Last week Gerlach's campaign said it stands by its ads that try to link one of Murphy's key supporters to the Taliban, rape, kidnapping and terrorist attacks. Murphy has repeatedly called on Gerlach to pull the ads, saying they are ‘pure fiction’ and ‘insulting to voters’ … A Gerlach spokesman, John Gentzel, said that in this hard-fought campaign, the ads are valid … ‘We think it is fair’ (Philadelphia Inquirer, 2 November 2004).
A new advertisement by Rick Perry takes a sharper tone than his previous ads, taking aim at President Obama's comment last weekend that American businesses had been ‘a little bit lazy’ in taking for granted that new investment would want to come to the United States … But the ad takes Mr. Obama's remark out of context (New York Times, 17 November 2011).
Election campaigns such as those in the United States are full of the kinds of negative claims described in these recent contests for various offices; perhaps increasingly so.Footnote 1 As political marketing techniques have come to characterize politics in other electoral systems too,Footnote 2 questions about the effects of negativity in political exchange are more compelling than ever. These newspaper accounts also suggest that some attacks go too far: they are unfair in some way. ‘Fairness’, it appears, constitutes one criterion that voters may use to evaluate campaign ad claims. Indeed, although agreement about the impact of negativity remains elusive,Footnote 3 political scientists appear to have settled on the idea that individuals hold principles of ‘fairness’ and what constitutes ‘civility’, as well as criteria such as ‘relevance’ in making judgments about political advertising claims. The consensus viewpoint goes further to claim that only when these principles are violated do we see distinct effects of negativity.Footnote 4
Our research questions the validity of this latter claim. We agree that individuals have principles regarding what is a fair negative claim, and we demonstrate that they are enduring standards, but we also show that voters apply such principles to actual ads in a real election campaign inconsistently. How do principles such as fairness and civility – or other standards of evaluation, such as the relevance of a campaign claim – figure into citizens’ definitions of acceptable election campaign rhetoric? We assess the influence of one criterion of evaluation – fairness – on the distinctions citizens make between acceptable campaign criticism and unacceptable ‘negative attacks’ in three US presidential campaigns. We discover that a principle of fairness matters: individuals who think a criticism is unfair in principle are generally more likely to regard a specific instance of such criticism as also unfair. Yet we also find that voters apply their principle of fairness inconsistently when they evaluate the actual ad claims made by real candidates for elective office. These findings lead us to conclude that: (1) previous research on fairness, civility or other standards of evaluation overestimates the role played by the content (compared to the source) of an attack and (2) the notion of a public inured to the harshness of modern campaigns is a simplification of reality that inaccurately characterizes individuals’ responses to negative critiques.
Our findings suggest that the presence of partisan cues can alter how negative advertising information is processed and used to evaluate candidates, thus affecting decisions about whether (and for whom) to vote.Footnote 5 These results comport with studies that have acknowledged the importance of the congruence between an ad's sponsor and the partisanship of a viewer. Our work suggests, however, that a continuing focus on content over source will overlook an important factor in voter decision making.Footnote 6 Whether partisan cues only influence attitudes toward information (for example, beliefs about its veracity, fairness or relevance) or indeed change how information is processed (for example, by varying the modality of cognition), empirical studies of assessments of fairness that omit information about the sponsoring party will sufficiently alter the context for judgment from that of an actual electoral contest so as to produce significantly biased conclusions.
To be clear, we are not denying that other studies have acknowledged the importance of the congruence between an ad's sponsor and the partisanship of a viewer;Footnote 7 their foci are different to ours. We are saying that many studies continue to focus on content over source, however, for example by conducting experiments in which partisan cues are absent.
Principles, Motivated Reasoning and Perceptions of Fairness
Political practitioners and academics have been attracted to the notion that perception of the fairness or acceptability of a criticism is a major source of variation in the influence of negative political messages. Ron FaucheuxFootnote 8 insists on the acceptability of ‘going negative’ – until it crosses the line and becomes unfair, deceptive or hateful; Larry BartelsFootnote 9 contends that the ‘largely misplaced’ research and media focus on so-called negativity fails to distinguish the merely contentious from the ‘nasty, inaccurate or unfair’ in advertising. Kim Fridkin Kahn and Patrick Kenney claim that ‘people distinguish between legitimate and tempered criticisms, on the one hand, and acrimonious and unjust criticisms, on the other hand’.Footnote 10 Researchers not only contend that citizens make nuanced judgments about political advertising; they also note that they often learn from – and are stimulated by – negative information.Footnote 11 Accordingly, when negative information fails to engage voters, it is because the attack goes too far and is perceived to be unfair or unacceptable, reducing the likelihood that offended voters will go to the polls.Footnote 12
Recent studies of the impact of modern campaign rhetoric claim that citizens hold principles – including fairnessFootnote 13 and civilityFootnote 14 – that inform their judgments about various types of criticisms that are made about candidates for elective office. The association between fairness (as an indicator of an advertising claim's acceptability) with political participation has been demonstrated in the context of a real campaign.Footnote 15 These principles, as well as other standards of evaluation such as the relevance of a given campaign claim,Footnote 16 can inform citizens’ judgments about which claims are acceptable or useful in evaluating the candidates. These studies have also helped distinguish between what citizens say about various types of claims and how claims they deem unacceptable may, nevertheless, influence their candidate evaluations. This distinction suggests that citizens may hold standards of evaluation, but that their principles also permit violations.
