Tone Versus Information: Explaining the Impact of Negative Political Advertising

Abstract

Recent findings from the literature on negative advertising conclude that its impact on campaign knowledge, interest, and turnout is beneficial. These conclusions are usually justified theoretically by the arguments that negativity prompts close attention in a way that positivity does not and that negative advertising contains more issue information than positive advertising. However, if negative advertising contains more issue information, we cannot separate the impact of its tone from information effects. Using an experiment that presented equal issue information in parallel negative and positive ads, little or no evidence is found of a greater impact of negative than positive ads. The conclusion is that the impact of negative advertising may be due more to the issue information it contains than to the arresting power of its negative tone.

KEYWORDS:

• campaigns
• information
• negative advertising
• positive advertising
• turnout

Notes

Note. **p < .01. *p < .05. ^# p < .10.

Note. **p < .01. *p < .05. ^# p < .10.

Note. **p < .01. *p < .05. ^# p < .10.

Note. **p < .01. *p < .05. ^# p < .10.

^a Positive Bush ad differs from negative Bush ad at < .10.

^b Positive Kerry ad differs from negative Kerry ad at < .10.

Note. **p < .01. *p < .05.

Studies of the impact of advertising tone have also been conducted in other systems like Britain's (e.g., Sanders and Norris 2003; Norris et al. 1999), where there is some evidence of growing negativity in political advertising (Hodess, Tedesco, and Kaid 2000; Pipkin 2001). Indeed, Norris et al. (1999) draw on some of the same arguments about asymmetric attention and information effects. The research described here focuses exclusively on the United States, although the findings about the asymmetric attention effect may also have implications for political advertising elsewhere.

“A Good Dirty Fight.” New York Times, November 4, 2004.

For references to research before Ansolabehere and Iyengar, see Lau et al. (1999). For a critique of much of that research for flawed research designs see Ansolabehere and Iyengar (1995).

Previous research is not always clear about whether “information” pertains to issues or to the candidates’ characters, as in this example. Given that analysis such as Geer's (2006) shows that negative and positive advertising differ in their attention to issue considerations but not in the information they provide about the personal attributes of candidates, it is assumed here that “information effects” are the result of issue information.

Sides et al. (2010) find no difference in peoples’ perceptions of the quality of information—knowing about policies and the candidates’ plans—from exposure to more or less positive and negative advertising. However, they examine perceptions of information rather than what people actually know; it may still be that people actually learn more from negative advertising. Indeed, Koch (2008) finds just such a distinction between perception and reality: exposure to negative advertising enhances confidence in one's knowledge but actually leads to greater perceptual error.

Some prefer the term attack advertising in order to distinguish negativity that provides useful and legitimate information from negativity that is illegitimate and unfair. However, Stevens et al. (2008) demonstrate that voters do not make such distinctions with any consistency (see also Sigelman and Kugler 2003). This article therefore continues to refer most often to negative advertising.

Eighteen percent of the sample claimed to have seen the broadcast before. When their answers were compared to the answers of those who claimed not to have seen the broadcast there were no differences. It is also worth noting that the proportion of respondents saying they had already seen the newscast was lower than the proportion claiming to have seen the political ad before. Given that respondents could not possibly have seen the ad before, these memories should be treated with a grain of salt.

The first presidential debate took place toward the end of the 8 days. A total of 93 subjects took part in the study after the first debate, adding a question about whether they had watched it—roughly three-fifths of these subjects said that they had. When their answers were compared to those that had not no significant differences were found. Moreover, they were randomly assigned to the experimental groups (see below).

Any effects of different group composition or group sizes were assessed in two ways: (1) by adding group size, operationalized in various ways—as the number of subjects in the group, as “small” and “large” groups, and as a 4-point scale—as a variable to all model estimates in Tables 1 through 3 as well as examining whether group size affected subjects’ likelihood to accurately recall stories in the local news broadcast and (2) by adjusting the standard errors of estimates to incorporate the possibility of non-independence within groups (using the “cluster” command in Stata 11.0). Neither test indicated that there was any influence of group composition or size in the study.

This should allay one of the typical concerns about student samples: that they are disengaged from politics. The proportion was almost the same as in the random adult national sample in the American National Election Study of 2004. There were more Independents, however.

