > 0 and σ B (x0 |h) = σ B (x0 |h) = 0. Let r0
be a function derived from some voter belief function π 0 such that UB (σ A , x, r|h) =
UB (σ A , x0 , r0 |h), and define
ψ B (x, x0 , r0 |l) = UB (σ A , x, r|l) − UB (σ A , x0 , r0 |l) .

27
Using the equality for policy motivated types, this reduces to:
⎡ ³¡ ¢ ¡ h ¢2 W ´ ⎤
h 2 γ
1 − e + βh
e −β −
ψ B (x, x0 , r0 |l) = β l [r0 (σ A , x) − r (σ A , x)] ⎣ ³¡ ¢ ¡ ¢ h ´ ⎦.
l 2 γ l 2 W
1−e + β e − β
l l

The proof of Lemma 4 established that the term in square brackets is greater than zero,
therefore:
ψ B (x, x0 , r0 |l) > 0 iff r0 (σ A , x) − r (σ A , x) > 0.
Suppose that |α − x0 | < |α − x|. As UB is strictly decreasing and continuous in
r, the indifference requirement for a type h implies that r0 (σ A , x) < r (σ A , x) and
ψ B (x, x0 , r0 |l) < 0. As utility is also continuous and monotonic in |α − x0 |, ρ (x0 |l) ⊂
ρ (x0 |h), proving part (i). Analogously, |α − x0 | > |α − x| ⇒ ρ (x0 |h) ⊂ ρ (x0 |l), proving
part (ii). Part (iii) follows by continuity.¥
It follows immediately from the lemma that the support of both candidates’ strategies
are in [−α, α]. The monotonicity results of Lemma 4 then imply that there can be at
most one policy position for any j such that σ j (x|t) > 0 ∀t. A final observation, which
is stated as Lemma 7, is that the probability with which each type of candidate wins
election is bounded above zero. This implies that each candidate’s strategy support has
a most extreme point, and that positive mass is placed on this point.

Lemma 7 If σ ∗A (x|t) > 0 for some t, then r∗ (x, σ ∗B ) > ε for some ε > 0. The equivalent
holds for candidate B.

Proof of Lemma 7: Suppose not for candidate B: therefore, ∃ a sequence xn

s.t. σ ∗B (xn |h) > 0 and r∗ (σ A , xn ) → 1 as n → ∞. For this sequence, set m =
{n| min r∗ (σ A , xn )}. Recall the assumptions that in equilibrium candidate A locates to
the left of B and each candidate wins with positive probability; therefore, r∗ (σ A , xm ) <
1. This implies that UB (σ ∗A , xm , r∗ |h) > UB (σ ∗A , xn , r∗ |h) for large enough n as UB is
strictly decreasing in r and |α − x|, thereby violating equilibrium (as B prefers xm to
any policy of candidate A).¥

Lemma 5: Set q ∈ (0, 1). If in equilibrium σ j (x|l) = σ j (x|h) ∀x, j, then σ A (−α|t) =
σ B (α|t) = 1 ∀t.

Proof of Lemma 5: If candidates are pooling, then by monotonicity they each

pool at a single position: for some x and y, σ A (x|t) = σ B (y|t) = 1 and π A (x) =
π B (y) = q. If y < α consider the deviation to ỹ = y + ε ∈ (y, α]. By Lemma 6,
π B (ỹ) = 0, and as ε → 0, r (σ A , x + ε) ≤ r (σ A , x) < 1, and the deviation is profitable
as UB is decreasing in |α − x|; the lemma follows.¥

Theorem 1: Set q ∈ (0, 1). For all α ∈ [0, α1 ] the unique equilibrium is σ ∗A (−α|t) =
1 and σ ∗B (α|t) = 1 ∀t. A pooling equilibrium does not exist if α > α1 .