We echo the previous findings that individuals hold notions of what constitutes fair criticism in political advertising.Footnote 17 Like other researchers, we find that respondents in our studies were more likely to tolerate issue-based criticisms than character-based criticisms, and that partisans tend to be more generous than non-partisans in their evaluations of what is fair game in a campaign. Our findings also show substantial agreement among respondents on the types of claims that, in the abstract, fall either within or outside the bounds of acceptability. This body of work indicates that it is appropriate to refer to norms such as fairness and civility as ‘principles’, and that important generalizations may be made about the articulated bounds of propriety in campaign rhetoric for American political culture.
Despite the significant advances of previous research on two of the most important questions asked about campaign discourse – (1) what constitutes ‘negative’, unfair, and uncivil rhetoric and (2) how such communication might influence electoral behavior – in many cases researchers continue to draw conclusions about ad effects from research conducted in a contextual vacuum, in which a candidate from one unidentifiable party attacks a candidate from another unidentifiable party. This research portrays a public that has been habituated to rough campaign rhetoric and is able to maintain principles of fairness and civility, resisting the effects of a negative campaign. Yet we actually know nothing about the ways that principles of acceptable rhetoric operate in the context of real campaign exchanges (or indeed, whether these principles matter at all). Nor do we know whether information about the source of a claim is relevant to the standard of evaluation that is used.
Suppose an individual believes that an advertisement criticizing a candidate's military record is, in principle, unacceptable. In a real campaign, this individual may be a partisan identifier and the candidate who is being criticized for his military record may be Democratic presidential candidate, John Kerry or Barack Obama, or the Republican incumbent (in 2004) George W. Bush. What happens then? We consider four competing models.
First, her judgment that criticizing a candidate's military record is beyond the pale may lead her to perceive such attacks on either candidate as unfair. We know that individuals have notions of what is fair in public discourseFootnote 18 and political advertising, and that adherence to norms influences perceptions of what constitutes fair treatment in other realms such as legal conflict.Footnote 19 Moreover, there is a great deal of evidence, both old and new, that individuals try to maintain cognitive consistency in their attitudes and behaviors.Footnote 20 For these reasons, we might expect norms of fairness to be consistently applied, perhaps not by everyone but by many citizens.
We call this a ‘No Partisan Bias’ or Principles = Practice model. It holds that the distinctions that individuals make between fair and unfair criticisms in principle (that is, in the absence of partisan labels) are applied evenly in practice: if the attack is unfair in principle it will also be unfair in practice, and if it is fair in principle it is also fair grounds for critique in a real campaign.Footnote 21 Returning to our hypothetical, consider an individual who is a Republican identifier viewing, in one scenario, criticism of John Kerry's military service record and, in another scenario, criticism of George W. Bush's military service record. She may be psychologically motivated to maintain a different kind of consistency, arriving at the conclusion that while the attack on George W. Bush's military record is objectionable, the attack on John Kerry's military record is in fact acceptable. This possible response reflects a type of consistency driven by a process of motivated reasoning in service of an evaluation that sits well with her partisan predispositions. Studies from political science and social psychology show that, rather than carefully weighing information to arrive at a judgment, individuals often adopt strategies of processing information that protect their predispositions.Footnote 22 Motivated reasoning suggests that individuals encountering criticisms made in negative political advertising will draw on the cue of partisanship, either to infer the fairness of the criticisms (characterized as fair if they criticize the opponent and unfair if they criticize the preferred candidate) or to ignore the content of the criticisms and rely only on partisan cues. For example, partisans receiving incongruent information may reason that the criticism is irrelevant to the election or that the source is too unreliable to be taken seriously.Footnote 23
Nevertheless, motivated reasoning has its limit. Individuals are ultimately ‘constrained by their ability to construct seemingly reasonable justifications for these conclusions’.Footnote 24 This possibility leads us to suggest three models of partisan bias. One potential constraint on motivated reasoning is that ‘norm violations should generate a negative reaction, much as they do when people are exposed to such violations in real life’.Footnote 25 This limitation suggests that where norms of fairness are violated (that is, individuals are exposed to criticism that they regard as unfair in principle) partisan motivated reasoning may be inhibited. Faced with a topic of criticism that they unequivocally consider unfair, these individuals may give no consideration to whether they share the partisanship of the sponsor or target; the subject matter is simply off limits for them. If there is no violation of such a normative constraint, however (that is, the criticism is seen as fair in principle), partisans for whom the criticism is incongruent are not so constrained: their affinity to the shared partisanship of either the sponsor or the target is an important consideration. They can rationalize that in this instance (an attack on their preferred candidate) the criticism is unfair, which apparently violates their original principled assessment that this topic of criticism is fair game. We term this, the mildest version of partisan bias, the Weak Partisan Bias model. It is similar to the Principles = Practice model in one respect: individuals who view a topic as fair in principle will, overall, be more likely than individuals who consider a topic as unfair in principle to judge actual criticism on that subject as fair in practice. However, when a criticism is judged as fair in principle, partisans for whom an actual criticism is congruent with their predispositions (for example, Republicans exposed to a criticism of Barack Obama) will be more likely to judge it as fair than partisans for whom an actual criticism is incongruent with their predispositions (for example, Democrats exposed to a criticism of Barack Obama).
