Candidate Attacks and Voter Aversion The Uncertain Effects of Negative Campaigning

First, drawing on a survey and a series of focus groups conducted during the 2002 California gubernatorial campaign, we examine voters’ response to negativity and its relationship to att

April 2005

John SidesUniversity of Texas at Austinjsides@gov.utexas.edu

Keena LipsitzMatthew GrossmannChristine TrostUniversity of California, Berkeley

Abstract

Researchers have found it difficult to show that negative

campaigning decreases voter turnout We propose a cognitive model of response to negative campaigning that helps explain this inconclusive research Any connection between negativity and the turnout decision requires a series of “steps” for voters: the actual reception of negative messages, an unfavorable evaluation of these messages, the generalization of this

evaluation to the candidates or campaign as a whole, and the decision to respond by staying home on Election Day Drawing

on survey data, content analysis of candidate advertising, and focus groups, we demonstrate that there is “slippage” at each step, undermining any relationship between negativity and

turnout

I Introduction

Americans love to hate both negative advertising and political

campaigns The conventional wisdom is that these aversions must be

related The contention among some scholars that negative advertising

leads to lower turnout (Ansolabehere and Iyengar 1995; Ansolabehere et al

1999) also implies this relationship: voters’ distaste for negativity produces dissatisfaction with the candidates and the campaign, which leads voters to stay home on Election Day However, the connection between negative campaigning, dissatisfaction with the candidates and campaigns, and lower

turnout is far from universal or certain, as a variety of studies attest (e.g.,

Clinton and Lapinski 2004; Finkel and Geer 1998; Freedman and Goldstein 1999; Goldstein and Freedman 2002; Kahn and Kenney 1999; Lau and

Pomper 2004; Lau et al 1999; Vavreck 2000; Wattenberg and Brians 1999).

We believe the current debate is inconclusive because it is largely

atheoretical in terms of how voters process campaign messages To move forward, we must unpack the cognitive process by which an individual’s experience with negative campaigning might affect her intention to vote

We argue that “demobilization” is not an inevitable consequence of negative campaigning because voters’ perceptions of and responses to negative campaigning are more complex than simple aversion Any causal connection between negative campaigning and lower participation entails a series of “steps” in the voter’s mind: the actual reception of negative

messages, an unfavorable evaluation of these messages, the generalization

of this unfavorable reaction to the candidates or campaign as a whole, and the decision to respond by staying home on Election Day We argue that there is the potential for “slippage” at each step, suggesting that negative tactics have no deterministic effect on attitudes toward the campaign and ultimately on turnout.

We employ two studies to assess this process empirically First,

drawing on a survey and a series of focus groups conducted during the

2002 California gubernatorial campaign, we examine voters’ response to negativity and its relationship to attitudes about the campaign and to vote intention Both the survey and focus groups reveal that voters do not

uniformly reject negativity They may even evaluate negative messages favorably if the messages contain information they deem useful

Furthermore, multivariate analysis reveals a striking paradox: the attributesassociated with more favorable views of negativity, notably political

involvement, are associated with more dismal assessments of the

candidates and campaigns Furthermore, these same attributes are

strongly associated with vote intention Thus, those most likely to find negativity acceptable are also the least satisfied with the campaign but yet the most likely to intend to vote The tension inherent in these results

suggests that, while negative advertising may be distasteful to many, it cannot wholly explain dissatisfaction with politicians and their campaigns orthe decision to participate on Election Day

Second, we use rolling cross-sectional surveys and television

advertising data from the 1998 California and Illinois gubernatorial races tocompare actual levels of negativity, as measured in advertising, with voters’perceptions of the candidates’ behavior over the course of the campaigns

We find that these perceptions are multi-dimensional: while actual levels of negativity do affect whether voters believe the candidates are being

“negative,” they do not necessarily affect whether votes believe the

candidates are providing “useful information.” This suggests that a

“negative” advertisement may have some redeeming value to voters

Moreover, there is again no significant relationship between actual or

perceived levels of negativity and vote intention Both sets of evidence illustrate how the relationship between negative campaigning and citizen dissatisfaction with campaigns and participation on Election Day can break down at any stage of the cognitive process we describe

