Assessing Accountability in a Post-Citizens United Era

Backlash

The overwhelming majority of the public dislikes negative campaigning and would like to see much less of it on the airwaves. A Gallup Poll conducted in 2000 found that only about one in five people agree that “negative advertisements have a place in campaigns” (The Roper Center for Public Opinion Research, University of Connecticut, 2000). Another poll conducted by the Institute for Global Ethics in 2002 found that 8 in 10 believe that “negative, attack-oriented campaigning is unethical and damaging our democracy.” Many other polls have obtained similar results. ¹¹ In an experiment, Brooks and Geer find that not all negative ads are equal in people’s minds: “civil” but opponent-focused messages are thought to substantially more “fair” than “uncivil” (i.e., especially harsh, mean-spirited) opponent-focused messages (Brooks & Geer, 2007). So negativity itself may not be a critical problem in people’s minds but the tenor of that negativity may be.

The average citizen may well be misguided in being concerned about negative advertising: Although specific defenses of uncivil campaigning are in short supply, many scholars have made compelling arguments that negative campaigning overall has inherent value to political dialogue with fewer societal costs than might be assumed (Geer [2006] most directly takes that position in In Defense of Negativity). The present study is neutral on that matter, however. The question at the core of this article is whether harsh attack messages work better for some types of sponsors than for others. That is likely to hinge, in part, on whether people penalize candidates more than independent groups for sponsoring those messages, regardless of whether people are “right” or “wrong” in trying to apply penalties for attacks.

Many scholars have hypothesized that voters do penalize candidates for being too negative, with a dynamic known as “backlash effects” (also sometimes called “boomerang effects” or “unintended effects”). Backlash may result from an explicit attempt to send a signal to the sponsoring candidate that such messages are unacceptable. It may result from anger or frustration with the candidate only talking about his or her opponent (or for talking about the opponent in an unacceptably derogatory manner) rather than using a message to highlight his or her own plans and achievements. There may be a motivated reasoning dynamic in which people who share the partisanship of the target of the ad are especially likely to penalize the ad sponsor because they feel protective of the target candidate or because they find it easier to denigrate the sponsor than to change their beliefs. A range of research has revealed that people penalize candidates for running negative ads (see, for example, Carrano, Gawronski, & Castelli, 2010; Kahn & Geer, 1994; Shultz & Pancer, 1997). Lau, Sigelman, and Rovner’s (2007, p. 1183) meta-analysis shows that 33 of 40 studies on the topic clearly confirm a backlash effect. This general line of research has not determined whether the intended/persuasive effects of negative ads outweigh their unintended/backlash effects (see Lau et al., 2007 for a review of this literature). However, the point here is that most people say they do not like attack ads and there is considerable evidence that people attempt to assess blame on sponsoring candidates for running them.

With the exception of Garramone’s research, the research that specifically addresses backlash effects has been conducted using only candidate-sponsored ads and therefore cannot speak to the sponsorship issue that lies at the core of the present article. Garramone (1984, 1985) and Garramone and Smith (1984) looked explicitly at backlash depending on ad sponsorship by a known, explicitly partisan group versus a candidate and found that backlash against the benefiting candidate was substantially reduced among undergraduates when the sponsor was an independent group rather than the candidate himself. This makes sense intuitively. When an independent group runs an attack ad against a particular candidate, voters have little opportunity to hold anyone accountable for the message in a direct manner. The candidate can potentially be penalized for the decisions of unaffiliated groups that work to get that candidate elected, and some voters are likely to do that. ¹² However, it is reasonable to assume that a fair portion of voters who would like to see negativity minimized will not be inclined to hold the candidate responsible for another group’s decisions to run an attack ad. ¹³ Thus, affect for the benefiting candidate may fall somewhat when an independent group sponsors a message or when no sponsorship information is provided at all, but it is unlikely to fall to nearly the degree that it will when the benefiting candidate explicitly sponsors the message and can be directly penalized for going on the attack. Based on that logic, we propose the following hypothesis:

Persuasion

Ad persuasiveness is the second element to consider in the ad sponsorship–ad effectiveness relationship. To be persuasive, viewers need to believe the ad is truthful. However, given that many people are only marginally informed about politics, how can voters possibly determine the veracity of campaign claims while viewing an ad? For a given ad, source credibility is one of the only simple heuristics voters can use to determine whether the information contained in the ad can be trusted.

