Variability in Citizens' Reactions to Different Types of Negative Campaigns

Variability in Citizens' Reactions to Different Types of Negative CampaignsAuthor(s): Kim L. Fridkin and Patrick J. KenneyReviewed work(s):Source: American Journal of Political Science, Vol. 55, No. 2 (April • 2011), pp. 307-325Published by: Midwest Political Science AssociationStable URL: http://www.jstor.org/stable/23025053 .

Accessed: 07/01/2013 06:56

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Midwest Political Science Association is collaborating with JSTOR to digitize, preserve and extend access toAmerican Journal of Political Science.

http://www.jstor.org

This content downloaded on Mon, 7 Jan 2013 06:56:34 AMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=mpsa

http://www.jstor.org/stable/23025053?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


Variability in Citizens' Reactions to

Different Types of Negative Campaigns

Kim L. Fridkin Arizona State University Patrick J. Kenney Arizona State University

Do negative advertisements lower voters' evaluations of the targeted candidate? We theorize that there is much to be gained

by examining the variance in the content and tone of negative campaign messages and the variance in voters' sensitivity

to negative political rhetoric. We employ data from the 2006 Cooperative Congressional Election Study to investigate the

impact of negative campaigning in U.S. Senate campaigns. We sampled 1,045 respondents in 21 of the 28 U.S. Senate races

featuring a majority party incumbent and challenger. In addition to the survey data, we collected contextual data regarding

the political advertisements aired during the campaigns and the news coverage of these campaigns in state newspapers. The

evidence suggests that the impact of negative information is multifaceted, and under some circumstances, substantial. We

find that uncivil and relevant negative messages are the most powerful, especially for people with less tolerance for negative

political rhetoric.

Is

negative campaigning effective? Do negative adver

tisements lower evaluations of the targeted candi

date? The preponderance of the evidence, after 25

years of scholarship, suggests that negative campaigning has limited effects on citizens' impressions of candidates

(e.g., Lau et al. 1999; Lau, Sigelman, and Rovner 2007). In

the most comprehensive and up-to-date review of the lit

erature, Lau, Sigelman, and Rovner conclude, "The find

ings reported in the research literature do not bear out

the proposition that attacking is an effective way to bol

ster one's own image relative to that of one's opponent"

(2007, 118). Such findings of the ineffectiveness of negative cam

paigning are not surprising given the plethora of evidence

suggesting that many Americans do not pay close atten

tion to political messages, negative or otherwise (e.g.,

Zaller 1992). And, when people do pay attention to cam

paign rhetoric, they are not easily persuaded (e.g., Kinder

2003). Furthermore, the ineffectiveness of negative cam

paigning is entirely consistent with the cottage industry

of scholarship demonstrating that the outcomes of pres idential elections can be predicted before the onset of

the fall campaign.1 All told, the conclusion that negative

campaigning is ineffective is consistent with theories and

evidence emanating from the voting and public opinion literature.

So, why do we feel uncomfortable concluding that

negative advertisements are ineffective? It seems unrea

sonable that the millions of dollars spent on negative

campaign messages are for naught. Negative messages are

pervasive in presidential and competitive congressional and senatorial elections (Franz et al. 2008; Geer 2006;

Jacobson 2009). In the final days of competitive cam

paigns, especially in the battleground states during presi dential elections, the campaign messages are often exclu

sively negative (Teinowitz 2008). Given the preponderance of negative advertisements

produced and disseminated by the candidates and the

political parties on television, on the radio, and online, the information costs for obtaining negative messages

Kim L. Fridkin is Professor of Government, Politics, and Global Studies, Arizona State University, Coor Hall, 6th Floor, P.O. Box 873902,

Tempe, AZ 85287-3902 ([email protected]). Patrick J. Kenney is Professor and Director in the School of Politics and Global Studies,

Arizona State University, Coor Hall, 6th Floor, P.O. Box 873902, Tempe, AZ 85287-3902 ([email protected]).

This article was supported by a grant from the Institute of Social Science Research at Arizona State University. We would like to thank

Pat Crittenden for her editorial assistance and Jill Carle for her research assistance. An online appendix with supplementary material for

this article is available at http://www.ajps.org/. The data and the syntax files used in this article will be available by January 10, 2012, at

http://pgs.clas.asu.edu/research.

'The most recent round of predictions appeared in PS: Political Science and Politics, October 2008. Ten articles, authored by 15 researchers,

appeared in a symposium entitled "Forecasting the 2008 National Elections."

American Journal of Political Science, Vol. 55, No. 2, April 2011, Pp. 307-325

©2011, Midwest Political Science Association DOI: 10.1111/j.l540-5907.2010.00494.x

307



KIM L. FRIDKIN AND PATRICK J. KENNEY

for voters is approaching zero. Since negative messages

are abundant during the final weeks of many presiden tial and subpresidential campaigns, we expect that such

saturation coverage would influence attitudes toward the

candidates.

Consistent with this expectation, some researchers

have found that negative commercials do produce critical

impressions of targeted candidates (e.g., Ansolabehere

and Iyengar 1997; Fridkin and Kenney 2004; Roddy and

Garramone 1988).2 Moreover, political consultants and

political candidates consider negative advertisements to be a valuable tool in campaigns (Abbe et al. 2001; Thurber and Nelson 2000).

Therefore, we believe that additional exploration into

how negative advertising affects citizens' attitudes about

candidates is worthwhile. Specifically, we believe that scholars have not fully investigated the variance in the content and tone of negative messages and the variance

in citizens' receptivity and tolerance of negative political rhetoric. We hypothesize that the variability in these di mensions is related to how negative information shapes

voters' attitudes toward candidates.

Theory and Expectations The Message

We know that negative information is likely to attract peo ple's attention, even when individuals are distracted by the demands associated with daily life (Fiske 1980). This"neg ativity bias" in impression formation is well established

(e.g., Fiske 1980; Taylor 1981). The reason that people attend readily to negative information is that the content of negative messages is filled with clues about events or situations that people should avoid (e.g., Kahneman and

Tversky 1979; Lau 1982; McGrawand Steenbergen 1997). In addition to the negativity of messages, people pay

more attention to messages that they perceive as relevant

to their daily lives. In social psychology, for example, a

variety of models of persuasion indicate that the relevance of the message is a prerequisite for persuasion (McGuire 1964, 1989; Petty and Cacioppo 1986). McGuire (1989), in his classic model of persuasion, argues that a message is persuasive only if people attend to and comprehend the message. And, when deciding which messages to pay attention to, people use the relevance of the message as

a decision rule, concentrating on the messages that are

relevant and ignoring those that are irrelevant.3

Although the social psychological literature is based

on messages that are personally relevant to individuals, we believe that the same logic extends to "relevance for

governing" in the political realm. We define "relevance"

as discussions about issues, personal traits, or other top

ics that people consider pertinent for an electoral cam

paign.4 We reason that voters find some negative mes

sages far more relevant (e.g., candidates' voting records) than others (e.g., candidates' drug use in college). A re

cent survey experiment examined subjects' impressions of fictitious candidates and the relevance of several com

mercials (Fridkin and Kenney 2008). Findings from this

study demonstrated that subjects readily rated certain

commercials as more relevant to governing than others.

And, the same data indicated that relevant negative mes

sages (e.g., a message about the candidate's health care

proposal) are more effective in shaping attitudes than

irrelevant negative messages (e.g., a message about the candidate's divorce).5

Negative commercials, in addition to varying in con

tent, also vary dramatically in tone. Some negative mes

sages are delivered in an uncivil and strident manner, while other negative commercials embrace a more mea

sured and courteous tone (Geer 2006). We expect that

negative messages differing in their civility will vary in their impact on citizens' evaluations of candidates. Our

expectations rest on research demonstrating that people

embrace norms that guide their interactions with other individuals. These norms about civil discourse extend to public discourse as well, with people expecting a cer tain level of civility from political actors, including candi dates running for office (Guttman 1993; Mutz and Reeves

2005). In fact, several studies have yielded consistent find

ings demonstrating that not only do citizens make pre dictable distinctions between civil and uncivil messages,

2In contrast, other scholars have found that negative advertisements are ineffective in lowering evaluations of the targeted candidate or that negative advertisements produce modest or inconsistent effects. See, for example, Basil, Schooler, and Reeves (1991), Lau and Pomper (2004), and Thorson, Christ, and Caywood (1991).

3Petty and Cacioppo (1986) also show that the relevance of the

message significantly influences people's willingness to process a

message (see also O'Keefe 2002; Perloff 2003).

4We theorize that issues and traits can be relevant or irrelevant for governing. For example, an irrelevant issue may be an issue no longer on the public agenda (e.g., a candidate's opposition to a war that has been over for several years). Similarly, an example of a relevant trait may be the candidate's prior political experience or the candidate's questionable integrity, as evidenced by the acceptance of gifts by a wealthy donor.

5To be sure, scholars have explored the role of the content of mes

sages by distinguishing between negative messages focusing on the candidate's personality and negative messages focusing on a candi date's policy positions (e.g., Brooks and Geer 2007; Geer 2006, Lau and Pomper 2004; Thorson, Christ, and Caywood 1991). However, the results of these studies are inconclusive.



Zoli

Highlight

NEGATIVE CAMPAIGNING

but also uncivil messages consistently produce more neg

ative views of politicians and the political process (e.g., Brooks and Geer 2007; Fridkin and Kenney 2008; Mutz

and Reeves 2005). Given the importance of civility as a norm in pub

lic discourse, how might people respond to political dis

course violating these established norms? According to re

search focused on product advertising, an advertisement

is a social object and is evaluated by common social norms

(Dahl, Frankenberger, and Manchanda 2003). When the

content or tone of an advertisement breaches norms for

decency, good taste, and personal moral standards, the

advertisement causes surprise, thereby attracting atten

tion to the novel stimulus (Meyer et al. 1991). In the political realm, a campaign advertisement vi

olating norms of civility is likely to be noticed. And,

consistent with theoretical and empirical research on

information-processing models, advertisements that vio

late the norms of civility are viewed as shocking, thereby

facilitating message comprehension and enhancing mes

sage retention (e.g., McGuire 1976). In exploring the impact of negative messages, it is

essential to consider the civility of those messages in con

junction with their relevance.6 In particular, in predicting how negative messages will influence evaluations of the

candidates, the relevance of the message is paramount.

Prospective voters are only interested in hearing criticisms

and critiques that provide information directly relevant

to how a candidate will perform in office. In contrast,

we expect that negative messages focusing on topics ir

relevant to governing will not influence evaluations of a

targeted candidate.