II Theory and Expectations

We argue that the effect of negative campaigning on voters depends crucially on how voters perceive and respond to it Few studies have

attempted to “unpack” voter perception and response, focusing instead on the relationship between negative campaigning and such phenomena as turnout, political trust, and political efficacy We believe, however, that there is value in understanding the cognitive process that may lead to a decision not to vote Two recent studies have begun to move in this

direction by arguing that voter perceptions of negativity may not accord with conventional wisdom Sigelman and Kugler (2003) show that

negativity may mean different things to different people They examine three campaigns (including two we examine here)—the 1998 California, Illinois, and Georgia gubernatorial races—and argue that because most Americans “perceive political campaigns dimly,” their assessments of

negativity are colored by numerous biases As a result, Americans simply

do not perceive negativity in ways that correspond to how political

scientists typically define it, nor do Americans agree with each other about what is negative In contrast, Kahn and Kenney (1999) argue that the

problem is not that citizens disagree about what is negative but that they

“distinguish between legitimate and tempered criticisms, on the one hand, and acrimonious and unjust criticisms, on the other hand” (884) The latter,they claim, is responsible for lowering turnout

We agree with these authors that there is a “complex linkage betweenreality and perceptions” (Sigelman and Kugler 2003: 158), and our

proposed model builds on this insight To begin, we define negative

campaigning simply as statements made about one’s opponent, as opposed

to about one’s self Our model of response to negativity then posits four keysteps, all of which must take place in order for negative campaigning to induce lower turnout Below we elaborate this model and the empirical expectations it generates

First, voters must actually “receive” the negative messages that the

candidates are broadcasting, e.g., in their paid advertising This exposure

stage is an obvious one, as it is unlikely that voters will be influenced by what they do not see or hear (see Zaller 1992) At this stage in the process,

we are not concerned about whether a respondent believes that an ad is

“negative”—however they might define that term—but whether they

recognize when candidates are talking mainly about themselves or their opponents The testable hypothesis is that voters’ perceptions of how muchthe candidates are criticizing each other will depend on the volume of this criticism (H1) That is, there is a relationship between the actual and

perceived levels of negativity in a campaign—a relationship that Sigelman and Kugler (2003) argue is lacking.1 We use new data sources that enable

us to test directly whether perceived levels of negativity indeed track actuallevels To our knowledge, this has not been done in the negative

advertising literature

Evaluation

Second, and perhaps most importantly, once voters have been

exposed to a campaign message, they must decide whether its content is

1 An obvious corollary is that the likelihood of reception will be greater among voters who are more exposed to the campaign In the supplementary materials for this paper, we investigate whether individual-level measures of advertising exposure and political

attentiveness mediate the effects we describe below We did not find consistent evidence that any such measure plays an important mediating role, though this may reflect in part the limitations of the available measures.

“negative” That is to say, their evaluation of the message must itself be

negative It might seem that any message featuring one candidate’s

criticizing the other would inevitably produce aversion among voters, but this is not necessarily the case

We believe that voters’ responses to negative campaigning are dimensional Voters certainly evaluate the conduct of a campaign in terms

multi-of its negativity, but they have other criteria as well One important

criterion, we argue, is whether they believe they are learning information about the candidates, including their qualities, experiences, and stances on issues ([authors] 2005) This does not mean that voters diligently seek out such information (see Hibbing and Theiss-Morse 2000); instead, voters are confronted with information when, usually inadvertently, they see or hear communication from a candidate The key is whether they find that

information helpful in terms of learning about the candidates Many

scholars (Finkel and Geer 1998; Geer 2000; Jamieson, Waldman, and Sherr 2000; West 1993) have argued that political ads contain useful information, such as discussion of the candidates’ issue positions In fact, negative ads

may often contain more of such information (Geer 2000) Consequently,

voters actually learn valuable information from advertising (Freedman,

Franz, and Goldstein 2004; Ridout et al 2004; Zhao and Chaffee 1995).

As a result, voters may perceive campaign ads both as negative and as

providing useful information Indeed, negativity may actually trigger

political learning Negative information may be more salient amidst an

otherwise positive world (Lau 1985) It may illuminate the potential costs and benefits of electoral choices (Kahn and Kenney 1999; Lau 1985) It mayalso produce anxiety in viewers, encouraging them to stop relying on

political habit and to reevaluate their beliefs (Marcus, Neuman, and

MacKuen 2001) Thus, voters may find negative messages edifying and ultimately helpful in deciding between the candidates If so, then negativitymay facilitate participation more than it inhibits it One important

contribution of this analysis is that we are able to measure various

dimensions of voters’ evaluations of the candidates’ conduct during the campaign, including its negativity and informativeness

Our empirical expectations are two First, we expect voters’

evaluations of campaign conduct to be multi-dimensional (H2) That is, various indicators, such as perceptions of whether the candidates are being negative and providing useful information, tap different underlying

constructs This should be evident cross-sectionally—e.g., in a factor

analysis—and also over time, which implies that different indicators will not trend in the same fashion Second, we expect these indicators to respond differently to changes in advertising tone More precisely, we expect that voters will deem the criticism they see in advertising as “negative” but not necessarily as “uninformative” (H3 & H4).2

2 Two other factors that could affect how voters evaluate negative messages are the content

of the ad and the source of the ad (i.e., the party of the candidate who sponsored it) The

supplementary materials for this paper evaluate the role of each factor, focusing on

whether issue- or character-based negative appeals have a larger effect and on whether voters’ party identification shapes how they evaluate in- and out-party messages We find little conclusive evidence that, defined thusly, the content and source of the ad have

systematic consequences.