28
 Proof of Theorem 1: Set α ∈ [0, α1 ]. If σ B is not pooling then there exists y < x
s.t. σ B (x|h) > 0, σ B (y|l) > 0, π B (x) < q, and π B (y) = 1 (as, by the arguments of
Lemma 5, if types pool at all then it must be at α). By the definition of α1 , this implies
r (σ A , x) < r (σ A , y), violating monotonicity.
The equilibrium must be pooling, and, by Lemma 5, σ B (α|t) = σ A (−α|t) = 1 ∀t.
Lemma 6 implies that π B (y) = 1 for all y 6= α (likewise for A). For α < α1 this implies
by definition that r (σ A , y) = 1, the policy outcome is −α, and the deviation is not
profitable. As utility is continuous, there exists for α = α1 an r (σ A , 0) sufficiently small
s.t. the deviation is not profitable. Thus, the pooling equilibrium exists.
If α > α1 then a deviation by candidate B to 0 implies r (σ A , 0) = 0. As B is
risk averse and utility is decreasing in r (.), the deviation is profitable and a pooling
equilibrium does not exist.¥

Theorem 2: Set q ∈ (0, 1). For all α ∈ (α1 , α2 ) the unique equilibrium is symmet-
ric and given by: σ ∗B (α|h) = 1 and σ ∗B (0|l) + σ ∗B (α|l) = 1, s.t. Um (0|π ∗ ) = Um (α|π ∗ ).

Theorem 3: Set q ∈ (0, 1). For α ≥ α2 the unique equilibrium is symmetric where
σ ∗B (α2 |h) = 1 and σ ∗B (0|l) = 1, s.t. Um (0|π ∗ ) = Um (α2 |π ∗ ).

Proof of Theorems 2 and 3: I begin by ruling out equilibria other than those in
the theorems. Consider policy motivated candidates first and define x = max {x|σ B (x|h) > 0}.
If there is an x0 < x s.t. σ B (x0 |h) > 0 then, by Lemma 6 and monotonicity, π B (x̃) = 0
∀x̃ ∈ (x0 , x). By Lemma 7, there must exist an action y, that is played with positive
probability by candidate A and is such that r (y, x) ∈ (0, 1). Therefore, a deviation by
B to x̃ ∈ (x0 , x) implies that r (y, x̃) = 0, and for x̃ → x, candidate B is strictly better
off. Therefore, there is a pair, x and y, s.t. σ B (x|h) = σ A (y|h) = 1, and by Lemma 7,
Um (x|π ∗ ) = Um (y|π ∗ ).
The next step is to find the values of x and y, and the equilibrium strategy of
office motivated candidates. The strategies of candidates creates a distribution over the
possible policy positions. As candidates are risk averse, at least one of them (and for
symmetric strategies, both) prefer the certain outcome of 0; without loss of generality, let
this be candidate B and suppose there is a z where 0 < z < x and σ B (z|l) > 0. Consider
two possible cases for the strategy of A. If σ A (0|l) + σ A (y|l) = 1 then r (0, z) = 1, which
implies r (0, x) = 1 by monotonicity. If B deviates to 0 then he increases his probability
of winning and guarantees a policy outcome of 0 (as r (0, x) = 1 ⇒ r (y, 0) = 0); z can’t
be an equilibrium action. If σ A (0|l) + σ A (y|l) < 1 then a deviation to 0 is similarly
profitable. Therefore, in equilibrium σ B (0|l) + σ B (x|l) = 1. An analogous condition
holds for candidate A. If σ B (x|l) > 0 then by analogous arguments to Lemma 5, x = α;
similarly, σ A (y|l) > 0 ⇒ y = −α.
Suppose again that candidate B prefers the certain outcome of 0 over the lottery if
of type h. If Um (0|π ∗ ) > Um (x|π ∗ ) then r (0, x) = 1, and a deviation to 0 is profitable:

29
it guarantees a policy of 0 and increases B’s probability of winning. A similar condition
holds for candidate A, and in equilibrium Um (0|π ∗ ) = Um (y|π ∗ ) = Um (x|π ∗ ).
Theorem 2: α ∈ (α1 , α2 ). If strategies are fully separating then Um (0|π ∗ ) <
Um (x|π ∗ ), violating equilibrium. By Theorem 1, equilibrium can’t be pooling as if
pooling then Um (0|π ∗ ) > Um (α|π ∗ ). Thus, equilibrium requires a symmetric strategy
where σ B (0|l) = k, σ B (α|l) = 1 − k, and σ B (α|h) = 1, for some k ∈ (0, 1). By the
continuity of utility, a k 0 exists s.t. Um (0|π∗ ) = Um (α|π ∗ ). Also by continuity, there
exists an r∗ (.) s.t. candidates of type l are indifferent over committing to policies 0 and
their ideal points. By Lemma 6, type h candidates strictly prefer committing to their
ideal points. Finally, consider possible deviations. Lemma 6 implies that πB (z) = 1
∀z 6= α, which guarantees r (σ A , z) = 1; no profitable deviation exists for candidate B,
and analogous arguments apply to candidate A.
Theorem 3: α ≥ α2 . If strategies are partially separating then x = α and Um (0|π ∗ ) >
Um (x|π ∗ ), violating equilibrium. Thus, the equilibrium is fully separating, which re-
quires σ A (0|l) = σ A (−α2 |h) = σ B (0|l) = σ B (α2 |h) = 1 to maintain voter indifference.
By Lemma 6 there exists an r (.) s.t. each type of candidate prefers his own strategy
to imitating the other. Consider deviations by candidate B. For z ∈ (0, α2 ), π B (z) = 1,
and z ∈ (α2 , α), π B (z) = 0; for all z ∈ / {0, α}, r (σ A , z) = 1 and the deviation isn’t
profitable. Analogous arguments apply to candidate A.¥

Proof of Corollary 1: (i) As q → 1, simple algebra shows α1 → 0. The result

follows from the equilibrium condition that Um (0|π ∗ ) = Um (α̂|π∗ ).
(ii) As q → 0, simple algebra shows α1 → α2 , and the results follows from the equilibrium
requirement that σ ∗B (α̂|h) = 1.¥

Proof of Corollary 2: (β h − β l ) → 0 implies by algebra that α2 → 0. As α > α2

for β h − β l sufficiently small, the result follows from the equilibrium conditions that
σ B (α̂|h) = 1 and σ B (0|l) = 1.¥

Theorem 4: For the conditions stated in the text, a unique equilibrium exists and
is σ A (αA |t) = σ B (αB |t) = 1 ∀t.

Proof of Theorem 4: Analogous arguments to Lemma 6 and Theorem 1 im-

ply that the only possible equilibrium is that in the theorem. As αA +α 2
B
< ω then
r (αA , αB ) ∈ (0, 1). If candidate B deviates to x̃ 6= αB then π B (x̃) = 1 and, by the
definition of α1 , r (αA , x̃) = 1, and the deviation isn’t profitable. The same logic applies
to deviations by A, and the result follows.¥

30
References
[1] Alesina, Alberto. 1988. “Credibility and Policy Convergence in a Two-Party System
with Rational Voters.” American Economic Review 78: 796-806.

[2] Ansolabehere, Stephen, and James M. Snyder, Jr. 2000. “Valence Politics and Equi-
librium in Spatial Election Models.” Public Choice 103 (June): 327-336.

[3] Aragones, Enriqueta, and Thomas R. Palfrey. 2002. “Mixed Equilibrium in a Down-
sian Model with a Favored Candidate.” Journal of Economic Theory 103: 131-161.

[4] Aragones, Enriqueta, and Thomas R. Palfrey. 2005. “Electoral Competition Be-
tween Two Candidates of Different Quality: The Effects of Candidate Ideology
and Private Information.” In David Austen-Smith and John Duggan (eds.), Social
Choice and Strategic Decisions. Berlin, Germany: Springer.

[5] Banks, Jeffrey S., and Joel Sobel. 1987. “Equilibrium Selection in Signaling Games.”
Econometrica 55: 647-62.

[6] Banks, Jeffrey S. 1990. “A Model of Electoral Competition With Incomplete Infor-
mation.” Journal of Economic Theory 50: 309-25.

[7] Banks, J. and J. Duggan. 2006. “A Dynamic Model of Democratic Elections in

Multidimensional Policy Spaces.” Working paper.

[8] Barro, Robert J. 1973. “The Control of Politicians: An Economic Model.” Public
Choice 14: 19-42.

[9] Bernhardt, D., D.E. Ingberman. 1985. “Candidate Reputations and the ‘Incum-
bency Effect’.” Journal of Public Economics 27: 47-67.