The significance of source credibility has been recognized for thousands of years; Aristotle, for example, argued persuasively in Rhetoric that people consider it when evaluating arguments: “It is not true, as some writers assume in their treatises on rhetoric, that the personal goodness revealed by the speaker contributes nothing to his power of persuasion; on the contrary, his character may almost be called the most effective means of persuasion he possesses” (2156a10-15 in Barnes, 1984). Along these lines, we would expect a source seen as “good” to be more credible than a source seen as “bad.” At the same time, Pornpitakpan’s recent review of five decades of research reveals that a more credible source will be more persuasive than a less credible one (see Pornpitakpan, 2004). Trustworthiness is central to the concept of credibility and many have found that attitude change is greater when a source is viewed as trustworthy rather than untrustworthy (see, for example, Bauer, 1967; Druckman, 2001; Garramone, 1985; Hovland & Mandell, 1952; Hovland & Weiss, 1951-1952; Petty & Wegener, 1998). The question is whether candidates or independent groups will be more trustworthy in the minds of voters. The answer is not entirely clear.

It is reasonable to expect that a message will be viewed as more trustworthy as the stakes of lying increase. For candidates, the stakes of getting caught are potentially very high as it could affect their future in politics. The stakes for long-standing interest groups with national reputations are less than those for candidates (after all, the organization will likely still exist even if the average voter does not trust it), but the size of their membership base and their ability to credibly lobby legislators may be endangered if they are seen as untrustworthy; as such, many “Repeater” groups such as Sierra Club, AARP, EMILY’s List, or other known groups will face at least some incentives to be truthful. “First Timer” groups, however, rarely face those incentives. With generically appealing, nonpartisan names, many “First Timer” independent groups are designed to sound vaguely compelling to voters without inciting skepticism. Little to nothing of value beyond money is at stake for newer independent groups: they are not running for office, they may be funded by just a small handful of individuals, and they have no reputation to uphold. They will not be lobbying officials as a group after the election and as such they do not need to uphold a credible reputation to that end. They have the freedom to disappear quickly and even reappear under a different name with a similar roster of donors if and when the need arises. As the stakes of lying are far lower for unknown independent groups than they are for candidates, we should expect unknown independent group messages to be perceived as less truthful and credible, and less persuasive as a result.

On the basis that candidates should be viewed as more credible than unknown independent groups because they are more accountable to voters, we forward what we term the “Candidate Trust Hypothesis”:

Regarding persuasiveness, there are theoretical reasons to expect a countervailing force, however. Despite the stakes of dishonesty, there are reasons why candidates may not be perceived as reliable sources of information. A substantial literature suggests that less self-interested actors will be seen as more trustworthy than more self-interested actors (see, for example, Andreoli & Worchel, 1978; Hass & Grady, 1975; Hovland, Janis, & Kelley, 1953; Kiesler & Kiesler, 1964; Smith, Lasswell, & Casey, 1946; Walster, Aronson, & Abrahams, 1966). Although most political donors have an interest in the electoral outcome, they do not have the opportunity to enjoy the full trappings and power of holding office, as a candidate does. As such, it may be reasonable to assume that the public will view candidates as less than trustworthy sources (Garramone & Smith, 1984). In this regard, Abraham Lincoln perhaps said it best when he claimed that politicians are “a set of men who have interests aside from the interests of the people and who, to say the most of them, are taken as a class, at least one long step removed from honest” (quoted in Current, 1958, p. 197). Survey evidence also suggests that the public might be wary of trusting politicians. In response to the American National Election Studies question “Do you think that quite a few of the people running the government are crooked, not very many are, or do you think hardly any of them are crooked?” only about one in ten respondents in recent decades respond “hardly any of them,” whereas somewhere between one third to one half of respondents give the most cynical answer of “quite a few of them.” ¹⁴

