Strategic Campaigns and Redistributive Politics

STRATEGIC CAMPAIGNS AND REDISTRIBUTIVE POLITICS*

Christian Schultz

The article investigates strategic, informative campaigning by two parties when politics concernredistribution. Voters are uncertain about whether parties favour special groups. Parties will targetcampaigns on groups where most votes are gained by informing about policies. In equilibrium,campaigning will be most intensive in groups where the uncertainty is largest and where voters aremost mobile, most likely to vote, most receptive to campaigns and relatively uninformed initially.These groups will become more informed about policy. Parties will therefore gain more votes bytreating these groups well so these groups will gain from strategic campaigning. Welfare effects areassessed.

Political campaigns are an important part of electoral democracies. Candidates spendendless days on the campaign trail and parties spend huge amounts of campaignmoney. According to the Federal Election Commission, the parties and candidates areestimated to have spent roughly 3 billion US dollars in the 2000 US electoral cycle.Although, the US is probably the country where campaign money are most important,campaign expenditures are increasing in most developed democracies. Clearly, thecampaigns have a purpose: to influence voters so they are induced to vote for thecampaigning party. While it is well documented that they actually do work, see e.g.Green and Krasno (1988) or Gerber (1998), the way may differ. Some voters may beattracted just by the fact that a party campaigns intensely and perhaps use nice cam-paigns. In the literature such campaigns are usually called persuasive, cf. Baron (1994).On the other hand, parties also spend considerable resources on informative cam-paigns, informing about their policies or perhaps about other parties� policies. Suchcampaigns would seem beneficial from a social perspective as they allow voters to makemore informed choices. But they also have implications for the parties� strategicstruggle and the policies which are adopted in the end. In this article I consider theeffects of strategic, informative campaigning on redistributive policies when parties cantarget their campaigns on groups. The main questions I am interested in are: how willthe parties spend their limited campaign resources and which effects will this have onpolicies? Which groups will benefit? Are such campaigns to be considered good or bad?

The basic model of the article is the model of redistributive politics originallyintroduced by Lindbeck and Weibull (1987) and extended by Dixit and Londregan(1996). Here, parties favour groups, whose vote-response is high when they are offeredbetter policies.1 In my model, voters are uncertain about whether parties are biasedtowards special groups. The voters� information about politics differs. Some voters areuninformed and become informed through political campaigns. Parties will target

* I benefited from the comments of the editor, Leonardo Felli, two referees and discussions with GiuseppeBertola, Douglas Hibbs, Niels Henrik Morch von der Fehr, Peter Norman S�rensen and Jean Tirole, as well asseminar audiences in CORE, Goteborg, Florence, Oslo and Paris and EEA in Venice, while working on thisarticle.

1 Dahlberg and Johansson (2002) find that this prediction is supported in data from a Swedish temporarygrant programme.

The Economic Journal, 117 (July), 936–963. � The Author(s). Journal compilation � Royal Economic Society 2007. Published by

Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

[ 936 ]

campaigns to those groups where campaigns are most effective. These are groups withmany swing voters, with a high turnout (and a high increase in turnout in response tocampaigning) and groups where the initial level of information is low. The campaignswill increase the information level in these groups and this makes the parties offer thesegroups better policies. In the end, these groups will benefit from the strategic cam-paigning. There is a complementarity between campaigns and spending – a campaign-multiplier – campaigns reinforce the effects of voter mobility.

Informative campaigns are often considered good since they allow voters to makeinformed choices. However, the results of this article point to another effect.Informative campaigning creates a particular distribution of information in the elec-torate, which in turn affects real policies. A group’s information level is determined bythe parties� strategic incentives to gain most votes from the limited campaign moneyrather than considerations of fairness. This favours groups, who already gain from thestrategic struggle among the parties, the mobile groups and the groups with highturnout. In this respect informative campaigning makes the distribution of policiesover groups more unequal than if all groups had the same level of information and alimit on campaign spending (or finances) would make for a more equal distributionin society. Since the model is one of pure redistribution, all allocations are Paretoefficient. If, however, one takes a utilitarian perspective and evaluates the sum ofutilities in society, the fact that campaigns reinforce the skewness due to differences invoter mobility and participation rates is bad. However, there is a countervailing effect;campaigns are also targeted at groups where the level of information initially is low.These groups are initially disadvantaged and therefore equality may be enhanced bycampaigning. In reality one would often think of such groups being less educated, lowincome groups and campaigns may therefore benefit such weak groups. Morereceptive groups also gain. Whether these are less educated or more educated groupsis not obvious. Presumably, the well-educated are better at receiving information, butif the information level of the well-educated is very high from news coverage, thenthere is not much to learn. Results reported by Alvarez (1998, p. 172) suggest that thelatter effect is the most important. Although the model provides specific predictionsabout which characteristics make a group an attractive target for campaigning andtherefore make the group gain, the total welfare effect depends on the strength of thedifferent forces. A general welfare assessment is complicated by the fact that themodel does not allow closed form solutions, but for the case where the ideologicalcomponent of the parties� preferences is almost egalitarian, it is shown that the welfareeffect of campaigns is negative or positive depending on which of the above men-tioned effects dominate.

The model predicts that only one party will campaign in a given group. The reason isthat a party’s campaign informs voters about policies. A party will campaign in groupswhere it has good news to reveal: either that its policy is better than expected (positivecampaigning) or that the other party’s policy is worse than expected (negative cam-paigning). The parties will therefore never campaign in the same groups. The modelthus predicts that parties should have a tendency to focus on issues where theirstanding among voters is strong.

In reality, one often observes that parties campaign in the same groups. This doesnot necessarily contradict our results since I focus on informative campaigning. As is

[ J U L Y 2007] 937S T R A T E G I C C A M P A I G N S

� The Author(s). Journal compilation � Royal Economic Society 2007

well known from for instance Baron (1994) parties will have incentives to direct per-suasive campaigns to the same groups of voters.

A large literature considers the effects of lobby groups� influence on politics throughcampaign donations. The effects of lobbying are by now well described and under-stood, see e.g. Baron (1994) and the book by Helpman and Grossman (2001). Thelobbying literature usually assumes that campaigning is persuasive, so money directlybuys votes. I show that campaign funds are also valuable for parties if campaigns areinformative, and hence the incentive to induce lobby group payments remains.Although lobbies will not be considered in this article, the model could serve as abuilding block in a model of lobbying.

I assume that the parties commit to their policy proposals before the election andthat they inform truthfully about them. Clearly, this is a benchmark case. In the realworld, politicians may lack a commitment device; however, reputational forces willpunish politicians who grossly disregard pre-electoral promises. Most of the worldremembers Bush Sr�s (in)famous �Read my lips� statement. Furthermore, I also want toinvestigate the effects of informative campaigning under the positive presumption, thatthey really are informative. Clearly, if they are cheap talk, they are either akin topersuasive campaigns or just a waste of money.

The organisation of the article is as follows: Section 1 briefly discusses some relatedliterature. Section 2 introduces the model. Equilibrium is defined in Section 3. Section4 derives equilibrium policies and campaign choices. Section 5 discusses welfare, andSection 6 offers some concluding remarks. Most proofs are in the Appendix.

1. Some Related Literature

A number of recent papers argue that parties offer favourable policies to well-informedparts of the electorate. Stromberg (2004a) considers the interplay between informationand redistributive policies. In his model, information is provided by mass media ratherthan by parties. Voters are uncertain about the parties� policies because they are unsureabout the size of the different groups. Thus, mass media have a role informing votersabout the policies of the parties. As in the present article parties will target informedgroups. Stromberg also shows that there will be a bias in the policy, since the massmedia will have an own interest in informing particular groups that are attractive toadvertisers. Stromberg (2004b) considers the impact of the spreading of radio for theallocation of funds under the New Deal in the 1930s in the US. The data show a clearpattern: more informed groups – with a high share of households owning a radio –receive larger funds as predicted by the theory. Although in a different context, Besleyand Burgess (2002) report the same kind of result for India. Indian states with largernewspaper circulation tend to receive more public food distribution and calamity reliefexpenditure. Compared with the papers of Stromberg and Besley and Burgess, theimportant new element in the present article is that the parties choose the distributionof information strategically as part of the electoral struggle.

Informative campaigning is also considered by Austen-Smith (1987) in a model witha uni-dimensional policy space. He considers a model where there is uncertainty aboutparties� policies. Risk adverse voters are therefore reluctant to vote for parties whosepolicies they only have stochastic information about. Informing the voters reduces the

938 [ J U L YT H E E C O N O M I C J O U R N A L


risk. Austen-Smith shows that in equilibrium informative campaigning makes ideo-logical parties choose policies closer to the median voter’s preferred policy. Whilehighly innovative, a problem with the analysis is that the source of the uncertainty isnot explained (in equilibrium, the positions of the parties are actually certain) and it isnot explained how campaigns contribute to reducing the uncertainty. In my model,voters are unsure about the ideological component of parties� preferences and thisgives uncertainty about the policies.

Prat (2002) is interested in whether a ban should be put on lobbies� contributions toparties. He considers informative campaigning in a model where two office-seekingcandidates may have different competencies. An interest group receives non-verifiableinformation about the competency of the candidates. The interest group offers moneyto the parties in return for a favourable policy position. The money is used on non-informative campaigning but a separating equilibrium exists in which voters infer thecandidates� competency from the amount of campaign spending. Campaigns are thusboth good and bad for voters, good since they provide information about the candi-dates� competency and bad since the parties distort policy in order to obtain moneyfrom the interest group. Therefore, whether a ban on campaigns will benefit voters isnot clear. Potters et al. (1997) also discuss signalling through campaigning.