Although all relevant messages should influence vot

ers' assessments of a targeted candidate, we expect that

the civility of the message will alter its impact. An un

civil message—on a relevant topic—will heighten peo

ple's attention to the message and will more likely be

remembered (Lau 1982; Mutz and Reeves 2005). There

fore, relevant and uncivil messages should produce the

most powerful negative images of a targeted candidate.

Relevant and civil messages should produce more mod

erate negative impressions. And, irrelevant messages—

whether they are civil or uncivil—should not influence

impressions of a targeted candidate.

The Receiver

Just as negative messages vary in terms of their relevance

and civility, we think that citizens' tolerance of negative

messages varies. In particular, we expect that certain peo

ple may be more tolerant of messages delivered in a stri

dent fashion or focused on topics only tangentially related

to governing. That is, for some people, "anything goes" when it comes to negative campaigning. These people believe that candidates should be free to attack their op

ponents in whatever way they choose. For other people,

certain types of negative messages "cross the line" and in

sult their sensibilities regarding public decency and fair

ness.

We contend that people's predispositions regarding the appropriate nature of political discourse will deter

mine their receptivity to certain types of negative mes

sages delivered during campaigns. For people who have

a high tolerance for negative messages (i.e., people who

think any type of attack advertising is appropriate), the

relevance and civility of the message will be less influ

ential. In contrast, citizens who have a low tolerance for

certain types of negative political messages will be more

affected by the (ir)relevance and (in)civility of negative

messages.

We draw an analogy for illustrative purposes between

tolerance toward political messages and tolerance of phys

iological conditions. In the medical field, for example, we

know that people who have a low tolerance for pain are

more likely to feel pain when they are hurt physically. So,

even if two people in a laboratory setting are exposed to

the same amount of pain (e.g., a certain level of pressure

on a subject's finger), the person with a lower tolerance

for pain will report a higher level of discomfort (see, for

example, Ellermeier and Westphal 1995). Similarly, we

expect that people with a low tolerance for certain types of negative messages are more sensitive and will be more

influenced by such messages. Therefore, the impact of

these messages will be greater.

We are not the first to hypothesize that people differ

in their sensitivity to negative stimuli. For example, Ulbig and Funk (1999) theorize that people differ in their "con

flict avoidance" tendencies and people who are more likely to dislike conflict are less likely to participate in politics. The authors find that conflict avoidance is significantly and inversely related to political participation.7

'Consistent with our notion regarding the importance of relevance

and civility, Sides, Lipsitz, and Grossman (2010) find that citi

zens' evaluations of the negative tone of a campaign are distinct

from their evaluations of the informational nature of the cam

paign messages. Brader (2006) and Brooks and Geer (2007) look at

the interaction of different message dimensions in their research.

Brader explores the combination of visual and sound stimuli in

an advertisement, along with the content. Brooks and Geer (2007)

investigate the intersection of the civility of the advertisement with

the content of the advertisement (i.e., issues vs. traits).

7In the field of psychology, researchers have found individual

differences in people's attention to threatening social cues (e.g.,




These findings suggest that individual differences ex

ist in people's reactions to negative or threatening stimuli.

We extend this logic to negative political messages: cer

tain people will be more sensitive to irrelevant and uncivil

messages and may react more strongly when exposed to

these types of messages during campaigns. We think that

these differences in sensitivity to negative information go

beyond standard political variables, such as political so

phistication, partisanship, or political interest.8 Instead, we contend that people have distinct attitudes concern

ing their sensitivity or receptivity to negative political rhetoric. Some citizens have an easier time tolerating

negative political messages, while others dislike public criticism and grow weary of such messages as campaigns

progress.

In this article, we hope to advance our understanding

of the impact of negative messages on people's impres sions of campaigns by examining the intersection between

the relevance and civility of negative messages and how

people differ in their reactions to these types of mes

sages. We rely on a representative survey sample of over

1,000 respondents along with data assessing the content

of political communications during the 2006 senatorial

campaigns. We turn now to a discussion of our data and

measurements.

Data

This study uses data from the 2006 Cooperative Congres sional Election Study (CCES) to investigate the impact of negative campaigning in U.S. Senate races. The CCES is a preelection/postelection Internet survey, conducted

by Polimetrix, Inc., on behalf of a consortium of scholars

at 37 colleges and universities. Each group of scholars in

the consortium designed a "module" with unique sur

vey questions asked of 1,000 respondents. In other words, each group of scholars controlled the content of ques

tions for 1,000 citizens. In addition, these respondents

completed a "common content" survey. The common

content survey included 30,000 respondents.

Polimetrix uses a sample matching methodology to

produce a sample that is representative of the overall U.S.

electorate (Rivers, n.d.). Comparisons of the CCES sam

ple with census data suggest that the CCES sample is

somewhat more educated and younger than the popula

tion at large. However, the CCES sample matches census

data in terms of gender, race, and ethnicity composition (Hill et al. 2007). And, analysis conducted by Hill et al.

(2007) suggests that the 2006 CCES produces an Internet

sample that looks similar to existing RDD phone surveys. For our "module," Polimetrix sampled about 75 re

spondents in 21 of the 28 U.S. Senate races featuring a majority party incumbent and a majority party chal

lenger, for a total of 1,045 respondents.9 The races varied

dramatically in terms of competition, in the amount of

money spent, and in the amount of media attention given to the contests.

In addition to the survey data, we also collected con

textual data regarding the political advertisements aired

during the Senate campaigns and the news coverage of

these campaigns in state newspapers. Turning first to

the political advertising data, we collected advertisements

from the National Journal website. The National Journal website included downloadable links to almost all U.S.

Senate advertisements publicly available (Grose and Glo

betti 2007).10 We searched the candidates' own websites, as well as other websites, to increase the completeness of

candidates' advertisement samples. With this additional

search, we were able to include advertisements from the

following races: Utah, California, Texas, and Wyoming. We coded 302 advertisements for the 42 candi

dates.11 Some contests featured a great number of political threatening facial expressions). For example, anxious individuals

appear to be more sensitive to threat stimuli and are quicker at

perceiving threatening messages (Bradley et al. 1998). Also, psy

chologists relying on functional magnetic imaging find that indi

viduals who score higher on neuroticism demonstrate higher brain

reactivity to negative pictures (Canli et al. 2001). Researchers have

also suggested that there are cultural differences in people's desire

to avoid conflict. In particular, many East Asian cultures value har

mony and conflict avoidance, compared with European Americans

(see Lind, Huo, and Tyler 1994). In addition, some feminist scholars

(e.g., Gilligan 1982) suggest that women prefer less confrontational

procedures compared with men.

8To be sure, researchers have explored who is most susceptible to negative campaigning. For example, some scholars suggest that

Independents are more likely than partisans to be "demobilized" by negative campaigning (e.g., Ansolabehere and Iyengar 1997; Kahn

and Kenney 1999; Lau and Pomper 2004). These explorations, however, have yielded somewhat inconsistent findings (see Brooks

2006; Sigelman and Kugler 2003).

9In selecting the sample of races, we stratified by the competitive ness of the incumbent/challenger contest. In particular, we sampled 10 races classified as competitive (toss-up or leaning) by the Cook

Political Report and 11 races classified as noncompetitive (solid Democrat or solid Republican). We did not include open races in

our sample because of limited resources and the small number of

open races contested (n = 5). In the online Supporting Informa

tion, we present information about the races in our sample.

10Ad Spotlight at the National Journal website featured 544 adver

tisements for 15 of the 21 races included in our sample. We stratified

the population of advertisements on Ad Spotlight by candidate and

collected a maximum of 21 ads per candidate.

11 Overall, the mean number of advertisements coded per candidate

was 7.2, with a minimum of 0 and a maximum of 21. We cannot

estimate the completeness of our advertising sample because we




advertisements aired repeatedly over the months preced

ing Election Day. In contrast, in some Senate campaigns, there were no advertisements on television at all. In cod

ing the political advertisements, we assessed a variety of

factors associated with the tone of the advertisement, in

cluding the civility of each commercial, the relevance of

its content, and the substance of its message.12

Turning to the news coverage, we collected newspaper

data from each state's largest circulating newspaper from

October 1 to Election Day. We coded 2,077 news articles

for the 21 Senate races. In coding the news content, we

assessed a variety of factors associated with the tone of

the coverage, including the number, content, and source

of criticisms published about each candidate.13

Measurement of Key Concepts

There are no established measures for defining the rele

vance and civility of commercials. In addition, we know

of no prior studies that have measured respondents' toler

ance for irrelevant and uncivil messages. Consequently, in

this section we explain how we developed new measures

for these key concepts. We spend a significant amount

of time and effort determining the validity of these mea

sures. We employ three tests to determine the validity of

the measures assessing the relevance and civility of the

commercials, and we conduct two validity tests for the

survey questions measuring citizens' tolerance of nega

tive messages.

Relevance and Civility in Negative Commercials

To content analyze the political advertisements, coders

distinguished between messages that were relevant and

irrelevant for governing.14 For example, negative com

mercials classified as relevant by coders contained the fol

lowing examples of content: "The Senator voted to give

tax breaks for companies that move overseas..." or "The

Senator voted 96% of the time with Bush." In contrast,

negative commercials classified as irrelevant contained

content such as the following: "The candidates' support ers are from Hollywood" or "My opponent parties with

Playboy playmates." Coders also assessed the civility of the message in each

of the political advertisements.15 We adopted a generous

interpretation for determining civility given the rough and-tumble nature of campaign messages in U.S. Senate

campaigns (Kahn and Kenney 1999). That is, there needed

have no measure of the population of commercials aired during the 2006 campaigns. Unfortunately, we do not have access to how

often or where the ads were shown during the 2006 election. The

Wisconsin Advertising Project (WiscAds), the main source for data

regarding the frequency and timing of political advertisements, did

not collect data for the 2006 election cycle. This is clearly a limita

tion of our advertising data. To measure the reach of the political

advertising in our study, we employed campaign spending and es

timated a multiplicative term where each candidate's advertising was interacted with the candidate's campaign spending. However,

this alternative specification did not improve the performance of

our models in the upcoming analyses.

12 In conducting our content analysis, we followed the procedures described by Neuendorf (2002). A team of eight research assistants

(graduate and undergraduate political science students) coded the

advertisements. Coders were provided with the codesheet and a

codebook containing detailed instructions for coding each variable

(the codesheet is available in the online Supporting Information).

The comprehensive guidelines in the codebook helped to reduce

individual differences among coders. Coders were instructed to

view the advertisements as if they were "the average citizen." For

example, while a political science student may be aware that favor

ing increases in the size of government is a liberal position, a typical

person viewing the advertisement may not make such an inference.

