ANRV377-PL12-15 ARI 2 May 2009 12:49

Negative CampaigningRichard R. Lau and Ivy Brown RovnerDepartment of Political Science, Rutgers University, New Brunswick, New Jersey 08901;email: ricklau@rci.rutgers.edu

Annu. Rev. Polit. Sci. 2009. 12:285–306

The Annual Review of Political Science is online atpolisci.annualreviews.org

This article’s doi:10.1146/annurev.polisci.10.071905.101448

Copyright c© 2009 by Annual Reviews.All rights reserved

1094-2939/09/0615-0285$20.00

Key Words

attack advertising, demobilization, meta-analysis, negativity

AbstractThe past two decades have seen an explosion of social science researchon negative political advertising as the number of political observerscomplaining about its use—if not negative campaigning itself—hasalso grown dramatically. This article reviews the literature on negativecampaigning—what candidates are most likely to attack their opponent,under what circumstances, and most importantly, to what effect. Wealso discuss the many serious methodological issues that make studyingmedia effects of any kind so difficult, and make suggestions for “bestpractices” in conducting media research. Contrary to popular belief,there is little scientific evidence that attacking one’s opponent is a par-ticularly effective campaign technique, or that it has deleterious effectson our system of government. We conclude with a discussion of whethernegative political advertising is bad for democracy.

285

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

INTRODUCTION

Political candidates invest heavily in strategiccampaign communications. Whereas press re-leases, public appearances, stump speeches, andinterviews are geared toward garnering freemedia coverage, carefully crafted commercialor paid communications represent an indus-try unto themselves. Substantial percentages ofcampaign war chests are dedicated to such var-ied forms of political communication as lawnsigns, bumper stickers, direct mail, candidateweb sites, and television and radio appeals. Al-though all of these elements are integral to asuccessful campaign media strategy, televisionadvertisements remain the most visible, expen-sive, and presumably cost-effective form of paidpolitical communication.

As one might expect, television advertising isby far the most pervasive communicative tech-nique studied in the political communicationsliterature. In just half a minute, a candidate candeliver a precise message to a broad audience. Asingle ad can increase a candidate’s name recog-nition, inform voters of salient issues and can-didates’ policy concerns, or motivate previouslyreluctant individuals to vote through emotionalappeals—if it is repeated often enough. Thereinlies the rub: effective political advertisementsare expensive to produce, and they become in-creasingly costly with each ad buy. The moreoften an ad is shown, the wider its audience,and the more often people are exposed to anad, the more likely they are to process and re-member its message. In theory, there should bea point of decreasing marginal utility at whicheveryone who is going to see an ad has alreadyseen and processed it, and additional exposuresare ignored or, worse, become annoying. Howoften (if ever) this point is reached in a politicalad campaign is not clear. But even the wealthiestand best financed candidates are limited by thesize of their war chests, and consequently mustdecide which messages will be conveyed to thepublic, what media tactics should be employed,and how often they can afford to present eachmessage. If the electoral district is large enoughto include multiple media markets, candidates

also must decide what messages to provide inwhich media markets.

One of the most important decisions can-didates make is whether to run on their ownmerits—that is, their own policy ideas, past ac-complishments, and personal strengths, whichmost observers would call a “positive” appeal; orif instead their campaign will concentrate on theperceived weaknesses of their opponent’s policyproposals, prior policy failures, and/or personalpeccadilloes, which in this essay we call a “neg-ative” appeal. Candidates are not restricted toone or the other, of course, and most campaignsemploy a combination of both techniques. Butin our review, at a very general level, if the ap-peals a candidate makes are primarily positive,we refer to their campaigns as positive, whereasif the appeals they make are primarily nega-tive, we refer to their campaign as negative.1

We use the word “tone” to refer generically tothe balance of positive and negative appeals ina campaign.

Why a Focus on NegativePolitical Campaigns?

This essay focuses on negative campaign com-munications, a widespread phenomenon thathas attracted increasing attention from the pressand social scientists alike. Although politics inAmerica was never a gentleman’s sport, an ac-celerated proliferation of negative advertisingover the past 30 years—or at least a rise in thenumber of people complaining about it—seemsapparent. But that is the most we can say. Before1996 there are not, in fact, any good data on theactual prevalence of negative campaigning in

1Campaigns that employ approximately equal numbers ofpositive and negative appeals could be called “mixed,” butfor simplicity we assume here that all campaigns are eitherprimarily positive or primarily negative. How one could ac-tually get an overall negativism “score” for a campaign thatincluded television ads aired different numbers of times in dif-ferent media markets, another set of radio ads similarly aireddifferent numbers of times in different markets, newspaperadvertisements, billboards, lawn signs, bumper stickers, andpersonal appearances by the candidate, etc., is a question noone we know of has ever tried to address.

286 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

the United States. Geer (2006) reports a steadygrowth in the percentage of negative appeals inthe television advertisements produced by themajor party presidential candidates from 1960through 2004 but has no evidence on how of-ten each ad was shown. Buell & Sigelman (2008)report no growth in the negativism of presiden-tial elections between 1960 and 2004, as judgedby the tone of statements from all presiden-tial candidates reported in the New York Timesbetween Labor Day and election day. Simi-larly, Lau & Pomper (2004) report no growthfrom 1988 to 2002 in the proportion of state-ments reported in newspapers from major partysenatorial candidates (or their spokespersons)criticizing the opponent. But this is secondaryevidence, filtered through the lens of the news-paper reporter, and tracks only what the can-didates are saying. It is only with the adventof “ad detector” technology developed by theCampaign Media Analysis Group, which trackssatellite-based feeds of political advertisements,that we have been able to gather any solid evi-dence on how often different televised ads areshown in different media markets. As seen inFigure 1, in this relatively short timeline, neg-ative political advertising in the United Stateshas not grown at all (at least at the federallevel).2

We can be more definitive about increas-ing interest in negative campaigns, however.Figure 2a displays the number of storiesabout negative advertising or negative cam-paigns appearing in the New York Times, theWashington Post, the Christian Science Moni-tor, and the Associated Press newswire dur-ing presidential election years since 1980.Only 17 stories about either of these topics

2These data are available from the Wisconsin AdvertisingProject at the University of Wisconsin. We thank John Sidesfor providing the numbers in Figure 1, and John Geer andShana Gadarian for data in Figure 2a. We may soon have newevidence on how much money presidential candidates frombefore 1996 spent advertising in the different states, and evenpossibly how often specific ads aired in different locales, fromdata currently being gathered by Scott Althaus and DaronShaw from presidential libraries around the country. Staytuned. . . .

appeared in any of these newspaper sourcesin 1980, but this figure grew to 210 storiesin the millennium year before falling back abit in 2004. Figure 2b shows the number ofacademic articles providing some evidence onthe effects (broadly construed) of negative po-litical advertisements or negative campaigns.The first article we could find in the social sci-ence literature was published in1984, and thefirst study by a political scientist appeared in1990. At last count (near the end of 2006) therewere 110 books, chapters, dissertations, and ar-ticles addressing this broad topic empirically,and many more exploring other aspects of neg-ative campaigns. The upshot of the great ma-jority of these articles and news stories is thatnegative campaigning has become a pervasiveand often corrosive aspect of the American po-litical scene—“the electronic equivalent of theplague” (West 1993, p. 51), to quote one of themore colorful characterizations. Negative cam-paigning and the use of attack ads have beencriticized for reducing politics to its least com-mon denominator. Some scholars have gone sofar as to attribute diminishing public trust andever-decreasing turnout rates to the growinguse of negative ads.

Why all this attention to campaign nega-tivism if there is no real evidence it is increas-ing? The obvious answer is that coordinatedcandidate messages appear to grow increasinglynegative with each campaign cycle even if theyactually are not, making negativism increas-ingly difficult for any voter to avoid or ignore.This is due primarily to biases in the news me-dia that make stories about negative advertise-ments much more likely to be written and/oraired than stories about positive ads. For exam-ple, West (2005) reports that almost two thirdsof the stories about political advertisementsbroadcast on the CBS evening news between1972 and 2004 involved a negative advertise-ment. Unless predicated on false information,positive communications provide little if anyentertainment value. After all, how many storieson a candidate’s love of God, country, family,and all-American values can a single media out-let run? By contrast, negative communications

www.annualreviews.org • Negative Campaigning 287

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

dangle like red meat in front of journalists hun-gry for a new angle. It is therefore not surprisingthat negative ads get covered (and consequentlyreplayed by the media) much more frequentlythan positive ads—leading to the common im-pression that negative ads are more pervasivethan they are. To the extent that political prac-titioners recognize these media biases, they cer-tainly take advantage of them, running negativeads within the Washington Beltway media mar-ket in order to capture the attention of the po-litical news media.

These same biases contribute to the impres-sion that most negative ads are ad hominem at-tacks and misleading exaggerations if not out-and-out lies. Some, surely, are, and the media dotheir best to bring the most outrageous exam-ples to our attention, but it would be a mistaketo assume that all or most or even a large mi-nority of negative political advertisements havethese characteristics.

Even if it could be definitively shown thatnegative ads are increasing, such evidencewould not explain why candidates attack theiropponents. We can assume the answer is thatcandidates and political practitioners believe at-tack ads are effective—and there are sound the-oretical reasons to think they may be right. So-cial psychologists have pointed to “negativity”:the tendency for negative information to bemore influential than equally extreme or equallylikely positive information (Kanouse & Hanson1971). The “equally likely or equally extreme”part of the definition is crucial, and is whatmakes negativity an interesting psychologicalphenomenon rather than mere common sense.If we describe a candidate as having a pleas-ant smile but holding foreign policy positionsthat will lead to nuclear war, of course no onepredicts that the combination of these two bitsof information will result in neutral evalua-tions. Kernell (1977) and Lau (1982) have care-fully documented negativity effects in candidateevaluation and voting behavior, and Lau (1985)has provided evidence for two different expla-nations for negativity, one perceptual and onemotivational. The perceptual “figure-groundhypothesis” states that negative information

stands out against a generally more positivebackground; the motivational “cost-orientationhypothesis,” which is presumably part of our ge-netic makeup, states that most people are inher-ently more strongly predisposed to avoid coststhan they are to seek gains. By this account,negativity explains both the apparent growth ofcampaign negativism and its alleged corrosiveeffect on our political system.