Coate (2004a) offers an interesting analysis of informative campaigns in a model ofa uni-dimensional policy space. Two parties are policy-motivated; they each propose acandidate, which may be moderate or extreme (partisan). Party members prefer apartisan candidate, while moderate voters prefer a moderate candidate. While votersknow a candidate’s party affiliation, they do not know his type. Informative cam-paigning can inform about the characteristic of a candidate. Such campaigningbenefits moderate candidates as it is possible to attract moderate voters from the otherparty. Parties receive contributions from partisan interest groups. Coate shows that alimit on contributions reduces expected campaign spending but it also decrease thelikelihood that parties will select moderate candidates. Coate (2004b) considersinformative campaigning regarding candidates� qualities or competencies. Politics isuni-dimensional and extreme lobby groups are informed about the qualities of can-didates. A trade off arises, since qualified candidates will be able to raise money fromlobby groups and can inform the electorate about their superior quality, which is agood thing. However, the candidates modify their policy in a more extreme direction inorder to cater to the extreme lobby groups, which is a bad thing. Coate shows thatunder some conditions a limit on campaign contributions may then be Paretoimproving. While the negative effects of informative campaigns in Coate’s papersderive from the power of extreme lobby groups, my argument is different. There are nolobby groups in the present model; the unequal distribution, which results frominformative campaigning, derives from the unequal distribution of information thecampaigns induce. Ashworth (2006) studies the effect of incumbency advantages infund-raising in model similar to Coate’s. He finds that public matching funds improvewelfare in districts where parties campaign in the absence of matching funds andreduces welfare in other districts.

Ortuno-Ortin and Schultz (2000) consider informative campaigning in a model of auni-dimensional policy space where policy motivated parties receive public funding forcampaigning. They show that informative campaigns contribute to making the parties�

2007] 939S T R A T E G I C C A M P A I G N S


policies converge since a more informed electorate responds more to policy moder-ation.

All of the above-mentioned papers with informative campaigning consider uni-dimensional policy spaces. Their aim is to consider different legislative regimesregarding campaign finance, in order to discuss the socially best regime. The presentarticle, on the other hand, focuses on the distributional aspects of the effects of stra-tegic informative campaigning. It shows that there are distributional consequences ofinformative campaigning.

Konrad (2004) considers inverse campaigning in a model of redistributive politics,where a party deliberately informs the whole electorate that a particular group is goingto lose from its proposal or benefit from the opponent’s proposal. This may be sensiblein situations where it is common knowledge that a particular fraction of the electoratewill gain from the proposal, but where the identity of the winners and losers isunknown to (parts of) the electorate. An uninformed voter, who just knows that somefraction of voters will win, will revise the probability that she is a winner if she learnsthat particular other groups will lose for sure. In this way, inverse campaigning canindirectly be positive for a party. Since the present article assumes that parties caninform a group directly about its policy through targeted campaigning, the effectKonrad focuses on is not present here.

2. Basics

I consider a society with n groups i ¼ 1, . . ., n. Each group contains a continuum ofcitizens of size 1. Two parties A and B each proposes a distribution of per capita con-sumption before an election, cA ¼ (cA1, . . ., cAn), and cB ¼ (cB1, . . ., cBn), respectively.The proposals have to fulfil the budget constraintX

i

cAi ¼ X: ð1Þ

The parties are committed to these proposals; the winner of the election implements itsproposal.

Voters become informed about policies in a variety of ways: through media, con-versation, attending meetings, etc., and through parties� campaigns. For the model, theimportant feature is that the parties can affect the share of informed voters by cam-paigning. Each party has an amount, M, of campaign money, which can be usedinforming the different groups of voters about the policies.2

2 There is ample evidence to this effect starting with a famous panel study by Lazarsfeld et al. (1944).Alvarez finds (1998, p. 172) that �there are substantial reductions in voter uncertainty of the positions of thecandidates during the campaign that are directly related to the flow of information�. Ansolabehere andIyengar (1995, p. 42) note that �exposure to advertising makes voters much more likely to refer to issues asreasons for supporting or opposing a candidate�. This is echoed by many authors: Popkin (1991) stressesinformation about candidate positions, Gelman and King (1993) stress the role of media and West (1993)stresses the importance of TV-ads. Holbrook (1996) stresses that media’s coverage of the campaigns is mostlyabout campaign events (conventions, debates etc.). So although a lot of campaigning is done throughadvertisements, direct mail, telephone calls and canvassing, the ways campaigns work are manifold. Holbrok(1996) answers his title �Do campaigns matter?�, with a resounding �yes�, both for the electorates� informationand voting. Brians and Wattenberg (1996, p. 172) find that �citizens recalling political advertising have themost accurate knowledge of the candidates� issue positions�. Just et al. (1990) report experiments showing thatpolitical ad viewing conveys more accurate candidate issue positions than televised debates.



I will assume that information is hard so a party cannot lie about policies. Evidently,this is a benchmark case. But as touched upon in the Introduction, it is a benchmarkcase of interest. Political life has its reputations, a politician caught lying will be frontpage news, journalists and independent experts enjoy such prey.3 Furthermore, as alsoargued in the Introduction, I am interested in evaluating informative campaigns on thepresumption that they actually are informative. Voters, who remain uninformed aboutthe policies, rely on their expectations. They are uncertain about the exact policieschosen by the parties but derive whatever inferences can be drawn from the fact thatthey remain uninformed.

I model the uncertainty as originating from an uncertainty about the parties� pre-ferences. The voters may be unsure about which faction of the party is the mostinfluential, how the different factions strike deals within the party and they may also beunsure about leading politicians� preferences. As an example, consider the newlyelected Margaret Thatcher in UK. Everybody knew that she was a right-winger but howstern and how tough she would be for instance in relation to trade unions wasuncertain. Likewise, the newly elected Tony Blair for sure had changed some prioritiesin Labour but how much?

While voters are uncertain about the parties� preferences, it is assumed that bothparties know each other’s preferences. The motivation is that parties are lead by pro-fessional politicians, who spend time and effort to understand their competitors, theyare assisted by full time employees, whose job it is exactly to gather such informationand make the relevant assessments, and they are often in daily contact with leadingpoliticians from other parties. For the leaders of the parties it is an important part ofthe strategic struggle to know the opponent. The incentive for a voter is much smaller(actually zero in our model, where there is a continuum of citizens in each group). Inshort, there is asymmetric information among voters and parties.

In principle, a party can choose to inform about its own policy only, about the otherparty’s policy only, or about both parties� policies. However, rational voters, whoobserve a campaign message from a party, should ask themselves why the party with-holds some information, if it indeed does. I will assume that the cost of reaching a voteris the same regardless of whether the information is about one of both of the parties�policies. The idea is that the (significant part of the) cost of campaigning is associatedwith reaching the voters. A voter therefore knows that the party could provide moreinformation costlessly, if it �hides� information. Assume for instance, that party Ainforms about its own policy only. Had B�s policy been bad, party A would happily haveinformed about it. As there are only two types of party B in the model, there are onlytwo possible equilibrium policies of party B, so a voter seeing A�s campaign will inferthat B�s policy is good. Similarly, it does not help a party only to inform about the otherparty’s policy. Consequently, when a party decides to campaign in a group, it revealsinformation about both parties� policies. Notice that the same unravelling argument

3 A recent Danish experience is the March 1998 general election. Polls showed a very close race betweenthe ruling Social democrats and the right wing opposition. Prime Minister Poul Nyrup Rasmussen made a verypopular pledge not to change an early retirement scheme, which most economists found very unwise. TheSocial democrats won the election and Prime Minister Nyrup reneged on his promise half a year later,changing the early retirement scheme. He immediately made front page news as a liar; the Social democratsplummeted in the polls and never regained the loss. They lost the next election in 2001.



applies even if there are more types of parties. A party would always want to reveal themost positive information, if it does not voters will exclude this case etc.

In the end a party therefore only has two options: inform about both parties� policiesor not inform at all, in this sense a party’s campaign is both positive (�my policy isgood�) and negative (�his policy is bad�).

Some voters receive information from newspapers, television etc. regardless ofwhether the parties campaign or not. Let the fraction of voters in group i, who areinformed by other means than campaigns, be Ci, where 0 � Ci < 1. This fraction maydiffer across groups, a highly educated group would presumably have a high Ci.