[10] Bernhardt, D., S. Dubey, and E. Hughson. 2004. “Term Limits and Pork Barrel
Politics.” Journal of Public Economics 88: 2383-2422.

[11] Besley, Timothy, and Stephen Coate. 1997. “An Economic Model of Representative
Democracy.” The Quarterly Journal of Economics 112 (February): 85-114.

[12] Black, D. 1958. The Theory of Committees and Elections (Cambridge University
Press: Cambridge, U.K.).

[13] Blomberg, S.B. and J.E. Harrington. 2000. “A Theory of Rigid Extremists and
Flexible Moderates with an Application to the U.S. Congress.” American Economic
Review 90: 605-620.

[14] Budge, Ian, David Robertsen, and Derek J. Hearl, eds. 1987. Ideology, Strategy, and
Party Change: Spatial Analyses of Post-war Election Programmes in 19 Democra-
cies. Cambridge, United Kingdom: Cambridge University Press.

31
[15] Calvert, Randall L. 1985. “Robustness of the Multidimensional Voting Model: Can-
didates’ Motivations, Uncertainty and Convergence.” American Journal of Political
Science 29: 1056-1070.

[16] Cho, I.K. and D.M. Kreps. 1987. “Signaling Games and Stable Equilibria.” Quar-
terly Journal of Economics 102: 179-221.

[17] Crawford, Vincent P., and Joel Sobel. 1982. “Strategic Information Transmission.”
Econometrica 50 (November): 1431-51.

[18] Duggan, J. 2000. “Repeated Elections With Asymmetric Information.” Economics

and Politics 12 (July): 109-135.

[19] Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public
Choice 50: 5-26.

[20] Gilligan, Thomas, and Keith Krehbiel. 1987. “Collective Decision Making and
Standing Committees: An Informational Rationale for Restrictive Amendment Pro-
cedures.” Journal of Law, Economics, and Organization 3 (2): 287-335.

[21] Groseclose, Timothy. 2001. “A Model of Candidate Location When One Candidate
Has a Valence Advantage.” American Journal of Political Science 45: 862-886.

[22] Hotelling, Harold. 1929. “Stability in Competition.” The Economic Journal 39:
41-57.

[23] Kartik, N., and P. McAfee. 2006. “Signaling Character in Elections.” Working pa-
per.

[24] Levitt, Steven D. 1996. “How Do Senators Vote? Disentangling the Role of Voter
Preferences, Party Affiliation, and Senator Ideology.” American Economic Review
86(3): 425-441.

[25] Osborne, Martin J., and Al Slivinski. 1996. “A Model of Political Competition with
Citizen-Candidates.” Quarterly Journal of Economics 111 (February): 65-96.

[26] Persson, Torsten, and Guido Tabellini. 2000. Political Economics: Explaining Eco-
nomic Policy. Cambridge, Massachusetts: MIT Press.

[27] Poole, Keith T., and Howard Rosenthal. 1997. Congress. New York, NY :Oxford
University Press.

[28] Prat, Andrea. 2002a. “Campaign Advertising and Voter Welfare.” Review of Eco-
nomic Studies 69 (October): 999-1017.

[29] Prat, Andrea. 2002b. “Campaign Spending with Office-Seeking Politicians, Rational
Voters, and Multiple Lobbies.” Journal of Economic Theory 103: 162-189.

32
[30] Reed, R. 1994. “A Retrospective Voting Model With Heterogeneous Politicians.”
Economics and Politics 6 (March): 39-58.

[31] Roemer, John E. 1999. “The Democratic Political Economy of Progressive Income
Taxation.” Econometrica 67 (January): 1-19.

[32] Schultz, Christian. 1996. “Polarization and Inefficient Policies.” Review of Economic
Studies 63 (April): 331-343.

[33] Stokes, Donald E. 1963. “Spatial Models of Party Competition.” American Political
Science Review 57: 368-377.

[34] Wittman, Donald. 1977. “Candidates with Policy Preferences: A Dynamic Model.”
Journal of Economic Theory 14 (February): 180-189.

[35] Wittman, Donald. 1983. “Candidate Motivation: A Synthesis of Alternative The-

ories.” American Political Science Review 77: 142-157.

33