In this article, we review the evidence on justhow effective negative advertising is in achiev-ing its goals, and just how corrosive those at-tacks are on our political system. We begin bytrying to carefully define what a “negative polit-ical campaign” is. Is our very general definitionof attacking one’s opponent (rather than talk-ing about oneself) sufficient, or do we need amore precise and nuanced definition? We mightbelieve (echoing Justice Stewart’s famous state-ment about pornography) that we know it whenwe see it, but negative campaigning turns out tobe a bit more difficult to define precisely thanone might think. Having established a work-able definition of the phenomenon of inter-est, we then look at the literature on whencandidates choose to “go negative.” There isa fair amount of agreement between the for-mal and the empirical literature, but we alsofind some clear theoretical predictions that havebeen disconfirmed by empirical evidence.

Then we turn to arguably the most impor-tant question: Do negative campaigns work?Clearly, paid campaign consultants and the re-porters who cover campaigns believe they do—otherwise they would not continue to use them.As Laura Mansnerus, writing in the New YorkTimes about the highly negative 2005 NewJersey gubernatorial campaign, explained: “Thepeople who produce these ads and the consul-tants who hire them know that negative cam-paigning works. These people are paid way toomuch to be mistaken about whether poison iseffective” (Mansnerus 2005). As social scien-tists, we seek more convincing evidence thanthis.

There is a long history in political science ofdoubting that the media affect the political pro-cess. But as the “minimal effects” view has been

288 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

replaced by the “media matter” view (Iyengar &Kinder 1987), some within the academy havecome to believe that negative political cam-paigns matter so much that they can affect thevery nature or our political system, discourag-ing voters from participating in individual elec-tions and weakening their faith in the politicalprocess itself over time. In fact, this is one ofthe few instances that we can think of wherean academic finding has been picked up by thepress and become part of the conventional wis-dom about political campaigns. The impact ofnegative campaigns on the American politicalsystem is the fourth major topic of our review.

Although we mention some methodologi-cal issues as we discuss our first four topics,the penultimate section of the article focuseson a collection of methodological concerns thatplague the study of political campaigns (and ofmedia effects more generally). In the conclud-ing section we ask a normative question: Arenegative campaigns bad for democracy? Somepolitical scientists argue that, far from demo-bilizing the electorate, negative political cam-paigns may even stimulate participation. Theybelieve that negative ads educate voters by pro-viding critical information about a candidate’spolicy positions, character, and personal his-tory, and in so doing, allow voters to makemore informed political decisions. A review ofthe literature suggests that categorizing nega-tive campaigns as patently good or bad is at thevery least an oversimplification and at most agross oversight. The question of whether nega-tive campaigns are good for democracy, like thedecision to launch a negative attack, is compli-cated and multilayered. Nonetheless, as nega-tive campaigns continue to have a sizeable pres-ence in our political system, an assessment oftheir intended and unintended influence on po-litical behavior is warranted.

WHAT IS A NEGATIVECAMPAIGN?

In order to examine the role of negative cam-paigning, we need to carefully define it. Theliterature has emphasized analysis of the use

and effectiveness of negative political adver-tisements, trending toward a simple positive-negative dichotomy in which ads that focus ontheir sponsors are branded “positive” and thosethat address the opposition are “negative.” Eachadvertisement is treated as a single unit ofanalysis and weighted equally. This dichotomymay lead many people intuitively to catego-rize positive ads as patently “good” and nega-tive ads as inherently “bad,” but Jamieson et al.(2000) warn that doing so is both incorrect andmisleading:

Many scholars mistakenly assume that “at-tack” is both “negative” and “dirty.” Conflat-ing these terms obscures the important dis-tinction between legitimate and illegitimateattack and minimizes the likelihood that thedeceptions found in supposedly “positive” dis-course will be probed. (p. 44)

Although the positive-negative dichotomy maybe conducive to experimental research—whereisolating and manipulating a single variable(e.g., negative communication) in order to mea-sure its effect is the end goal—Jamieson et al.suggest that content analysis that fails to treat“contrast” or “comparative” advertising as aseparate and unique category of political dis-course not only excludes a key communica-tive ingredient, but also portrays a simplerlandscape than what, in fact, exists. We agree,when it comes to creating typologies of indi-vidual political advertisements. This distinctionloses most of its power when applied to en-tire campaigns, however. Campaign commu-nications include television, radio, and printads, as well as the content of official campaignweb sites, speeches, debates, interviews, and di-rect mail (fund-raising) appeals. Unless a can-didate completely ignores his or her opponent(as may occur when a challenger fails to rep-resent a legitimate threat), all campaigns bytheir very nature are comparative. Not forget-ting the tremendous difficulties of measuringand combining the positive or negative tone ofevery different aspect of a major political cam-paign, a unidimensional scale (100% positive to

www.annualreviews.org • Negative Campaigning 289

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

Table 1 Perceived fairness of different types of attacks

CriticismPercent sayingcriticism is fair

Talking one way and voting another 80.7His/her voting record 75.8His/her business practices 71.0Taking money from special interest groups 70.7Taking money from individuals with ethical problems 63.0Current personal troubles with drugs or alcohol 56.1Current extramarital affairs 45.1Political actions of his/her party’s leaders 37.1Past extramarital affairs 27.8Past personal troubles with drugs or alcohol 25.9Personal lives of his/her party’s leaders 19.1Behavior of his/her family members 7.7

Source: 1998 Sorenson Institute for Political Leadership Survey of Virginia Voters, asreported by Freedman et al. (1999).

100% negative) seems in theory the best way tothink about the negativism of a campaign.

But these are all subtle differences in aca-demic definitions that may be meaningless tothe general public. Do laypeople know whenthey have seen a negative ad? Is there a limitto what they will tolerate in a political cam-paign? Drawing on two surveys conducted inVirginia, Freedman et al. (1999) maintain thatvoters are indeed capable of distinguishing be-tween mudslinging and fair play, and that theyneither condemn nor condone campaign nega-tivism outright. When asked to rate the fairnessof 12 charges that one candidate might makeagainst another, respondents revealed a clear hi-erarchy of tolerance for attacks (see Table 1).Generally speaking, voters believe that attack-ing a candidate for his actions in office is fairgame, whereas bringing up prior personal prob-lems should be out of bounds.

Freedman et al.’s findings also reveal educa-tional and partisan biases. Respondents’ thresh-olds for tolerating different types of attacks in-creased with education, political information,and Republican party identification. That is,more educated, better informed, and Repub-lican voters were all somewhat less likely tosay that any given charge was unfair. Freedmanet al. did not, however, find significant varia-tion with respect to either gender or race. A

follow-up study examining the repercussions ofcounterattacks reveals that voters ascribe sim-ilar standards of fairness to rebuttals. The less“fair” the response, the more willing voters areto vote for the initial attacker, regardless of theperceived fairness of his or her original charge.

But do voters’ theoretical opinions of cam-paign negativism translate into the reality ofmass political behavior? Not necessarily, ac-cording to Sides et al. (2003). They report astudy of the 1998 California gubernatorial elec-tion, in which Democrat Lieutenant Gover-nor Gray Davis opposed Republican AttorneyGeneral Dan Lungren. They link rolling cross-sectional survey data from 2902 respondentsinterviewed during the last six weeks of the cam-paign with data on the actual tone of the adver-tising aired by the two candidates during eachday of that period. This design allows Sideset al. to determine how closely citizens’ per-ceptions of a campaign’s negativism are tied tothe actual nature of that campaign. Not sur-prisingly, there was a very strong partisan com-ponent to these perceptions, with Democratsperceiving Davis’s campaign to be much lessnegative than Lungren’s campaign, and Repub-licans perceiving the opposite pattern. What ismuch more surprising is that the realities of thecampaign—how negative Davis’s and Lungren’scampaigns actually were—had no effect on per-ceptions. Indeed, the only significant predictorin the equations for both Davis and Lungrenwas partisanship. Clearly, negativism remains inthe eye of the beholder. It seems that few vot-ers during actual political campaigns come any-where close to the objectivity that a researcherwould apply in judging the content or tone of acampaign.

Sigelman & Kugler (2003) agree, suggest-ing that scholars’ systematic examinations andeven our definitions of campaign negativismare out of sync with voter perceptions. Ourobjective, systematic coding of ads implicitlyassumes that scientific realities guide percep-tions, but that assumption suggests that vot-ers consume campaign information uniformlyand as judiciously as scholars conduct their ownresearch. Sigelman & Kugler point out that

290 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

whereas scholars seek out the breadth of po-litical communications and afford each state-ment equal weight, voters do not; instead theyseek the information that is most concise andreadily available, often at the expense of accu-racy, and pay disproportionate attention to thatwith which they most agree. Given those dif-ferences, it is not surprising that Sigelman &Kugler uncover inconsistencies between voterperceptions of negativity and scholarly classi-fications. Their comparison of ANES surveydata against the newspaper coverage of three1998 gubernatorial races reveals asymmetricalimpressions of campaign tone, as well as incon-sistent perceptions among voters within eachstate. However, although they attribute differ-ences among voters to partisanship, politicalefficacy, and above all, political sophistication,Sigelman & Kugler stop short of prescribing asolution for this problem.

The challenge lies in reconciling the rigorsof systematic social science research with thereality that voters consume political informa-tion erratically and with bias. Perhaps our def-inition of tone should go beyond written tran-scripts of televised ads and include the visualimages and accompanying music often aimedat triggering voters’ emotions. For example,Brader’s (2005, 2006) analysis of emotional ap-peals reveals frequent, deliberate attempts toevoke enthusiasm or fear among the electoratethrough the strategic use of visual images, mu-sic, and voiceover tone. These components donot change the ads’ message but rather clar-ify that message so that it “strikes an emotionalchord,” as his title suggests. Unlike analyses thatmeasure tone in terms of ad content, Bradermeasures the ads’ intended emotional outcome.He describes enthusiastic ads—such as RonaldReagan’s famous 1984 “Morning in America”spot—as those which appeal to hope, enthusi-asm, sentimentality, and national pride. Suchads aim to capture the attention of a candidate’sbase by motivating participation and ultimatelyturnout. By contrast, fear ads do not automati-cally elicit a transfer of allegiance from one can-didate to another, but rather trigger anxietiesamong undecided or weakly committed voters,

causing a re-evaluation of both candidates’ trackrecords and capabilities.