But perhaps motivated reasoning is not so constrained by norm violations. Indeed, Redlawsk et al.Footnote 26 show that motivated reasoning is resistant to large amounts of seemingly inconsistent information, suggesting a stronger kind of partisan bias. For individuals behaving according to a Strong Partisan Bias model, those who view a topic as fair in principle will again, overall, be more likely to view actual criticism on the topic as fair in practice than those who view a topic as unfair in principle. But beyond that, partisanship moderates the application of principle for the categories of unfair and fair criticism alike. Although a topic is judged off limits as a matter of principle, individuals exhibiting a strong partisan bias will be less likely to judge a criticism of their preferred candidate as fair than will be individuals for whom the attack is congruent with their partisanship. The Strong Partisan Bias model postulates that an individual who is confronted with an attack on the opposition may reason that, although the subject matter of the attack is usually out of bounds, in this case, applied to this (opposing) candidate, the topic of attack is legitimate, and the attack is fair: biased processing is likely to overrule the strict application of their principles, regardless of whether those principles say that the topic is unfair or fair grounds for criticism.
A variation on such partisan bias is the Only Partisan Bias model, which is similar to the Strong Partisan Bias model in that partisanship influences all individuals’ judgments of the fairness of criticisms – both those who say a criticism is fair in principle and those who say a criticism is unfair in principle. But with the Only Partisan Bias model, principles of fairness themselves have no influence on judgments of the fairness of actual criticisms. Thus in the Only Partisan Bias model, knowing that a respondent viewed a topic of criticism as unfair or fair in principle is not predictive of their judgment of actual criticism on that topic as either unfair or fair.
Research Design and Analysis
Testing the four models requires that we ascertain individuals’ evaluations of the fairness of criticisms in principle and in practice. We conducted three random sample national telephone surveys that accomplished this aim in 2002, 2004 and 2008, completing more than 2,000 interviews (for details of the surveys, see Appendix I). Thus our findings have the virtue of not being tied to a particular election context, such as an incumbent facing re-election, or particular candidates. The first survey took place from March to June 2002, the second in October 2004 and the third in October 2008. In 2002 we interviewed 705 American citizens who were eligible to vote; in 2004 we surveyed 510 citizens eligible to vote and re-interviewed 210 of the 705 respondents from the 2002 sample; in 2008 we interviewed 520 citizens eligible to vote and re-interviewed 261 respondents from the earlier samples. (See Table A-2 of online Appendix III for sample characteristics and comparisons to census and American National Election Study data). We pooled the new respondents and re-interviews in 2004 and 2008. Analyzing only the new respondents makes no difference to our substantive conclusions about each election.
Our surveys first replicated the questions used by Freedman and Lawton:Footnote 27 whether it is fair or unfair for candidates to criticize an opponent in campaign ads on various issue positions, matters of character and other candidate constructs. In the next stage of questioning we injected partisanship into candidates’ criticisms by asking about the fairness of some specific examples of these criticisms of the opponent made by the presidential candidates and their supporters during the 2000, 2004 and 2008 elections. We randomized the order in which we asked these questions. We did not inform respondents that they were specific examples of the same charges whose fairness they had previously been asked to assess in principle; to the extent that respondents recognized that they were, it is likely that they would try to keep their answers consistent, biasing results toward the Principles = Practice model. Table A-3 of online Appendix III provides a summary of the abstract and specific attacks presented to respondents across the surveys: thirty-one specific criticisms on nine subjects whose fairness they had first been asked to evaluate in principle (Appendix IV presents the exact wording). We thus obtained judgments of fairness in principle and in practice, which enable us to evaluate the validity of the four models.