During the content analysis training procedure, the coders com

pleted a pilot coding of 10 political advertisements. Differences

among coders were discussed and, when necessary, adjustments were made to the codesheet and codebook. The pilot procedure

helped standardize the coders' techniques so that all coders were

assessing the content in the same way. Given the comprehensive ness of the training procedure, maturation effects by coders during

the content analysis process were reduced. In addition, coders were

kept blind to the purpose of the study as a way to reduce potential bias. Eighty advertisements (10 per coder) were subject to reliability

analysis. One of the authors conducted the reliability checks, com

paring the author's coding of the advertisements with the coding by

the research assistants. Cohen's kappa was used to assess intercoder

reliability, with a resulting score of .91 (p < .001), indicating a high

level of agreement among coders.

13A team of six research assistants (graduate and undergraduate

political science students) coded the news coverage. The procedure

outlined in footnote 12 for the political advertising content analysis

was employed for the newspaper content analysis. An intercoder

reliability analysis using Cohens kappa was performed; kappa =

.93 (p < .001), indicating a high level of agreement among coders.

A copy of this codesheet is available in the online Supporting In

formation.

14The codebook instructions for coding the relevance of the adver

tisement read, "Do you consider the content of the advertisement

relevant for governing? Advertisements focus on diverse subjects, such as: a candidate's past drug use (e.g., in college), a candidate's

vote for a tax increase while serving in the House of Represen

tatives, a former spouse of a candidate accusing the candidate of

marital infidelity in a divorce proceeding a decade earlier, or an

opponent criticizing a candidate for his or her lack of electoral ex

perience. The question is: do you think MOST PEOPLE viewing an

advertisement would consider the content very relevant, somewhat

relevant, somewhat irrelevant, or very irrelevant?"

15For the civility code, the codebook instructions read, "Some ads,

even if negative, present the information in a civil manner (diplo

matically, without derision, etc.), while other ads rely on a more

uncivil tone (e.g., overly strident, rude, discourteous). Do you think

MOST PEOPLE watching the ad would consider it somewhat civil,

very civil, somewhat uncivil, or very uncivil?"




to be an explicit use of harsh, shrill, or pejorative adjec tives describing candidates, their policies, or their per sonal traits for an advertisement to be considered uncivil.

The following are sentences taken from advertisements

categorized as civil by coders: "The candidate voted 18

times to raise taxes," or "He talks conservative, but his

record says otherwise." In comparison, advertisements

with the following messages were coded as uncivil in tone:

"The Senator shows a disgusting display of arrogance," or

"After all these years, can't he offer more than smears and

distortions?"

These examples are from actual commercials aired

during the 2006 senatorial campaigns and illustrate dif

ferences between irrelevant and relevant messages and

differences between civil and uncivil messages. According to our coders, most of the negative commercials in our

sample were viewed as focusing on a relevant topic.16 The

coders scored 88% of the ads as relevant and only 12%

as irrelevant.17 This is not surprising given that campaign

managers and insiders have argued that attacks need to

resonate with voters' concerns and worries (Fenno 1996;

Kahn and Kenney 1999). The negative messages, on the

other hand, generated more variance in terms of civility.

Fifty-three percent of the negative advertisements were

classified as civil, while 47% were considered uncivil.

We perform three tests to assess the validity of the

measures of relevance and civility. First, we look at

whether the number of criticisms per advertisement in

creases as the relevance and civility of the advertisement

decreases. We expect that the nastiest commercials (com

mercials that are less relevant and less civil) will also in

clude the greatest number of criticisms.18 The content

analysis confirms our expectations; the number of crit

icisms per advertisement is statistically significant and

negatively correlated with the civility of the commercials

(Kendall's tau b = —.27, p < .01) and relevance of the

commercials (Kendall's taub = —.26, p < .04).19

Second, we examine whether the relevance and civil

ity of political commercials are related to the competitive ness of the campaigns. We expect that candidates running in competitive races would be more likely to produce neg ative messages embracing an uncivil tone and focusing on irrelevant messages, since prior research shows that

negativity—in general—increases in hard-fought races

(e.g., Kahn and Kenney 2004; Lau and Pomper 2004). We find that competitive races (i.e., fewer than 20 points

separating the candidates in preelection polls) generate

messages with a less civil tone and with less relevant con

tent when compared with lopsided races (i.e., more than

20 points separating the candidates in preelection polls). In particular, the average civility score is —.23 in com

petitive races and .05 in noncompetitive races (p < .10, one-tailed test). The average relevance score is .79 in com

petitive races and 1.0 in noncompetitive races (p < .05, one-tailed test).20

As a final validity check, we look at whether the rel

evance and civility of the negative commercials (as mea

sured by the content analysis) predict people's assessment

of the tone of the Senate campaigns in their state. If we

have captured the tone and content of the advertisements

properly, then these indicators should be strongly cor

related with respondents' perceptions of the tone of the

16If we use the more traditional operationalization of negative ad

vertisements as "attack ads" or "comparative ads" (e.g., Geer 2006;

Jamieson et al. 2000) and further distinguish between negative advertisements focusing on traits and issues, we find that attack

advertisements focusing on issues and comparative advertisements

emphasizing issues are the most common types of negative com

mercials. In particular, 26% of the negative advertisements were

categorized as issue attack ads, and 26% of the ads were classified

as comparative issue ads. In contrast, negative advertisements fo

cusing on personality were less common. In our sample, only 6%

of the advertisements are "comparative trait advertisements," and

15% of the advertisements are categorized as "trait attack ads."

And, 15% of the negative ads are issue/trait attack ads and 12%

are issue/trait comparative advertisements. The preponderance of

negative issue ads, compared with negative trait ads, has been doc

umented by other scholars. For example, Franz et al. (2008) find

that the majority of negative ads aired during U.S. House, Senate, and presidential elections in 2000 and 2004 focused on issues.

17While measuring relevance of negative commercials is less

straightforward than simply distinguishing between negative com

mercials focusing on issues or traits, we think that the effort is

worthwhile. First, prior work examining the impact of negative trait versus negative issue advertisements has been inconclusive.

Some researchers find that negative issue advertisements are more

effective than negative trait advertisements (e.g., Min 2004), while

others find that negative trait advertisements are more influential

than negative issue appeals (e.g., Brooks and Geer 2007). Therefore, the trait/issue dichotomy may not be a critical dimension. Second,

negative trait advertisements are not necessarily irrelevant. Accord

ing to our content analysis, 40% of the negative trait commercials

are irrelevant, while 60% are relevant. Second, Brooks and Geer's

(2007) study shows that respondents do not differ in their view of

the informational value of civil negative trait commercials and civil

negative issue commercials, suggesting that the issue/trait dimen

sion is not the key dimension determining relevance.

18All commercials were scored on a 5-point scale: —2 (very uncivil

or very irrelevant), —1 (somewhat uncivil or somewhat relevant), 0 (neutral), +1 (somewhat civil or somewhat relevant), +2 (very civil or very relevant).

19We rely on a one-tailed test since our hypothesis is directional. We

use Kendall's tau b because of the ordinal nature of the relevance

and civility variable.

20We looked at whether the relevance and civility of political dis

course increase with the seniority of the senator. Senior senators

are less vulnerable and are less likely to generate quality challengers (Jacobson 2009). We find a significant positive correlation between

the seniority of the senator and the civility (r = .51, p < .01) of the

commercials, and the seniority of the senator and the relevance (r = .55, p < .02) of the political advertisements.




campaigns. We asked people during the postelection sur

vey to describe the tone of the U.S. Senate campaign as

"very positive," "somewhat positive," "somewhat nega

tive," or "very negative."21 To predict people's views of the

tone of the campaign, we include the measures of civility and relevance of the candidates' campaign commercials.22

We expect that as civility of the negative commercials in

creases, people will view the campaign discourse as more

positive. In contrast, we expect that the relationship be

tween the relevance of negative messages and the tone of

the campaign to be negative. That is, as the relevance of

negative messages increases, we expect people's view of

the campaign to become more negative since we expect

relevant negative messages will be more influential than

irrelevant negative messages. To avoid specification error, we include several stan

dard control variables that are correlated with people's

views of the overall tenor of a senatorial campaign. We

include a simple measure of the number of criticisms aired

in the candidates' political commercials.23 This control is

important to be certain that the measures of the relevance

and the civility of the ads are not simply picking up overall

negativity. We also include a measure of the number of

criticisms published about the candidates in the largest

circulating newspaper in the state.24 We also control for

the context of the campaign by including a measure of the

closeness of the campaign based on preelection polls.25

The results of the MLM/MLE analysis are presented in Table l.26 The dependent variable, tone of the cam

Table 1 MLM/MLE Estimates Predicting Respondents' Assessments of the Overall Tone of the U.S. Senate

Campaign1

Campaign Tone Fixed Components

Advertisement Characteristics

Relevance of Negative Advertisement — ,03(.01)***

Civility of Negative Advertisement ,03(.01)***

Campaign Characteristics

Competition ,005(.008) Criticisms in Candidates' Ads — ,04(.02)** Criticisms in the Newspapers

— ,001(.0007)* Constant 1.84(.09)*** Variance Components

Intercept .05(.01)** N (survey) /N (states) 482/21

'Parameter estimates are followed by standard errors in parenthe ses.

Note: The dependent variable campaign tone is a scale ranging from

1 (very negative) to 4 (very positive). The advertisement charac

teristics run from low (civility or relevance) to high (civility or

relevance). See text for additional information about the opera tionalization of the variables, and see Supporting Information for

exact question wording. P-values are based on one-tailed tests:

***p < .01; **p < .05; *p < .10.

paign, ranges from 1 (very negative) to 4 (very positive). The results demonstrate that the tone of the candidates'

commercials significantly influences people's perceptions of the tone of the campaign. As the candidates' negative commercials become increasingly civil, respondents view

the campaign as more positive. And, as expected, the rele

vance of the negative advertisements is negatively related

to people's assessment of the tone of the campaign. When

candidates' negative commercials focus on more relevant

topics, people see the campaign as more negative.

Turning to the control variables, the number of crit

icisms presented in the candidates' commercials leads

people to view the campaign as more negative. It is im

portant that this variable, as well as the relevance and

civility of the advertisements, reaches statistical signifi

cance, demonstrating that our measures of relevance and

civility are not simply tapping the number of criticisms in

the ads. Similarly, as the number of criticisms in the press

increases, people are significantly more likely to view the

campaign tone as negative.

21Twenty-five percent of respondents characterized their campaign as "very negative," 22% answered "somewhat negative," 13% an

swered "somewhat positive," 5% said "very positive," and 15%

answered "don't know." Respondents who answered "don't know"

were eliminated from the analysis in Table 1.