If people do not trust candidates much, then an appealing-sounding independent group with a less obvious agenda may be seen as more trustworthy. Ads without sponsors should be expected to lack credibility as well; however, without an appealing name of the kind often chosen by 527 and equivalent groups, we would expect them to be less trusted than an independent group but more trusted than a candidate. These expectations form the basis of what we term the “Candidate Distrust Hypothesis”:

It is possible that these two effects could cancel each other out: Accountability could benefit the credibility of candidates, whereas perceptions of trust could benefit the credibility of independent groups, yielding no net difference in credibility and, therefore, persuasiveness. As it is very likely that both of these dynamics operate, we have strong reasons to expect that our “Countervailing Forces Hypothesis” will be confirmed:

Partisanship could potentially affect credibility as well. Through a process of motivated reasoning, a candidate from one’s own partisan group could logically be viewed as being more credible than a candidate of the other party. However, it is possible that an independent group that is clearly trying to support a candidate of one’s own party will also benefit from the same kind of motivated reasoning regarding its credibility. In addition, while people who share the partisanship of the benefiting candidate might be more inclined to believe a negative message about a target candidate of another party than individuals who share the partisanship of the target of the ad, it is also true that a target candidate’s partisans have farther to fall from their initial predispositions toward that candidate. As such, it is not clear that there would be strong partisan effects with respect to credibility.

Effectiveness

The relationship between backlash and ad persuasiveness for negative ads has been neglected within the literature overall, with a tendency for scholars to examine the existence of backlash and persuasiveness in separate studies. ¹⁵ The basis of the relationship is straightforward: The overall effectiveness of a negative ad will consist of Hypothesis 1 (change in affect for the benefiting candidate) minus Hypothesis 2 (change in affect for the target). In other words, if persuasion is greater than backlash, there will be net positive effectiveness for the benefiting candidate. Assuming that Hypothesis 1 is correct (backlash will be stronger for candidate ads), ¹⁶ there are three different hypotheses (Hypothesis 3a, Hypothesis 3b, and Hypothesis 3c) depending on whether Hypothesis 2a, Hypothesis 2b, or Hypothesis 2c is confirmed.

If Hypothesis 2a is accurate and candidates are seen as more trustworthy and credible than unknown independent groups, greater backlash and greater credibility of candidate-sponsored ads should cancel each other out, and then we would expect the following:

If Hypothesis 2b is accurate, then reduced backlash effects and greater credibility effects for ads sponsored by independent groups as compared with ads sponsored by candidates would lead to the following hypothesis:

If Hypothesis 2c is accurate, then the combination of reduced backlash effects and similar credibility levels for independent-group-sponsored ads and candidate ads would lead to the following hypothesis:

Table 1 provides a summary of our directional expectations for each type of ad for each hypothesis.

Study Design

Advancing our understanding of ad effectiveness requires a design that contrasts three possibilities for a negative ad: no sponsorship, candidate sponsorship, and sponsorship by an unknown independent group. We designed an experiment that allows us to vary only the ad attribution, holding all other ad content constant. We needed to randomly assign viewers to each condition. Candidate partisanship also needed to be randomized to avoid partisan effects. Moreover, the relationship should ideally be tested with a large and representative national sample of adults. We accomplished these goals with an Internet experiment fielded by Polimetrix with a post-election module added to the 2006 Cooperative Congressional Campaign Election study. ¹⁷ Our study had a sample of 1,500 people (and an N size of approximately 1,370 respondents for most of the core analyses ¹⁸ ) who were sampled through a stratified procedure and weighted to approximate a geographically and demographically representative sample of U.S. adults. ¹⁹

We created a fictional state assembly race between “Tim Clark” and “Michael Norris” to test our hypotheses. We used a fictional state-level race for several reasons. First and foremost, we wanted to be able to change the partisanship of the candidates to randomize any partisan effects, so we needed unknown candidates and messages without any identifiable partisanship. That required fictional candidates with no preexisting partisan associations. Moreover, to increase external validity, we wanted a race that many respondents would be inclined to think could occur in their area so that they might be more psychologically vested in it. State assembly races were ideal in that respect. An added benefit of using a state-level race is that while BCRA already governs ads for federal office, many states are actively considering changing their rules for state-level office to be more like those of BCRA. Focusing on that level of office thus has especially immediate policy relevance.