Our approach to analyzing the data is to estimate models that predict perceptions of fairness for each of the thirty-one specific criticisms covered in the three surveys, where the perception of attacks on George W. Bush (in 2000 and 2004), John McCain, Al Gore, John Kerry and Barack Obama is the dependent variable, coded as ‘fair’ or ‘unfair’.Footnote 28 The key predictors that test the different models are, with variable labels in parentheses:
-
(1) Perceptions of the fairness of the criticism in principle (Fairness of the Criticism in Principle, a dichotomous 0–1 variable)
-
(2) Whether the partisanship of the respondent is congruent or incongruent with the criticism (Partisanship Congruent with Criticism, coded 1 if a criticism is congruent and 0 if it is not: a criticism is congruent if a respondent is a Democrat and the target of the criticism is George W. Bush or John McCain, or if the respondent is a Republican and the target of the criticism is Al Gore, John Kerry or Barack Obama)
-
(3) The interaction between fairness of the criticism in principle and congruency of partisanship (Fairness in Principle × Partisanship Congruent with Criticism)
We also include variables for Independents (Independent) and for the interaction between Independents and perceptions of the fairness of a criticism in principle (Fairness in Principle × Independent).
Fairness in Principle is the effect of fairness principles on partisans for whom a criticism is incongruent (that is, a Democrat when the target of the criticism is Al Gore, John Kerry or Barack Obama, or a Republican when the target of the criticism is George W. Bush or John McCain). Thus the interaction term Fairness in Principle × Partisanship Congruent with Criticism shows how sharing the partisanship of a criticism's sponsor moderates the influence of principles in evaluating ad claims by affecting the judgments of those who identify a criticism as fair in the abstract. (Fairness in Principle × Independent shows the moderating influence of identifying as an Independent). Including interaction variables means that the ‘main effects’ of the variables comprising the interactions (the constituent terms) capture the effects of a variable when the other variable in the interaction equals zero.Footnote 29
Results
Rather than present the results from all thirty-one specific criticisms, we display illustrative findings from each survey for two types of criticisms: that the candidate either 1) talks one way but votes another or 2) accepts campaign contributions from special interests. Focusing on two examples of criticisms permits us to present a detailed analysis of the effects of our three key predictors on perceptions of the fairness of actual ad claims and the implications of this analysis for the role of principles in fairness judgments. We display these detailed findings in Tables 1 and 2 and in Figure 1, and summarize the analysis of the remaining nineteen criticisms in Figure 2.
Table 1 Influences on Perceptions of the Fairness of Criticizing Candidates for Talking One Way and Voting Another (Logit Models)
Note: standard errors in parentheses. **p < 0.01, *p < 0.05, #p < 0.10 (one-tailed test).
Table 2 Influences on Perceptions of the Fairness of Criticizing Candidates for Campaign Contributions (Logit Models)
Note: standard errors in parentheses. Ns are smaller in 2008 because the sample was divided into experimental groups (the subject of another paper). The 2008 results are from the control groups. **p < 0.01, *p < 0.05, #p < 0.10 (one-tailed test).
Fig. 1 Simulation results for Bush 2000 and Bush 2004 (Table 1 equations)
Fig. 2 Summary of biases in perceptions of fairness for the remaining nineteen criticisms
Table 1 presents estimates for criticisms of George W. Bush and Al Gore from the 2000 election, George W. Bush and John Kerry from the 2004 election, and John McCain and Barack Obama in 2008 that were related to ‘talking one way but acting or voting another’. These critiques covered charges such as George W. Bush claiming credit in 2000 for a law he had opposed as governor of Texas, John Kerry touting fiscal responsibility while having voted for the largest tax increase in American history and John McCain opposing and then supporting Bush administration tax cuts.
The key variables appear in the first three rows of Tables 1 and 2. We indicate whether the coefficients for variables are statistically significantly different from zero at levels of up to p < 0.10 in one-tailed tests. Our estimates control for other individual-level variables that the literature indicates are important: trust in the federal government as an indicator of views of government, political expertise and generation. Studies of trust in government suggest two plausible effects: (1) citizens who view government positively may also perceive the kinds of charges that are the common currency of negative political advertising to be among the most unfairFootnote 30 or (2) to the extent that trust in the federal government is incumbent based,Footnote 31 its influence on perceptions of fairness may depend on the target of the attack. Political expertise, which we measured by political knowledge and frequency of newspaper readership, may foster a greater tolerance of negative criticisms.Footnote 32 Generation, measured by age cohorts, may show signs of ‘life-cycle effects’ in which, along with electoral interest and participation, perceptions of the fairness of negative exchanges increase as a cohort ages; or perhaps older cohorts may have ‘seen it all’ but still reject caustic public discourse and a younger generation that has perhaps become habituated to the vitriolic public criticism that is modeled in the media to which they pay attention. All variables were coded from 0 to 1 (see online Appendix II).Footnote 33
Beginning with Table 1, it is immediately apparent from all six sets of estimates that principles of fairness influence judgments of specific criticisms, which are represented by the positive and statistically significant coefficients in five of the models for Fairness of the Criticism in Principle.