But do purely factual, unemotional ads evenexist? Doesn’t every ad seek to arouse eitherconfidence in its sponsor or anxiety about thecompetition? Even “objectively” pointing outan opponent’s weak record on crime is an ef-fort to create anxiety about his or her ability toaddress the issue in the future. Similarly, high-lighting one’s own proven track record is in-tended to instill confidence and security. Brader(2006) maintains that, though rare, “unimpas-sioned ads” do exist—if only so that their spon-sors may appear to function above the politi-cal fray. But isn’t that an enthusiastic appeal,painting oneself as a candidate of whom we canall feel proud? Brader points to Ross Perot’s1992 advertisements and 30-minute “infomer-cials” as an example of such authentic commu-nications, yet Perot’s ads tapped into a disgustand frustration with the Republican status quoand garnered enthusiasm for a more respon-sible fiscal conservatism—so much enthusiasmthat he received more votes than any third-partycandidate in U.S. history. Even his dry formattriggered emotions among the electorate. To befair, Brader concedes that “ads are almost neverstripped completely of emotional content”; hecategorizes as “unimpassioned” those ads whichdo not overtly appeal to voters’ emotions, ac-knowledging that minimal residual sentimentsare inevitable (Brader 2005).

Every candidate’s goal is to win. Therefore,just as candidates understand that it is unreal-istic to expect all voters to make decisions onthe merit of facts and the content of campaigncommunications alone, scholars must appreci-ate the role of audiovisual elements in emotion-ally priming vote choice. This is not to sug-gest that political decisions are fundamentallypurely emotional, but rather that emotionalcues can take the voter to a place that facts alonemay not always reach. As Brader explains, emo-tional elements may reinforce content, but theydo not override it. Ultimately, Brader’s researchreminds us that researchers who code the toneof a political ad or campaign based solely onthe direction of the candidates’ appeals, without

www.annualreviews.org • Negative Campaigning 291

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

also considering the role of emotional overlays,are accepting a broad and less precise defini-tion of campaign negativism, one that can bepotentially misleading. On the other hand—avery important hand—simple directional cod-ing of campaign appeals can be accomplishedvery easily and reliably. The more subjectiveand complicated a coding scheme, the moredifficult (and thus expensive) it is to use, andtypically the less reliable are its results.

THE DECISION TO ATTACK:WHEN, HOW, AND WHO?

When does a candidate decide to inform vot-ers of his or her opponent’s alleged negative at-tributes, despite the risk of alienating potentialand existing supporters? The decision to attackis a political calculation, based on the presump-tion that its execution will damage the intendedtarget more than it will jeopardize the status ofthe candidate sponsoring the attack.

Skaperdas & Grofman (1995) use formaltheory to identify factors that motivate a candi-date to go negative. Assuming that candidatesare aware of their relative status in the pollsand that future vote shares remain indetermi-nate until the day of the election, and most im-portantly that campaign attacks provide a netbenefit to the attacking candidate, Skaperdas &Grofman examine the use of negative advertis-ing in both two-candidate and three-candidatecontests. In a two-candidate race, they find thatthe frontrunner typically uses a greater propor-tion of positive ads and his opponent more fre-quently goes negative. However, as the gap be-tween the two candidates narrows, the leadingcandidate’s use of negative advertising will in-crease in order to maintain his or her front-runner status. Multi-candidate races are lessstraightforward. A candidate who is trailing bya definitive margin and thus only serving as a“spoiler” is predicted to run positive ads exclu-sively. Further, in any three-way race, no can-didate should launch negative attacks againstthe weaker of the two opponents. Instead, allattacks should be directed toward the frontrun-ner or, in the case of the frontrunner him- or

herself, against the greatest threat. This was ev-ident in the 1992 presidential election: GeorgeH.W. Bush and Bill Clinton focused their at-tacks on each other, rather than against third-party candidate Ross Perot. However, as Perotbegan to chip away at Bush’s base toward theend of the campaign, he found himself on thereceiving end of Bush-sponsored attacks, whichis counter to the predictions of the model.

The rebuttal calculus also matters. Howdoes a candidate’s decision to go negative with-out provocation compare to his willingness toretaliate against a negative attack, in light ofpoll positioning, closeness of the race, or thenumber of opponents? Why, for example, didJohn Kerry fail to rebut a series of negative adsthroughout August 2004? Where is the line be-tween staying above the fray and being an oppo-nent’s punching bag? There are costs either way.On the one hand, a candidate’s failure to refuteattacks can leave the information environmentdominated by the charges from the opponent.On the other hand, responding to the allega-tions gives them credibility, and it can also putthe candidate on the defensive and take him orher off message, which can prove equally dam-aging (Kern 1989, Jamieson & Campbell 1983).After the candidate is attacked, the campaignfaces not only a question of strategy (whetherto go negative) but also a question of tactics(how).

Harrington & Hess (1996) disagree with theassertion that a candidate’s willingness to attackreflects his or her status in the polls. Instead,their theoretical model suggests that a candi-date’s personality and ideology are stronger pre-dictors. Any connection between the use of neg-ative ads and poll position, they maintain, ispurely coincidental; both are driven by a candi-date’s weakness in terms of personal attributes.Advertising tactics, they argue, are driven by theideological position of undecided voters ratherthan the relative strength of an opponent. Can-didates will use positive messages to repositionthemselves as ideologically compatible with un-decided voters and will likewise use negative adsto create an ideologically based wedge betweentheir opponent and potential swing voters.

292 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

Inherent to this theory is an assumption thatthe candidate with the more desirable personalattributes or larger “valence index score” willpresent the more positive campaign, regardlessof his position in the polls.

In order to reconcile these opposing ar-guments and identify whether actual cam-paign activity follows theoretical prediction,Theilmann & Wilhite (1998) ran a series ofexperiments using members of the AmericanAssociation of Political Consultants, exposingthis unique sample to a series of hypotheti-cal campaign scenarios and assessing their rec-ommendations regarding the willingness toattack. Theilmann & Wilhite report confir-matory evidence for Skaperdas & Grofman’stheory of relative strength, but they find noconnection between personal attributes andnegative ads, as suggested by Harrington &Hess. They also examine the role of partisan-ship, testing Ansolabehere & Iyengar’s (1995)hypothesis that positive ads are more effec-tive with Democratic voters whereas negativeads more often resonate with Republicans. TheRepublican consultants in their sample were in-deed more open to implementing attack strate-gies than were their Democratic counterpartsoverall, more inclined to employ them early ina campaign, and more likely to use them whena client is comparatively underfunded.

Intrigued by this decision-execution calcu-lus, Buell & Sigelman (2008) examined all ma-jor party presidential tickets between 1960 and2004 to identify how competing tickets de-cide not only when, but also how, to attack.Buell & Sigelman find support for some ofSkaperdas & Grofman’s predictions, but thissupport is primarily limited to races they con-sider blowouts—the elections of 1964, 1972,and 1984. Although the Skaperdas-Grofmanmodel can play a useful organizing function,when put to empirical test it “served pri-marily as a foil for one contrary finding af-ter another.” The model “abstracts away toomuch and thereby ignores a multitude of fac-tors that figure importantly in candidates’ deci-sions to go negative” (Buell & Sigelman 2008,pp. 441–42).

Rather than examining the campaign inits entirety, Haynes & Rhine (1998) focuson those variables—namely “competitive po-sitioning, reward factors, and media relatedconditions”—that predict negative communi-cations in the early primary season, a timewhen the intraparty candidate field is crowdedwith ideologically similar candidates, and lackof funding may prevent some candidates frompurchasing any airtime. Using content analy-sis, they discover that underfunded candidatesseek to increase name recognition throughpersonality-driven, self-promotional (positive)ads, and utilize a variety of “intermediated”vehicles—including press releases, televisioninterviews, stump speeches, and debates—tolaunch attacks on their opponents. These al-ternative media provide a low-cost opportunityfor lower-tier candidates to garner free (somewould call this earned ) media coverage by com-paring themselves to the frontrunners whilepresenting a positive and more dignified imagethrough advertising.

All of the empirical data on the nature ofcampaigns reported thus far have come fromstudies of relatively few campaigns, typicallyat the presidential level. In contrast, Lau &Pomper (2001, 2004) examine the tone of thecampaigns of virtually every contested U.S.Senate election between 1988 and 2002. Theypropose seven hypotheses about when candi-dates will go negative. In particular, they pro-pose that candidates who are behind, candidatesin close elections, challengers, candidates withrelatively few campaign resources, Republicans,males, and candidates whose opponents are at-tacking them have higher levels of negativism intheir campaigns. Lau & Pomper report signif-icant support for all seven of these hypothesesat the bivariate level. Only three of these hy-pothesized effects maintain their strength in amultivariate analysis, however. Controlling onthe other explanations, Republicans, candidateswith less money than their opponents, and can-didates whose opponents are attacking themare all significantly more likely to attack theiropponents. Particularly noteworthy in Lau& Pomper’s findings is the almost one-to-one

www.annualreviews.org • Negative Campaigning 293

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

correspondence between the tone of a candi-date’s campaign and that of his or her oppo-nent. These days, attacking your opponent in asenate election is almost guaranteed to provokean approximately equivalent counterattack.

In an unpublished manuscript, “Going Neg-ative in a New Media Age: Congressional Cam-paign Websites, 2002–2006,” Druckman, Kifer,and Parkin find similar patterns in political at-tacks seen on the internet. Having analyzed adataset that includes 730 candidate web sitesspanning three campaign cycles, Druckmanet al. concur that competition drives nega-tivism, but they provide a somewhat differ-ent explanation for this now familiar result.First, Druckman et al. argue that the audi-ences for campaign web sites are highly self-selective, falling into two distinct groups: jour-nalists and politically active supporters of acandidate. Drawing on the emotions literature(Brader 2006, Marcus et al. 2000), Druckmanet al. go on to argue that negative informa-tion raises anxiety levels and encourages vot-ers to seek out additional information aboutall candidates and engage in compensatory de-cision making, wherein candidates are com-pared against one another across a variety ofdimensions in order to “choose the best alter-native, regardless of cognitive demands.” Be-cause calm voters remain satisfied with the sta-tus quo, incumbents prefer to avoid campaignattacks except in highly competitive races. Thesame is true of both frontrunners and membersof the majority party. By contrast, challengers,nonfrontrunners, and members of the minorityparty not only have an incentive to go negative,but ultimately must go negative if they are toconvince the electorate that a change of courseis necessary. Nonetheless, as Druckman et al.and others have pointed out, the greater thecompetitiveness of the race, the more likely it isthat both candidates will go negative. Ultimatelycompetitiveness trumps incumbency, frontrun-ner, and majority-party status.