22In this analysis, we multiply the relevance rating (or civility rat

ing) by the total number of criticisms in the advertisements in

order to capture the intensity of the message. Then, we combine

the measure of relevant negative advertisements for incumbents

and relevant negative advertisements for challengers since we are

predicting general tone of campaign. Similarly, we combine the

measures of civil negative advertisements for incumbents and chal

lengers.

23This measure is a simple count of the number of criticisms aired

in the candidates' commercials. The average number of criticisms

in the advertisements in a Senate race is 13.94 (with a standard

deviation of 8.06).

24The number of press criticisms in a Senate race averaged 117.3

(with a standard deviation of 114.0).

25We relied on preelection polling data published in the 2006 New

York Times Election Guide, and we subtracted incumbent support

from challenger support.

26We employ multilevel modeling (MLM) with maximum likeli

hood estimation (MLE) because we have a cluster of survey re

spondents located in the same state or campaign. Under these

circumstances, the OLS assumptions of uncorrelated error terms

maybe violated (Luke 2004; Steenbergen and Jones 2002).




Individual Differences in Tolerance toward

Incivility and Irrelevance

We contend that people differ in their tolerance toward

negative campaigning, and we have developed four survey items to assess people's tolerance for negative campaign

ing. We presented respondents with four statements and

asked them to indicate whether they strongly agree, some

what agree, somewhat disagree, or strongly disagree with

each statement. The first two statements tap the person's

tolerance for uncivil discourse, and the second two state

ments measure the person's tolerance for irrelevant com

mentary. We vary whether an "agree" response means the

respondent likes or dislikes uncivil or irrelevant negative commercials.27 The four items are as follows:

Some negative advertisements are so nasty that I

stop paying attention to what the candidates are

saying.

Mean-spirited commercials attacking the oppo nent are appropriate during election campaigns.

Negative advertisements attacking a candidate's

personal life are inappropriate.

I find negative political commercials attacking the opponent's personal life as a young person to

be interesting.

Responses to the two civility questions suggest that

most people in our sample do not like mean-spirited

campaigns. For example, 47% of respondents "strongly

agree" and 35% "somewhat agree" that campaign ad

vertisements are ".. .so nasty that I stop paying at

tention." Similarly, 45% of respondents "strongly dis

agree" and 27% "somewhat disagree" that "mean-spirited

commercials..." are appropriate during a campaign.

Turning to the relevance of certain types of attacks, more

than half of the respondents (55%) "strongly agree" and

27% "somewhat agree" that attacking a candidate's per sonal life is inappropriate. Similarly, 61% of the people in the survey "strongly disagree" and 21% "somewhat

disagree" that negative advertisements attacking a candi

date's personal life as a young person are interesting. These

responses resonate with the electorate's long-standing an

tagonism toward negative campaign rhetoric (Kahn and

Kenney 2004). We conduct two tests to examine the validity of the

individual measures of tolerance toward incivility and

irrelevance. In our first test, we assess whether theoreti

cally relevant variables help predict people's level of toler

ance toward negative messages. To predict people's level

of tolerance toward negative messages, we look at two

sets of variables: demographic variables (i.e., the gen der of the respondent, the respondent's age, and the re

spondent's education level) and political variables (i.e., the respondent's political sophistication, the respondent's

ideology, and the strength of the respondent's party af

filiation).28 Research examining the connection between

negative campaigning and turnout suggests that women

and Independents may react most strongly to negative

messages. In particular, some scholars find that women

and Independents are more likely to become demobilized

as negativity increases (Ansolabehere and Iyengar 1997; Kahn and Kenney 1999; Kern and Just 1997). These in

dividuals may be more affected by campaign negativity because they are less tolerant of uncivil and irrelevant

messages.29

We estimate two OLS models, one predicting an in

dividual's tolerance toward irrelevant messages and one

predicting an individual's tolerance toward uncivil mes

sages.30 The results presented in Table 2 indicate that

certain people are significantly more tolerant of uncivil

27See the online Supporting Information, for a copy of the survey questions used in this study.

28Strength of party affiliation is measured by recoding the 7-point

party identification scale into 3 points (strong partisans = 3, weak

partisans = 2, and Independents = 1). Respondent's ideology is

measured on a 100-point ideological scale (where 0 is extremely liberal and 100 is extremely conservative), and we recode the 100

point scale into five points (0-20 = 1,21-40 = 2,41-60 = 3,61-80 = 4, 81-100 = 5). The measure of political sophistication takes

on one of three values: 0 (incorrectly placing both parties on the

ideological scale), 1 (correctly placing one of the parties on the ide

ological scale), 2 (correctly placing both parties on the ideological scale). Correct placements include locating the Democratic party on the liberal side of the scale and locating the Republican party on

the conservative side of the scale. Political interest is measured with

a 3-point scale ranging from very interested (3), to somewhat inter

ested (2), to not very interested (1). Gender is coded 1 = male, 0 =

female; age is an interval measure; education is an interval measure

based on years of schooling. We look at ideology (as opposed to

strength of ideology) because we know that Republicans are more

likely to use negative advertising, compared with Democrats (e.g., Lau and Pomper 2001; West 2010), and researchers have found that

Republicans are less likely to be demobilized by negative campaign

ing, compared with Democrats and Independents (Ansolabehere and Iyengar 1997).

29We also looked at whether strength of ideology (like strength of

party) was related to tolerance of uncivil and irrelevant negative

messages. In both cases, we find strength of ideology is signifi

cantly related to tolerance, with strong ideologues being more tol

erant of irrelevant and uncivil negative messages than respondents with more moderate ideological profiles. These findings mirror the

findings for strength of party identification.

30We combine the two measures of tolerance toward irrelevant

messages into a single relevance index, and we combine the two

measures of tolerance toward uncivil messages into a single civility index. Both indices range from a low of 2 (not tolerant) to a high of 8




Table 2 OLS Regression Predicting Respondents' Tolerance of Irrelevant and Uncivil Negative Commercials1

Tolerance of Irrelevant

Commercials

Political Characteristics

Strength of Party

Ideology

Sophistication Political Interest

Demographics

Education

Age Gender

Constant

R2

N

. 11 (.04)***

24( 04)***

-,19(.07)***

,24(.07)***

.03(.03)

-.005(.003)**

.44(.09)***

1.95(.28)*** .10

979

Tolerance of Uncivil Commercials

08 .12(.04)*** .08

2 .20(.04)*** .16

09 —.09(.07) -.04

10 ,27(.08)*** .11

04 —,02(.03) -.02

06 — .01(.003)*** -.13

16 ,53(.09)*** .18

2.55(.29)*** .10

987

1 Unstandardized regression coefficients are followed by standard errors in parentheses; beta coefficients are reported after levels of statistical

significance. Note: The dependent variables measuring tolerance of irrelevant messages and tolerance of uncivil messages range from 2 (not very tolerant) to a high of 8 (very tolerant). See Supporting Information for exact question wording. P-values are based on one-tailed tests:

"*p < .01; "p < .05; *p < .10.

and irrelevant messages. We find that people who are

more strongly attached to the political parties are more

tolerant of uncivil and irrelevant messages, as are people who are more interested in political campaigns. Ideology is related to people's tolerance to negativity, with con

servatives being more tolerant of uncivil and irrelevant

messages, compared with more liberal respondents. We

also find that men are more tolerant of uncivil and irrel

evant negative messages when compared with women. In

addition, older people are less tolerant of uncivil and ir

relevant messages. And, politically sophisticated citizens, who are most skilled at sorting through relevant and ir

relevant discourse, appear to be less tolerant of irrelevant

messages.31

A second test to establish the validity of our tolerance

measures is to examine whether people's level of political tolerance predicts how they view political commercials.

In our survey, we asked respondents to rate the useful

ness and civility of negative commercials they reported

seeing during their state's Senate race. These questions are asked in the postelection wave, whereas the questions

assessing the respondents' levels of tolerance regarding relevance and civility were asked in the preelection wave

of the survey. We find that as people's tolerance for ir

relevant advertisements increased, they were more likely

to view the advertisements they saw as relevant. The cor

relation between people's tolerance for irrelevant com

mercials and their relevance ratings for the incumbent

and challenger commercials was .21 (p < .01) and .13

(p < .01), respectively.32 Similarly, as people's tolerance

for uncivil advertisements increased, people were more

likely to view the advertisements they saw as civil. The

correlation between people's tolerance for uncivil com

mercials and their civility ratings for the incumbent and

challenger commercials was .16 (p < .01) and .08 (p <

.05), respectively.33 Based on the series of tests reported here, we be

lieve that we have developed valid measures of the civility

(highly tolerant). The correlation between these two civility items is

.25, p < .01, and the correlation between these two relevance items

is .48, p < .01. We rely on the Spearman's reliability rho to calculate

the correlation between the measures because of the ordinal nature

of the tolerance measures.

3IWe reestimate the model with ordered logit regression given the

ordinal nature of the dependent variable. However, the results are

unchanged and this can be seen in Table SI in the online Supporting Information.

32The correlations are based on Kendall's tau b since the data are

ordinal and the p-values are based on one-sided tests since our

hypotheses are directional. The correlations do not change sub

stantively if we rely on Spearman's rho.

33We also find that people who are less tolerant of uncivil political discourse and people who are less tolerant of irrelevant political discourse are significantly more likely to view campaigns as negative

(Kendall's tau b = .08, p < .01; Kendall's tau b = .06, p < .02,

respectively).




and the relevance of the candidates' advertisements as

well as valid measures of respondents' tolerance toward

uncivil and irrelevant advertisements. We have demon

strated that these measures are related to a set of theoreti

cally important predictors. And, we have shown that these

measures predict citizens' views of the Senate campaigns as well as their views of the candidates' commercials.

Findings

We turn now to exploring the key question in this article:

do negative advertisements lower evaluations of a targeted candidate? As we have argued throughout the article, we

believe that certain types of negative messages are more

powerful than others. Specifically, we expect that the rele

vance of the message combined with the message's civility will influence evaluations of the candidates. In addition, we expect that the influence of the negative messages on

citizens' assessments of candidates will depend on peo

ple's tolerance for negative campaigning. With the important independent variables in place,

we focus on developing several key dependent variables

that capture citizens' evaluations of the candidates seek

ing U.S. Senate seats in 2006. We assess four traditional

dimensions that capture individuals' evaluations of the

candidates: (1) citizens' impressions of the candidates'

personality traits, (2) individuals' affective assessments of

the candidates, (3) voters' views regarding the candidates'

abilities to deal with several issues, and (4) citizens' overall

favorability ratings of the candidates (Abramowitz 1988;

Campbell et al. 1960; Kinder 1986; Wright and Berkman

1986). The questions used to tap the dependent variables

are derived primarily from standard NES measures.34

The analyses take place in two steps. First, we examine

how negative campaign messages shape people's evalua

tions of the Senate candidates. With this initial analysis, we

examine how the interplay between the tone (civility) and

content (relevance) of the messages shapes citizens' im

pressions of the candidates. Second, we examine whether

citizens' tolerance of negative campaigning affects how

the tone and content of negative messages shape citizens'

views of candidates.