Partisanship also moderates the influence of principles of fairness – shown by the impact of the variables Partisanship Congruent with Criticism and Fairness in Principle × Partisanship Congruent with Criticism. But which of the partisan models is most strongly supported?
Testing across all four models simultaneously using classical hypothesis testing is not straightforward. We cannot simply rely on the statistical significance of individual coefficients, or on joint tests on the parameters of Fairness in Principle (F) and Fairness in Principle × Partisanship Congruent with Criticism (F × P) because, while each of our models has expectations for the coefficients on F, P and F × P, we have two non-standard complications. First, we are not pitting one model against another. Rather, we want to sort through four competing models. Secondly, tests of the applicability of our models do not simply require joint parameter tests, but joint one-sided parameter tests. Testing whether the parameters for Fairness in Principle = 0 and Partisanship Congruent with Criticism = 0 is straightforward, as is testing whether Fairness in Principle > 0. But testing whether the parameters on Fairness in Principle > 0 and on Partisanship Congruent with Criticism > 0 is not straightforward.Footnote 34
To demonstrate the implications of the models, given these complications, we adopted the following approach: for each equation, we simulate 100,000 draws from our estimated parameters (using ClarifyFootnote 35 ). Figure 1 provides an example, representing two equations from Table 1. For each equation, the signs of the three parameters on F, P and F × P can each be negative or positive, meaning that their possible combinations (2 × 2 × 2) yield eight regions. The panels for each row in Figure 1 are defined by the sign of the interaction term; the left panel has a negative interaction term, while the right panel has a positive interaction term. In each panel, the x-axis reflects the range of the fairness parameter, whereas the y-axis reflects the range of the partisan congruence parameter. The quadrants are labeled with the model that is consistent with the configuration of signs of the parameters in that region. For example, if the coefficients on F and P are both negative, such results are not consistent with any of our four models. If the sign on F is positive and the sign on P is negative, then if the sign on F × P is negative, that is consistent with Principles = Practice, whereas if the sign on F × P is positive, that is consistent with Weak Partisanship. In this sense then, discriminating between Principles = Practice and Weak Partisanship depends on the sign of the interaction term. Simply testing the sign of the interaction term does not work for our purposes, however, because for either model to hold, the partisan congruence parameter needs to be negative and the fairness parameter needs to be positive. If those last two conditions do not also hold, the set of parameters will lie in a quadrant consistent with a different model (or no model).
From the 100,000 draws we calculate the percent of time the simulated parameters fall into each of the eight regions, then map those percentages into our four models.Footnote 36 Figure 1 compares two sets of simulated parameter estimates from Table 1, from Bush 2000 (top row) and Bush 2004 equations (lower row).Footnote 37 In the Bush 2000 results, the parameter estimates are consistent with the Strong Partisan Bias model 97 per cent of the time – in the top row of Figure 1, nearly all of the points are in the upper right quadrant. A handful of points lies in the lower right quadrant, consistent with the Weak Partisan Bias model. The Bush 2004 simulations, by contrast, display a different pattern. A majority of points (60 per cent) falls in the upper left quadrant, which is consistent with the Only Partisan Bias model. A substantial portion of the simulated parameters lies in the upper right quadrant, however, so the discrimination across models is not as strong in the Bush 2004 case as in the Bush 2000 case. We indicate with which model the simulations indicate consistency in the bottom row of Table 1.
The simulations show that the results are most consistent with Strong Partisan Bias in most cases, with the Only Partisan Bias model mildly dominating in one example, for judgments of the criticism of George W. Bush in 2004. In the other five examples, the simulations fall in the quadrant indicating Strong Partisan Bias well over 90 per cent of the time. The Weak Partisan Bias model holds that if a criticism is seen as fair in principle, real examples of such criticism are more likely to be judged as fair when you share the partisanship of the messenger, while partisanship should not moderate the influence of fairness principles when a criticism is viewed as unfair in principle. The results in Table 1 and Figure 1 are clearly inconsistent with this model.