After exposure and evaluation, the third step in this process is that

aversion to negativity must produce a generalized negative attitude about

the candidate(s) and/or the campaign.3 This may occur if a voter has

witnessed a large enough amount of negative campaigning that he or she evaluates unfavorably However, dissatisfaction with campaigns may also

derive from other sources—e.g., frustration with the role of money in

campaigns, a lack of enthusiasm about the candidates themselves, or a sense that one’s issue priorities are getting scant attention As a

consequence, we expect people who are more averse to campaign negativity

to be more dissatisfied with campaigns, but that aversion should only

weakly predict dissatisfaction given that dissatisfaction depends on many other factors (H5)

Vote Intention

The fourth and final step in any relationship between negativity and turnout is the decision not to participate in an election Potential voters must be so unhappy with the tenor of the campaign that this feeling

overwhelms factors, such as their sense of civic duty or their habitual

behavior (see Abramson and Aldrich 1982) Because many other

3 We remain agnostic on how attitudes about the candidates are related to attitudes about the campaign more generally It may be that these should be kept distinct, or it may be that they are so strongly interrelated that distinguishing them is unnecessary We employ indicators that refer to the candidates by name and that refer to campaigns more generally.

demographic and attitudinal factors affect turnout (see, e.g., Wolfinger and Rosenstone 1980), we expect that aversion to negativity will not be

associated with vote intention (H6) Finally, the ultimate test of our

theoretical model is whether there is any relationship between negative advertising—either actual or perceived—and vote intention Because the cognitive process that links these factors is complex, involving reception, evaluation, and generalization, we believe that the relationship between negative advertising and vote intention should be extremely weak or

nonexistent (H7)

III Research Design and Data

Our research design draws on multiple sources of data to evaluate these hypotheses The data derive from three different campaigns: the

2002 California gubernatorial race, and the 1998 California and Illinois gubernatorial races Each set of data has its strengths and weaknesses, but

in combination they allow us to pursue a strategy of triangulation in

assessing voters’ reaction to negativity in campaigns

The first set of data, which we refer to as “Study #1,” was collected during the 2002 California gubernatorial race This race featured relatively unpopular candidates: the incumbent Democrat Gray Davis, who had only a

39 percent favorable job rating in a September 2002 statewide poll, and Republican Bill Simon, who was plagued by questions about his business

ethics Davis dominated Simon in advertising volume and focused his

message on these ethical questions

During the campaign, we helped design a survey of Californians that was conducted between October 28 and November 4, 2002.4 It has a

sample size of 1,814 adult citizens aged 18 and older In the survey, we included a question measuring respondents’ general attitude toward

negativity in campaigns, measures of voters’ evaluations of the candidates and of campaigns, and a measure of vote intention, allowing us to gauge whether attitudes towards negativity affect satisfaction with the campaign and how both of these affect vote intention

In conjunction with this survey, we also helped design and conduct a series of six focus groups in September and October 2002 The focus

groups were held in three different California cities, Walnut Creek (a

suburb of the Bay Area), Fresno, and Los Angeles, though participants camefrom surrounding areas as well In Walnut Creek, one group was composed

of young people aged 18 to 35, and the other of older people aged 55 and above In both Fresno and Los Angeles, one group was composed of strong partisans and the other of “swing voters.”5 Each focus group had 9-15 participants, and every effort was made to recruit participants diverse with respect to education, income, ethnicity, and party identification Each

4 The survey’s design and implementation was a joint project of the Institute of

Governmental Studies at the University of California, Berkeley, and the Public Policy Institute of California.

5 “Swing voters” were defined as partisan independents who said that they usually voted a split ticket.

session lasted approximately two hours.6 The focus group format allowed participants to discuss their feelings about campaign negativity in a more nuanced fashion and this discussion provides insight into responses to the survey questions Study #1, with its quantitative and qualitative

components, addresses the evaluation, generalization, and vote intention steps of the cognitive model (steps 2 through 4) It provides insight into how people make judgments about negative advertising, whether these judgments affect their overall attitudes about campaigns, and whether theseattitudes are reflected in the decision to vote