As discussed earlier, we expect that only relevant neg ative commercials will depress people's impressions of the

targeted candidate. Among these relevant negative com

mercials, we hypothesize that uncivil advertisements will

be more powerful than civil advertisements. Based on

these theoretical expectations, we categorize the commer

cials of each candidate as relevant and uncivil, relevant

and civil, and irrelevant. In particular, if a candidate's

commercials are rated as high on relevance and low on

civility, the candidate is given a score of 3.35 If a can

didate's commercials are rated as high on relevance and

high on civility, the candidate is given a score of 2. Finally, if a candidate's commercials are rated as low on relevance, the candidate is given a score of l.36

To properly estimate the effect of the relevance and

civility of negative advertisements on people's views of the

candidates, we need to control for several rival factors. To

begin, we need to assess whether candidates who engage in negative campaigning are susceptible to a "backlash"

effect; that is, candidates who attack their opponents may hurt their own evaluations.37 Relying on the theoretical

34See the online Supporting Information for exact question word

ing. To measure traits, we combine five questions asking respon dents to rate the candidates on the following trait dimensions: intel

ligence, leadership, honesty, experience, and caring. The resulting index ranges from 5 to 20, with a mean of 14.4 for incumbents

and 12.98 for challengers. We relied on Cronbach's alpha to test for the internal consistency of the items in each trait index: Cron bach's alpha = .84 for incumbent traits, .84 for challenger traits.

To measure affect, we combine four questions asking respondents whether the candidates make them hopeful, worried, proud, and

angry. The resulting index ranges from 4 to 12, with a mean of 7.80 for incumbents and 7.53 for challengers. Cronbach's alpha =

.90 for incumbent affect and .87 for challenger affect. To measure

issues, we combine two questions asking respondents to rate the candidate's competence for dealing with the economy and health care. The resulting index ranges from 2 to 12, with a mean of 7.77 for incumbents and 7.20 for challengers. Cronbach's alpha = .95 for incumbent issues and .94 for challenger issues. Finally, the overall

favorability score is based on a 10-point scale. The mean is 5.36 for incumbents and 4.68 for challengers. For each of these four

measures, a high score represents a more positive score.

35 To develop an overall measure of the relevance of a candidate's commercials during a campaign, we utilize the coding from the content analysis of the individual advertisements to derive an over all race estimate for the candidates' commercials as low, medium, or

high in terms of relevance. We then multiply the relevance ratings by the total number of criticisms in the advertisements in order

to capture the intensity of the message. Since we expect that only relevant advertisements will be effective, we take the mean of the

resulting relevance scores and classify the candidate's message as

having a low (below the mean) or high (above the mean) score for

relevance. We repeat the same procedure for civility, producing a

low (below the mean) or high (above the mean) score for civility. Based on these scores, we create the 3-point relevance/civility in

dex. The mean for the relevance/civility advertising measure is 1.79 for incumbents and 1.65 for challengers.

36We examine the relationship between the 3-point rele

vance/civility index and respondents' assessments of the tone of

the campaign. We find a significant negative relationship between the relevance/civility index and people's views regarding the tone of the campaign. In particular, as advertisements become more relevant and more uncivil, people view the campaign as more neg ative. The Kendall tau b is —.39 (p < .01) for the incumbents' commercials and —.45 (p < .01) for the challengers' commercials.

37Two recent reviews of the literature on the likelihood and size of backlash effects find that candidates who engage in attack




framework introduced earlier, we expect to find a back

lash effect when candidates attack their opponents with

uncivil and irrelevant messages. We expect citizens to find

these messages offensive because they are delivered in an

uncivil manner and focus on topics not germane to gov

erning. To measure the backlash effect, we employ a binary

coding where the sponsoring candidates' commercials re

ceive a score of 1 if these commercials are rated as low on

relevance and low on civility. All other combinations of

commercials receive a score of 0.38

We include additional measures assessing the context

of the campaign as well as measures tapping the political

profiles of the respondents. We include a variable mea

suring the number of press criticisms published about the

candidate to capture the general negativity of the cam

paign.39 In addition, we include a measure assessing the

competitiveness of the campaign. Based on prior research

examining campaign intensity, we expect incumbents to

be evaluated more favorably in noncompetitive races and

challengers to be rated more positively in competitive races (Kahn and Kenney 1999; Westlye 1991). Finally, we

control for respondents' ideological and political pref erences since people are likely to rate a candidate more

positively when they share the candidate's party and ide

ological positions.40

The MLM/MLE results predicting evaluations of the

incumbents and challengers are presented in Table 3. The

findings demonstrate that the relevance and civility of the

political advertisements perform precisely as expected.

Overall, in each of the eight equations, the measure tap

ping the relevance and civility of political advertising is

statistically significant and is signed in the hypothesized direction. In each case, as relevance and incivility of the

advertisements increase, people's evaluations of the tar

geted candidates decline. This finding is consistent re

gardless of whether respondents are evaluating the can

didates in terms of (1) personality traits, (2) affective

impressions, (3) issue competence, or (4) overall impres sions.41 In fact, the relevance and civility of the adver

tisements are more powerful and more consistent in their

impact than the amount of press criticisms or the close

ness of the race, suggesting that the relevance and civility measure is capturing something beyond the general neg

ativity of the campaign and the competitiveness of the

contest.

In addition, the influence of the relevance and civil

ity measure is more powerful in the challenger models

than in the incumbent models, according to the param eter estimates. This indicates that the content and tone

of incumbents' advertising is more effective at lowering evaluations of the challengers than vice versa. For exam

ple, in the model predicting respondents' assessments of

candidates' personality traits, the parameter estimate for

the relevance and civility variable is —.76 in the challenger model and —.42 in the incumbent model.

To provide a clearer picture of how the relevance

and civility of negative messages influence citizens'

assessments of candidates' personal traits, we calcu

late the "first differences" for the maximum likelihood

strategies can expect to hurt their own evaluations (Lau and Rovner

2009; Lau, Sigelman, and Rovner 2007). And, the decline in their

own image may offset any damage done to their opponent. The

preponderance of the studies identifying a backlash effect employ

experimental designs (e.g., Chang 2003; Houston and Doan 1999).

A small number of studies examining survey data find some evi

dence for a backlash effect (e.g., Fridkin and Kenney 2004; Lau and

Pomper 2004).

38 We rely on the measures of relevance and civility developed for

the relevance/civility index when operationalizing the backlash ef

fect. The sponsoring candidate received a score of 1 on the backlash

measure if the sponsoring candidate's commercials scored below

the mean on relevance and civility. If the sponsoring candidate's

commercials received a mix score (e.g., above the mean on civil

ity, below the mean on relevance) or a high score (i.e., above the

mean on relevance and above the mean on civility), the sponsoring

candidate received a score of 0.

39We also included an additional measure of negative advertising: a

simple summation of the number of criticisms offered in each can

didate's political advertisements. This measure does not capture

the relevance and civility of the candidates' negative communi

cations. In the models predicting evaluations of incumbents, the

number of criticisms of incumbents in the challengers' advertise

ments never significantly depressed evaluations of the incumbents.

In the models predicting evaluations of challengers, the number of

criticisms of challengers in the incumbents' advertisements never

significantly depressed evaluations of the challenger.

40To code ideological proximity, we rely on the respondent's self

placement on a 100-point ideological scale and the respondent's

placement of the incumbent and challenger on the same ideolog

ical scale. We recoded the 100-point scales into five points (0-20

= 1, 21-40 = 2, 41-60 = 3, 61-80 = 4, 81-100 = 5). Then, we

calculated (1) the absolute value of the difference between the re

spondent's ideological self-placement and respondent's placement

of the incumbent's ideology and (2) the absolute value of the dif

ference between the respondent's ideological self-placement and

the respondent's placement of the challenger's ideology. Finally, we

subtracted the respondent's closeness to the incumbent from the

respondent's closeness to the challenger. The resulting ideologi

cal proximity score ranges from 4 (the respondent and incumbent

share the same location on the ideological scale) to —4 (the respon

dent and the challenger share the same location on the ideological

scale).

41 In addition to the four dependent variables displayed in Table 3,

we also develop a model estimating the impact of the relevance and

civility of the commercials on vote choice. We find that the measure

assessing the relevance and civility of the candidates' commercials

is significant (estimate = —.07, with a standard error of .02). Party

and ideological proximity are significant in the model, as is com

petition. However, news criticisms and the backlash effect fail to

reach statistical significance.



KIM I. FRIDKIN AND PATRICK J. KENNEY

Table 3 MLM/MLE Estimates Predicting Evaluations of U.S. Senate Candidates: The Impact of Relevance and Civility of Negative Commercials1

Affect1 Traits Issues Overall Evaluations

Incumbents

Fixed Components

Candidate Commercials '

Relevance/Incivility — .29(.16)** —,42(.23)** — .46(.21)** —.33(.14)*** Backlash Effect —.55(36)* — .93(.52)** —.50(.48) —,57(.32)**

Respondent Characteristics

Party Proximity ,41(.05)*** 48( 08)*** 43( 06)*** .51 (.05)***

Ideological Proximity ,64(.05)*** .98(08)*** .71 (.06)*** .75(.06)***


Competition —.003(.01) ,03(.02)* .02(.01)** .006(.009) Criticisms in the News ~.007(.003)*** —.01(.003)*** — ,003(.003) —.006(.002)*** Constant 7.67(.12)*** 14.28(.17)*** 7.73(. 16)*** 5.17(. 10)*** Variance Components

Intercept .10(.07)* . 19(. 16)* ,23(.13)* .04(.04) N (Survey)/N (states) 646/21 622/21 716/21

Challengers

793/21

Fixed Components

Candidate Commercials

Relevance/Incivility —.36(.25)* — .76(.39)** —.60(.22)*** —.36(.25)* Backlash Effect —.38(.36) — .06(.56) — .36(.31) — ,07(.35)


Party Proximity .55(.06)*** 69( 09)*** .51(.07)*** .63(.05)***

Ideological Proximity .47(.05)*** .82(09)*** ,52(.07)*** .61(.05)***


Competition — ,01(.02) — ,0007(.03) —.003(.02) — .006(.02) Criticisms in the News —.002(.004) — ,0005(.006) .0005(.004) — .0002(.003) Constant 7.57(.17)*** 12.88(.29)*** 7.21 (. 15)*** 4.75(.17)*** Variance Components

Intercept .24(.13)** .55(.34)** o * .23(. 13)**

N(Survey)/N(states) 462/21 374/21 479/21 621/21

'Parameter estimates are followed by standard errors in parentheses. P-values are based on one-tailed tests: ***p < .01; **p < .05; *p < .10. 2Affect is measured on a scale ranging from 4 to 12. Traits are measured on a scale ranging from 5 to 20. Issues are measured on a scale

ranging from 2 to 12. Overall evaluations are measured on a scale ranging from 0 to 10. See the text and Supporting Information for information about the operationalization of the variables.