A review of the circumstances under whichcandidates go negative would be incompletewithout assessing whether candidates adoptdifferent message strategies depending on their

gender and that of the opposition. Kahn andKenney (Kahn 1993, Kahn & Kenny 2000) findthat gender does indeed factor into a candidate’sdecision to attack. However, their research hasyielded somewhat inconsistent results. Kahn’s1993 study examining the campaign messagesof 38 candidates for U.S. Senate in the 1984 and1986 races revealed that although both male andfemale candidates prefer policy-based messagesover candidate-oriented appeals, men make useof negative ads less frequently and are partic-ularly reluctant to go negative against femaleopponents, for fear of being perceived as “beat-ing up on a woman” (Kahn 1993, p. 491). Sheargued that regardless of candidate status (in-cumbent or challenger), the messages deliveredby female candidates tend to reflect those ofunderdogs, exhibiting a greater proportion ofnegativism than their opponents demonstrate.

Less than a decade later, Kahn & Kenney(2000) again compared the communicativestyles of male and female senatorial candi-dates, now spanning the 1988, 1990, and1992 campaigns. They found female candidatesconsistently less likely than male candidates—regardless of candidate status—to go nega-tive. Identifying five categories of negative ads(attack, comparative, negative-trait, negative-issue, and criticism), Kahn & Kenney foundthat male challengers consistently launch morenegative appeals than their female counterparts,and in particular are 10% more likely to runnegative-issue ads. This finding is consistentwith the bivariate results reported by Lau &Pomper (2001, 2004). When closeness of raceis taken into account, the differences are evenmore stark; the frequency of male-sponsoredattacks greatly increases with competitiveness,whereas female candidates are more consistent.The two studies are not entirely contradic-tory, however. Kahn and Kenney confirm thatmale candidates remain far more aggressive to-ward other males than toward female oppo-nents, attacking 56% of the time as opposedto 39%.

Kahn and Kenney’s inconsistent findings,and the null effect of gender in a multi-variate analysis reported by Lau & Pomper

294 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

(2001, 2004), cast doubt on any speculation thatfemale candidates adhere to a softer style andmore stereotypically feminine cultural normswhile male candidates are more aggressive andcut-throat. This skepticism is supported byProcter et al. (1994), who examine the use ofnegative advertising in eight 1990 U.S. Senatecampaigns and eight gubernatorial campaignsfrom the same year, all of which featured a fe-male candidate against a male opponent. Proc-ter et al. find that male and female candidatesdo not differ in their use of negative advertising.Female candidates are no less likely to strikefirst, do not manipulate ad formats to softentheir own images, and do not shy away from at-tacking their opponent’s personal ethics or issuepositions. Procter et al.’s findings illustrate thatdecisions to go negative and decisions concern-ing how to strike are not reflections of a candi-date’s gender but rather of a formulaic calculusthat is unaffected by gender. We should men-tion one important caveat here: Research on thecontrasting campaign styles of male and femalecandidates when they oppose each other suf-fers from the absence of an important compari-son group, female-versus-female races. Thereare still too few examples of statewide raceswhere major party female candidates opposeeach other.

DO NEGATIVE CAMPAIGNSWORK?

If you ask observers of the American politicalscene to think of a negative political advertise-ment, they might mention the infamous WillieHorton ads from the 1988 presidential elec-tion campaign, or remember seeing MichaelDukakis riding around in a tank during thatsame campaign. Or they might come up withthe Swift Boat Veterans ads from the 2004 presi-dential campaign. Somewhat older voters mightremember the daisy ad from the 1964 presi-dential campaign. In each case, the target ofthose attacks—Michael Dukakis in 1988, JohnKerry in 2004, Barry Goldwater in 1964—lostthe election. These impressions are perfectlyconsistent with the conventional wisdom about

negative campaigning: It works. This is prac-tically a mantra among political practitioners(Kamber 1997, Swint 1998) and has reverber-ated throughout the scholarly literature as well.

The only reason a rational candidate shouldchoose any campaign strategy is because thatcandidate believes the strategy will win votes.Thus, if attacking one’s opponent is an effectivecampaign strategy, it must result in the attackerreceiving more votes than he or she would haveif some other (less negative) campaign strategyhad been adopted. Assuming the vote choice islittle more than selecting the more highly eval-uated candidate, attacks that lower mean evalu-ations of the target of those attacks would con-stitute additional evidence for the effectivenessof negative campaigns. However, there is alwaysa danger that evaluations of the attacking candi-date will also decrease (a so-called backlash ef-fect: see, for example, Garramone 1984, Roese& Sande 1993). Candidates who attack their op-ponent might be perceived as mean-spirited ornasty, particularly by voters who were taught “ifyou can’t say anything nice, don’t say anythingat all.” But even if backlash effects against thesponsors of campaign attacks are frequent oc-currences, attacking one’s opponent would stillbe an effective strategy if evaluations of the tar-get of the attacks went down more than evalua-tions of the sponsor, such that the net differen-tial evaluation of the two competing candidatesworked to the advantage of the attacker.

Lau et al. (2007) have just published a meta-analytic review of the social science literature onthe effects of negative political campaigns (seeLau et al. 1999 for an earlier review). A meta-analysis is a quantitative synthesis of researchfindings from independent studies on a partic-ular topic, where the reported findings them-selves, rather than the raw data on which theyare based, are the data analyzed. A meta-analysisis as comprehensive as possible, including bothpublished and unpublished studies so that well-known publication biases (e.g., only publishingpapers with statistically significant findings), orthe prejudices of the reviewer (e.g., deciding aparticular study is “junk” and therefore shouldnot be included in the review), do not influence

www.annualreviews.org • Negative Campaigning 295

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

the outcome of the meta-analysis. Any factorthe analyst believes could influence the magni-tude of the reported results (e.g., experimentalversus correlation designs; the perceived qual-ity of the reported study) can be coded andtested empirically—if there are enough relevantstudies.

Lau et al. (2007) located 43 studies examin-ing the effects of negative campaigns (or expo-sure to particular negative advertisements) onthe actual or intended vote choice of those ex-posed to the campaigns/advertisements. For ex-ample, King et al. (1998) exposed subjects to ei-ther a positive or negative ad from Bill Clintonor Bob Dole near the end of the 1996 presi-dential campaign. Controlling for initial likingof the candidates, King et al. found that expo-sure to a negative ad from Clinton strongly de-creased intention to vote for him, but exposureto a negative ad from Dole had no effect on like-lihood of voting for him. In neither case is thereany evidence that negative ads are particularlyeffective in winning votes. Of the 43 relevantstudies, 12 reported positive effects (that is, themore negative candidate received more votes),four of which were statistically significant; 28studies reported unfavorable effects (that is, at-tacking the opponent was counterproductive),six of which were statistically significant. Threestudies reported that there were “no significantdifferences,” without providing any numbers orhints as to which direction the nonsignificantresults leaned; in such cases, results can only becoded as having an effect size of 0. Looking atall of these studies together, no one could con-clude that negative campaigning is a particularlyeffective campaign strategy.

Many factors go into a vote decision, ofcourse, including habit, and it may be ask-ing a lot of a typical political campaign, nevermind exposure to a few ads in an experiment,to change that decision. But surely, one wouldthink, evaluations of the targets of political at-tacks must go down. Lau et al. (2007) located 31studies with relevant data. For example, Kahn &Kenney (2004) coded the tone of up to five adsaired by major party candidates in 97 competi-tive senate elections between 1988 and 1992.

Controlling for a large number of factors ina multivariate analysis, Kahn & Kenney re-port that both incumbents and challengers wereliked significantly less when they were attackedby their opponent than when their opponentstayed primarily positive. Twenty-three of the31 studies reported similar effects, 12 of whichwere statistically significant. Only seven stud-ies (two of which were significant) reportedthe opposite effect, with evaluations of the tar-get actually increasing after the attack—whichmight plausibly occur if many people perceivedan attack to be exaggerated, false, or otherwiseout-of-bounds, and responded in part by sym-pathetically increasing their evaluations of thetarget of the unjustified attacks. In any case, weare reasonably confident that attacking a po-litical opponent is generally likely to result inlower evaluations of that candidate.

But attacking one’s opponent is not a risk-free enterprise, as evaluations of the attackermight also decrease if voters are convinced theattacks are unjustified or mean-spirited. Thissame study by Kahn & Kenney (2004) reportedthat evaluations of both incumbents and chal-lengers decreased (the former significantly) af-ter they sponsored attacks against their oppo-nent. All told, 31 of the 40 relevant studies (19of which were statistically significant) reportedsimilar findings, whereas only six studies (twosignificant) found evaluations of the attacker in-creasing after the attacks. If anything, this back-lash effect appears to be a bit stronger and moreconsistent than lowering evaluations of the tar-get of the attacks. On balance, then, there issimply no support in the scientific literature forthe hypothesis that negative campaigns are anymore effective than any other type of campaignstrategy.

To this point we have not commented onthe different designs that researchers employto study the effects of negative campaigns. Weassume the reader is generally familiar with therelative advantages and disadvantages of experi-ments and correlational (or observational) stud-ies. Despite what we often tell our undergrad-uates, it is often possible to make reasonablyconfident causal statements from correlational

296 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

data. But the very fact that we can reliably iden-tify several factors that explain when candidatesdecide to go negative raises a difficult but veryimportant methodological problem when wetry to examine the success of different strate-gies with observational data from real electioncampaigns. In statistical terms, this problemarises because the choice of campaign strategyis endogenous to the subsequent outcome ofthe election itself, which makes it very difficultto determine the effect of the campaign on theoutcome of the election. Now, how can a cam-paign strategy, which logically must take placeprior to election day, be endogenous to (influ-enced by) the outcome of the election, whichwe only know after the campaign is over andthe polls close on election day? This is possi-ble because many of the factors that help deter-mine the outcome of the election are widelyknown long before election day, and there-fore are likely to influence candidates’ choiceof campaign strategies in the first place. Mostincumbents running for re-election are betterknown, better liked, and better financed thantheir challengers, and thus can reasonably ex-pect to win ( Jacobson 2004). As we have seen,challengers and incumbents unexpectedly find-ing themselves in close elections are more likelyto decide to go negative . . . but they are alsomore likely to lose. This makes it extremelydifficult to determine if such a candidate lostbecause of her choice of campaign strategy, orin spite of it. This statistical problem does havea statistical solution: instrumental variables ortwo-stage least squares. As Bartels (1991) warns,however, this solution is no better than thequality of the instruments. See Lau & Pomper(2002) for further discussion of this issue anddetails on one reasonably successful attempt toprovide such instruments.