The survey did ask respondents to list the ads they remembered seeing, an open-ended question, as well as whether they specifically remembered seeing a political ad, and also included a thought listing task. By these measures of memory and cognitive activity, the negative ads were no more memorable than the positive.

Differences in affect toward the candidates may be influenced by the issue and trait information offered in negative ads—as individuals learn about the policy positions of candidates they may feel sharper distinctions between them, for example—but also by straightforward persuasion. In an earlier version of the paper, net affect toward the candidates was included as a dependent variable. The model estimates showed no impact of negative or positive advertising on net affect, neither for the whole sample nor among partisans and Independents considered separately. To the extent that net affect toward the candidates is the result of persuasion this is confirmation that the experimental stimuli were equally persuasive regardless of the small variations in language and imagery (see above).

The term generally is used because Ansolabehere and Iyengar (1995) argued otherwise. However, the indicators that they claimed to capture internal efficacy are more commonly viewed as indicators of external efficacy.

In principle, the ads should also have a typical information effect but it is not clear what that would be. Online processing makes it difficult to know the meaning of no gains in knowledge from an ad, for example. However, Lau et al.'s meta analysis (2007) shows minor gains to be what research in the field finds overall. The impact of exposure to the positive or negative stimuli on the number of likes and dislikes of the candidates and the likelihood to place the candidates correctly on a liberal-conservative ideological scale was examined. In neither case did the ads seem to make a difference, nor was there a difference between the impact of the positive and negative stimuli.

Controlling for individual-level variation in partisan intensity, political knowledge, internal efficacy, and gender does not affect these results. Also, their effects on emotions are generally statistically insignificant, although stronger partisans appear more likely to experience the positive emotions of enthusiasm, pride, and hope.

While the different emotions in Table 1 are grouped under the general headings of “negative” and “positive,” the intention is not to claim that the emotions within these categories are indistinct. In particular, Marcus’ (2002) research, as well as more recent work by Valentino et al. (2006) and Brader and Corrigan (2007), suggests a difference between anxiety and anger. A rotated factor analysis of the seven emotions shows one factor for the emotions of anger, hatred, and disgust; another for enthusiasm, pride, and hope; and a weaker third factor on which anxiety loads the strongest. These factors look very much like the emotions of anger, enthusiasm, and anxiety these authors identify. However, the rationale for looking at emotions here is not to examine their potentially distinct effects but to demonstrate that the ads had the kind of impact associated with the asymmetric attention effect.

Of the control variables, the most consistent influences were strength of partisanship and internal efficacy. Strength of partisanship was associated with greater disagreement that the choices presented by the candidates were limited, greater interest in the campaign, and a higher probability of voting, while internal efficacy was also associated with interest in the campaign and turnout and, in addition, with an enhanced perception of external efficacy. Political knowledge and gender had little influence: both enhanced vote intention, while gender lowered perceptions of government responsiveness to elections.

Another common subgroup on which to focus is the politically unsophisticated, the assumption being that those who know less about the political world will be more affected by advertising. The models in Table 2 were reestimated with an additional interaction between knowledge and tone to capture this reasoning. However, the results indicated it was the politically knowledgeable who were more responsive to the ads. More important, this analysis did not reveal any new differences in the impact of the positive and negative ads.

Stevens et al. (2008) show that partisans are more responsive to the messages of political advertising than Independents. However, this is due to variation in the perception of the fairness of different messages. In the experiment described in this paper, subjects were exposed to a single message. It did not measure perceptions of the fairness of the experimental ads, but the results are consistent with ads that divided partisans in terms of their fairness.

Models were also estimated that replaced the advertising variables with emotional indicators of anxiety, enthusiasm, and anger to see whether there were stronger effects of the negative emotions evoked by negative advertising. This was not found to be the case.

Additional information

Notes on contributors

Daniel Stevens

Daniel Stevens is an associate professor in politics at the University of Exeter. He is interested in political communication and political psychology, especially as they relate to political advertising and the media in the United States and Britain. He has published articles on these themes in journals such as the American Journal of Political Science, Journal of Politics, and British Journal of Political Science.