Study #2 involves data from the 1998 California and Illinois

gubernatorial elections The former featured the Democratic Lieutenant Governor, Gray Davis, running against the Republican Attorney General, Dan Lungren Davis won in an uncharacteristic landslide, garnering 60 percent of the two-party vote Conventional wisdom held that Lungren lost

because he was “too conservative” for California, e.g., pro-life The latter

featured Democratic Congressman Glenn Poshard and Republican

Secretary of State George Ryan Ryan won the election with 52 percent of the two-party vote, a narrow margin attributed to a scandal in Ryan’s office that benefited Poshard even as Ryan outspent Poshard nearly three to one

6 The Walnut Creek focus groups took place on September 9, 2002 From this point on, we will refer to the group with participants who were 55 and over as “WC1” and to the 18-35 group as “WC2.” The Fresno focus groups took place on September 26, 2002 The group

of independents will be referred to as “F1” and the group of strong partisans as “F2.” The Los Angeles focus groups were held on October 1, 2002 Again, the group of independents will be referred to as “LA1” and the group of strong partisans as “LA2.”

The public opinion data for these races come from two rolling sectional surveys commissioned by the Annenberg School for

cross-Communication at the University of Pennsylvania. 7 This survey design consists of small daily cross-sectional samples Because this design relies

on daily interviews, its granularity is fine enough to track changes in

opinion very closely, and to capture the impact, whether immediate or

delayed, of campaign activity (see Johnston et al 1992; Johnston and Brady 2002; Romer et al 2004) The Annenberg Survey sampled respondents in

the San Francisco-Bay Area and Chicago media markets from September

22, 1998 to November 2, 1998 The total sample size was 2,902 in

California and 613 in Illinois

We also measured the tone of the television advertising that aired in these two media markets, drawing on data originally collected by the

Campaign Media Analysis Group (CMAG), based in Arlington, VA Their technology tracks satellite transmissions from the major national networks and recognizes the digital “fingerprint” of various television programs and advertisements With that fingerprint, CMAG then records when and whereeach political advertisement ran, and which candidate (or party or interest group) was its sponsor.8 For the 1998 California governor’s race, the CMAGdata span the period from September 7 to November 2 During this period, Davis and Lungren aired a total of 3,609 advertisements in the San

7 The California and Illinois surveys were pilot surveys for the 2000 National Annenberg Election Study (see Johnston, Hagen, and Jamieson 2004).

8 Freedman and Goldstein (1999) were the first to demonstrate the power of this data source for political science, using it to argue that negative advertising does not depress turnout (see also Goldstein and Freedman 2000, 2002)

Francisco market.9 Davis aired the majority of these (1,847 vs 1,762) For the Illinois race, the CMAG data span the period from July 7 to November 2.During this period, Ryan and Poshard aired 2,372 advertisements in the Chicago market, of which Ryan aired the vast majority (1,735).10

The combination of survey and advertising data in Study #2 produces

a dataset that is well-suited for exploring the reception and evaluation steps

of the cognitive model and for testing whether negative campaigning is related to vote intention First, Study #2’s precise measurement of

campaign advertising in respondents’ media market up to the day of their interview allows us to ascertain whether there is any relationship between actual and perceived levels of negativity—something Study #1 does not allow Moreover, the time-series nature of the data makes it possible for us

to observe campaign effects as they occur Second, the survey asks

respondents to rate the campaigns and candidates along several different

dimensions, e.g., their level of negativity and the amount of useful

information they were providing These responses will shed light on how people evaluate the information they receive from candidates More

importantly, the responses will reveal whether these evaluations are dimensional, as we hypothesize Third, these data allow us to test the

multi-9 In California, the CMAG data did not have clear labels for some ads and thus we could not code them for content There were also several Spanish-language ads, for which CMAG did not record their content However, the vast majority of ads (83%) could be identified and coded In Illinois, we had none of these problems, though there were two Spanish-

language ads whose content was not recorded and could not be coded We were able to code 93 percent of the ads that aired in the Chicago market

10 CMAG’s system also picks up television advertisements aired on behalf of the candidates

by third parties However, in these races their data show very few such advertisements In Illinois, the state Democratic Party did produce an ad on behalf of Poshard which aired 61 times We consider this ad as one of Poshard’s in the analyses that follow.

expectation that actual negative advertising will not be strongly related to vote intention due to slippage at each stage of the cognitive model To our knowledge, no one has examined perceptions of negativity and their

relationship to advertising tone using comparable data.11 Finally, by

comparing three quite different campaigns, we can be more confident aboutthe generalizability of our findings.12

IV Study #1: The 2002 California Gubernatorial Race

We begin with Study #1, which addresses the evaluation,

generalization, and vote intention steps of our theoretical model

Step 2: Evaluating Negativity in the Abstract

We included a question on the 2002 survey that asked voters whether they supported or opposed the candidates’ criticizing one another We