3Relevance/Incivility is measured on a 3-point scale where the candidate is given a score of 3 if the candidate's commercials are rated as

high on relevance and low on civility, the candidate is given a score of 2 if the commercials are rated as high on relevance and high on

civility, and the candidate is given a score of 1 if the commercials are rated as low on relevance. The Backlash Effect is measured with a

binary variable where the sponsoring candidates' commercials receive a score of 1 if these commercials are rated as low on relevance and low on civility, and 0 for all other combinations of the sponsoring candidates' commercials.

coefficients (King 1989, 107-8).42 We want to estimate the average probability (i.e., 0 to 1.0) that people will

lower their evaluations of candidates' personal traits as the negative messages change in terms of relevance and

civility. We turn first to an examination of the size of the effects when challengers attack incumbents. In

42The first differences are estimates of the probability that citizens will decrease assessments of candidates as the content and tone of

campaign messages change. In nonlinear equations "a single unit

change in X will have a different effect on the expected value of Y depending on the points at which the curve is evaluated" (King

1989,108). While calculating the first differences, all other variables are held at their means.




particular, if challengers move from producing irrelevant

commercials to producing relevant and civil commer

cials, there is a .22 probability that citizens will decrease

their trait assessments of incumbents' personal traits. If

challengers move from marketing relevant and civil com

mercials to relevant and uncivil commercials, then voters

have a .15 probability of decreasing their assessments of

the incumbents' personal traits. In comparison, changes

in the incumbents' commercials have a larger impact on

assessments of the challengers' personalities. In particu

lar, moving from irrelevant commercials to relevant and

civil commercials produces a .25 probability that citizens

will decrease evaluations of challengers' traits, and chang

ing from relevant and civil commercials to relevant and

uncivil messages produces another .25 probability that

citizens will decrease their impressions of the challengers'

personalities.43

Looking at the remaining variables in the models, the

variable estimating "backlash" is statistically significant in

three of the four incumbent equations. But, the backlash

variable does not reach statistical significance in any of

the challenger models. These findings suggest negative

campaigning by incumbents is risky.44 While incumbents

will be effective at depressing views of their challengers as their commercials become more relevant and more

uncivil, incumbents who air irrelevant and uncivil adver

tisements run the risks of offending potential supporters and hurting their own evaluations.

We also find that press criticisms significantly influ

ence impressions of the incumbent in three of the four

equations. In contrast, the parameter estimates for press

criticisms never reach statistical significance in the four

equations predicting evaluations of the challengers. Press

criticisms are probably more powerful for incumbents

because incumbents garner significantly more press cov

erage than challengers (Kahn and Kenney 1999; Westlye

1991).

Finally, and as expected, ideological and party prox

imity have a substantial and consistent impact on impres sions of incumbents and challengers. This is not surpris

ing. What is impressive, however, is that the relevance and

civility of advertising exert a strong impact on candidate

evaluations, even controlling for party and ideological

proximity. We turn now to our final analysis. Are the strong

effects of relevance and civility modified or altered by cit

izens' tolerance of negative attacks? As discussed earlier,

we expect that people with a low tolerance for uncivil

and irrelevant messages will be more influenced by these

types of communications. The impact of these messages

will be greater simply because these individuals are more

sensitive to the content and tone of the negative adver

tisements.

To test whether tolerance for uncivil and irrelevant

messages influences the impact of negative messages, we

combine the measures of tolerance for uncivil messages and tolerance for irrelevant messages into a single index,

ranging from 4 to 16.45 We then divide our sample at the

mean and reestimate the MLM/MLE equations presented in Table 3 for people low in tolerance of negative messages

(Table 4) and for people high in tolerance of negative

messages (Table 5).46

Turning first to Table 4, the top panel presents the

results for evaluations of incumbents while the bottom

panel presents the results for evaluations of challengers.

We are most interested in the strength, statistical signif

icance, and direction of the parameter estimates for the

measure tapping the relevance and civility of the nega tive advertisements. The relevance and civility variable is

statistically significant in three of the four equations for

incumbents and in four of the four equations for chal

lengers. In addition, the sign of the relevance and civility variable is in the hypothesized (negative) direction in each

of the eight models in Table 4.

As in Table 3, the size of the parameter estimates for

relevance and civility is larger in the challenger models,

compared with the incumbent models. These results indi

cate that for people sensitive to negative campaigning, the

incumbents' messages are more damaging than the chal

lengers' messages. Also, consistent with the findings in

Table 3, the backlash effect is evident for incumbents and

not for challengers. In addition, the criticisms published in the newspaper damage evaluations of incumbents in

each of the four models, while failing to significantly affect

impressions of challengers.

Turning to Table 5, these analyses focus only on peo

ple who have a high score on the tolerance scale (i.e.,

43We also included a measure assessing the proportion of posi

tive advertisements produced by the candidates during the cam

paigns, based on our content analysis. We find that the proportion

of positive advertisements produced by a candidate (incumbent

or challenger) does not significantly influence evaluations of the

sponsoring candidate. The coefficient representing positive adver

tisements fails to influence evaluations in each of the eight equations

presented in Table 3.

44Kahn and Kenney (2004) and Lau and Pomper (2004) also find

a greater backlash effect for incumbents, compared to challengers,

for candidates running in U.S. Senate elections.

45The mean tolerance level is 7.0 on a scale ranging from 4 to 16.

46Dividing the sample by level of tolerance toward negative mes

sages is an appropriate statistical procedure for examining how

tolerance influences the impact of these negative messages on eval

uations (see Brambor, Clark, and Golder 2005).



KIML. FRIDKIN AND PATRICK J. KENNEY

MLM/MLE Estimates Predicting Evaluations of U.S. Senate Candidates: The Impact of Relevance and Civility of Negative Commercials for People with Low Tolerance for Negative Campaigning1


Incumbents

Fixed Components

Candidate Commercials3

Relevance/Incivility —.28(.27) —.45(.27)** — .65(.25)*** ,37(.24)* Backlash Effect —1.00(.61)* —,44(.62) — ,77(.58)* —1.15(.53)**


Party Proximity ,70(.08)*** ,83(.ll)*** ,72(.08)*** .77(.08)***

Ideological Proximity ,46(.08)*** 79(H)*** ,53(.08)*** .62(.08)***


Competition —.03(.02)* —.02(.02) —.02(.02) —.03(.02)* Criticisms in the News — ,01(.004)*** —.01(.004)*** — .007(.004)** — ,01(.004)*** Constant 7.70(.20)*** 14.36(.20)*** 7.77(.19)*** 5.22(.18)*** Variance Components

Intercept .38(.21)** .24(.24) .31(.18)** .22(. 18) N (Survey) /N (States) 308/21 293/21 345/21

Challengers

380/21

Fixed Components


Relevance/Incivility — .46(.33)* —1.02(.45)*** — .97(.34)*** —.57(.30)** Backlash Effect —.55(.47) —.25(.67) —.36(.48) —.12(.43)


Party Proximity ,66(.08)*** ,75(.15)*** .56(.ll)*** ,82(.08)***

Ideological Proximity ,39(.08)*** ,70(.15)*** .37(.ll)*** .45(.08)***


Competition —,009(.02) —.02(.03) — .03(.02)* —.01 (.02) Criticisms in the News — ,004(.005) — .004(.007) — ,003(.005) — .003(.005) Constant 7.68(.23)*** 12.92(.34)*** 7.24(.25)*** 4.92(.21)*** Variance Components

Intercept .38(.23)** .47(.53) .25(.24) .27(.20)* N(Survey)/N(States) 229./21 189/21 228/21 291/21

'Parameter estimates are followed by standard errors in parentheses. P-values are based on one-tailed tests: ***p < .01; **p < .05; *p < .10. 2Affect is measured on a scale ranging from 4 to 12. Traits are measured on a scale ranging from 5 to 20. Issues are measured on a scale

ranging from 2 to 12. Overall evaluations are measured on a scale ranging from 0 to 10. See the text and Supporting Information for

operationalization of the variables.

3Relevance/Incivility is measured on a 3-point scale where the candidate is scored a 3 if the candidate's commercials are rated as high on relevance and low on civility, the candidate is given a score of 2 if the commercials are rated as high on relevance and high on civility, and the candidate is given a score of 1 if the commercials are rated as low on relevance. The Backlash Effect is measured with a binary variable where the sponsoring candidates' commercials receive a score of 1 if these commercials are rated as low on relevance and low on civility, and 0 for all other combinations of the sponsoring candidates' commercials.

above the mean). For these individuals, we expect that the relevance and civility of the negative advertisements will only minimally influence their evaluations of the can didates. People with a high tolerance for negativity are less sensitive to negative information, and we do not expect

variations in the tone and content of negative messages to alter their impressions of the candidates. For evalua tions of incumbents (the top panel in Table 5), the rel evance and civility measure is statistically significant in

only one of the models (i.e., overall evaluations). And, for




MLM/MLE Estimates Predicting Evaluations of U.S. Senate Candidates: The Impact of Relevance and Civility of Negative Commercials for People with High Tolerance for

Negative Campaigning1


Incumbents

Fixed Components

Candidate Commercials3

Relevance/Incivility — .10(. 15) — ,21(,33) — ,24(.19) — .23(. 17)* Backlash Effect -.42(34) —1.46(.76)** — .19(.46) —. 10( .41)


Party Proximity .13(.07)** .23(. 11)** 17( 07)*** ,25(.08)***

Ideological Proximity .81(.06)*** 1 i2( ii)*** ,89(.07)*** ,93(.08)***


Competition ,02(.01)** .06(.02)*** ,05(.01)*** .03(.01)*** Criticisms in the News — ,006(.002)*** -.01(.004)**' —.003(.003) — ,004(.003)* Constant 7.62(.10)*** 14.05(.25)*** 7.58(. 15)*** 5.14(. 13)*** Variance Components

Intercept .OOl(.OOl) ,38(.37) .07(.12) .OOl(.OOl) N (Survey)/N(States) 301/21 298/21 346/21

Challengers

370/21

Fixed Components


Relevance/Incivility — .23(.22) —.39( .51) —.17(.24) —.23(.30) Backlash Effect — .51 (.29)** . 11 (.73) — ,49(.30)* — .08(.42)


Party Proximity ,36(.08)*** .67(.12)*** _41(_09)*** .41(.07)***

Ideological Proximity ,60(.07)*** .91 (.12)*** ,67(.09)*** 67( 07)***


Competition — ,01(.02) ,007(.03) ,03(.02)* ,006(.02)

Criticisms in the News —.0003(.003) —,002(.008) ,003(.004) .004(.004)

Constant 7.55(. 15)*** 12.89(.36)*** 7.23(.15)*** 4.58(.20)***

Variance Components

Intercept ,02(.10) ,91(.54)** ■ 14(. 12) .28(.19)*

N (Survey)/N(States) 209/21 161/21 231/21 290/21

'Parameter estimates are followed by standard errors in parentheses. P-values are based on one-tailed tests: ***p < .01; **p < .05; *p < .10.