We used a pretest/posttest design for this experiment. Repeated measures designs provide extremely powerful control over experimental treatments. Even when randomly assigned into different conditions, natural variation between individuals can produce noise in an analysis of treatment effects. Rather than assuming that pretest results would have been comparable across groups (as one must do with a standard single measurement), repeated measures designs explicitly factor in the exact starting point for each individual so that relative change regardless of starting point becomes the focus of the analysis. To give respondents a baseline on which to judge our fictional candidates, positive ads for Tim Clark and Michael Norris were first watched by respondents. The photos of our candidates were of actual state assemblymen from California and Iowa ²⁰ and showed them in standard political poses (interacting with constituents, sitting at desks, sitting in a legislative chamber, and the like). The ad messages are standard biographic positive messages emphasizing “proven leadership,” “knowledge and integrity,” “devoted father and husband,” “common sense approach,” “works tirelessly for people like you,” and other kinds of messages that lack geographic and partisan cues. They are representative of the standard name identification ads that often begin campaigns (see, for example, Diamond & Bates, 1992; Joslyn, 1986; Powell & Cowart, 2003). Full ad transcripts and a description of the visual and audio cues can be found in Appendix A.

The only experimental element of the positive ads was variation in candidate partisanship with three randomly assigned conditions: (a) no party cue mentioned at all, (b) Clark is a Democrat and Norris is a Republican, and (c) Clark is a Republican and Norris is a Democrat. The results for our “no partisanship” condition are virtually identical to the results for our partisan candidate conditions and also for the combined sample (where there is a random mix of people exposed to the different partisanship cues in each condition). As such, we opted for the larger N analysis (all combined; all but one of our results is of similar magnitude and significance no matter how the analysis is conducted). ²¹ We also ran the results broken out by “partisan alignment” (i.e., whether the sponsor or target of the ad shared partisanship with the respondent or whether the respondent is an independent). That analysis does not change the main findings from this study; however, we allude to the main findings from that analysis below, and the results of the analyses are included as a web appendix for readers who would like to learn more about the underlying partisan dynamics.

After viewing the two positive ads, respondents were asked for whom they would lean toward voting “based on what they know about the candidates so far” on a 5-point scale and their favorability for each of the candidates on a 7-point scale (see Appendix B for question wording). A photo of each candidate with his name listed was presented on pages with questions relevant to the candidates so that lack of name recognition would not create unnecessary noise in the results.

At that juncture, we then showed the negative ad which included our experimental treatment. Aside from sponsorship information, that ad was identical for all respondents. After viewing that final ad, respondents then answered the same set of survey questions a second time.

We faced a choice as to which kind of negative ad to use as adding cells to test multiple kinds of ads would leave us unable to address null results convincingly due to power limitations. We designed a trait-based ad message about Tim Clark that includes accusations of absenteeism in previous office, investigations into tax evasion within a family business, and accusations of “bad debts and campaign violations” (see the third ad in Appendix A for the negative ad text and description). It concludes with the message “Is Tim Clark really the kind of man you can trust to represent you?” complete with ominous music, a strong, deep voiceover, sound effects, and black and white photos—all tricks of the trade that make an ad not only read negative but also feel emotionally negative (see, for example, Brader, 2005; Mark, 2006; Swint, 2006).

This kind of ad—termed attack ad here—was specifically selected for examination for several reasons. First, it avoids simple practical difficulties of the partisan and geographic cues that are inherent to the discussion of most issues. Second, it was realistically the only option for studying these dynamics while varying candidate partisanship; if an issue-based negative ad was used, the issue content would realistically need to be varied depending on the partisanship of the target candidate. Varying the ad content in the treatment ad would eliminate the kind of control that is central to establishing causality in this type of study. In contrast, a negative trait-based ad allows us to vary only candidate partisanship and ad sponsorship, while keeping the ad content exactly the same in all of the experimental conditions.