Instead, the simulations make it clear that even individuals who judged a subject to be off-limits in principle were affected by partisanship when it came to judging the fairness of real criticisms offered by real candidates. Compared to partisans for whom a criticism was incongruent with their dispositions, congruent partisans were considerably more likely to judge criticisms of George W. Bush, Al Gore, John Kerry, John McCain and Barack Obama as fair, even if they had deemed such criticisms to be unfair in principle. Partisans judged criticisms evaluated as fair in principle to be unfair when applied to their candidate, and similarly saw criticisms evaluated as unfair in principle to be acceptable when applied to the opposition. In four instances (George W. Bush in 2000 and 2004, John Kerry in 2004 and Barack Obama in 2008) the interaction terms show that the gap widened further between partisans for whom the criticism was congruent and incongruent when a criticism was judged as fair in principle; in the other two cases (Al Gore in 2000 and John McCain in 2008) the partisan gap was no wider when a criticism was judged as fair in principle than when it was judged as unfair in principle.
The meaning of the estimates in Table 1 can be further illustrated with some point estimates. We set the control variables at their means or modes, while allowing values for perceptions of the fairness of the attacks in principle to be set as fair or unfair, and partisanship as congruent with the attack or incongruent with the attack. Examining the estimates shows that although principles mattered in most of the examples, partisan differences were great regardless of principles. Among the respondents who viewed criticism of a candidate for talking one way and voting another as unfair in principle, there was a generally lower probability of judging the specific criticism as fair, but partisan gaps are nevertheless evident. For all candidates, the probability of an individual judging a criticism as fair was at least ten points (and as much as sixty points) greater among respondents for whom the criticism was congruent than for whom it was incongruent with their partisan affiliations. This finding demonstrates that partisan bias has the capacity to move large numbers of respondents to judge a criticism as fair rather than unfair and vice versa. The estimates also illustrate how partisan perceptions are contingent upon who is being attacked. Republicans were less likely than Democrats to view attacking Bush or McCain as fair in each election, but the relationship is reversed when the target of the criticism is Al Gore, John Kerry or Barack Obama; in these latter cases, Republicans are far more likely than Democrats to judge criticizing the candidate for ‘talking one way and voting another’ as fair. Moreover, the predicted probabilities when partisanship is congruent with the charge tend to be above the 0.5 threshold, which demonstrates a tendency to evaluate these criticisms as fair regardless of principles, whereas incongruent partisans fall on or below the 0.5 threshold, indicating a tendency to regard the criticisms as unfair.Footnote 38
To demonstrate that these patterns are not confined to criticisms of candidates’ voting records, Table 2 presents analysis of a different criticism in the three elections: taking campaign contributions from special interests such as ‘big oil’ or trial lawyers. Again, these results show that congruent partisans are consistently more likely to judge a criticism against the opposing party's candidate as fair, even among respondents who had described the topic of criticism as unfair in principle. In all six examples the simulations, the results of which are presented at the bottom of the table, show that the Strong Partisan Bias model dominates. Contrary to the Only Partisan Bias model, fairness principles influence judgments of the fairness of the actual criticisms; contrary to the Weak or No Partisan Bias models, congruency of the criticism with partisanship results in consistently more generous judgments of fairness.Footnote 39
Point estimates from this analysis, following the same procedure described above, show similar patterns of highly malleable application of principles. When a criticism is viewed as fair in principle it is also more likely to be judged fair in practice, but there are partisan differences both among respondents who said criticizing a candidate for taking campaign contributions from special interests is unfair and among respondents who said it is fair. The partisan gaps are also comparable to the estimates from Table 1 – about 10–50 points. Moreover, when we examined respondents’ underlying probabilities of perceiving criticism of campaign contributions as fair, there was ambivalence about this kind of attack – large numbers of respondents were close to the 0.5 probability threshold. Thus once again the gaps are sufficient to change perceptions of the acceptability of an attack in large numbers.Footnote 40
Up to this point we have focused on a subset of results – criticisms of a candidate's record or the special interests to whom he is alleged to be beholden for money – in which the public tends to view the criticism as fair in principle (see Appendix Table A-4). Figure 2 shows the results from all the other criticisms. Of particular interest here are more personal criticisms, such as of the behavior of family members, which most members of the public view as off-limits. The analysis in Figure 2 shows two cases in point. The criticism is related to the behavior of family members in 2000. Very few respondents considered this subject to be a fair topic of attack in the abstract; only 3 per cent thought it a legitimate subject for a critical ad, rising to about 10 per cent for specific examples of misbehavior by Bush and Gore family members. Figure 2 shows that even in this case, the Strong Partisan Bias model is dominant for judgments of the criticism of Bush, while the Strong and Weak Partisan Bias models are about equally applicable to the criticism of Gore.Footnote 41 Thus even in such extreme cases, in which attacks violate nearly universally held norms, motivated partisans are prone to allow their predispositions to override their sense of the legitimacy of criticisms that they find so clearly objectionable in principle. This is less true of the Gore example; there is some indication that individuals who judge such a criticism unfair in principle also judge it to be unfair in practice, regardless of partisanship, but the Strong Partisan Bias model is still most applicable in half the simulations (full model results are in online Appendix III Table A-5).