11 In analyzing these races, Sigelman and Kugler (2003) employ the 1998 NES Pilot Study, which included a sample in each of three states, California, Illinois, and Georgia Their measures of campaign content come from newspaper coverage and from a small sample of political ads from the University of Oklahoma’s Political Commercial Archive Kahn and Kenney (1999) rely on campaign consultants’ subjective ratings of how much “mud-

slinging” took place The CMAG data are better in that they provide more complete

collection of political ads as well as information on when and how often those ads aired (see Prior 2001 for why this information is crucial) Sigelman and Kugler were also limited

to a single measure of voter perception (how they rated the negativity of their respective gubernatorial campaign on a four-point scale) Furthermore, the Annenberg data provide temporal leverage that a cross-section such as the 1998 NES Pilot Study cannot

Candidate tone may vary significantly during a campaign, with consequences for voters’ perceptions—something Sigelman and Kugler acknowledge (154) Thus it is possible that the people who were interviewed early on thought the campaign was less negative than those interviewed at the end The only potential disadvantage of the Annenberg data is that they were collected in select markets, San Francisco and Chicago, which are not demographically representative of the rest of the state, though we do not believe they are systematically different in a way that would affect our ability to test the model proposed herein

12 Given the assessments of these campaigns offered by consultants and the media, all three would likely be considered “mudslinging” campaigns using the measure developed by Kahn and Kenney (1999) If any campaigns were likely to generate a relationship between

negative advertising and turnout, these campaigns would be prime suspects.

strove to develop a question that would not elicit knee-jerk opposition and would give respondents reasons to oppose and support negative

campaigning This question also contained a wording experiment, in which

a random half of respondents were asked about “political candidates” and the other half were asked about “Gray Davis and Bill Simon.” Thus we can compare how voters feel in the abstract to how they feel when specific candidates are mentioned The question wording was:

Version 1: Some people say that in general, political candidates

should never be critical of their opponents because campaigns

have gotten too negative, while others say that candidates need

to criticize their opponents because it is important to know the

strengths and weaknesses of all candidates Which of these

comes closest to your view?

Version 2: Some people say that Gray Davis and Bill Simon

should not be critical of one another because campaigns have

gotten too negative, while others say that they need to criticize

one another because it is important to know the strengths and

weaknesses of both candidates Which of these comes closest to

your view?

The first version of the question produced an evenly divided response:

51 percent said that candidates should not criticize and 49 percent that they need to criticize The second version produced a more skewed

distribution: 62 percent said that candidate should not criticize and 38

percent that they need to criticize These results suggest that voters are not universally or reflexively opposed to criticism in campaigns, at least in the abstract, and thus may not react to every “negative” ad they see with disgust But it is also true that this opposition increased when voters were reminded of a specific campaign, in particular one that featured a great deal of criticism between unpopular candidates, as the 2002 California gubernatorial race did.

The focus group discussions revealed several reasons that voters may not be universally averse to criticism Although focus group participants often expressed their dislike for negativity, citing a variety of reasons,

including the irrelevance of an attack and its inaccuracy, many participants indicated a willingness to accept a negative campaign message if it

provided useful information Thus, they appear to evaluate campaign

communications along several dimensions They defined useful ads as thosemaking important information available, those divorced from partisan

bickering, and those that helped voters assess a candidate’s character:

Margaret: When I learned or I believe I learned that Simon had

a lawsuit against his company, that was critical information to

me, because I didn’t know Simon (WC1)

John: If it’s pertinent and important, you know, but not just

mudslinging It’s got to be truthful and honest, and not biased

as a result of party separations [sic] (WC1)

Doug: I don’t think all negativity is bad in this campaign,

because it really casts doubt upon somebody’s character And if

they’re going to grease their own pockets, then, you know, let’s

face it, that’s one of the best ways to know the guy’s character,

if he’s in it for himself… If they’re falsifying their taxes and it’s

public, someone is going to call them on it, and then you want to

know that negative information (WC2)

The format of the attack also affects whether participants viewed an attack

as acceptable:

Jeff: I don’t mind when they’re saying, you know… ‘We stand this way

on this and they stand that way on that.’ It’s when, it’s just, you

know, the ad and the dramatic music behind it, and then show a

close-up on it, and like freeze-frame it in black and white, try to make it all super-dramatic (F2)

Donald: You can’t… if somebody is mudslinging in a 20-second

[sound]bite, you’re not going to get any information He’s going to say Simon did this or Gray Davis says that—what did he really do? Wedon’t know In 20 seconds or whatever it is, you get no information (WC1)

Finally, some focus group participants stated that no matter how much theydisliked negativity, it would not necessarily affect their assessments of the candidates One participant, Charles, said that for him negativity cannot overcome issue differences: “No matter how ‘positively’ someone says

they’re going to raise taxes, I probably won’t vote for them” (F2)