Does this methodological problem accountfor the lack of empirical evidence for the ef-ficacy of negative campaigning? Probably not;in their meta-analytic review of the litera-ture, Lau et al. (2007) report that experiments(which avoid the methodological problem dis-cussed above by randomly assigning campaign

strategies to different candidates) were no morelikely to produce positive results than werestudies using various nonexperimental designswhere the endogeneity problem must be con-fronted. Twelve of the 25 experimental stud-ies showed positive effects (that is, the morenegative candidate received more votes), whichwere statistically significant in four of the stud-ies; three of the 13 negative findings were sig-nificant, and the overall mean effect was –0.02.Again, there is no way one could conclude thatnegative ads or negative campaigns are anymore or less effective than more positive adsor campaigns. Still, this methodological issuemust be kept in mind when reviewing the avail-able literature or when thinking about the suc-cess of any particular negative campaign.

At this point one may reasonably ask why,if attacks are not particularly effective, politicalconsultants continue to urge candidates to at-tack their opponents. There are several quickanswers. First, as discussed above, few politicalcampaigns are in fact overwhelmingly negative.We are more likely to hear about negative cam-paigns in the media, but relatively few ads areentirely negative, and most campaigns, on bal-ance, are probably more positive than negative.Second, to say that all or even most politicalattacks are not effective is not to say that at-tacks are never effective. Even if all electionsinvolved only two candidates, half of all polit-ical campaigns would have to fail. Rarely arepolitical consultants trained as scientists, andthey have few incentives—and even less timeand money—to study the efficacy of their ac-tions. Like most of us, they are likely to repeatwhat they have done in the past. Furthermore,the time horizon of a consultant for judging thesuccess of a particular ad campaign is probablyshorter than that of a researcher, who inevitablyis not looking at the campaign until the electionis over. All of these factors combine to makenegative campaigns a steady part of the politi-cal scene, even if they sometimes, or even mosttimes, do not ultimately produce the resultsthat are desired by the candidates who employthem.

www.annualreviews.org • Negative Campaigning 297

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

THE DEMOBILIZATIONHYPOTHESIS

The scholarly debate over the impact of nega-tive advertising has centered on the demobiliza-tion hypothesis, developed by Ansolabehere,Iyengar, and colleagues (Ansolabehere et al.1994, Ansolabehere & Iyengar 1995). Seekingto establish a causal link between the presumedrise of negative advertising and the steadilydeclining turnout in national elections, An-solabehere et al. suggested three reasons whynegative campaigns might lower turnout. First,negative campaigns could be successful in low-ering the probability of voting for the targetof the attacks without simultaneously increas-ing the probability of voting for the sponsor ofthe attacks, resulting in a decision to stay homeon election day. Second, negative campaignscould lower evaluations of both the target andthe sponsor of the attacks, thus decreasing theprobability of voting for either of them. Third,negative campaigns could influence turnout in-directly through an effect on civic attitudes suchas trust in government and political efficacy,which in turn are often associated with turnout.

Ansolabehere et al. (1994, 1995) tested thesehypotheses by conducting a series of controlledexperiments on the effects of exposure to a sin-gle positive or negative political ad. Partici-pants, all of whom were drawn from the LosAngeles area, were shown a 15-minute localnewscast and systematically exposed to one oftwo experimental political ads (either positiveor negative) embedded midway through thebroadcast. While manipulating tone by chang-ing key words within the script, both ads usedthe same announcer and were visually and the-matically identical. In order to accurately re-flect real-world ad exposure, Ansolabehere et al.used authentic political ads spanning a variety ofCalifornia campaigns, including the 1990 gov-ernor’s race, the 1992 senate races (primariesand general election), and the 1993 Los Angelesmayor’s race. Upon comparing ad exposure toeach individual’s intention to vote and generallevel of political participation, Ansolabehereet al. found that participants exposed to the

negative ad were 5% less likely to intend to votein the upcoming election compared to thoseexposed to a positive ad. A subsequent study,combining analysis of raw aggregate turnoutand ballot roll-off data from all thirty-four 1992senate races with a tone-specific content anal-ysis of the news coverage of those campaigns,yielded similar results. (Roll-off occurs when anindividual only votes in the race at the top of theballot—president—and abstains from voting inthe senate election that was on the same ballot.)Ansolabehere et al. go on to report that politicalindependents seem particularly susceptible todemobilization from exposure to negative ads,and further report a general increase in polit-ical cynicism and lower sense of political effi-cacy among those exposed to negative ads. Inconcluding that exposure to negative advertis-ing does in fact reduce turnout and erode con-fidence in the political process, Ansolabehereet al. warn that because “candidates with suffi-cient resources can, through the use of negativemessages, keep voters away from the polls,” thephenomenon poses a unique threat to democ-racy (Ansolabehere et al. 1994, p. 835).

This is wonderful research, with clear theo-retical predictions and amazingly consistent re-sults across two very different research designs,and it did what all wonderful research shoulddo: generate a lot more research. Finkel & Geer(1998) were among the first to respond. Whileacknowledging the logic of Ansolabehere et al.’sthree reasons why negative campaigns mightdecrease turnout, Finkel & Geer counter withthree reasons why negative campaigns mightactually stimulate turnout. First, negative ad-vertising often conveys large amounts of pol-icy and retrospective performance informationto voters, and knowledgeable citizens are morelikely to participate. Second, negativity effectscould make attack ads more useful than com-parable positive ads in helping voters discrim-inate between the two candidates, and peoplewith distinct alternatives are more likely to vote.Third, negative ads may be more likely to pro-duce strong emotional responses that can getpeople to care more about the outcome of theelection, which again should increase turnout.

298 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

Finkel & Geer (1998) then turn to an empir-ical investigation of whether the negativism ofpresidential election campaigns affects turnout.They measured the tone of presidential elec-tion campaigns by the percentage of negativethemes in the campaign ads aired by the majorparty candidates between 1960 and 1992. Finkel& Geer use this variable to predict turnout ina thoroughly specified multivariate equation.Controlling for the standard demographic andpolitical variables, they find no systematic rela-tionship between turnout and tone. Thus, hereis one study that adopted a different method-ology and studied a different type of elections,but found no support for demobilization due toexposure to negative campaigns.

The major weakness of this study, as Finkel& Geer (1998) readily admit, is that campaigntone only varies across election years, provid-ing very gross measurement of the crucial in-dependent variable. When this study was pub-lished, there were no better data available. Geer& Lau (2006) try to overcome this limitationwith a Bayesian analysis that treats these yearlyaverages as a baseline around which one canestimate plausible variations of tone across dif-ferent electoral contexts (states)—and find verysimilar results.

But there have been many more tests of thedemobilization hypothesis. The meta-analyticreview of the literature by Lau et al. (2007)mentioned above identified 55 additional testsof demobilization. Of the results of those tests,25 were in the demobilization direction (ninestatistically significant results) and 29 were inthe mobilization direction (eight of which weresignificant). This hypothesis has been such a fo-cus of research in the field that we want to verycarefully illustrate the lack of any clear find-ings. Figure 3 displays the adjusted effect sizefor all 56 findings reported in the social sci-ence literature. An effect size is a way of ex-pressing the results of many different studiesin the same metric. For the prototypic study, itrepresents the difference in turnout between agroup exposed to a positive ad (or campaign)and another group exposed to a negative ad(or campaign), expressed in standard deviation

units. These raw effect sizes can then be ad-justed for sampling error (so that studies withlarger sample sizes count more than studies withsmaller sample sizes) and unreliability of mea-surement of the dependent variable. By meta-analytic standards, there is one very large pos-itive effect in this literature, one moderatelylarge positive effect, one moderately large neg-ative effect, and 53 findings that hover a littleabove or a little below zero but do not amountto much. In sum, the research literature pro-vides no general support for the hypothesis thatnegative political campaigning depresses voterturnout. In fact, if you had to bet on one orthe other, mobilization is a bit more likely thandemobilization.

Some of the individual findings in this liter-ature are based on very large samples. As is al-ways the case with statistics, particular findingscan be statistically significant but not representvery large effects. For example, the experimen-tal work reported by Ansolabehere & Iyengar(1995) is easily statistically significant with a tvalue of −2.56, but because it is based on a sam-ple of 2216 subjects, its effect size is only −0.10.At the same time, the accompanying analy-sis of 34 senate election campaigns from 1992(Ansolabehere et al. 1999) has a larger t value(–3.11), but this represents a much larger effect(–1.27) because it is based on only 34 cases. Ofcourse we might trust a data point based on>2000 cases a lot more than one based on 34cases, and it is conventional in the meta-analyticliterature to adjust all effect sizes for samplingerror, which has the effect of weighting findingsbased on larger samples relatively more thanthose based on a smaller number of cases. Lauet al. (2007) report the raw, unadjusted averagesalong with adjusted means; never do the variousmeasures differ by very much.

It could still be the case that although votersare not turned off by all campaign negativism,when legitimate criticism crosses the line intomudslinging, demobilization occurs. But whenexactly does a critical ad cross the line?Goodman (1995), writing in Campaigns andElections, a magazine for political consultants,attempts to clarify the point: “Does it feel

www.annualreviews.org • Negative Campaigning 299

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

negative? Is it mean-spirited? Is it crudely pro-duced? Does it go beyond the pale of good tasteand appropriate manners? Cross any of theselines, and you’ve crossed the viewer” (p. 23).Besides the fact that production sophisticationis hardly an operational measure, the author’sdescription of inappropriate negative advertis-ing reflects another problem: Like the percep-tion of campaign negativism, the “point of noreturn” is very much in the eye of the beholder.Although we can agree that presentations ofdocumented facts are fair game, we are alsoaware that talented strategists are capable ofmanipulating facts in such a way that they dis-tort the truth. Even topics that may be consid-ered “off limits” (for example, those listed in thebottom panel of Table 1) can be framed in sucha way that they are perceived by some voters tobe relevant.