2Affect is measured on a scale ranging from 4 to 12. Traits are measured on a scale ranging from 5 to 20. Issues are measured on a scale

ranging from 2 to 12. Overall evaluations are measured on a scale ranging from 0 to 10. See the text and Supporting Information for

operationalization of the variables.

3Relevance/Incivility is measured on a 3-point scale where the candidate is scored a 3 if the candidate's commercials are rated as high on

relevance and low on civility, the candidate is given a score of 2 if the commercials are rated as high on relevance and high on civility, and

the candidate is given a score of 1 if the commercials are rated as low on relevance. The Backlash Effect is measured with a binary variable

where the sponsoring candidates' commercials receive a score of 1 if these commercials are rated as low on relevance and low on civility,

and 0 for all other combinations of the sponsoring candidates' commercials.

evaluations of challengers, the relevance and civility do

not reach statistical significance in any of the four equa tions.47

Finally, we expect that the magnitude of the rele

vance and civility coefficients will be larger for people with low tolerance (i.e., Table 4) than for people with high

47In addition to the four dependent variables displayed in Tables

4-5, we also develop a model estimating the impact of the rele

vance and civility of the commercials on vote choice for high- and

low-tolerance respondents. We find that the measure assessing the

relevance and civility of the candidates' commercials is significant

(estimate = —.13, with a standard error of .03) for low-tolerance

respondents, but the measure is not significant (estimate = —.008,

with a standard error of .03) for high-tolerance respondents.



KIM L. FRIDK1N AND PATRICK J. KENNEY

tolerance (i.e., Table 5). And, this is precisely the case. In

eight of the eight models, the size of the relevance and

civility coefficient is larger in the low-tolerance sample than in the high-tolerance sample. In fact, the differences

are quite dramatic. In seven of the eight comparisons, the

relevance and civility coefficients are over twice as large in the low-tolerance sample, compared with the high tolerance sample.

The results presented in Tables 4 and 5 provide strong

support for our hypothesis. The analyses, conducted

across four different dependent variables and two sub

populations, are consistent and impressive. People who

do not like uncivil and irrelevant discourse in negative communication are more responsive to the variation in

the content and tone of negative commercials. These mes

sages directly influence their assessments of incumbents

and challengers. This finding stands in stark contrast to

those people who are unperturbed by messages presented

in an uncivil manner and focused on irrelevant content.

For people with higher tolerance for negativity, the vari ance in the relevance and civility of messages has little

influence on their evaluations of candidates running for the U.S. Senate.48

Conclusion

Communication between citizens and candidates dur

ing political campaigns is an essential element of rep resentative democracies. A fundamental aspect of these

discussions is for candidates to "make a case" for their

candidacies. This involves disseminating a variety of

messages, including biographical statements about the

candidates, messages intended to build trust with con

stituents, discussions about the candidates' political phi

losophy, and statements describing the candidates' policy plans for solving contemporary problems. In addition, candidates need to explain why they will be better repre sentatives than their opponents. Candidates often deliver these messages by disseminating negative advertisements

aimed at their rivals. This campaign strategy is extremely common in the last days of statewide and national elec tions.

Although negative campaigning is common, demon

strating the impact of negative campaigning has proven to be elusive. In this article, we introduce and test a more

sophisticated theory of negative campaigning, exploring the conditions contributing to the effectiveness of neg ative messages. In particular, we examine how the rele

vance and civility of negative messages influence citizens' assessments of candidates. The data analyses focus on the

2006 U.S. Senate campaigns and demonstrate that neg ative messages centered on topics relevant to governing

and disseminated in an uncivil manner (e.g., harsh or shrill language) strongly influence citizens' evaluations of candidates. Irrelevant messages, regardless of whether

they are presented in a strident or civil manner, have little

impact on voters' views of the candidates. However, neg

ative advertising is not without risk. When incumbents

stray away from relevant messages and produce and dis

seminate irrelevant and uncivil messages, citizens react

by lowering their evaluations of these incumbents. Chal

lengers, on the other hand, do not face as fierce a backlash

from potential voters.

We also demonstrate that not all citizens are equally influenced by negative campaigning. In particular, the

impact of negative messages is especially strong for cit izens who have limited tolerance for "attack politics." In contrast, for people with a high tolerance for rough and-tumble campaign rhetoric, negative messages are less influential.

All told, the evidence suggests that the effect of neg ative information on targeted politicians is not minimal.

Rather, the effects are multifaceted, and under some cir

cumstances, substantial. These effects hold up in the face of stiff controls for partisan and ideological attachments.

48Scholars have examined how partisanship and political sophisti cation condition the impact of negative campaigning. For example, Fridkin and Kenney (2004) find that political novices are more af fected by negative political advertising, compared with political experts. We reestimate the analysis in Table 3 for political novices and political elites, and we find that novices are more responsive to the variance in relevance and civility. In particular, the rele

vance/incivility coefficient is statistically significant in seven of the

eight equations for novices, but it is statistically significant in only two of the eight tests for experts. However, we contend that toler ance toward negativity is something more than sophistication. As Table 2 shows, sophistication does not influence people's tolerance of uncivil negative messages. And, while sophistication does signif icantly influence people's tolerance of irrelevant negative messages, it is not the most important variable, according to the standardized coefficients. Researchers have also found that partisanship affects the impact of negative advertising, with Independents being more

responsive to negative appeals than partisans (e.g., Ansolabehere and Iyengar 1997). In our analysis, because of the small number of

Independents (n =118), it is difficult to examine whether Indepen dents are more responsive to the relevance and civility of negative messages. However, we compare weaker partisans (Independents and leaners) and stronger partisans (strong and weak identifiers) in their responsiveness to the relevance and civility of negative mes

sages and find no differences. In particular, six of the coefficients for relevance/civility are statistically significant for weaker parti sans and six of the coefficients for relevance/ civility are statistically significant for stronger partisans.




In addition, we have identified these effects in real cam

paigns. While the present article focuses on evaluations

of candidates, we think that the relevance and civility of negative messages—and people's tolerance for these

messages—may also contribute to the resolution of the

ongoing debate regarding the (de)mobilizing effect of

negative campaigning. Embracing the variance in the con

tent and tone of messages may help explain whether nega tive messages enhance or depress turnout. We expect that

not all negative messages will demobilize citizens. Instead,

messages focusing on irrelevant material—and presented in an uncivil manner—may be the most likely type of

message to "turn off' voters. In contrast, messages focus

ing on relevant topics and presented in a civil manner

may actually engage citizens and increase participation. In addition, we expect that the impact of these different

messages will depend on people's tolerance for negative

campaigning. In particular, irrelevant and uncivil mes

sages may be more effective at depressing turnout among

people with little tolerance for negative rhetoric. We en

courage researchers to consider the variability in the con

tent and tone of negative messages when exploring how

negative campaigning affects turnout.

We believe that this article advances our understand

ing of negative campaigning by embracing the variability in the content and tone of negative messages as well as

the variance in people's tolerance for negative messages.

However, this article represents a first step. Future work

should explore variance in additional dimensions of cam

paign rhetoric, such as variance in the medium (e.g., In

ternet vs. television advertising), variance in the timing of the negative messages (e.g., early vs. late in the cam

paign), variance in the presence of a counterattack (e.g., how often, how fierce), and variance in the repetition of

the message (e.g., how many repeated attacks before effec

tiveness is detected). We believe that all of these avenues

will shed more light on how negative messages shape cit

izens' attitudes and, ultimately, their choices at the ballot

box.

References

Abbe, Owen G., Paul S. Herrnson, David B. Magleby, and Kelly

Peterson. 2001. "Are Professional Campaigns More Nega

tive?" In Playing Hardball: Campaigningfor the U.S. Congress, ed. Paul S. Herrnson. Upper Saddle River, NJ: Prentice Hall,

70-91.

Abramowitz, Alan I. 1988. "Explaining Senate Election Out

comes." American Political Science Review 82: 385-403.

Ansolabehere, Stephen, and Shanto Iyengar. 1997. Going Neg

ative: How Negative Advertisements Shrink and Polarize the

Electorate. New York: The Free Press.

Basil, Michael, Caroline Schooler, and Byron Reeves. 1991.

"Positive and Negative Advertising: Effectiveness of Ads and

Perceptions of Candidates." In Television and Political Ad

vertising: Psychological Processes, ed. Frank Biocca. Hillsdale,

NJ: Lawrence Erlbaum, 245-62.

Brader, Ted. 2006. Campaigning for Hearts and Minds—How

Political Ads Use Emotion to Sway the Electorate. Chicago:

University of Chicago Press.

Bradley, Brendan P., Karin Mogg, Sara J. Falla, and Lucy R.

Hamilton. 1998. "Attentional Bias for Threatening Facial Ex

pressions in Anxiety: Manipulation of Stimulus Duration."

Cognition and Emotion 12: 737-53.

Brambor, Thomas, William Roberts Clark, and Matt Golder.

2005. "Understanding Interaction Models: Improving Em

pirical Analysis." Political Analysis 14: 63-82.

Brooks, Deborah Jordan. 2006. "The Resilient Voter: Moving Toward Closure in the Debate over Negative Campaigning and Turnout." The Journal of Politics 68: 684-96.

Brooks, Deborah Jordan, and John G. Geer. 2007. "Beyond Neg

ativity: The Effects of Incivility on the Electorate." American

Journal of Political Science 51: 1-16.

Campbell, Angus, Philip E. Converse, Warren E. Miller, and

Donald Stokes. 1960. The American Voter. New York: Wiley.

Canli, Turhan, Zuo Zhao, John E. Desmond, Eunjoo Kang,

James Gross, and John D. Gabrieli. 2001. "An MRI Study

of Personality Influences on Brain Reactivity to Emotional

Stimuli." Behavioral Neuroscience 115: 33-42.