The comments of the focus group participants suggests that blanket terms like “negative campaigning” or “mudslinging” may conceal

distinctions that voters make between acceptable and unacceptable

criticism Some participants viewed criticism—even in the form of a pure attack—as a necessary part of campaigns, as long as it is fair, relevant, and truthful.13 The variation in response to the survey question about candidate criticism, as well as the comments offered by focus group participants, suggest how “slippage” may occur at the evaluation stage of the model Voters perceive candidate criticisms of their opponent as “negative” but do not always find those criticisms unhelpful or objectionable

Another potential reason for slippage is that different kinds of voters may evaluate criticism differently (Sigelman and Kugler 2003) We

hypothesize that educated and politically attentive voters will be more

accepting of candidate criticism for several reasons First, these voters tend to know more about the political process and to be savvier about its nature.14 Second, the educated and politically attentive may be more likely

to appreciate what can result from a critical exchange between candidates, such as a better understanding of their character or issue positions Third, the educated and politically attentive may feel more politically efficacious (Craig, Niemi, and Silver 1990) and in particular capable of parsing a

13 While participants generally agreed on these criteria, they occasionally defined concepts like fairness, relevance, and truth in different ways As one participant put it, “Because what I think is a personal attack, you might not I mean, we have to define that” (Laverne, WC1).

14 Some participants in the focus group voiced the notion that critical exchanges between

the candidates are de rigueur As one participant, Mark, who professed a great deal of

interest in politics, put it: “Mostly, it’s basically just the essence of marketing, you know, the true essence, is basically to sell… It’s about ‘I’m better than the next guy’” (WC2)

critical exchange between candidates and extracting useful information, whereas the less attentive may feel unable to do so Explaining why she dislikes campaigns, one of the focus group participants, Linda, who

professed little interest in politics, said, “There is a lot of lying, and those of

us who are not really politically minded, like myself, end up more confused when it comes to actual voting” (F1)

To investigate this hypothesis, we used the data from Study #1 to estimate the effect of attentiveness on support for criticism between the candidates, controlling for various demographic and political attributes The dependent variable is coded such that 1 indicates support for

candidates’ criticizing each other and 0 opposition to criticism; we

combined into an omnibus measure both the generic version of this questionand the version referencing Davis and Simon.15 Following Sigelman and Kugler (2003) our chief measure of attentiveness is political information (the number of political figures a person could identify correctly, out of four).16 We also include a measure of education as well as age, gender, race, and partisanship (the seven-point scale folded)

Table 1 presents the results As hypothesized, political information and education are associated with a greater likelihood of endorsing

criticism Holding other variables at their means or median, a shift from theminimum to the maximum value of information produces a 07 increase

15 When these two versions are modeled separately, the only major difference is that the effect of gender is much stronger in the version mentioning Davis and Simon explicitly.

16 The measure of political information is based on questions asking respondents to identify the offices held by Dianne Feinstein, Tony Blair, and Colin Powell In this sample, the mean number correct was 1.7 (s.d.=1.2) The overall reliability of the index is 60.

(s.e.=.03) in the probability of believing criticism necessary.17 The same increase in education is associated with a 12 increase (s.e.=.05) Besides political information and education, the only other variables that exert any impact are gender and ethnicity The coefficient for gender suggests that women dislike criticism more than men do Blacks and Latinos are also more supportive of criticism than whites.

[insert Table 1 about here]

How do all of these results speak to the possible disjuncture between campaign discourse and voters’ perceptions of it? We find that, when

provided with reasons for both opposing and supporting criticism, voters are roughly evenly split in their response, although their opposition to

criticism increases when the two gubernatorial candidates, Davis and

Simon, are mentioned directly This suggests that “negative campaigning”

is not always noxious The focus group responses shed light on why this is the case Moreover, we find some cross-sectional differences among voters,with the politically informed more likely to support criticism This suggests,

as Sigelman and Kugler (2003) also find, heterogeneity among the

electorate in how they evaluate campaigns

Step 3: Generalizing about the Candidates and the Campaign

The next question is whether voters’ reactions to negative

campaigning lead to negative assessments of campaigns and the

17 These changes in probability were calculated using the Clarify program (King, Tomz, and Wittenberg 2000).

candidates To assess these connections, we rely on two indicators from thesurvey conducted during the 2002 California gubernatorial campaign When respondents were asked, “Would you say you are satisfied or not satisfied with the choices of candidates in the election for Governor on November 5?” a majority of respondents (58%) said they were “not

satisfied” with the candidates The second indicator asked “More generally,

do you think that election campaigns in California have gotten better in the last 10 years, gotten worse, or stayed about the same?”18 In this sample 49 percent said “worse,” 40 percent said “the same,” and only 10 percent said

“better.”