Following the literature on informative advertising (Butters, 1977; Grossman andShapiro, 1984) it is assumed that the fraction of informed voters is increasing in theamount of campaigning. If party A spends mAi on campaigns in group i, then thefraction

wi mAið Þ ¼ ciw mAið Þ

of the uninformed voters in group i will see A’s campaign, where ci > 0, w0> 0, w

0(0) ¼

1, w0 0< 0, and ciw(M) < 1. This in accordance with the �ceiling effect� discussed by

Bartels (1988). He notes that there is evidence that campaign effects are not linearlycorrelated with exposure to campaign messages but that the marginal effect declines.Some groups may be easier to inform than others, for a discussion of this see Zaller(1991). A group with a high ci is a group, which is easy to reach through campaigning;perhaps because it is well educated, watches a lot of TV or reads a particular magazine.In principle one could also imagine that the wi function depends on the party; it may bethat a party is better to inform a particular group than the other party. I will howeverdisregard such complications. I assume that the probability that an individual voterbecomes informed by party A�s campaign is the same for all uninformed voters in groupi and therefore equal to wi(mAi). Similarly, a voter becomes informed by party B �scampaign with probability wi(mBi

). Let mA ¼ (mA1,. . .,mAN).For simplicity, I assume that the events that an uninformed voter becomes informed

from A or from B are independent. The fraction of informed voters in group i there-fore becomes

Wi ¼ fwi mAið Þ þ ½1� wi mAið Þ�wi mBið Þg 1� Cið Þ þ Ci : ð2Þ

If consumption in group i is ci, a voter in the group derives utility

uðciÞ ¼ lnci : ð3Þ

Voters may be ideologically attached to one of the parties. Voter v in group i has aideological component in favour of party B, equal to xiv þ y, xiv is a personal compo-nent, y a general component shared by all voters in the electorate. It represents acommon shock to preferences in the electorate occurring just before the election,which may be positive or negative.4 When voter v is informed about the parties� policies,she prefers party A if

4 In the recent (2002) German election, the Social democratic Kansler, Schroder, gained support in pollsjust before the election following severe flooding in southern Germany. Allegedly, Schroder showed lead-ership. In the model this would be good realisation of y for the Social democrats.



xiv þ y � uðcAiÞ � uðcBiÞ: ð4Þ

There is a distribution of ideological components in each group. In group i, theideological component is given by the distribution function Ui. To keep the analysissimple, I follow Persson and Tabellini (2000) and assume that Ui is the uniformdistribution, with support [�1/2di, 1/2di] and density di. This restriction is in-essential, as the results by Lindbeck and Weibull (1987) show. It will save us someconcavity assumptions later. The fact that there is a zero mean is also inessential aslong as the support is sufficiently large so that u(cAi) � u(cBi) (or the expectedcounterpart considered below) is contained in the support of Ui. As I show below, itis only the density that matters. The general ideological component is alsouniformly distributed, with mean 0 and density h. The parties and the voters areinformed about the distribution of the ideological components, but the parties donot know the realisation of the general component y before choosing policies andcampaign expenditures.

A voter who is not informed about the parties� policies will form expectations. Herexpected utility from consumption if party A wins the election is denoted EAui. Whenvoter v is uninformed, she will prefer party A if

xiv þ y � EAui � EBui : ð5Þ

Informed and uninformed voters thus cast their vote differently, viz. (4) and (5).5

Voters may be more or less inclined to actually go to the voting booth and cast a vote.The inclination to vote may differ across groups. Let ni be the fraction of voters ingroup i, who actually cast a vote at the election, the voter turnout. For simplicity, I firstassume that voter turnout is exogenous, the same among informed and uninformedvoters, and independent of the ideology of particular voters. If informed voters weremore inclined to vote, the vote response from informing voters would be larger thanwhat is the case in the model at present. Later I will consider the case where voterturnout is responsive to campaigning.

We can now find the votes for party A in district i

Vi ¼ ni WiUi ½uðcAiÞ � uðcBiÞ � y� þ 1�Wið ÞUi EAui � EBui � yð Þf g: ð6Þ

Since the individual ideological component is uniformly distributed with density di andmean 0, we have

Vi ¼ ni1

2þ di Wi uðcAiÞ � uðcBiÞ½ � þ 1�Wið Þ EAui � EBuið Þ � yf g

� �: ð7Þ

The total number of votes cast for party A is therefore

5 This is in accordance with the results of Bartels (1996). Using probit analysis on data from US NationalElection Study surveys from six presidential elections, he concludes that individual voting differs significantlydepending on whether the individual is informed or not. Aggregation does not eliminate this effect:according to his estimates, incumbents did almost five percentage points better than in his hypothetical �fullyinformed� electorate. Shaw (1999) estimates campaign effects from an extensive data set on state-wide tele-vision advertising and candidate appearances from the 1988, 1992 and 1996 presidential campaigns in the USand concludes that campaigns influence electorates and electoral votes.



VA ¼Xn

i¼1

Vi : ð8Þ

The probability party A wins the election, pA, equals the probability that VA � Rini=2.This depends on the realisation of the general shock to preferences, y. Since y isuniformly distributed with mean zero and density h, we have after a few manipulations

pA ¼1

2þ h

d

Xn

i¼1

nidi Wi uðcAiÞ � uðcBiÞ½ � þ 1�Wið Þ EAui � EBuið Þf g ð9Þ

where d �Pn

i¼1 nidi , i.e. sum of densities weighted by voter turnout. In each groupnidi is a measure of the �effective� mobility of votes, i.e. the increase in votes for aparty if it improves the policy of the group so much that a voter in the group gains oneutil.

The parties are partly motivated by power and partly by ideology. If a party wins theelection it receives rents R. Furthermore, the party cares about a weighted average ofvoters� utilities in the different groups. The uncertainty of uninformed voters stemsfrom that they do not know how the parties weigh the utility of different groups.There are two types of each party. One type is neutral, type N, and weighs all groupsequally, while the other is biased towards particular groups and put larger weights onthe utility of these groups, which we could think of as the party’s constituency.6 I callthis type, the biased type P. When party A is of type t 2 fN, Pg, its objective is tomaximise

U tA ¼ pAR þ

Xn

i¼1

ati ½pAuðcAiÞ þ 1� pAð Þu cBið Þ�: ð10Þ

The party cares for the rents of office and a weighted sum of the expected utility ofdifferent groups. I normalise so that the average weight

Pni¼1 at

i=n ¼ 1. The neutraltype of party A has aN

i ¼ 1 for all i, while the biased type weighs the groups differentlyas explained in detail below.

Party B of type t seeks to maximise

U tB ¼ 1� pAð ÞR þ

Xn

i¼1

bti ½pAuðcAiÞ þ 1� pAð Þu cBið Þ�: ð11Þ

AgainP

i bti=n ¼ 1, and the neutral type has bN

i ¼ 1 for all i.In order to facilitate the analysis I focus attention on a symmetric economy. Group i

(i < n/2) has the same characteristics as group n þ 1 � i, i.e. ci ¼ cnþ1�i,ni ¼ nnþ1�i, Ci ¼ Cnþ1�i and di ¼ dnþ1�i. The biased parties� preferences are sym-metric, in the sense that aP

i ¼ bPnþ1�i . The symmetry implies that when both parties

are of the same type they are in fact in a symmetric situation. The biased type of aparty attaches a weight larger than one to some groups, one can think as those asthe party’s constituency. Party A�s constituency consists of groups i ¼ 1, . . . , na � n/2

6 Of course, it would be more realistic to assume that there were infinitely many types of a party corres-ponding to varying degrees of bias. However, this would also complicate the analysis to follow; we follow thelead of a large literature and restrict attention to the two-type case.



and party B�s of groups i ¼ n þ 1 � na, . . ., n. The constituencies are thereforenon-overlapping. The parties attach a weight less than one to the groups outside theirconstituencies.

I assume that either both parties are biased or they are both neutral. An inter-pretation is that some voters are unsure about how partisan politics really is. Theprobability that the parties are biased and attached to their constituencies is p, where0 < p < 1. Voters know the different ai and bi and the probability p, but they do notknow parties� types. Alternatively one could have assumed that it was independentevents whether the parties were biased or not. Then there would be four possibleconfigurations of types and this would make the formulas longer and some proofsmore intricate without adding qualitatively to the article. The important issue is thatvoters in the different groups are unsure about which party proposes the best policyfor them, so that campaigning is potentially important. This feature is also secured inour formulation.

The timeline is as follows. First the parties� types are realised and observed by bothparties. Parties then decide on policies and the distribution of campaign expendituressimultaneously. Some voters receive information about the policies and some do not.The uninformed voters form expectations about the policies and voters vote. Finally,the winning party implements the policy it proposed.

3. Equilibrium

I solve the model for a Perfect Bayesian Equilibrium. In this equilibrium each party (ofeach type) maximises its utility. A voter takes as given the equilibrium strategies of theparties and prefers party A, if she is informed and (4) is fulfilled, or she is uninformedand (5) is fulfilled. Otherwise she prefers B. Let, for example, mt

Ai be the amount ofcampaign money party A spends informing voters in group i, when parties are of type t.An uninformed voter rationally derives whatever information is contained in the factthat she is uninformed. For instance, if the biased type of party B favours her group, shemay think that since she was not informed by party B, party B is not favouring her groupafter all and is of the neutral type. Formally, she takes the campaign strategies as givenand uses them to update her prior belief using Bayes� rule. Her posterior belief thatparties are biased is therefore

�pi ¼1� wi mP

Ai

� �� 1� wi mP

Bi

� �� p

1� wi mNAi

� �� 1� wi mN

Bi

� �� 1� pð Þ þ 1� wi mP

Ai

� �� 1� wi mP

Bi

� �� p: ð12Þ

This is the probability of being uninformed when parties are biased, divided with thetotal probability of being uninformed. As wi(M) < 1, I always have that 0 < �pi < 1.Let, for example, the policy of party A of type t be denoted ct

A. With posterior belief �pi ,the expected utility to a voter in group i if party A wins the election is

EAui ¼ 1� �pið Þu cNAi

� �þ �piu cP

Ai

� �: ð13Þ

The expected utility if party B wins the election is

EBui ¼ 1� �pið Þu cNBi

� �þ �piu cP

Bi

� �: ð14Þ



Since voters do not observe the distributions of the parties� campaign expenditures,the parties take the uninformed voters� beliefs as given. Therefore they also take theexpected utilities in (13) and (14) as given.7

In equilibrium parties also take each others� strategies as given. Party A of type t’sproblem is:

Given t and �pi ; ctBi ;m

tBi

� �n

i¼1

maxcA ;mApAR þ

Xn

i¼1

ati pAuðcAiÞ þ 1� pAð Þu ct

Bi

� �� sub

Xi

cAi ¼ X;X

i

mAi ¼ M ; cAi ;mAi � 0;

ð15Þ

while B�s is:

Given t and ð�pi ; ctAi ;m

tAiÞ

ni¼1

maxcB ;mB1� pAð ÞR þ

Xn

i¼1

bti pAuðct

AiÞ þ 1� pAð Þu cBið Þ� �

subX

i

cBi ¼ X;X

i

mBi ¼ M ; cBi ;mBi � 0:

ð16Þ

I can now define an equilibrium.