The overall message of the analysis in the tables and figures is clear: partisan bias moderates principles in the vast majority of cases; the main variation relates to whether fairness principles have any discernible influence on judgments of real campaign criticisms. In most cases they do; the results are most often consistent with the Strong Partisan Bias model, in which principles of fairness affect judgments of fairness in practice, but partisanship colors those judgments regardless of whether the topic of criticism is deemed fair or unfair in principle. For eighteen of the thirty-one criticisms, more than 90 per cent of the simulations are consistent with Strong Partisan Bias. Of the remaining thirteen criticisms, Strong Partisan Bias dominates 50–89 per cent of the time for nine of them. Only Partisan Bias is the dominant model for the other four, although in only two of these do over 90 per cent of simulations fall in the Only Partisan Bias region.
On closer examination, the pattern in these four examples – criticisms of George W. Bush in 2004 for talking one way and voting another and of his voting record, of John Kerry via his vice presidential nominee and of Barack Obama's voting record – is that partisans who rated the topic of criticism as fair in principle but for whom the criticism was incongruent were somewhat more likely to see the actual criticism as unfair than incongruent partisans who rated the topic as unfair in principle. Among partisans for whom the criticism was congruent with their predispositions, on the other hand, we see the same patterns as in the Strong Partisan Bias model – clear partisan bias for those who judged a criticism to be fair in principle; they were also more likely to view it as fair in practice. Thus where we see results consistent with the Only Partisan Bias model, it is true to say, as we did in our description of this model, that fairness principles exert no discernible influence in our estimates. However, what emerges is more nuanced: there is an influence of fairness principles, but it is confined to congruent partisans in these examples. It is unclear why we see this pattern in these four examples and no others, but it may be that partisan supporters of a candidate are so convinced that these particular charges are untrue that fairness principles play no role in their judgments.Footnote 42
The range of cases in which the type of criticism apparently invites partisan bias is considerable. Are there aspects of the claims or presentation of information that either encourage or discourage motivated processing and partisan bias? One important element of motivated processing theories concerns uncertainty about stimuli. Eileen Braman and Thomas Nelson find that ‘predispositions leave their clearest mark on ambiguous stimuli’.Footnote 43 Uncertainty about the factual basis for claims may be more common than certainty for most individuals in attacks that contain claim and counterclaim. If we are right – that motivated reasoning is behind the patterns of judgments we observe – we would expect that manipulating their certainty, for example by presenting individuals with additional context to a criticism that suggests it is untrue, would moderate partisan bias.
In our 2004 survey we conducted an experiment in which half of the sample was asked to judge the fairness of an attack on John Kerry's military service in Vietnam (control group) – a ‘Swift Boat Veterans for Truth’-style charge that Kerry did not deserve his three Purple Hearts – while the other half of the sample was presented with the same attack and given additional information from an ‘independent organization’ that questioned the factual accuracy of the attack (treatment group).Footnote 44 To gauge the impact of this manipulation, we estimated a model identical to those in Tables 1 and 2, with the addition of a dummy variable for the half sample condition; a three-way interaction between the condition, the fairness of the criticism in principle and partisanship; and all the combinations of two-way interactions (see Appendix III Table A-6 for the full results). Figure 3 presents point estimates to illustrate our findings.
Fig. 3 Point estimates of perceptions of fairness attacking John Kerry's military record: Swift Boat experiment
Figure 3a shows estimated judgments of individuals in the control group, and Figure 3b shows the treatment group, again computed separately for respondents judging such a criticism as fair game or unfair game. The patterns are consistent with the Strong Partisan Bias model, showing that principles matter – individuals who think attacking a candidate's military record is fair game are also more likely to see the Swift Boat Veterans’ attack as fair – but they are applied differently depending on their congruence with partisanship. Thus Figure 3a shows that Republicans in the control group who say such an attack is unfair in principle have a greater probability of judging the specific criticism as fair than do Democrats who express the same principle (and these differences are statistically significant). Republicans in the control group who say such an attack is fair in principle also have a greater probability of perceiving the specific criticism as fair than Democrats who think such criticisms are fair in principle. For Democrats, the predicted probabilities fall below the 0.5 threshold, indicating objection to these criticisms, while the predicted probabilities for Republicans are on or above the 0.5 threshold. The differences in perceptions cover 40–50 points of the probability scale, representing sizable gaps in perceptions.