To explore the sources of campaign satisfaction, we estimated models

of each indicator, including the same covariates described above as well as people’s support for criticism between the candidates (combining the two versions of this question into a single indicator) If views of criticism are generalized to views of the campaign, we would expect those who support criticism to be more satisfied with the campaign as a whole Our other mainhypothesis concerns the role of political information Although the better-informed are more likely to support criticism in campaigns in the abstract,

we hypothesize that they are actually less likely to be satisfied with the

18 This is not an ideal question, in that respondents could think campaigns in California were awful ten years ago and awful now, which would lead them to say “the same” even though their opinion of the campaign was hardly positive Moreover, the question does not reference the 2002 campaign specifically Nevertheless, it is difficult to believe that

responses to this question would bear no relationship to responses to the 2002 campaign This question also randomized respondents into two subsamples The first heard the

question wording above The second heard a question that said, “Overall, in terms of

ethics and values, do you think election campaigns in California etc.” This experiment

produced only a little variation in response and so we combined those questions into a single measure.

specific candidates and with campaigns as a whole Because they are more attentive to political events and news, they are exposed to more informationabout the candidates and campaign, and it is likely that this familiarity breeds contempt (see Sigelman and Kugler 2003).

Table 1 contains these results from these models First, we find that views of negative campaigning affect satisfaction with the campaign Thosewho endorsed criticism were more likely to say they were satisfied with the candidates and to express more positive views of campaigns generally.19 This suggests that views of negative campaigning are linked to views of the campaign as a whole, though the substantive magnitude of this linkage is not very large Holding the other variables constant, a shift from opposition

to support of negative campaigning increases the probability of expressing satisfaction with the candidates by only 08 (s.e.=.03).20

The hypothesis that political attentiveness breeds dissatisfaction with the candidates and with campaigns is strongly confirmed in these results For example, a shift in information from its minimum to maximum value decreases the likelihood of satisfaction with the candidates by a whopping -.35 (s.e.=.04) These findings suggest that familiarity does indeed breed

19 In models with the two versions of this question entered separately, support for criticism

by Davis and Simon is more strongly related to satisfaction with the candidates than

generic support for criticism However, both versions of this question were significantly associated with evaluations of California campaigns overall.

20 Satisfaction with the candidates and campaign increases among strong partisans, Asians, and Latinos, but declines with age Independents are likely more alienated from politics than are partisans, while older people are more likely than young people to recall past elections when the candidates were more likable The effect of ethnicity comports with the finding that Latinos tend to hold more trusting attitudes towards government (Uhlaner and Garcia 2002)—a finding that appears generalizable to Asians as well We also investigated whether political involvement conditions the relationship between views of negative

campaigning and satisfaction with campaigns This was not the case.

contempt: those who are knowledgeable about politics find the proceedings more distasteful Thus, although aversion to negativity is related to

dissatisfaction with the campaign, those voters who are most likely to find

negative campaigning acceptable are also those most likely to find the

candidates and campaigns unacceptable.

Step 4: From Dissatisfaction to Lack of Participation

The final question is whether voter attitudes toward negativity and campaigns generally result in a disinclination to vote Our indicator is a dichotomous measure of whether the respondent intended to vote, coded 1

if yes and 0 otherwise (see also Ansolabehere and Iyengar 1995).21 Our model of vote intention includes the measure of opposition to negativity as well as the two indicators of satisfaction with the campaign along with the traditional correlates of participation Table 1 reports the results The

model of vote intention reveals that, despite their dissatisfaction with

campaigns, the politically informed are more likely to intend to vote

Perhaps most strikingly, neither support for criticism nor campaign

satisfaction has any significant relationship to vote intention.22

21 The number of respondents who “intend” to vote is much higher than the number who actually voted (In this sample, about 65 percent indicated such an intention, while actual turnout was only 50 percent of those registered to vote.) We did attempt to mitigate social desirability bias by making respondents feel more comfortable admitting that they did not intend to vote The actual question wording was, “Some people who plan to vote can’t always get around to it With your own personal schedule in mind, are you absolutely certain to vote, will you probably vote, are the chances about 50-50, less than 50-50, or don’t you think you will vote in the November election?” Using ordered probit models, we found no significant differences between any of the intermediary categories, and thus we collapsed the measure Those who said they were absolutely certain to vote were coded 1, and all other categories were coded 0.