Third, a frequently expressed concern about the rise of unknown independent groups on the advertising landscape is that these groups face no disincentives to delivering that kind of very hard-hitting, often controversial negative spots with which campaigns do not want to be associated. It is important to determine if that concern has any merit, in part, because these kinds of ads are well positioned to cut through a cluttered media environment and thus may play an outsized role in setting the terms of a campaign. The kind of message we used is precisely the kind of negative ad that produces the most hand-wringing by pundits and politicians about the (assumed to be) detrimental effects of campaign advertising, with its unproven accusations (“investigations” and “charges”) and an emphasis on distrust and an impending sense of doom. Thus, although trait-based attack ads do not tend to dominate the airwaves, the kind of ad we examine here is characteristic of the type that tends to generate the most concern with respect to independent group advertising. ²²

The experimental treatment consisted of whether an attribution was provided for the attack ad and, if so, what type (the specific attributions are delineated in Appendix A). One fourth of respondents saw no attribution; the sponsorship of the ad was left unstated. For the ads sponsored by an independent group or a candidate, we fully complied with BCRA sponsorship requirements. To be BCRA compliant, it is now impossible to hide the sponsor of an ad; this constitutes a marked contrast to the pre-BCRA era, in which advertisers often masterfully distracted viewers from noticing the minimal sponsorship information present in the most extremely negative ads. In keeping with those requirements, the attribution was both written and spoken for both candidates and independent groups, and a photo of the candidate appears in the candidate-sponsored ad (sponsor-identifying photos are not required for independent group ads). One quarter of the sample saw the independent group attribution, whereas the remaining half of the sample saw ads sponsored by a candidate (Michael Norris). ²³

For our independent group sponsor, we selected a fictional name that has the generic, nonpartisan, feel-good connotations shared by many of the recent “First Timer” 527 groups. “Citizens for a Better State Government” is quite similar to many of the 527 group names used in recent years (e.g., “Citizens for Strength and Security,” “Citizens for the Future,” “Citizens for Truth,” “Citizens for a Strong Senate,” etc.), yet was not the name of any existing group, so respondents would have no prior associations with that particular name. After the attack ad, respondents were asked the same questions they had been asked after the positive ads (see Appendix B for question wording).

This design does not incorporate any potential additional effects of free media coverage on ad reception. Although it is possible that ads that garner intense media analysis and criticism could generate directionally different effects than low-profile ads, it is not clear that would be the case. Analyses of so-called “ad watches” indicate that media scrutiny tends to amplify rather than minimize the intended effects of ads (Ansolabehere & Iyengar, 1995; Iyengar & Simon, 2000; McKinnon & Kaid, 1999; Pfau & Louden, 1994).

Findings

With a pre/post design, we are able to compare the candidate scores a respondent provided after viewing the attack ad with the original benchmark score the respondent provided after viewing the two nonexperimental ads. Specifically, with analysis of covariance (ANCOVA), we can predict the postexperiment score based on our experimental treatment (attack ad with no sponsorship information vs. independent group/Citizens for a Better State Government sponsorship vs. candidate/Norris sponsorship) while controlling for the preexperiment score by including it as a covariate. ²⁴ As compared with the approach of only administering a single test after the experimental intervention and comparing those scores across individuals (where random differences between individuals can cause variance and unnecessary noise), we hold the individual constant with a benchmark measurement prior to our experimental intervention.

In the first analysis, we examine the effects of ad sponsorship on candidate backlash. In Hypothesis 1, we hypothesized that candidate-sponsored messages would generate a substantial degree of backlash toward the candidate, whereas ads sponsored by independent groups and nonsponsored messages would generate relatively less backlash. For this measure, movement downward between the pre- and posttreatment measurements means that Norris’ favorability fell after exposure to the attack ad, and the change is displayed by holding the pretreatment measurement constant at 4.24. In this case, the effects of our experimental treatment are significant (see Table 2, row 2A) and are substantively large. The change in means in Table 2 (row 2A) shows that the candidate is penalized by .85 of a point on a 7-point scale for sponsoring an attack ad himself. Candidates are not entirely off the hook even when they do not sponsor the ad: Voters still penalize the nontarget candidate somewhat in the other conditions (the backlash effect is significantly greater than 0 in all three conditions), but the penalty is significantly lower than when the ad is sponsored by a candidate (see Table 3, row 3A for the significance of the group differences using Tukey-Kramer post hoc testing). Thus, Hypothesis 1 is confirmed.