Definition. An electoral equilibrium with strategic campaigns consists of policies and cam-paign expenditures in all groups for both parties ðct

Ai;mtAiÞ

ni¼1, ct

Bi;mtBi

� �ni¼1; t 2 fN ;Pg and

expected utilities for the voters of all groups EAui;EBuið Þni¼1 such that

1 Each party (of each type) chooses policies and campaign expenditures to maximise utility asin (15) and (16).

2 Each uninformed voter forms beliefs using the parties� strategies and Bayes� rule andevaluates expected utilities using these beliefs. For instance EAui is given by (13), wherethe beliefs are given by (12). The number of votes cast for party A is then given by (8).

Applying a standard fixed point argument, I show in the Appendix

Lemma 1. An equilibrium exists.

4. Campaigns and Policies

I now consider equilibrium choices of campaigns and policies. For party A of type t thefirst order conditions for an interior maximum are

@pA

@cAfR þ

Xn

i¼1

ati ½uðct

AiÞ � uðctBiÞ�g þ pAat

iu0ðct

AiÞ ¼ k ð17Þ

@pA

@mAifR þ

Xn

i¼1

ati ½uðct

AiÞ � uðctBiÞ�g ¼ l ð18Þ

7 If voters were able to observe the campaign expenditures, a party would take into account that it couldaffect beliefs through the distribution of expenditures.



where k and l are Lagrange multipliers. The amount of campaign spending is non-negative. If the non-negativity constraint binds in a group i, then the correspondingderivative, opA/omAi, is negative and the optimal choice is mAi ¼ 0.

Using (9), (3) and the budget constraint I get

ctAi ¼

Winiðdi=dÞhGtA þ pAat

iPnj¼1½Wjnjðdi=dÞhGt

A þ pAatj �

X: ð19Þ

where

GtA � R þ

Xn

i¼1

ati ln

ctAi

ctBi

� �ð20Þ

is the gain from winning for party A of type t. Party B�s policy is given by

ctBi ¼

Winiðdi=dÞhGtB þ 1� pAð Þbt

iPnj¼1½Wjnjðdi=dÞhGt

B þ 1� pAð Þbtj �

X; ð21Þ

where

GtB � R þ

Xn

i¼1

bti ln

ctBi

ctAi

� �: ð22Þ

Although I have not solved for an equilibrium yet, (19) and (21) convey a nice insight.As in the models of Lindbeck and Weibull (1987) and Dixit and Londregan (1996), theparties favour a district, where the density of voters at the cutpoint, di, is high and wherethe inclination to vote, ni, is high. Shifting resources to such a district will give manyvotes. However, for this effect to work, voters have to be informed about the parties�policies, Wi has to be high. The term Wi(nidi/d)h equals the expected increase in votesfrom offering the group a policy which is one util better, so Wiðnidi=dÞhGt

A is theincrease in expected gain to the party from offering such a better policy. Groups withmany informed voters will gain; this is just as in Stromberg (2002b).8

When both parties are neutral they both weigh all groups equally. When they arebiased, they are also in a symmetric situation. A symmetric equilibrium therefore existsin which, cP

Ai ¼ cPBnþ1�i , cN

Ai ¼ cNBi (for all i), they win with probability 1

2 each, and wherethe gain from winning is the same for the parties, equal to R when they are neutral andG (¼ GP

A ¼ GPB Þ when they are partisan. In such an equilibrium the policies of the

neutral types are given by

cNAi¼ cN

Bi¼

Winiðdi=dÞhR þ 12Pn

j¼1½Wjnjðdi=dÞhR þ 12�X; ð23Þ

and the policies of the biased types are given by

8 As touched upon in the Introduction, Dahlberg and Johansson (2002) find support for the Lindbeck andWeibull (1987) model from Swedish data. Milligan and Smart (2003) consider a federal programme forinterregional development transfers distributed by two government agencies covering different parts ofCanada. For the Atlantic regions, the pattern of spending accords with the predictions of Lindbeck-Weibull,spending is larger in electoral districts with close races and where plurality in the previous vote was small.Cadot et al. (2002) find that the regional allocation of spending on roads in France also shows such a pattern.Wright (1974) reports that US federal spending in the period 1933–40 was positively related to a state’s�political productivity�, which is a measure depending on the number of electoral votes per capita, variability inthe incumbents vote share in past elections, and closeness of presidential elections.



ctAi ¼

Winiðdi=dÞhG þ 12 at

iPnj¼1½Wjnjðdi=dÞhG þ 1

2 atj �X and ct

Bi ¼Winiðdi=dÞhG þ 1

2 btiPn

j¼1½Wjnjðdi=dÞhG þ 12 bt

j �X: ð24Þ

In the Appendix it is verified, that this constitutes an equilibrium. I will focus on thisequilibrium. Here I have that

cPAi

cPBi

¼Winiðdi=dÞhG þ 1

2 aPi

Winiðdi=dÞhG þ 12 bP

i

: ð25Þ

From which it is clear that

aPi > bP

i , cPAi > cP

Bi : ð26Þ

If party A is more concerned with group i than party B is, it offers the group a betterpolicy than B does. Party A is interested in revealing that its policy is better for thevoters of group i. In fact, I have

Lemma 2. At most one party campaigns in a group. This is the party whose policy is relativelybetter than expected for the group’s voters. If parties are biased, then A campaigns in groups whereaP

i > bPi and B campaigns in groups where aP

i < bPi . If parties are neutral, A campaigns

where aPi < bP

i and B where aPi > bP

i . No campaigning occurs in groups where aPi ¼ bP

i .

All proofs, which are not in the text, are in the Appendix.There will be campaigning in all groups where the uninformed voters are uncertain

about the policies, and no campaigning where there are not. Parties will campaignwhere they have positive news to reveal. A biased type campaigns in the groups itfavours, and a neutral type campaigns in groups, whom its biased type would disfavour.

I now show that there is an equilibrium, where the amount of campaigning, andtherefore the information level of a group is independent of types of the parties. In thisequilibrium, the posterior belief of the uninformed voters in all groups thereforeequals the prior.

Lemma 3. There exists an equilibrium, where the posterior belief in all groups equals the prior.In this equilibrium, the identity of the party who campaigns in a group depends on the type of theparties, but the amount of campaigning in a group is the same independent of the types.

The key to the Lemma is that the parties are in a symmetric situation. If parties arebiased and aP

i > bPi , then party A has the incentive to campaign in group i as A�s

policy compares more favourably with B�s policy than expected. The incentivedepends on the size of the positive �utility surprise� in group i compared with othergroups j, where A campaigns. If, on the other hand, parties are neutral, party Bcampaigns in group i (and the other groups biased A would campaign in). B�sincentive is also governed by the size of the surprise in group i compared with theother groups, and this comparison is the same as when biased A campaigns. The onlydifference is the sign. Therefore the two parties have equally strong incentives tocampaign in group i, and the amount of campaign spending is independent of types.Because of this there is no new information in being uninformed. Hence, an unin-formed voter does not revise her prior.



In what follows I focus on the equilibrium of Lemma 3, where the posteriorequals the prior. In this equilibrium, the parties do no influence the beliefs ofthe uninformed voters. This feature distinguishes it from other potential equili-bria in the model where the Bayesian updating using the parties� strategies wouldlead to different beliefs. The fact that beliefs stay at the prior also makes forsimplicity.

Let mi denote the total level of campaigning in group i. We now have

Proposition 1. The level of campaigning and information is larger in groups, which aremobile, receptive, have high voter turnout, where the uncertainty about the parties� preferences islarge, and where there are many initially uninformed voters.

The parties campaign most intensively where many votes can be gained. This is ingroups where the uncertainty is large and the policy surprise therefore large, in groupswhere voters are receptive, so that the informational effect of a given amount ofcampaigning is large, in mobile groups and in groups with high voter turnout where agiven surprise changes many voters� vote.9

Groups where the initial information level is low are also attractive to campaign in.The party which offers the group a relatively good policy will want to inform about itand campaigns are effective, since there are so many uninformed voters. The incentivesto campaign in such a group are therefore strong.

In the model Ci and ci are independent parameters. If, however, it is the case thatvery well informed groups are hard to inform further, so that a high Ci goes hand inhand with a low ci, then groups with low Ci will, of course, experience even morecampaigning.

In order to assess which groups gain from campaigning, I use as a point of referencethe case where there are no campaigns. Then Wi ¼ Ci for all groups i. Let ~ct

Ai denotethe policy party A offers group i if parties are of type t.

From (23) and (24), we know that the consumption level offered to a group isincreasing in its information level. Combining this with Proposition 1 directly gives

Theorem 1. Groups which are mobile, receptive, initially relatively uninformed, where theuncertainty about the parties� preferences is large, or have high voter turnout gain from strategiccampaigning.

The groups where campaigns attract many votes will be targeted by the parties andthis makes these groups well-informed, which in turn makes the parties offer themfavourable policies.