Yet Kahn & Kenney (1999) maintain thatvoters know the difference. They examine threedistinct measures of the nature of the 1990 sen-ate election campaigns: the tone of up to five adsaired by each candidate, the negativism of news-paper coverage of the campaign, and, of particu-lar concern here, a judgment by the media con-sultants involved in these campaigns of whetherthe campaign degenerated into mudslinging.Kahn & Kenney use all three of these measuresin a multivariate equation predicting turnout.Controlling on a variety of demographic, psy-chological, and political orientations, Kahn &Kenney find that both the negativism of thecandidates’ political ads and the amount ofcriticism they received in newspaper cover-age were associated with higher turnout. Cam-paigns involving mudslinging, however, hadsignificantly lower turnout. More fine-grainedanalyses suggest that political independents andthose with little interest in politics were partic-ularly affected by mudslinging. Although otherresearchers have had difficulty replicating thisfinding (see Brooks & Geer 2007, Jackson &Sides 2006), Kahn & Kenney’s research pro-vides some evidence that voters are capable ofdiscerning malice from legitimate critiques, andthat it is the former that turns voters off frompolitics, whereas legitimate, campaign-relevant

negative information mobilizes participation,particularly among independents and thosewho are generally more interested in or moreknowledgeable about the political process.

METHODOLOGICAL CONCERNS

The major methodological issues plaguing thestudy of political campaigns can be placed intothree broad categories: the lack of good dataon the very nature of political campaigns, thedifficulties of measuring exposure to politicalcampaigns, and the statistical assumptions thatare sometimes unwittingly made in our analysisof campaign effects. We are making progress onall of these fronts, however. The designs thatresearchers use to study political campaigns areone important vehicle for addressing some ofthese problems.

Data

As discussed above, the most fundamental prob-lem in the study of political campaigns involvesdata: the lack, until very recently, of any goodevidence on exactly what candidates actuallydo when they are running for office. With re-cent technological advances this information isbecoming easier to gather, but there are stillmajor parts of campaigns that remain largelyhidden from the researcher’s view. Before wehad more objective measures of the actual con-tent of political campaigns, some researchersasked survey respondents to recall an adver-tisement from some campaign, code the na-ture of the remembered ad, and then try tojudge the impact of that ad on the voter. Alter-natively, researchers asked respondents to ratehow positive or negative a particular campaignhas been. But memory is notoriously unreliable,and we have already seen that subjective per-ceptions of how negative a campaign is sufferfrom serious partisan biases. Noisy data about aphenomenon are always better than no data atall, but more objective, systematically gatheredmeasures should almost always be preferred toless reliable subjective measures. Today thereshould be no reason to have to rely on memory

300 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

of political advertisements from any campaignas the primary measure of that campaign’s tone.

We have already mentioned the CampaignMedia Analysis Group (CMAG), which since1998 has tracked every political advertisementaired on television in the 75 largest media mar-kets (which include 80% of the population) inthe United States.3 Among campaign profes-sionals, the television and radio advertisementsare often called the “air war.” But there is also a“ground war,” which consists of the speechescandidates deliver and campaign visits theymake to different constituencies, the bulk mail-ings they send out, the billboards, lawn signs,bumper stickers, and get-out-the-vote phonecalls. Any candidate blessed with sufficient re-sources to advertise on television also typicallyinvests heavily in different aspects of the groundwar. In fact, because ground war activities arenot included in the definition of electioneeringcommunication by the Bipartisan CampaignReform Act of 2002, there are now incentivesfor parties to shift their campaign activities to-ward the ground (Dwyre & Kolodny 2003).It is usually possible to obtain a fairly com-plete record of the campaign stops a candidatemakes from media accounts (see Shaw 1999),but we have not seen anyone systematically col-lect this information except at the presidentiallevel. Magelby et al. (2006) describe a new datacollection effort in which a random sample ofregistered voters were asked to collect every bitof campaign mail and log every political phonecall and personal contact they received duringthe last three weeks of the 2004 presidentialelection (see also Hillygus & Monson 2006).But this is only one election, and the focus is onthe presidential campaign. Today we must alsospeak of the “cyber war,” the candidates’ offi-cial web pages and the less official pages of their

3The Wisconsin Political Advertising Project also hasCMAG data from the 1996 presidential election, but for noother races that election year. CMAG also records radio adsand the text of many local and cable news programs (to theextent their content has been transcribed for the hearing im-paired), but these data are not stored at Wisconsin and arethus much more difficult for political scientists to obtain. Wehave never seen anyone utilize these other CMAG data.

supporters (Cornfield 2004; Druckman et al.,unpublished manuscript). Limiting attention toonly one aspect of a political campaign providesat best an incomplete image, and at worst a mis-leading picture, of the entire campaign. Yet fewresearchers have the resources to consider morethan one or (rarely) two of these different as-pects of a campaign. It is vitally important thatany researcher who gathers data on the natureof any of the multiple components of a politi-cal campaign make that data available to otherresearchers studying campaigns.

Exposure

In gauging the effects of any aspect of a cam-paign, one also must be careful to distinguishthe content or tone of the campaign from itsvolume. In using the CMAG data to mea-sure negative advertising campaigns, for ex-ample, some researchers have mistakenly em-ployed the number of negative ads shown ina media market as their primary measure ofcampaign negativism. Without a comparablemeasure of the number of positive ads shownin that same market, this approach confoundsthe nature of the advertising campaign with itsmagnitude, as Dan Stevens details in his unpub-lished manuscript, “More than Volume: Pro-portion, Volume, and the Effects of Exposureto Negative Advertising.” The resulting regres-sion coefficient will be estimating the effect ofone more negative ad, all else (in the equation)equal, that is, irrespective of the number of pos-itive ads that have also been shown. This makeslittle sense. We recommend devising separatemeasures of the nature or tone of the campaign(e.g., the proportion of all ads shown in a par-ticular market that are negative) and of its vol-ume (the total number of ads shown in thatmarket).

Even if we have a very good idea about howmuch a candidate advertises in a particular me-dia market, we still do not know how manycampaign messages any individual in that me-dia market has been exposed to. There is a largerandom element to individual exposure levels,but there is a larger nonrandom component to

www.annualreviews.org • Negative Campaigning 301

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

exposure because most of us are creatures ofhabit. Some of us watch three or four hoursof television almost every day whereas oth-ers watch an hour or less. Some of us watchtelevision in the morning while getting readyfor the day to start, others in the afternoonbefore the kids come home from school, andothers exclusively during the prime-time hoursin the evening. Freedman & Goldstein (1999)have devised survey-based measures of typi-cal viewing habits that can be combined withobjective measures of the number and mix ofads that were shown at various times of day.This is analogous to, although more compli-cated than, asking survey respondents which (ifany) newspaper they typically read before es-timating the impact of editorial endorsementson the vote choice. Best practice now dictatestrying to combine some measure(s) of the con-tent of a media message with measures of itslikelihood of being heard, either through globalmeasures of volume or, ideally, with the addi-tion of more idiosyncratic measures of individ-ual viewing habits.

Statistical Assumptions

Whenever we study a random sample of vot-ers but measure media content at some ag-gregated level (such as the media market), weintroduce another source of potential bias toour studies: data clustering. Inevitably withsuch research designs, individual respondentsare nested within media markets, which vio-lates basic assumptions of our most commonmultivariate statistical techniques (Steenbergen& Jones 2002). Unfortunately, following “bestpractices” in media research does not yet in-clude explicitly modeling the multilevel struc-ture of our data, and the conclusions from thatresearch must consequently be somewhat sus-pect. One recent study found that for the mostpart, the results of multilevel analysis convergewith those of conventional methods (Lau &Steenbergen 2006), but this result could be dueto the small intraclass correlations observed inthe three datasets examined in that paper. How

general this result will prove to be is still anopen question.

We have already discussed the problem ofthe endogeneity of the decision of how to cam-paign with the outcome of the election, whichmakes causal statements about the likely effectsof those campaign strategies much more diffi-cult to make. Holding that problem aside, thereis also the issue that every effort by one can-didate to convince citizens to vote for her istypically offset by the efforts of another can-didate trying to convince those same citizensto vote for him. Cast in this light, it may seemmiraculous whenever we can detect any effect ofa campaign on its outcome. Fortunately, thereis still a great deal of variance in the conductof campaigns that gives us leverage to esti-mate their effects. Even if candidates had thesame amount of money to spend (as presiden-tial candidates do in the United States, if theyaccept public financing), they do not have tospend it in the same way, nor in the same lo-calities. Competing candidates typically do nottry to deliver the same message or focus on thesame issues in their advertising, nor are indi-vidual candidates required to deliver the samespeeches or use the same advertisements in ev-ery locale within their constituency. Such vari-ance is great for analytic purposes, but we mustremember to take it into account in our data-gathering efforts. The Federal Election Com-mission, for example, has since its inceptionkept very good records on how much moneycandidates and parties raise, who they raise itfrom, and how much they spend. But there isalmost no reporting on where the money isspent.

The study of political campaigns is madeeven more difficult by the fact that in mostelections there are multiple races on the bal-lot. Every two years all citizens face an elec-tion selecting their member of the House ofRepresentatives, but usually there is at leastone and sometimes more simultaneously oc-curring statewide races for senator, governor,lieutenant governor, and so on. And every fouryears the presidential election is also on the

302 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

ballot. How much information are we losingby the almost universal practice of focusingon only one of these campaigns and ignoringevery other race on the ballot? If our goal isto study the effectiveness of a particular cam-paign strategy, the practice might be justified.If instead we are interested in the indirect andpresumably unintentional effects of negativecampaigns on the political system as a whole(e.g., political efficacy, turnout), the practiceseems much harder to justify. Most voters donot make race-by-race decisions of whether tovote—they make one universal decision to go tothe polls or stay home, and to fully understandthat universal decision, we must consider all ofthe political stimuli to which they have beensubjected.