Figure 3b shows the point estimates of individuals presented with exculpatory information intended to undermine the veracity of this criticism. Beyond a general effect, in which all respondents are less likely to view the criticism as fair, partisan differences are attenuated. Among respondents who reject such attacks in principle, the gap in predicted probabilities of judging the attack as fair is more than halved – and the attenuation of the influence of congruent partisanship is statistically significant. There is also a much greater propensity, regardless of partisanship, to accept the exculpatory information and perceive the Swift Boat attacks as unfair. Although the results are less dramatic for respondents who said such criticism was fair in principle, the findings also show a lower probability of judging a criticism as fair in these circumstances and a diminishment of partisan bias.
The simulations show that in the control condition, almost all the outcomes fall in the Strong Partisan Bias region (99.6 per cent). With the additional information the pattern changes: 64 per cent of the simulations now fall in the Strong Partisan Bias region, while the other 36 per cent are consistent with Only Partisan Bias. As Figure 3 illustrates, these results reflect that once they are exposed to exculpatory information about Kerry, Democrats do not think this negative ad is fair regardless of whether they believe such criticisms are fair or unfair in principle. Given the same information, Republicans tend to think it is a fair ad only if they believe such ads to be fair in principle, and in far lower numbers than in the absence of the additional information. This finding is consistent with the notion that partisan bias can be attenuated by providing information that is clearly contrary to the assumptions individuals need to make to reach desired conclusions.
Discussion and Conclusion
We have taken a hard look at how the principles that citizens hold about acceptable political speech affect judgments about negative ads in the heat of a campaign. We replicate the findings of past research on dimensions of acceptability such as fairness, civility and relevance by showing that citizens hold principles of the fairness of criticisms that accord with conventional wisdom. We go beyond earlier acknowledgment of a likely impact of partisanship and speculation about its form, however, using research designed to explicitly examine the manner and extent to which these principles affect perceptions of the fairness of criticisms of real candidates in real campaigns. We have tested four possible models: (1) that citizens object to norm violations and thus translate principles faithfully into their perceptions of real campaign criticisms, (2) that they engage in weaker or (3) stronger forms of partisan-motivated processing in which the impact of principles is moderated by partisanship or (4) that citizens engage in a form of partisan-motivated processing in which principles of fairness have no overall impact.
Our analysis has demonstrated that for most types of real campaign criticisms, partisan bias is evident among both partisans who judge the topic of criticism to be legitimate and those who view a given type of criticism as unacceptable in principle. Motivated processing associated with strong partisan bias appears to be the default behavior leading to evaluations of election campaign advertising. Only when the topic of an attack so clearly deviates from norms of fairness that very few will say it is ever fair in principle – criticizing a candidate for the behavior of his family members or for past personal troubles with substance abuse – do we see much evidence of weaker partisan bias, but there is still partisan bias and we do not see this reduction in bias for all candidates.
Our research has focused solely on criticisms in presidential campaigns, but we would expect to see the same kind of partisan bias in judgments of campaign rhetoric in down-ballot races such as for the Senate and House. Indeed, to the extent that down-ballot races can be more personal in their attacks, provided those attacks do not grievously violate norms of fairness in the way that attacks on the behavior of family members appear to, there may be more room for partisan disagreement and partisan bias may be even stronger than that shown here.
What do our findings imply for previous research on perceptions of acceptable campaign rhetoric? First, they suggest that abstract fairness judgments, rendered in a partisan vacuum, tell us relatively little about perceptions in a real partisan campaign. Normative principles are not meaningless – on the whole, citizens who view an attack as beyond the pale in principle are also likely to object to such an attack in practice – but these principles are inconsistently applied. Our research points to an important disconnect between what citizens judge to be acceptable or unacceptable negativity and their application of such principles in reality.Footnote 45
Secondly, our research leads us to question the portrayal of the public as tough and resilient to campaign rhetoric, inured to violations of the principles of fairness and civility as applied to the types of campaign ad claims they regard as fair. Although the end result of partisan bias may be the same, the process suggested by our findings is quite different: the public is partisan and thin-skinned when it comes to attacks on the candidate the partisan prefers and callous when it comes to attacks on the opposition, but in either case it is a rigid (rather than resilient) public applying principles in a very partial manner.
Thirdly, our findings suggest that different groups of voters process information about negative advertising in specific ways, and that these distinct processing strategies need to be identified and assessed in both experimental and survey studies. For example, our work shows that a very powerful form of motivated information processing operates among partisans when they encounter negative advertisements about candidates. It is likely that in studies that lack partisan cues, the substantive information presented to subjects and respondents is processed differently than in an actual electoral setting. Therefore until we are able to better identify and test these alternative information-processing mechanisms, it seems likely that many of our conclusions about the importance of negative campaign ads must remain tentative.