Lindbeck and Weibull (1987) showed that mobile groups would gain, since they offereda solid vote response to better policies. As is clear from Theorem 1, this effect is reinforcedby informative campaigning. Groups with a solid vote response are interesting targets forcampaigning. The campaigning makes these groups better informed and therefore even

9 These results are in line with the empirical results of Nagler and Leighley (1992), who investigatecampaign expenditures in different states in 1972. They find these to be higher in states with close races andmore voters.



more attractive for the parties to cater to. There is a complementarity between cam-paigning and spending, which gives rise to a campaign-multiplier.

The strategic distortion identified by Lindbeck and Weibull is thus aggravated byinformative campaigning. From an egalitarian perspective, this feature of informativecampaigning is not good and a reduction in campaign funds would improve on thesituation. A social planner, maximising the sum of utility in society, would find thisaspect of campaigning as welfare decreasing. Clearly, as the model is one of pureredistribution all allocations are Pareto Optimal.

The evaluation of campaigns may change when we consider asymmetries in initialinformation. Groups, who are initially relatively uninformed, will experience morecampaigning and their level of information is thus relatively much increased, whichmake the parties offer them better policies. In this respect campaigning may lead to amore even distribution in equilibrium, which is beneficial from a welfare perspective.In the next Section, where we consider welfare more in detail in a tractable example, itis shown that this may indeed be the case.

An important objective of real world campaigning, which is not included in themodel at present, is that campaigns, beside being informative, are often directed atmaking people vote.10 It is straightforward to include such an effect in the model.Suppose that turnout depends positively on campaigns and that the effect differs acrossgroups so that

ni ¼ jin mð Þ;

where n0(m) > 0 and n

0 0(m) < 0 and ji > 0 is a parameter representing the group’s

responsiveness. A group with a high turnout response to campaigning has a high ji. Forthe model to be well behaved it is necessary that the total response to morecampaigning in a group is concave. A sufficient condition ensuring this in all groups isthat

@2½wðmÞnðmÞ�@m2

< 0: ð27Þ

Theorem 2. Suppose condition (27) is fulfilled. Groups where turnout is more responsive tocampaigning benefit from strategic campaigning.

If campaigns not only inform but also induce people to vote, groups witha responsive turnout win. This conforms to the general principle behind allresults of the model. The more responsive a group is in a broad sense, the more itwins.

10 Rekkas (2002) estimate a random coefficients discrete choice model using Canadian elections data. Herempirical results confirm the importance of campaign expenditure for the outcome of elections. Interest-ingly, the own-and cross-expenditure vote share elasticities show that political campaign spending not onlyredistributes voters across parties but also decreases the size of the abstaining group of the electorate.Matsusaka and Palda (1999) also find that campaign spending increases voter turnout. On a similar note,Gerber and Green (2000) report that in a field experiment, voter turnout was increased substantially bypersonal canvassing.



5. Welfare

An assessment of the welfare consequences of campaigning is difficult, since the modeldoes not allow closed form solutions.11 In this Section I investigate the special casewhere biased types of parties have an almost egalitarian ideology, so that aP

i ¼ 1 þ efor all groups i ¼ 1, .., na in A�s constituency and aP

i ¼ 1 � ðna=nÞ e in the rest.Similarly bP

i ¼ 1 þ e for groups in B�s constituency and bPi ¼ 1 � ðna=nÞ e in the rest.

I look at the limit as e ! 0, here the parties have no particular preference for anygroup, so differences in policies across groups are due to the parties� strategic concerns.I assume that w mið Þ ¼

ffiffiffiffiffimip

andffiffiffiffiffiMp

� max1=ci so

Wi mið Þ ¼ ci

ffiffiffiffiffimip

1� Cið Þ þ Ci : ð28Þ

So far I have not discussed where the campaign resources stem from. In the realworld they stem from private donations and public funding of political parties. I willhere – for simplicity – consider the case where the campaigns are publicly financed sothat when each party receives campaign funds equal to M, the resources left for dis-tribution among the groups are X � 2M.

The level of consumption offered by party A of type P in district i is given by

cPAi¼

Winidiðh=dÞG þ 12aP

iPnj¼1 Wjnjdjðh=dÞG þ 1

2aPi

� � X� 2Mð Þ!Winidiðh=dÞR þ 1

2Pnj¼1 Wjnjdjðh=dÞR þ n

2

X� 2Mð Þ for e! 0:

In fact, as e ! 0: cPAi! cN

Ai¼ cN

Biand cP

Bi! cN

Ai¼ cN

Bias can be seen from (23) and (24). In

the limit as e ! 0, both parties offer the same policies to a particular group. The gain fromwinning therefore approaches R as e ! 0. Let ti � ½ci

ffiffiffiffiffimip

1 � Cið Þ þ Ci �nidiðh=dÞR þ 12,

then

ci ¼tiPnj¼1 tj

X� 2Mð Þ: ð29Þ

Total welfare equals

W ¼Xn

i¼1

ln ci :

Letting t 0i � dti=dM the marginal welfare effect of increasing M can be written as

dW

dM¼Xn

i¼1

t 0iti� n

Pnj¼1 t 0jPnj¼1 tj

� 22

X� 2Mð30Þ

The amount of available resources, X, and campaign resources, M, are important forthe results. If X is low and M high, the welfare effects of increasing M are bound to benegative. In the following I will assume that X is large and M relatively small so that Ican disregard the term �2[2/(X � 2M)]. Evidently, when I find a negative welfareeffect even disregarding this term, the total effect is more negative. When I find a

11 From a technical point of view, the problem is that the expressions for the policies of the biased types(24) are not on closed form, as G depends on the policies, cf (20). Similarly, there are no closed formexpressions for campaign spending.



positive effect disregarding this term, the total effect is positive provided X is suffi-ciently large and M sufficiently small. Disregarding �2[2/(X � 2M)] (30) gives thatthe welfare effect of increasing the campaign resources is positive if and only if

Xn

i¼1

t 0iti� n

Pnj¼1 t 0jPnj¼1 tj

> 0 ð31Þ

Using this condition I find

Proposition 2. When the biased parties are almost egalitarian

– if groups only differ in voter turnout and/or in mobility (nidi 6¼ njdj),– or if groups only differ in receptiveness (ci 6¼ cj), then an increase in campaign resources

M decreases welfare.– If groups only differ in initial level of information (Ci 6¼ Cj) and M is sufficiently small,

then an increase in campaign resources M increases welfare.– If groups only differ in initial level of information (Ci 6¼ Cj) and M is sufficiently large,

then it depends on parameter values whether an increase in campaign resources Mincreases or decreases welfare.

Campaigns decrease welfare, when the important difference between the groupsstems from difference in turn out (n), density of the swing voters (d), or receptivenessfor campaigns. In these cases, campaigns reinforce the skewed distribution alreadypresent in the original Lindbeck and Weibull model – the parties� strategic concernsinduce them to exploit the campaign-multiplier, and this decreases overall welfare.This is so even though the biased types are almost egalitarian in their ideology.

Campaigns increase welfare when some groups have a low fraction of initiallyinformed voters and campaign resources are sufficiently small. The low informationlevel makes these groups an interesting target for campaigns, and in equilibriumthe groups gain and total welfare is increased. When campaign resources becomeabundant a welfare assessment is more complicated and it depends on parametervalues whether an increase in M increases or decreases welfare. Welfare effects can benegative for high M since the initially uninformed groups are so attractive to campaignin that their equilibrium share of consumption exceeds that of the initially wellinformed groups. Increasing M further will benefit the initially uninformed furtherand make the distribution even more skewed.

When M becomes large enough, the conditionffiffiffiffiffiMp

� max ci is violated. Consider aslight change in w, so that

wðmÞ ¼ffiffiffiffimp

if m � 1=c2

1 if m > 1=c2

Suppose groups only differ in Ci but are almost alike, so that half of each party’sconstituency has Ci ¼ CI and the other half Ci ¼ CI þ g, where g > 0 is small (g ! 0).In the Appendix I show that increasing M decreases welfare for all R, n, d, C, c, if M issufficiently large, i.e. when

c

ffiffiffiffiffiM

n

r>

1

2ffiffiffi2p � 0:35:



One may ask, which policies would be Pareto improving? If the positive welfare effectsof campaigns dominate, it obviously helps to give public funding to parties� campaigns.If the negative effects dominate, this is not a good idea and a cap on campaignresources would improve welfare. Ideally, one would want to increase the informationlevel in groups which are disadvantaged in the distributional game, but it is not clearwho should implement this policy; the parties have a different agenda. If, however, itwas somehow done, the parties would react. If information were increased in a par-ticular group, the parties would partly offset this by campaigning less in the group.

One could imagine general laws stipulating that public campaigns should be direc-ted at increasing the information level in all groups. Regardless of whether it would infact be implementable, it may give negative results. For example one finds in a numericexample with four groups where na ¼ 2 and where R ¼ 1, h ¼ 1, n1 ¼ 0.6 n2 ¼ 0.5 andd1 ¼ d2 ¼ c1 ¼ c2 ¼ 1, that increasing C1 ¼ C2 from 0.5 to 0.7 decreases welfare.

6. Concluding Remarks

Strategic informative campaigning benefits groups that are responsive to campaigning.The reasons for this responsiveness may be that there are many swing voters; manyvoters, who listen to campaigns; many voters, who actually cast a ballot and whoseturnout increases much in response to campaigning. Groups where uncertainty aboutthe parties� preferences and thus policies are particularly large also offer a solid voteresponse to campaigns. So do groups with relatively many initially uninformed voters.Furthermore, the parties will never direct informative campaigns at the same groups.Parties want to reveal the positive message that their policy is better than expectedcompared with the rival’s and, in the nature of things, only one party has somethingpositive to reveal to a particular group.