22 Even if only those respondents who received the version of the negative campaigning question that referenced Davis and Simon specifically are included in the model, this

These results provide initial confirmation of several hypotheses First,the public’s evaluation of negative campaigning is mixed Second, while those averse to negative campaigning in the abstract are more dissatisfied with the candidates and campaigns, the conflicting role of political

information complicates this relationship While the politically informed aremore tolerant of criticism between candidates, they are remarkably less satisfied with the gubernatorial candidates and with campaigns generally, and yet still much more likely to intend to vote Thus, even though

opposition to criticism is associated with more negative assessments of candidates and campaigns, the dual effect of information suggests that it is not the primary reason people dislike campaigns or choose not to vote At each stage of the cognitive process, different groups of voters do not

“proceed” to the next step No group appears to make it through all of the stages

IV Study #2: The 1998 California and Illinois Gubernatorial Races

Our study of the 1998 California and Illinois gubernatorial races

allows us to test three steps of the cognitive model directly: reception,

evaluation, and vote intention To analyze reception and evaluation, we firstcoded the tone of ads aired in the San Francisco and Chicago media

markets during 1998 California and Illinois races, and then compared our content-based measure of advertising tone to voters’ perceptions We

measure still has no significant effect on vote intention.

coded each claim in each ad as referring to the candidate sponsoring the ad(a “positive” claim) or to the opponent (a “negative” claim) Coding the claim as opposed to the entire advertisement provides a more nuanced reading of the ad’s content (see Jamieson, Waldman, and Sherr 2000) It also avoids slippery assessments of whether an entire ad can be categorized

“positive,” “negative,” or some combination thereof (i.e., “contrast” ads).23

Having coded the ads in this fashion, we counted the number of

positive and negative claims that aired on each day (by multiplying the number of claims in each ad by the number of times that ad aired) Figure

1 displays the trend in the number of positive and negative claims for each candidate in both races The vertical line indicates the date when the

survey began (September 22)

[insert Figure 1 about here]

In California, both Davis’ and Lungren’s advertising follow the same basic pattern Initially, advertising was on balance positive But during September and the first half of October, both candidates aired a larger amount of negative advertising As Election Day drew nigh, both Davis and Lungren turned positive again, perhaps intending to end the campaign on a warmer and fuzzier note In Illinois, Ryan’s advertising was initially

positive, with a large peak in positive appeals in late September

Thereafter, his advertising became much more negative and continued to be

so until Election Day Poshard’s advertising was far less numerous because

23 Three of us coded the ads separately These initial assessments were all correlated at 90

or better We then resolved any discrepancies to finalize the coding.

he did not accept PAC contributions and thus had less money to spend In terms of tone, his advertising was a roughly equal mixture of positive and negative claims, with no clear secular trends.24

How do these trends in advertising compare to voters’ perceptions? The Annenberg surveys included several questions that measure how votersperceived the candidates’ behavior Before answering these questions, voters were randomized into one of three conditions The first condition referred to “the candidates for governor” and did not specify either

candidate by name The second condition referred to the Republican

(Lungren or Ryan), and the third to the Democrat (Davis or Poshard) This battery of questions began “Thinking about the [candidates for governor|governor’s race], overall, would you say the [candidates for governor are|Lungren, the Republican, is|Davis, the Democrat, is]…” The question

wording of the specific measures was:

 “Responsible”: “…conducting campaigns that are very responsible, somewhat responsible, somewhat irresponsible, or very

irresponsible.”

 “Useful Information”: “…giving voters a great deal of useful

information, some, not too much, or no useful information at all?”

24 We do not present the Illinois ads that aired before September 7 so that the two graphs will be comparable Almost all of these ads were aired by Ryan and the tone of these ads was on balance negative This spate of ads stopped by mid-August and there was a two-

week lull before advertising began again after Labor Day—i.e., at the traditional “start” of

the campaign

 “Encouraging”: “…is encouraging people to vote and participate in politics, or is it turning people off from politics? Is it [encouraging people|turning people off] a great deal or somewhat?”

 “Negative”: “…conducting a campaign that is very, somewhat, not too,

or not very negative?”

 “Criticizing”: “How much time would you say the [candidates for governor are|Lungren is|Davis is] spending criticizing [their|his]

opponent? A great deal of time, some, not too much, or aren’t they doing this at all?”

In terms of voters’ perceptions of negative campaigning, these last two indicators are obviously the most relevant measures Our hypothesis is thatboth indicators will shift in response to the actual level of negativity in the campaign But we also hypothesize that the measure of “useful

information” will not shift in this fashion since respondents may believe some of this negative information is useful In addition, if negative

campaigning has the potential to “demobilize” voters, we might expect voters exposed to negative advertising to say that the candidates are not running “responsible” campaigns or “encouraging people to vote.”

To track voter perceptions, we took the daily averages of each of these indicators under the three experimental conditions and then

smoothed these averages to separate true opinion change from sampling fluctuation.25 For simplicity’s sake, in Figure 2 we present perceptions of

25 Specifically, we employed lowess procedure in Stata 8, with a bandwith of 0.3.