In the second analysis, we examine the effects of ad sponsorship on persuasion and confirm our Hypothesis 2c (Countervailing Forces Hypothesis) that messages do not vary in persuasiveness depending on the sponsor. Table 2 (row 2B) shows the favorability change from the pre- to postexperiment measurements, with the value of the posttreatment mean estimated by holding the pretreatment covariate constant across individuals at its grand mean (4.04). A more persuasive attack ad will produce lower favorability toward Tim Clark and greater movement downward. Overall, our attack ad was quite persuasive; respondents in all conditions moved at least three-fourths of a point after viewing the attack against Tim Clark, a movement that represents a significant change. However, we see no differences in persuasion depending on the message sponsor. We are thus able to confirm Hypothesis 2c with the finding that the nature of the ad sponsor—even the presence of having an ad sponsor at all—does not affect the persuasiveness of the negative message.

With no differences in terms of persuasion between the three types of sponsors, but greater backlash toward the candidate when a candidate sponsors the ad, we would expect a confirmation of Hypothesis 3c (effectiveness will be higher when ads are sponsored by independent groups or contain no sponsorship information than when they are sponsored by candidates). This hypothesis was indeed confirmed by our findings (see Table 2, row 2C for the model results and Table 3, row 3C for comparisons between the groups) using the net effectiveness of the ad. ²⁵ To use candidate sponsorship as an example, affect for the target (Clark) drops by –.750 between the pre- and postmeasures and also drops for the sponsor (Norris) by –.848 (using actual, not covariate-adjusted, means). As everything is ultimately relative in a zero-sum two-candidate race, the relative effect on favorability is the change for Norris minus the change for Clark (–.11, which means that Norris ended up slightly worse off for having run the ad). In other words, we see no net benefit—instead, possibly a very slight cost, albeit one that is not statistically significant from zero—to relative candidate favorability when a candidate sponsors the ad.

There is a substantial net favorability benefit for Norris, however, when an ad either contains no sponsorship information or when an independent group sponsors the ad. Both produce a similar result: a net benefit of about half a favorability point in favor of Norris. As we noted before, both types of ads result in significant backlash toward Norris even though they are not technically sponsored by him, but the persuasive benefit of the ad offsets the backlash effect, thus benefiting Norris overall.

Favorability is an interesting variable because it allows us to observe the separate mechanisms of persuasion and backlash operating beneath the surface. For a race involving fictional candidates, we are probably safest focusing on a leading-indicator variable like favorability. However, candidates ultimately try to improve favorability scores because they are commonly thought to be leading indicators of vote choice and so voting preference is also an outcome measure of importance. To address this question as a secondary confirmation of Hypothesis 3c, we also asked for vote preference on a 5-point scale for each candidate after the positive ads and again after the attack ads. The findings for vote preference conform to the findings from our analysis of favorability differences, albeit in a slightly less dramatic manner (see Table 2, row 2D for the results).

As mentioned earlier, we also ran the results depending on whether the respondent shared partisanship with the ad sponsor or ad target. The results are presented in the web appendix. The bottom line of that analysis, however, is that the presence of partisan alignment in the model does not change any of the core findings from this study: Overall, ads sponsored by independent groups are more effective than candidate-sponsored ads. Persuasiveness (and, by extension, credibility and trustworthiness) does not appear to be particularly conditional on partisan alignment: The sponsor has relatively little effect on whether partisans who share the party of the target candidate or who share the party of the sponsoring candidate are persuaded by the content of the negative ad. There are more differences for backlash: As expected, respondents who share the partisanship of the target candidate are more likely to penalize the sponsoring candidate for going negative. With reference to net effectiveness, ads sponsored by independent groups are moderately more effective than candidate-sponsored ads regardless of whether the respondents share or oppose the partisanship of the candidate who benefits from the ad; however, the difference is somewhat more dramatic for partisans who share the party of the target of the ad. With reference to vote preference, sponsorship by independent groups rather than candidates is especially effective for moving respondents of the opposite party over to the side of the benefiting candidate.