Groups, who are the subject of intensive campaigning, end up being very well--informed. Parties wish to offer attractive policies to such groups as many votes can begained this way. In equilibrium, such groups therefore gain from the strategic cam-paigning. From a welfare point of view, this may be good or bad. If some groups aredisadvantaged because of few initially informed voters, campaigns may improve welfare.However, if the main differences between groups are in turnout, swing voters andreceptiveness to campaigns, campaigns reinforce a skewed distribution and decreasewelfare.

In the particular setting of our model, the uncertainty is about how much the partiesfavour different groups. Groups where this uncertainty is large gain. However, theuncertainty may concern other issues such as the size of the group or the distribution ofideological slant in the group. In these cases, it will still be true that groups where theuncertainty is particularly large will benefit from strategic campaigning. These aretestable predictions of the model: informative campaigns are intense in groups wherethere is large uncertainty about policies; they are intense in groups, which containmany mobile voters who are inclined to vote, and in groups which are initially relativelyuninformed and who are receptive, and parties will never use informative campaigns inthe same groups. Furthermore, groups which are subject to intense campaigning gainin the distributional conflict.



The model does not include organised lobbies. Since strategic campaigning willbenefit particular groups, these groups will have a stronger incentive to form lobbygroups and support parties with campaign finance. It will be a subject of furtherresearch to investigate the interplay between lobbying and strategic campaigning.

Informative campaigns are often considered good since they allow voters to makemore informed choices. One may value this feature per se, and one may also conceive ofkinds of imperfect information, where more informed choices certainly are prefer-able.12 In the present case, where politics concern redistribution among heterogeneousgroups, campaigns may enhance welfare, but they may also easily end up aggravatingthe skewness resulting from the strategic behaviour of the parties and then they do notenhance welfare.

Appendix

Proof of Lemma 1Since u is strictly concave and Ui is the uniform distribution, it follows from (6), (13), (14), (11)and (10) that the objective functions of the parties are strictly concave and continuous in theparties’ choice variables and continuous in the vector of beliefs �p � ð�piÞni¼1. Furthermore, thechoice sets of the parties’ are strictly convex and compact. Hence the vector of optimal choices ofthe parties given by (15) and (16), r � cN

A ; mNA

� �; cN

B ; mNB

� �; cP

A ; mPA

� �; cP

B ; mPB

� �� , is a con-

tinuous function of ð�p; rÞ. Call this function s(Æ). I also have that the vector �p is a continuousfunction of r, which takes values in a convex compact set, call this function p(Æ).

Now consider the compound mapping r ! r0 defined by

r0 ¼ s pðrÞ; r½ �

This is a continuous function mapping a compact, convex set into itself, hence it has a fixedpoint. This fixed point is an equilibrium. n

Proof of the existence of a symmetrical equilibriumWe need to check that ct

Ai ¼ ctBnþ1�i , pA ¼ 1

2 and GPA ¼ GP

B . From (24) I have that

ctAi ¼

Winiðdi=dÞhG þ 12 at

iPnj¼1 Wjnjðdj=dÞhG þ 1

2 atj

h iX ¼Wnþ1�innþ1�iðdnþ1�i=dÞhG þ 1

2 btnþ1�iPn

j¼1 Wjnjðdj=dÞhG þ 12 bt

j

h i X ¼ ctBnþ1�i

and using (20) and (22) we get

GPA ¼ R þ

Xn

i¼1

aPi ln

cPAi

cPBi

� �" #¼ R þ

Xn

i¼1

bPnþ1�i ln

cPBnþ1�i

cPAnþ1�i

!" #

¼ R þXn

i¼1

bPi ln

cPBi

cPAi

� �" #¼ GP

B ¼ G :

Inserting into (9) it is easily checked that indeed pA ¼ 12 when cP

Ai ¼ cPBnþ1�i ; and cN

Ai ¼ cNBi .

Hence the symmetric equilibrium exists. n

Proof of Lemma 2If t ¼ P, the difference between the actual and the expected utility difference in group i is

12 See for instance Schultz (2002) for the case where politics is uni-dimensional.



uðcPAiÞ � uðcP

BiÞ � EAui � EBuið Þ ¼ 1� �pið Þ uðcPAiÞ � uiðcP

BiÞ� �

; ð32Þ

and if t ¼ N it is

uiðcNAiÞ � uðcN

BiÞ � EAui � EBuið Þ ¼ ��pi uiðcPAiÞ � uðcP

BiÞ� �

: ð33Þ

Using (9) and (18), the first order condition for an interior choice of campaign expendituresfor party A of type P in group i, mP

Ai , can be written

@Wi

@mAinidi

hd

GPA uðcP

AiÞ � uðcPBiÞ � EBui � EAuið Þ

� �¼ l: ð34Þ

If the left-hand side is negative, the non-negativity constraint binds and mPAi equals zero.

Using (19), (21) and (32), I can rewrite (34) as

@Wi

@mAinidi

hd

1� �pið Þ ln cPAi

cPBi

� �GP

A ¼ l: ð35Þ

Hence, a biased party A, only campaigns in groups where where cPAi > cP

Bi . Recall that oWi/omAi ¼ [1 � wi(mBi)]owi/omAi, and [1 � wi(mBi)] > 0. Furthermore, w0(0) ¼ 1, so party Acampaigns in all groups where cP

Ai > cPBi . From (26) these are groups where aP

i > bPi :

Party B ’s first order condition, when it is biased is

@Wi

@mBidini

hd

1� �pið Þ ln cPBi

cPAi

� �GP

B ¼ l: ð36Þ

Party B campaigns where cPBi > cP

Ai :When the parties are neutral, they propose the same policies, and the first order condition in

groups where A campaigns is

� @Wi

@mAinidi

hd

�pi lncP

Ai

cPBi

� �R ¼ l: ð37Þ

The left hand side is positive in groups where cPBi > cP

Ai , i.e. where A’s policy would have beenworse than B’s if the parties were biased.

Finally, party B’s first order condition, when parties are neutral is

� @Wi

@mBinidi

hd

�pi lncP

Bi

cPAi

� �R ¼ l: ð38Þ

We see that B then campaigns in groups where cPBi < cP

Ai .

Proof of Lemma 3Consider a group where aP

i > bPi . Here party A campaigns if t ¼ P and B campaigns if t ¼ N.

Using (12) the posterior belief for an uninformed voter in the group is

�pi ¼1� wi mP

Ai

� �� p

1� wi mPAi

� �� pþ 1� wi mN

Bi

� �� 1� pð Þ

: ð39Þ

In a group where aPi < bP

i B campaigns when parties are biased, and A when they are not, andthe posterior belief is therefore



�pi ¼1� wi mP

Bi

� �� p

1� wi mNAi

� �� 1� pð Þ þ 1� wi mP

Bi

� �� p: ð40Þ

Suppose that �pi ¼ p for all i. Consider a group i in B’s constituency, where bPi > 1 > aP

i . Ifparty B is biased, party B campaigns in this group and the campaign expenditure is determinedby the (set of) first order conditions (36). If on the other hand t ¼ N, then party A campaigns inthe group and the campaign expenditure is determined by the set of first order conditions (37).When �pi ¼ p for all i, then these two set of equations are equivalent and the solutions mP

Bi

� �ijbi>1

and mNAi

� �ijbi>1

are the same. Hence, mPBi ¼ mN

Ai . Using (40) we see that in this case it is indeedtrue that �pi ¼ p is a solution to the Bayesian updating of the voters. Clearly, the same argumentcan be applied for all groups. In groups where aP

i ¼ bPi , there is no campaigning so trivially,

�pi ¼ p. n

Proof of Proposition 1I will focus on groups in which party A campaigns when the parties are biased. By symmetry thesame results are true in groups where B campaigns when parties are biased. Lemma 3 implies thatI can write the first order condition (35) of party A of type P in a group where it campaigns as

ci

@w@mAi

1� Cið Þhnidi 1� pð Þd

lnWiðnidi=dÞhG þ aP

i

Wiðnidi=dÞhG þ bPi

¼ l ð41Þ

and there are similar expressions for the other first order conditions. Notice that p is the sameacross groups, independent of the other variables.

Let

XAi �@2w

@ mAið Þ2ln

Wiðnidi=dÞhG þ aPi


" #

þ @w@mAi

� �2 ðnidi=dÞhG


� ðnidi=dÞhR


" #;

then XBi < 0 for aPi > bP

i . Applying the implicit function theorem on (41), we then get that foraP

i > bPi ,

dmPAi

daPi

¼ �

@w@mAi

1Wi

nidi

d hG þ aPi

XAi> 0 and

dmPAi

dbPi

¼

@w@mAi

1Wi

nidi

d hG þ bPi

XAi< 0:

Hence, if I compare two groups i and j which differ only in that aPi � bP

i > aPj � bP

j > 0, so ci ¼cj, Ci ¼ Cj, ni ¼ nj and di ¼ dj, then the solutions to the set of first order conditions (41) must fulfillmP

Ai > mPAj , i.e. the biased type of A campaigns more in group i than group j. If parties are neutral,

party B campaigns in these groups and we know from Lemma 3 that mNBi ¼ mP

Ai and mNBj ¼ mP

Aj . Sowhen parties are neutral I also have mN

Bi > mNBj .