Research Designs

All of the methodological issues raised so farapply to the problems of studying real-worldpolitical campaigns. Most of these problemsare solved by experiments, although now thenatural focus becomes studying the effects ofparticular campaign events (e.g., individualcampaign ads) rather than the effects of moreextended campaigns. This review is not theplace to discuss the relative merits of experi-mental and nonexperimental designs, or eventhe desirability of conducting “realistic” asopposed to more artificial or “ideal setting”experiments. Some degree of artificiality is in-evitable in almost any controlled environment,but particularly in a discipline such as politicalscience, many of whose practitioners have notbeen trained as experimental scientists, the ex-perimental methodology developed by Iyengar& Kinder (1987) is about as good as it gets,providing high levels of internal and externalvalidity. In the Iyengar & Kinder protocol, acarefully controlled experimental manipulationis unobtrusively delivered in a very familiarsetting—for example, an experimental ad isinserted into the normal commercial break ina tape of an actual television program.

Experiments may not be as useful in tryingto understand the decisions made by candidates

and their political consultants during an actualpolitical campaign, however. If this is one’s goal,then longitudinal designs, which gather infor-mation (e.g., relative standing in the polls) andmeasure behavior (the volume and tone of cam-paigning in different media markets) at multi-ple points in time across an entire campaign, arecrucial. We as scientists can wait to analyze thesuccess of any advertising campaign until afterit is over, but campaigners must make their de-cisions during the heat of battle, so to speak,and undoubtedly under conditions of consider-able uncertainty. Fortunately the CMAG dataare collected with very precise time stamps, sowe know exactly when and where different adsare shown. It is typically much harder to getcomparably precise measures of the public’s re-sponse to any advertising campaign, but manycampaigns conduct tracking polls that provideexactly those data—if we can get a hold of them.Developing personal relationships with polit-ical consultants who often manage high-levelpolitical campaigns can prove invaluable to suchan enterprise [see Sides (2006) for a recent studyutilizing this type of data].

CONCLUSION: ARE NEGATIVECAMPAIGNS BAD FORDEMOCRACY?

The literature on the use and utility of negativecampaigns is extensive, including descriptiveaccounts, systematic experimentation, surveyresearch analysis, and formal theoretical pre-scriptions. Although we have not had the spaceto present a comprehensive review of all of thisresearch, we do believe we have presented athorough one. Descriptive accounts, many ofwhich are by political consultants rather thanacademics, address the strategic concerns guid-ing the decision to attack. Empirical investi-gations include both systematic manipulationsof hypothesized causal factors, analysis of me-dia content—real, as well as fictitious—and sur-vey analysis of voter reactions to actual politicalcampaigns.

Yet there is nothing in what we have pre-sented so far that would allow us to say that

www.annualreviews.org • Negative Campaigning 303

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

negative campaigns are inherently good or bad.Many recognize that voters are capable of sep-arating vicious attacks from legitimate compar-isons but also appreciate that even justifiableattacks are not always effective, nor are theyalways desirable. The standards for makingsuch judgments should be the public’s inter-ests, not the candidates’. Elections are, afterall, about choices, not courtesy (Lau & Pomper2004), so questions about the effectiveness ofnegative campaigns become irrelevant.

The literature does reflect an overall in-dictment of baseless attacks and unwarrantedcharacter assaults. Beyond that, however, existsa spectrum of views. The widespread concernthat negative campaigns can reduce turnout ap-pears unwarranted. There is less concern butstronger evidence that negative campaigns candeleteriously affect the political system in otherways, for example, by reducing political efficacyand decreasing trust in government (Lau et al.2007). But negative campaigns also stimulateinterest and learning about the candidates andissues in an election.

Is it unfair to objectively point out an oppo-nent’s shortcomings? Conversely, are so-called“positive” ads that boast of a candidate’sstrength to the point of exaggerated self-aggrandizement fair, simply because they arenot negative? Mayer (1996) points out that nocandidate can speak honestly and substantivelyabout what he intends to do upon election with-out drawing a contrast to his opponent and ex-plaining why he is a superior choice over theother candidate. If we assume that voters wantto know why a given candidate is better suitedor more qualified than his or her opponent, wecannot legitimately fault a candidate for tact-fully juxtaposing him- or herself against thatopponent.

Geer (2006) takes this argument even far-ther. All advertisements, by their very nature,exaggerate the truth. But this point appliesequally to positive campaigns as well as to neg-ative. We cannot expect candidates to discusstheir own policy failures and shortcomings, anymore than we can expect Chevrolet to tell usabout the problems with its cars, or Hoover

to tell us its vacuum cleaners sometimes fail tosuck. That is the job of the opposition, in pol-itics as well as in commerce. Geer argues thatdemocracy requires negative campaigning if cit-izens are to become fully informed about thechoices they face.

Although Ansolabehere & Iyengar (1995)conclude their book with policy recommenda-tions aimed at limiting the extent of negativecampaigning, we see no need to impose any fur-ther restrictions on negative advertising. Votersare clearly capable of distinguishing legitimatecriticism from unfair attacks and punish can-didates who are out of line. We do approve ofrecent legislation requiring federal candidatesto appear in and publicly approve of the adsthey sponsor, as this increases accountability inelection campaigns and exposes those individu-als who would prefer to shield themselves fromboomerang effects by allowing surrogates to dothe attacking. Of course what this change hasreally done is push most of the attacking off toparty-sponsored ads, but as long as voters canhold the parties accountable for their actions,that too is fine.

Although it is difficult to convincingly docu-ment the widely presumed increase in negativecampaigning over recent decades, no one is sug-gesting the practice is on the decline, and thereis little doubt that negative campaigns are animportant part of the current political scene.Thus we are sure that negative campaigningwill remain fertile ground for political scienceresearch for years to come. Future researchshould explore the impact of negative ads thatare sponsored by “independent” (504) groupsand individuals other than the candidates them-selves. It would be good to examine the ex-tent to which the messenger matters, and howattacks from third parties are weighted rela-tive to those launched personally by candidates.How does the timing of attacks during a cam-paign, or the medium through which attacksare delivered, influence their effectiveness? Isthe internet a medium where almost anythinggoes, and candidates can “get away with” morevicious attacks of their opponents? What is therole of music, production quality, or particular

304 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

types of images in the effectiveness or effects ofnegative advertisements? Despite the volumeof research already available on the decision to

attack and its impact on political behavior, amyriad of questions remain for future researchto address.

DISCLOSURE STATEMENT

The authors are not aware of any affiliations, memberships, funding, or financial holdings thatmight be perceived as affecting the objectivity of this review.

LITERATURE CITED

Ansolabehere S, Iyengar S. 1995. Going Negative. New York: Free PressAnsolabehere S, Iyengar S, Simon A. 1999. Replicating experiments using aggregate and survey data: the case

of negative advertising and turnout. Am. Polit. Sci. Rev. 93:901–9Ansolabehere S, Iyengar S, Simon A, Valentino N. 1994. Does attack advertising demobilize the electorate?

Am. Polit. Sci. Rev. 88:829–38Bartels L. 1991. Instrumental and “quasi-instrumental” variables. Am. J. Polit. Sci. 35( July):777–800Brader T. 2005. Striking a responsive chord: how political ads motivate and persuade voters by appealing to

emotions. Am. J. Polit. Sci. 49(Apr.):388–405Brader T. 2006. Campaigning for Hearts and Minds. Chicago: Univ. Chicago PressBrooks D, Geer J. 2007. Beyond negativity: the effects of incivility on the electorate. Am. J. Polit. Sci. 51( Jan.):1–

16Buell E, Sigelman L. 2008. Attack Politics: Negativity in Presidential Campaigns since 1960. Lawrence: Univ. Press

KansasCornfield M. 2004. Politics Moves Online: Campaigning and the Internet. New York: Century Found. PressDwyre D, Kolodny R. 2003. National parties after BCRA. In Life after Reform: When the Bipartisan Campaign

Reform Act Meets Politics, ed. M Malbin, pp. 95–112. Lanham, MD: Rowman & LittlefieldFinkel S, Geer J. 1998. A spot check: casting doubt on the demobilizing effect of attack advertising. Am. J.

Polit. Sci. 42:573–95Freedman P, Goldstein K. 1999. Measuring media exposure and the effects of negative campaign ads. Am. J.

Polit. Sci. 43:1189–208Freedman P, Wood W, Lawton D. 1999. Do’s and don’ts of negative ads: what voters say. Campaigns Elections

20:20–25Garramone G. 1984. Voter response to negative political ads. J. Q. 61:250–59Geer J. 2006. In Defense of Negativity: Attack Ads in Presidential Campaigns. Chicago: Univ. Chicago PressGeer J, Lau R. 2006. A new approach for estimating campaign effects. Br. J. Polit. Sci. 35:269–90Goodman A. 1995. Producing TV: a survival guide. Campaigns Elections 16:22–24Harrington J, Hess G. 1996. A spatial theory of positive and negative campaigning. Games Econ. Behav. 17:209–

29Haynes A, Rhine S. 1998. Attack politics in presidential nomination campaigns: an examination of the frequency

and determinants of intermediated negative messages against opponents. Polit. Res. Q. 51:691–721Hillygus S, Monson J. 2006. Campaign microtargeting and presidential voting in 2004. Presented at Annu. Meet.

Am. Polit. Sci. Assoc., 102nd, PhiladelphiaIyengar S, Kinder D. 1987. News that Matters. Chicago: Univ. Chicago PressJackson R, Sides J. 2006. Revisiting the influence of campaign tone in senate elections. Polit. Anal.

14(Spring):206–18Jacobson G. 2004. The Politics of Congressional Elections. New York: Longman. 6th ed.Jamieson KH, Campbell C. 1983. The Interplay of Influence: Mass Media and their Publics in News, Advertising,

Politics. Belmont, CA: WadsworthJamieson KH, Waldman P, Sherr S. 2000. Eliminate the negative? Categories of analysis for political adver-

tisements. In Crowded Airwaves, ed. JA Thurber, CJ Nelson, DA Dulio, pp. 44–64. Washington, DC:Brookings Inst.

www.annualreviews.org • Negative Campaigning 305

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

ANRV377-PL12-15 ARI 2 May 2009 12:49

Kahn KF. 1993. Gender differences in campaign messages: the political advertisements of men and womencandidates for U.S. Senate. Polit. Res. Q. 46:481–502

Kahn KF, Kenney P. 1999. Do negative campaigns mobilize or suppress turnout? Clarifying the relationshipbetween negativity and participation. Am. Polit. Sci. Rev. 93:877–89

Kahn KF, Kenney P. 2000. How negative campaigning enhances knowledge of senate elections. In CrowdedAirwaves, ed JA Thurber, CJ Nelson, DA Dulio, pp. 65–95. Washington, DC: Brookings Inst.