For independents, the attack ad has essentially a neutral effect on them, regardless of who sponsors it. However, digging beneath the surface, it is clear that independents are affected by ad sponsorship: They have stronger reactions to candidate-sponsored ads than ads sponsored by independent groups. However, the stronger backlash and stronger persuasion that is produced by candidate-sponsored ads rather than ads sponsored by independent groups still combines to produce essentially zero net effectiveness, so the difference for independents in reactions to sponsorship is ultimately inconsequential.

Conclusion

The primary finding from this analysis is that a trait-based attack ad sponsored by an unknown independent group is more effective than an identical ad sponsored by a candidate in the eyes of the public overall. Although our experiment allows us to carefully address the effects of the kinds of ads that have generated the most concern among pundits and the public, many of the negative ads on the airwaves are not as harshly negative or as personally negative as the ad used in this experiment. What might this experiment have to say about responses to other kinds of opponent-focused ads? This study shows that it is largely differences in backlash, not persuasion, which produce sponsorship effects. As long as an ad produces backlash, there is little theoretical reason to think that the tenor of a negative ad would affect the relative difference in candidate backlash associated with candidate-sponsored versus independent-group-sponsored ads. Kahn and Kenney (2004) find that while “mudslinging” (extremely negative campaigning, as defined by campaign managers) generates a particularly high degree of backlash for incumbents, other negative ads produced significant backlash as well. ²⁶ In sum, there are reasons to expect that these findings are likely to be generally applicable, although the differences between sponsorship conditions may become more modest for less negative ads.

Of course, negative ads and campaigns come in many different flavors. Even the seemingly clean distinction between an issue-based ad and one that focuses on traits is, in reality, often very blurry. Some issue-based ads focus on issues pure and simple, but many of them do not: The underlying purpose seems to be not to discuss issues per se but rather for voters to regard the target of the ad as having personal characteristics unfit for the office in question. The infamous independent ad featuring Willie Horton is an example of how this can be done implicitly; the text of the ad focuses completely on issues and no direct effort is made to portray the target of the ad (Michael Dukakis) as being someone who lacks appropriate personal characteristics for the presidency, yet it clearly seemed to coincide with—and reinforce—the general effort to portray him as being “not tough enough.” Using an ad that centers on issues to raise questions about a candidate’s personal characteristics can also be done explicitly: For example, the text of the controversial independent ad linking Sarah Palin with aerial wolf hunting is almost completely focused on a discussion of policy issues, but the last sentence of the ad—“Do we really want a vice president who champions such savagery?”—seems to indicate that the true purpose of the ad is really to undermine her on a personal level. The bottom line is that although the present study is the first one to analyze how the public responds to attack television ads by unknown independent groups, it should not be the last: To fully confirm whether the theory delineated in this study as well as its specific findings apply generally, scholars in the future should examine how sponsorship influences the public response to the entire spectrum of negative ads and campaign situations. Scholars should examine the role of rebuttals and one-sided versus two-sided information flows, races with known incumbents versus open-seat races, races at lower levels of office versus races at higher levels of office, and so on. ²⁷ Furthermore, examining the length of time that these differences in effectiveness might persist for individuals could also be an interesting avenue for future research. Examining such a wide range of campaign scenarios will be very resource intensive but is worthwhile, given how important independent groups have now become in American politics today. If such an effort is undertaken, my predication is that the underlying theories presented here will hold for most of the negative ads on the airwaves and across many different campaign contexts, but we can only know for sure once a range of different campaign and message dimensions have been specifically tested.