Similarly,

dmPAi

ddi¼

� @w@mAi

lnWi

nidid hG þ aP

i

Winidid hG þ bP

i

0@

1Aþ Wi

nidid hG

Winidid hG þ aP

i

�Wi

nidid hG

Winidid hG þ bP

i

24

35

XAið42Þ

The sign of this expression equals the sign of the numerator. For aPi and bP

i close to one, thenumerator approximately equals13

13 We use the approximation ln x � x � 1 for x close to one.



@w@mAi



� 1

" #þ Wiðnidi=dÞhG


� Wiðnidi=dÞhG


which equals

@w@mAi

1

Wiðdini=dÞhG þ aPi

1

Wiðdin=dÞihG þ bPi

aPi � bP

i

� �aP

i

which is positive for aPi > bP

i . Furthermore it is straightforward to check that the numerator of(42) is increasing in aP

i . I conclude that for aPi > bP

i ;dmPAi=ddi > 0 and accordingly that if

groups i and j only differ in that di > dj, then mPAi > mP

Ai . Since ci, ni and di enter symmetrically inthe first order conditions, Proposition 1 is also true for these variables.

Finally, consider two groups where the only difference consists in Ci > Cj. Totally differenti-ating

dmPAi

dCi¼ �

ci@w@mAi

hnidið1� pÞd ð1� CiÞ

@ lnWi

nidid hG þ aP

i

Winidid hG þ bP

i

0@

1A

@Ci� ln

Winidid hG þ aP

i

Winidid hG þ bP

i

0@

1A

26666664

37777775

XAi:

As

@ lnWi

nidid hG þ aP

i

Winidid hG þ bP

i

0@

1A

@Ci¼ 1

Winidid hG þ aP

i

Winidid hG þ bP

i

nidid hR Wi

nidid hG þ bP

i

�� nidi

d hR Winidid hG þ aP

i

�Wi

nidid hG þ bP

i

�2 < 0

we getdmP

Ai

dCi< 0

and it follows that mPAi < mP

Aj . n

Proof of Theorem 2With ni ¼ jin(m), o[Wi(mBi)ji n(mBi)]/omBi the first order condition (41) becomes

ciji@2 wðmÞnðmÞ½ �

@2m

1� Cið Þhdi 1� pð Þd

lnWiðnidi=dÞhG þ aP

i


" #¼ l:

It is straightforward to check that under condition (27) the steps of Proposition 1 as well asTheorem 1 can be retraced.

Proof of Proposition 2In order to shorten notation, let zi � nidi(h/d).

Inserting for ti I get

t 0iti¼

ci 1=ð2 ffiffiffiffiffimip Þ

� �ð1� CiÞziR

ciffiffiffiffiffimip ð1� CiÞ þ Ci

� �ziR þ 1

2

dmi

dM:

Letting



ai �Ci þ 1=ð2RziÞ

cið1� CiÞ

we get

t 0iti¼ 1

mi þ aiffiffiffiffiffimip

dmi

dM: ð43Þ

In groups where the biased type of party A campaigns (i ¼ 1, . . ., na) (or in groups where thebiased type of B campaigns (i ¼ n þ 1 � na, . . ., n)), the first order condition (35) for choice ofmi becomes (using (25) and (28)).

ci

1

2ffiffiffiffiffimip 1� Cið Þzi 1� pð Þ ln

Wi ziGðeÞ þ 12 ð1þ eÞ

Wi ziGðeÞ þ 12 1� ðna=nÞe½ �

( )GðeÞ ¼ lðeÞ:

As e ! 0, I can use the approximation ln (x) � x � 1, so that

lnWi ziGðeÞ þ 1

2 ð1þ eÞWi ziGðeÞ þ 1

2 1� ðna=nÞe½ �

( )�

12 ðn þ naÞ=n½ �e

Wi ziGðeÞ þ 12 1� ðna=nÞe½ �

and letting aiðeÞ ¼1 � na

n e2GðeÞ þCi zi

cið1 � CiÞziI can rewrite the first order condition

mi þ aiðeÞffiffiffiffiffimip ¼ ð1� pÞn þ na=ne

4lðeÞ : ð44Þ

In the limit as e ! 0 the choice of campaign expenditures fulfill

mi þ aiffiffiffiffiffimip ¼ mj þ aj

ffiffiffiffiffimjp ¼ k: ð45Þ

The implicit function theorem and (45) gives

dmi

dM¼ 1

1þ ai1=2ffiffiffiffiffimip

dk

dM: ð46Þ

Using the fact thatPna

i¼1 @mi=@M ¼ 1 this gives for i ¼ 1,. . ., na and by symmetry also for i ¼n þ 1 � na, . . .n

dmi

dM¼

11þ ai

12ffiffiffiffiffimip

Pna

j¼11

1þ aj1

2ffiffiffiffiffimjp

0B@

1CA: ð47Þ

If there exists i such that na < i < n þ 1 � na, then for those i : dmi/dM ¼ 0.Using the first order condition (45), I can rewrite (43)

t 0iti¼ 1

k

dmi

dM:

We can therefore rewrite condition (31), as

dW

dM> 0,

Xn

i¼1

1

k

dmi

dM� n

Pni¼1ð1=kÞðdmi=dM ÞtiPn

j¼1 tj> 0

using the fact thatPna

i¼1 @mi=@M ¼ 1 andPn

i¼nþ1�na@mi=@M ¼ 1 this gives



dW

dM> 0, 2

n>Xn

i¼1

dmi

dM

tiPnj¼1 tj

or

dW

dM> 0, 2

n>Xn

i¼1

@mi

@M

ci

X: ð48Þ

Let e.g. Dmi denote the difference between dmi=dM and the mean, then

Dmi �

dmi

dM� 2

nand Dc

i �ci

X� 1

n

andPn

i¼1 Dmi ¼

Pni¼1 Dc

i ¼ 0. Then I can rewrite (48)

dW

dM> 0, 2

n>Xn

i¼1

Dmi þ

2

n

� �Dc

i þ1

n

� �

or

dW

dM> 0, 2

n>Xn

i¼1

Dmi Dc

i þ2

nDc

i þ Dmi

1

nþ 2

n2

� �

which, gives

dW

dM> 0, 0 >

Xn

i¼1

Dmi Dc

i :

From (45) ai > aj ) mi < mj. Using this, the definition of ai and (47) we get that

Ci > Cj or ci < cj or zi < zj ) ai > aj ) mi < mj ;dmi

dM<

dmj

dMand Dm

i < Dmj : ð49Þ

Recall that

ti � ci

ffiffiffiffiffimip

1� Cið Þ þ Ci½ �ziR þ1

2: ð50Þ

Together with (49) and (29), this gives us

ci < cj or zi < zj ) ti < tj and Dci < Dc

j : ð51Þ

Suppose groups differ only wrt ci. Wlog renumber the groups so that c1 � . . . � cn, where atleast one inequality is strict. Then by (49) and (51) we have that

Dmi � Dm

iþ1 and Dci � Dc

iþ1

for all i < n, and there exists at least one i such that Dmi < Dm

iþ1 and one i such that Dci < Dc

iþ1. SincePni¼1 Dm

i ¼ 0 andPn

i¼1 Dci ¼ 0 there is an i, im, such that for i < im, Dm

i < 0 and for i � im,Dm

i > 0. Similarly, there is ic, such that for i < ic, Dci < 0 and for i � ic, D

ci > 0:Wlog assume that

ic � im. AsPn

i¼1 Dmi ¼ 0 we have that

Pim�1i¼1 Dm

i ¼ �Pn

i¼imDm

i . Now

Xn

i¼1

Dmi Dc

i ¼Xim�1

i¼1

Dmi Dc

i þXn

i¼im

Dmi Dc

i >Xim�1

i¼1

Dmi Dc

im�1 þXn

i¼im

Dmi Dc

im

where the inequality follows from the fact that Dcim�1 � 0; Dc

im � 1 � Dcj , i ¼ 1...., im�2,Pim�1

i¼1 Dmi < 0 and similarly that 0 < Dc

im� Dc

i ; i ¼ im þ 1; . . . ; n andPn

i¼imDm

i > 0.



Moreover,

Xim�1

i¼1

Dmi Dc

im�1 þXn

i¼im

Dmi Dc

im¼ Dc

im� Dc

im�1

�Xn

i¼im

Dmi � 0

which proves that

Xn

i¼1

Dmi Dc

i > 0

so

dW

dM< 0:

If groups differ wrt zi only, an identical proof shows that also in this case dW =dM < 0.

Consider the case where groups differ wrt Ci only. Wlog assume that C1� . . . � Cn, thenDm

i � Dmiþ1 for all i < n. As is clear from (50) the relation between Dc

i and Dciþ1 is not obvious.