Chang, Chingching. 2003. "Party Bias in Political-Advertising

Processing." Journal of Advertising 32: 55-67.

Dahl, Darren W., Kristina D. Frankenberger, and Rajesh V. Man

chanda. 2003. "Does It Pay to Shock? Reactions to Shocking and Nonshocking Advertising Content among University Students." Journal of Advertising Research 43: 268-81.

Ellermeier, Wolfgang, and Wolfgang Westphal. 1995. "Gender

Differences in Pain Ratings and Pupil Reactions to Painful

Pressure Stimuli." Pain 61: 435-39.

Fenno, Richard. 1996. Senators on the Campaign Trail: The

Politics of Representation. Norman: University of Oklahoma

Press.

Fiske, Susan T. 1980. "Attention and Weight in Person Percep

tion: The Impact of Negative and Extreme Behavior." Journal

of Personality and Social Psychology 38: 889-906.

Franz, Michael M., Paul B. Freedman, Kenneth M. Goldstein,

and Travis N. Ridout. 2008. Campaign Advertising and Amer

ican Democracy. Philadelphia: Temple University Press.

Fridkin, Kim L., and Patrick J. Kenney. 2004. "Do Negative

Messages Work? The Impact of Negativity on Citizens' Eval

uations of Candidates." American Politics Research 32: 570

605.

Fridkin, Kim L., and Patrick J. Kenney. 2008. "The Dimen

sions of Negative Messages." American Politics Research 36:

694-723.

Geer, John G. 2006. In Defense of Negativity: Attack Ads in Pres

idential Campaigns. Chicago: University of Chicago Press.

Gilligan, Carol. 1982. In a Different Voice: Psychological The

ory and Women's Development. Cambridge, MA: Harvard

University Press.

Grose, Christian R., and Suzanne Globetti. 2007. "Valence Vot

ers: Images, Issues and Citizen Vote Choice in U.S. Senate




Elections." Presented at the annual meeting of the Midwest

Political Science Association.

Guttman, Amy. 1993. "The Disharmony of Democracy." In

Democratic Community, ed. John W. Chapman and Ian

Shapiro. New York: New York University Press, 411-21.

Hill, Seth J., James Lo, Lynn Vavreck, and John Zaller. 2007. "The

Opt-in Internet Panel: Survey Mode, Sampling Methodology and the Implications for Political Research." University of

California, Los Angeles. Unpublished paper.

Houston, David A., and Kelly Doan. 1999. "Can You Back That

Up?" Media Psychology 1: 191-206.

Jacobson, Gary. 2009. The Politics of Congressional Elections.

New York: Longman.

Jamieson, Kathleen Hall, Paul Waldman, and Susan Sherr. 2000.

"Eliminate the Negative? Categories of Analysis for Political

Advertisements." In Crowded Airwaves: Campaign Advertis

ing in Elections, ed. James A. Thurber, Candice J. Nelson,

and David A. Dulio. Washington, DC: Brookings Institution

Press, 44-64.

Kahn, Kim Fridkin, and Patrick J. Kenney. 1999. "Do Negative

Campaigns Mobilize or Suppress Turnout? Clarifying the

Relationship between Negativity and Participation." Ameri

can Political Science Review 93: 877-90.

Kahn, Kim Fridkin, and Patrick J. Kenney. 2004. No Holds

Barred: Negativity in U.S. Senate Campaigns. Upper Saddle

River, NJ: Pearson Prentice Hall.

Kahneman, Daniel, and Amos Tversky. 1979. "Prospect The

ory: An Analysis of Decision under Risk." Econometrica 47:

263-91.

Kern, Montague, and Marion Just. 1997. "A Gender Gap among Viewers?" In Women, Media, and Politics, ed. Pippa Norris.

New York: Oxford University Press, 99-112.

Kinder, Donald R. 1986. "Presidential Character Revisited." In

Political Cognition, ed. Richard R. Lau and David O. Sears.

Hillsdale, NJ: Lawrence Erlbaum, 233-56.

Kinder, Donald R., 2003. "Communication and Politics in the

Age of Information." In Oxford Handbook of Political Psy

chology, ed. David O. Sears, Leonie Huddy, and Robert Jervis. New York: Oxford University Press, 357-93.

King, Gary. 1989. Unifying Political Methodology: The Likeli hood Theory of Statistical Inference. Ann Arbor: University of Michigan Press.

Lau, Richard. 1982. "Negativity in Person Perception." Political

Behavior 4: 353-77.

Lau, Richard, and Gerald M. Pomper. 2001. "Negative Cam

paigning by U.S. Senate Candidates." Party Politics 7:69-87.

Lau, Richard R., and Gerald Pomper. 2004. Negative Campaign

ing: An Analysis of U.S. Senate Elections. Lanham, MD: Row

man and Littlefield.

Lau, Richard R., and Ivy Brown Rovner. 2009. "Negative Cam

paigning." Annual Review of Political Science 12: 285-306.

Lau, Richard R., Lee Sigelman, Caroline Heldman, and Paul

Babbitt. 1999. "The Effects of Negative Political Advertise

ments: A Meta-Analytic Assessment." American Political Sci

ence Review 93: 851-76.

Lau, Richard R., Lee Sigelman, and Ivy Brown Rovner.

2007. "The Effects of Negative Political Campaigns: A

Meta-Analytic Reassessment." The Journal of Politics 69:

1176-1209.

Lind, Allan E., Yuen J. Huo, and Tom R. Tyler. 1994. "And

Justice for All: Ethnicity, Gender, and Preferences for Dis

pute Resolution Procedures." Law and Human Behavior 18:

269-90.

Luke, Douglas A. 2004. Multilevel Modeling. Thousand Oaks,

CA: Sage.

McGraw, Kathleen M., and Marco Steenbergen. 1997. "Pictures

in the Head: Memory Representation of Political Candi

dates." In Political Judgment: Structure and Process, ed. Mil

ton Lodge and Kathleen M. McGraw. Ann Arbor: University of Michigan Press, 15-42.

McGuire, William J. 1964. "Inducing Resistance to Persuasion:

Some Contemporary Approaches." In Advances in Experi mental Social Psychology, Vol. 1, ed. Leonard Berkowitz. New

York: Academic Press, 191-229.

McGuire, William J. 1976. "Some Internal Psychological Fac

tors Influencing Consumer Choice." Journal of Consumer

Research 2: 302-19.

McGuire, William J. 1989. "Theoretical Foundations of Cam

paigns." In Public Communication Campaigns, ed. Ronald E.

Rice and Charles K. Atkin. Thousand Oaks, CA: Sage, 43

65.

Meyer, Wulf-Uwe, Michael Niepel, Udo Rudolph, and Achim

Schutzwohl. 1991. "An Experimental Analysis of Surprise."

Cognition and Emotion 5: 295-311.

Min, Young. 2004. "News Coverage of Negative Political Cam

paigns: An Experiment of Negative Campaign Effects on

Turnout and Candidate Preference." Harvard International

Journal of Press/Politics 9: 95-111.

Mutz, Diana C., and Byron Reeves. 2005. "The New Video

malaise: Effects of Televised Incivility on Political Trust."

American Political Science Review 99: 1-16.

Neuendorf, Kimberly A. 2002. The Content Analysis Guidebook.

Thousand Oaks, CA: Sage.

O'Keefe, Daniel J. 2002. Persuasion Theory and Research. Thou

sand Oaks, CA: Sage.

Perloff, Richard M. 2003. The Dynamics of Persuasion: Com

munication and Attitudes in the 21st Century. Hillsdale, NJ: Lawrence Erlbaum.

Petty, Richard E., and John T. Cacioppo. 1986. "The Elabora

tion Likelihood Model of Persuasion." In Advances in Social

Psychology, Volume 19, ed. Leonard Berkowitz. Orlando, FL:

Academic Press, 123-205.

Rivers, Douglas, n.d. "Sample Matching: Representative

Sampling from Internet Panels." Unpublished paper:

www.polimetrix.com.

Roddy, Brian L., and Gina M. Garramone. 1988. "Appeals and Strategies of Negative Political Advertising." Journal of

Broadcasting and Electronic Media 32:415-27.

Sides, John, Keena Lipsitz, and Matthew Grossman. 2010. "Do

Voters Perceive Negative Campaigns as Informative Cam

paigns?" American Politics Research. 38: 502-30.

Sigelman, Lee, and Mark Kugler. 2003. "Why Is Research on

the Effects of Negative Campaigning So Inconclusive? Un

derstanding Citizens' Perceptions of Negativity." Journal of Politics 65: 142-60.

Steenbergen, Marco, and Bradford Jones. 2002. "Modeling Multi-level Data Structures." American Journal of Political

Science 46:218-37.




Taylor, Shelly E. 1981. "A Categorization Approach to Stereo

typing." In Cognitive Processes in Stereotyping and Intergroup

Behavior, ed. David L. Hamilton. Hillsdale, NJ: Lawrence

Erlbaum, 83-114.

Teinowitz, Ira. 2008. "Study: Obama Gains on Mc

Cain in Negative-Ad Race." Advertising Age.

http://adage.com/campaigntrail/post?article Jd= 132167

(accessed October 10, 2008).

Thorson, Esther, William G. Christ, and Clarke Caywood. 1991.

"Selling Candidates Like Tubes of Toothpaste: Is the Com

parison Apt?" In Television and Political Advertising, Vol. 1,

ed. Frank Biocca. Hillsdale, NJ: Lawrence Erlbaum, 145-72.

Thurber, James A., and Candice J. Nelson. 2000. Campaign Warriors: Political Consultants in Elections. Washington, DC:

Brookings Institute.

Ulbig, Stacy G., and Carolyn L. Funk. 1999. "Conflict Avoid

ance and Political Participation." Political Behavior 21: 265

82.

West, Darrell M. 2010. "A Report on the 2008 Presiden

tial Nomination Ads: Ads More Negative Than Previ

ous Years." Brookings Institution. http://www.brookings.

edu/papers/2008/0630_campaignads_west.aspx. (accessed

March 2, 2010).

Westlye, Mark C. 1991. Senate Elections and Campaign Intensity. Baltimore: Johns Hopkins University Press.

Wright, Gerald C., Jr., and Mark B. Berkman. 1986. "Candi

dates and Policy in U.S. Senate Elections." American Political

Science Review 80: 567-90.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press.



Variability in Citizens' Reactions to Different Types of Negative Campaigns

Documents

exact question wording

ideological proximitycampaign characteristics

reach statistical significance

negative political rhetoric

cam paign rhetoric

negative messages focusing

negative trait advertisements

negative mes sages