Kahn KF, Kenney P. 2004. No Holds Barred: Negativity in U.S. Senate Campaigns. Upper Saddle River, NJ:Pearson Educ.

Kamber V. 1997. Poison Politics: Are Negative Campaigns Destroying Democracy? Cambridge, MA: PerseusKanouse D, Hanson L Jr. 1971. Negativity in evaluations. In Attribution: Perceiving the Causes of Behavior, ed.

EE Jones, DE Kanouse, HH Kelley, RE Nisbett, S Valins, B Weiner, pp. 47–62. Morristown, NJ: GeneralLearning

Kern M. 1989. Thirty Second Spots: Political Advertising in the Eighties. New York: PraegerKernell S. 1977. Presidential popularity and negative voting: an alternative explanation of the midterm con-

gressional decline of the president’s party. Am. Polit. Sci. Rev. 71(Mar.):44–66King E, Hendersen R, Chen H. 1998. Viewer response to positive vs negative ads in the 1996 presidential campaign.

Presented at Annu. Meet. Midwest Polit. Sci. Assoc., ChicagoLau R. 1982. Negativity in political perception. Polit. Behav. 4(Dec.):353–78Lau R. 1985. Two explanations for negativity effects in political behavior. Am. J. Polit. Sci. 29(Feb.):119–38Lau R, Pomper G. 2001. Negative campaigning by U.S. Senate candidates. Party Polit. 7( Jan.):69–87Lau R, Pomper G. 2002. Effectiveness of negative campaigning in U.S. Senate elections. Am. J. Polit. Sci.

46:47–66Lau R, Pomper G. 2004. Negative Campaigning: An Analysis of U.S. Senate Elections. New York: Rowman &

LittlefieldLau R, Sigelman L, Heldman C, Babbitt P. 1999. The effects of negative political advertisements: a meta-

analytic assessment. Am. Polit. Sci. Rev. 93:851–75Lau R, Sigelman L, Rovner IB. 2007. The effects of negative political campaigns: a meta-analytic reassessment.

J. Polit. 69(Nov.):1176–209Lau R, Steenbergen M. 2006. Reconsidering “best practices” in political communications research. Presented at Annu.

Meet. Midwest Polit. Sci. Assoc., 64th, ChicagoMagelby D, Monson JQ, Patterson K. 2006. Mail communications in political campaigns: the 2004 campaign

communications survey. Presented at Annu. Meet. Midwest Polit. Sci. Assoc., ChicagoMansnerus L. 2005. Forrester tosses his “Hail Mary” ad. NY Times, Oct. 23, p. NJ14Marcus G, Neuman WR, MacKuen M. 2000. Affective Intelligence and Political Judgement. Chicago: Univ.

Chicago PressMayer W. 1996. In defense of negative campaigning. Polit. Sci. Q. 111:437–55Proctor D, Schenck-Hamlin W, Haase K. 1994. Exploring the role of gender in the development of negative

political advertisements. Women Polit. 14(2):1–22Roese N, Sande G. 1993. Backlash effects in attack politics. J. Appl. Soc. Psychol. 23(Aug.):632–53Shaw D. 1999. The effect of TV ads and candidate appearances on statewide presidential votes, 1988–1996.

Am. Polit. Sci. Rev. 93( June):345–61Sides J. 2006. Swing your partner? The importance of interactions in campaigns. Presented at Annu. Meet. Am.

Polit. Sci. Assoc., Washington, DCSides J, Grossmann M, Trost C, Lipsitz K. 2003. Candidate attacks and voter aversion: the uncertain link between

negativity and campaign satisfaction. Presented at Annu. Meet. Am. Polit. Sci. Assoc., PhiladelphiaSigelman L, Kugler M. 2003. Why is research on the effects of negative campaigning so inconclusive? Un-

derstanding citizens’ perception of negativity. J. Polit. 65:142–60Skaperdas S, Grofman B. 1995. Modeling negative campaigning. Am. Polit. Sci. Rev. 89:49–61Steenbergen M, Jones B. 2002. Modeling multilevel data structures. Am. J. Polit. Sci. 46( Jan.):218–37Swint K. 1998. Political Consultants and Negative Campaigning: The Secrets of the Pros. Lanham, MD: Univ. Press

Am.Theilmann J, Wilhite A. 1998. Campaign tactics and the decision to attack. J. Polit. 60:1050–62West D. 1993. Air Wars: Television Advertising in Election Campaigns, 1952–1992. Washington, DC: CQ PressWest D. 2005. Air Wars: Television Advertising in Election Campaigns, 1952–2004. Washington, DC: CQ Press.

4th ed.

306 Lau · Rovner

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

www.annualreviews.org ● Negative Campaigning C-1

Figure 1

Positive, negative, and contrast televised political ads in recent U.S. federal elections.

Negative

Positive

Contrast

60

50

40

30

20

10

1998 2000 2002Te

levi

sed

polit

ical

ads

(%)

Year

Figure 2

(a) Number of newspaper stories about negative political campaigns. (b) Academic studies of negative campaigning, by discipline.

Aca

dem

ic s

tudi

es o

fne

gati

ve c

ampa

igni

ng (%

)

Year1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004

5

10

15

20

New

spap

er s

tori

es a

bout

nega

tive

cam

paig

ning

Year1980 1984 1988 1992 1996 2000 2004

50

100

150

200

250a

b Political scienceCommunications/marketingPsychology economics

ANRV377-PL12-15-Lau.qxd 4/7/09 12:21 Page 1

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

C-2 Lau ● Rovner

Figure 3

Do negative campaigns demobilize the electorate? This figure displays the effect size associated with every study testing the demobi-lization hypothesis, adjusted for sampling error and measurement unreliability (Lau et al. 2007). Data have been sorted by whetherthe dependent variable is actual (reported) vote or intended vote, and within that distinction, by methodology. Negative effect sizes areconsistent with the demobilization hypothesis.

1.5000

1.0000

0.5000

0.0000

–0.5000

= Surveys= Experiments= Aggregate

–1.0000

–1.5000

S S S S S S

SXA

S S S S S S S S S S S S S S S S S S S SX X X X X X X X X X X X X X X X X X X XA A A A A A A A A A A

Actual turnout Intended turnout

Effec

t siz

e (Δ

)

Studies testing the demobilization hypothesis

ANRV377-PL12-15-Lau.qxd 4/7/09 12:21 Page 2

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

AR377-FM ARI 7 April 2009 11:25

Annual Review ofPolitical Science

Volume 12, 2009Contents

A Conversation with Robert A. DahlRobert A. Dahl and Margaret Levi � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 1

Neorepublicanism: A Normative and Institutional Research ProgramFrank Lovett and Philip Pettit � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �11

Domestic Terrorism: The Hidden Side of Political ViolenceIgnacio Sánchez-Cuenca and Luis de la Calle � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �31

Women in Parliaments: Descriptive and Substantive RepresentationLena Wängnerud � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �51

Self-Government in Our TimesAdam Przeworski � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �71

Social Policy in Developing CountriesIsabela Mares and Matthew E. Carnes � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �93

Variation in Institutional StrengthSteven Levitsky and María Victoria Murillo � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 115

Quality of Government: What You GetSören Holmberg, Bo Rothstein, and Naghmeh Nasiritousi � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 135

Democratization and Economic GlobalizationHelen V. Milner and Bumba Mukherjee � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 163

Has the Study of Global Politics Found Religion?Daniel Philpott � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 183

Redistricting: Reading Between the LinesRaymond La Raja � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 203

Does Efficiency Shape the Territorial Structure of Government?Liesbet Hooghe and Gary Marks � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 225

Bargaining Failures and Civil WarBarbara F. Walter � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 243

Hobbesian Hierarchy: The Political Economy of PoliticalOrganizationDavid A. Lake � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 263

v

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.

AR377-FM ARI 7 April 2009 11:25

Negative CampaigningRichard R. Lau and Ivy Brown Rovner � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 285

The Institutional Origins of Inequality in Sub-Saharan AfricaNicolas van de Walle � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 307

RiotsSteven I. Wilkinson � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 329

Regimes and the Rule of Law: Judicial Independence in ComparativePerspectiveGretchen Helmke and Frances Rosenbluth � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 345

Field Experiments and the Political Economy of DevelopmentMacartan Humphreys and Jeremy M. Weinstein � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 367

Laboratory Experiments in Political EconomyThomas R. Palfrey � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 379

Field Experiments on Political Behavior and Collective ActionEline A. de Rooij, Donald P. Green, and Alan S. Gerber � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 389

Experiments on Racial Priming in Political CampaignsVincent L. Hutchings and Ashley E. Jardina � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 397

Elections Under AuthoritarianismJennifer Gandhi and Ellen Lust-Okar � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 403

On Assessing the Political Effects of Racial PrejudiceLeonie Huddy and Stanley Feldman � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 423

A “Second Coming”? The Return of German Political TheoryDana Villa � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 449

Group Membership, Group Identity, and Group Consciousness:Measures of Racial Identity in American Politics?Paula D. McClain, Jessica D. Johnson Carew, Eugene Walton, Jr.,and Candis S. Watts � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 471

Opiates for the Matches: Matching Methods for Causal InferenceJasjeet Sekhon � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 487

Indexes

Cumulative Index of Contributing Authors, Volumes 8–12 � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 509

Cumulative Index of Chapter Titles, Volumes 8–12 � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 511

Errata

An online log of corrections to Annual Review of Political Science articles may be foundat http://polisci.annualreviews.org/

vi Contents

Ann

u. R

ev. P

olit.

Sci

. 200

9.12

:285

-306

. Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Rut

gers

Uni

vers

ity L

ibra

ries

on

11/0

1/10

. For

per

sona

l use

onl

y.