However, if M is sufficiently small, thenffiffiffiffiffimip

is small and I have that ti < tiþ1 and thereforeDc

i � Dciþ1 for all i < n. This implies that

Xn

i¼1

Dmi Dc

i < 0 sodW

dM> 0

if M is sufficiently small.For larger M a general result is not available (since I do not have a closed form solution for mi,

it is difficult to assess which term dominates in ti). An increase in M decreases welfare if ti � tiþ1

so that Dci � Dc

iþ1 for all i < n (with at least one strict inequality).Consider an example where na ¼ n/2 (n even). Let Ci ¼ CI for i ¼ 1, . . ., n/4 and Ci ¼ CII ¼

C1 þ g for i ¼ n/4 þ 1, . . ., n/2, where g > 0 is small. In the limit as g ! 0, the districts arealike. Let mI denote campaign expenditures in groups i ¼ 1, . . ., n/4 and mII campaign expen-ditures in groups i ¼ n/4 þ 1, . . ., n/2. Then Dm

I > DmII and an increase in M decreases welfare if

DcI > Dc

II , i.e. if tI > tII. I have that tI > tII if

cffiffiffiffiffiffimIp

1� CI½ Þ þ CI � cffiffiffiffiffiffiffimIIp

1� CIIð Þ þ CIIð � > 0: ð52Þ

Using the implicit function theorem on the first order condition (45) and evaluating in g ¼ 0, wefind that

dmI

dg¼ 1

c

ffiffiffiffiffiffiffiffi2Mn

q1� CI

1þCI þ 1

2Rz1� CI

1c

1ffiffiffiffiffiffiffiffi2Mn

q CI þ 12Rz

1� CIþ 2

:

Furthermore, differentiating the left hand side of (52) wrt g and evaluating at g ¼ 0 gives

dðLHSÞdg

¼ c1ffiffiffiffiffiffiffiffiffiffiffiffiffi

2M=np 1� CIð ÞdmI

dgþ c

ffiffiffiffiffiffiffiffi2M

n

r� 1:

For small g, I have that tI > tII if d(LHS)/dg > 0. Inserting for dmI/dg we get



dðLHSÞdg

¼1þ

CI þ 12Rz

1� CI

2þ 1c

1ffiffiffiffiffiffiffiffi2Mn

q CI þ 12Rz

1� CI

þ c


n

r� 1: ð53Þ

By assumption, the maximal value of M is 1/c2. Evaluating d(LHS)/dg at M ¼ 1/c2 gives

dðLHSÞdg

¼ffiffiffi2

n

rþ

1þCI þ 1

2Rz1� CI

2þffiffiffin2

q CI þ 12Rz

1� CI

� 1

which is positive if

2�ffiffiffin

2

r� �CI þ 1=ð2RzÞ

1� CI> 1:

This necessarily requires that n < 8.If n � 8, (53) can only be fulfilled if M > 1=c2, implying that c

ffiffiffiffiffiMp

> 1, which does notmake economic sense. If, however,

wðmÞ ¼ffiffiffiffimp

if m � 1=c2

1 if m > 1=c2.

then (53) is positive if

c


n

r� 1

2

!1þ CI þ ð1=2ÞRz

1� CI

� >

1

2� c


n

r !c


n

r

which is fulfilled for all relevant CI, R and z if M is so large that

c

ffiffiffiffiffiM

n

r>

1

2ffiffiffi2p � 0:35:

University of Copenhagen

Submitted: 9 August 2004Accepted: 22 July 2006

ReferencesAnsolabehere, S. and Iyengar, S. (1995), Going Negative: How Political Advertisements Shrink and Polarize the

Electorate, New York: The Free Press.Ashworth, S. (2006). �Campaign finance and voter welfare with entrenched candidates, American Political

Science Review, vol. 50, pp. 214–31.Alvarez, R. M. (1998). Information and Elections, Ann Arbor: The University of Michigan Press.Austen-Smith, D. (1987). �Interest groups, campaign contributions and probabilistic voting�, Public Choice,

vol. 54, pp. 123–39.Baron, D. P. (1994). �Electoral competition with informed and uninformed voters�, American Political Science

Review, vol. 88, pp. 33–47.Bartels, L. M. (1988). �Electioneering in the United States�, in (D. Butler and A. Ranney eds.) Electioneering:

A Comparative Study of Continuity and Change, pp. 244–77, Oxford: Clarendon.



Bartels, L. M. (1996). �Uninformed votes: information effects in presidential elections�, American Journal ofPolitical Science, vol. 40(1), pp. 194–230.

Besley, T. and Burgess, R. (2002). �The political economy of government responsiveness: theory and evidencefrom India�, Quarterly Journal of Economics, vol. 117(4), pp. 1415–52.

Brians, C. L. and Wattenberg, M. P. (1996). �Campaign issue knowledge and salience: comparing receptionfrom TV commercials, TV news, and newspapers�, American Journal of Political Science, vol. 40(1), pp. 172–93.

Butters, G. (1977). �Equilibrium distribution of sales and advertising prices�, Review of Economic Studies, vol.44(3), pp. 465–91.

Cadot, O. H., Roller, L.-H. and Stephan, A. (2002). �Contribution to productivity or pork barrell? The twofaces of infrastructure investment�, Discussion Paper FS IV 02-09, Wissenschaftszentrum Berlin.

Coate, S. (2004a). �Political competition with campaign contributions and informative advertising�, Journal ofEuropean Economic Association, vol. 2(5), pp. 772–804.

Coate, S. (2004b). �Pareto improving campaign finance policy�, American Economic Review, vol. 94(3), pp. 628–55.

Dahlberg, M. and Johansson, E. (2002). �On the vote-purchasing behavior of incumbent governments�,American Political Science Review, vol. 96(1), pp. 27–40.

Dixit, A. and Londregan, J. (1996). �The determinants of success of special interests in redistributive politics�,Journal of Politics, vol. 58, pp. 1132–55.

Friedman, J. W. (1983). �Advertising and oligopolistic equilibrium�, Bell Journal of Economics, vol. 14, pp. 464–73.

Gelman, A. and King, G. (1993). �Why are American presidential election polls so variable when votes are sopredictable?�, British Journal of Political Science, vol. 23, pp. 409–51.

Gerber, A. (1998). �Estimating the effect of campaign spending on senate election outcomes using instru-mental variables�, American Political Science Review, vol. 92, pp. 401–11.

Gerber, A. and Green, D. P. (2000). �The effects of canvassing, telephone calls, and direct mail on voterturnout: a field experiment�, American Political Science Review, vol. 94(3), pp. 653–63.

Green, D. P., and Krasno, J. S. (1988). �Salvation for the spendthrift incumbent: reestimating the effects ofcampaign spending in House elections�, American Journal of Political Science, vol. 32, pp. 884–907.

Grossmann, G. M. and Shapiro, C. (1984). �Informative advertising with differentiated products�, Review ofEconomic Studies, vol. 51, pp. 63–81.

Helpman and Grossman (2001). �Integration vs. outsourcing in industry equilibrium�, Papers 2001-7, Tel Aviv.Holbrook, T. M. (1996). Do Campaigns Matter?, Thousand Oaks: Sage Publications.Just, M., Crigler, A. and Wallach, L. (1990). �Thirty seconds or thirty minutes: what viewers learn from spot

advertisements and candidate debates�, Journal of Communication 40, pp. 120–33.Konrad, K. (2004). �Inverse campaigning�, Economic Journal, vol. 114, pp. 69–82.Lazarsfeld, P., Berelston, P. B. and Gaudet, H. (1944). The Peoples Choice, New York, Columbia University Press.Lindbeck, A. and Weibull, J. (1987). �Balanced budget redistribution as the outcome of political competition�,

Public Choice, vol. 52, pp. 273–97.Matsusaka, J. G. and Palda, F. (1999). �Voter turnout: how much can we explain?�, Public Choice, vol. 98,

pp. 431–46.McKelvey, R. and Ordeshook, P. (1987). �Elections with limited information: a multidimensional model�,

Mathematical Social Science, vol. 14, pp. 77–99.Milligan, K. and Smart, M. (2003). �Regionalism and pork barrell politics�, mimeo, Universities of British

Columbia and Toronto.Nagler, J. and Leighley, J. (1992). �Presidential campaign expenditures: evidence on allocations and effects�,

Public Choice, vol. 72, pp. 319–33.Ortuno-Ortin I. and Schultz, C. (2000). �Public funding of political parties�, CES-ifo Working Paper no. 368.Persson, T. and Tabellini, G. (2000). Political Economics, Cambridge MA: MIT Press.Popkin, S. L. (1991). The Reasoning Voter, Chicago: The University of Chicago Press.Potters, J., Sloof, R. and van Winden, F. (1997). �Campaign expenditures, contributions, and direct

endorsements: the strategic use of information and money to influence voter behavior�, European Journalof Political Economy, vol. 13, pp. 1–31.

Prat, A. (2002). �Campaign advertising and voter welfare�, Review of Economic Studies, vol. 69(4), pp. 997–1017.Rekkas, M. (2002). �The impact of campaign spending on votes in multiparty elections�, mimeo, Simon Fraser

University.Schultz, C. (2002). �Policy biases with voters� uncertainty about the economy and the government�, European

Economic Review, vol. 46(3), pp. 487–506.Shaw, D. R. (1999). �The effect of TV ads and candidate appearances on statewide presidential votes, 1988–

96�, American Political Science Review, vol. 93(2), pp. 345–61.Stromberg, D. (2004a). �Mass media competition, political competition and public policy�, Review of Economic

Studies, vol. 71(1), pp. 265–84.



Stromberg, D. (2004b). �Radio’s impact on public spending�, Quarterly Journal of Economics, vol. 119(1),pp. 189–221.

Stromberg, D. (2002). ‘Optimal campaign spending in presidential elections: the probability of being Flor-ida’, Seminar Paper 706, Institute for International Economic Studies, Stockholm University.

West, D. (1993). Air Wars: Television Advertising in Election Campaigns, 1952–1992, Washington DC: CQ Press.Wright, G. (1974). �The political economy of New Deal spending : an econometric analysis�, Review of Economics

and Statistics, vol. 56, pp. 30–8.Zaller, J. (1991). The Nature and Origins of Mass Opinion, Cambridge: Cambridge University Press.



Strategic Campaigns and Redistributive Politics

Documents

Transcript of Strategic Campaigns and Redistributive Politics