What are the wider implications of the specific findings from this study for American politics? The particular kind of attack ads examined here clearly have the potential to have a significant influence on the dynamics of elections in part because, to the extent that they may generate controversy, they may garner free media coverage which can extend their reach well beyond actual ad expenditures. The advertisements sponsored in 2004 by Swift Boat Veterans for Truth are the most memorable example along these lines. This study indicates that candidates have every reason to hope for an unofficial division of labor, in which independent groups that are unaccountable to voters will do the dirty work of running these kinds of harsh attack ads that the candidates would rather not do themselves. To the extent that these kinds of ads can be outsourced to independent groups, candidates can best put themselves in a position to gain the benefits associated with this kind of harsh negative advertising while avoiding the costs. ²⁸

Of course, outsourcing the harshest kinds of ads to independent groups could be a problematic strategy if the groups themselves have clearly identifiable origins, such that voters would be more easily able to link the activities of such groups to particular candidates or to parties more generally. Yet as was noted previously, the percentage of outside groups that disclosed the names of their donors fell to only 32% in 2010 and, moreover, such groups frequently choose generic names to obscure their origins. The fact that the public cannot identify the contributors to so many of these groups thus makes it easier for these groups to go on the attack and thus helps to explain why negative ads by these groups are now so much more prevalent than in previous eras. In this regard, it is telling that in their prominent 1995 book arguing that negative ads are concerning for American democracy, Ansolabehere and Iyengar (1995) did not see independent group advertisements as a cause for anxiety principally because the ads sponsored by these groups were overwhelmingly positive at the time they were writing. ²⁹

In turn, outsourcing these kinds of ads to independent groups would be hard to pull off if campaigns and independent groups were truly “independent” of each other in their advertising decisions. Yet although independent groups are explicitly barred from coordinating and communicating with campaigns and parties, staffers and consultants often cycle between working for independent groups and campaigns. Although such movement has generally been found by the Federal Election Commission (FEC) to be legal, there is legitimate concern about how independent many of the groups really are from the campaigns they benefit. The Willie Horton ad is one example in this regard: The ad was funded by a group that was independent of the George H. W. Bush campaign—the National Security Political Action Committee (NSPAC)—but the media consultant who created the ad, Larry McCarthy, had previously been a senior vice president of the Ailes Communications, the main media group for the Bush campaign. McCarthy later admitted during an FEC investigation that he had been in contact with Ailes at various points during the campaign (the FEC split 3-3 on whether such coordination was illegal). Swift Boat Veterans for Truth is another prominent illustration of this dynamic. In August 2004, the lead election lawyer for the Bush campaign team resigned once it was made public that Swift Boat was one of his clients. A member of the steering committee for veterans’ issues on the Bush campaign also resigned after appearing in one of the Swift Boat ads and Jeb Bush sent a letter of thanks to the largest donor to the Swift Boat 527 for his “willingness to stand up against John Kerry.” Although these activities were not deemed by the FEC to constitute illegal coordination, they are of course highly questionable from the standpoint of accountability; in public, President Bush distanced himself from the group by strongly criticizing the initial ad run it ran, albeit a week after it stopped running. The overall findings from the present study provide further indication that there are considerable incentives for questionable coordination between campaigns and independent groups.

Moreover, independent groups need not be independent of one another, and recent elections have seen systematic coordination efforts between groups who share similar ideological goals. During the 2010 election cycle, for example, Ruttenberg (2010) reports that conservative groups participated in regular strategy sessions to divvy up races to avoid duplicating their efforts. The strategy sessions were coordinated by several independent groups which shared office space and were founded, in part, with Karl Rove’s assistance (Ruttenberg, 2010). Through such efforts, like-minded independent groups can maximize their power within individual campaigns and within elections more generally.

To be sure, outsourcing attack ads to independent groups is not cost free for candidates. With no control over the content of such ads, it can make it harder for a candidate to stay “on message” and control the campaign dialogue. ³⁰ However, candidate control of the campaign dialogue is often an elusive goal anyway. Ceding some control over the message may seem to be a small price for candidates, nearly all of whom would like to have more ads on the air than they can afford to run and would likely prefer to be in a position to distance themselves from the harshest negative messages.

For those concerned about the accountability issues raised by independent group advertising, the future does not look promising: In the wake of the Citizens United decision, corporations and unions can now contribute to independent groups to air campaign-related ads and they will generally have substantial incentives to hide their identities from the public to avoid alienating consumers, shareholders, and membership bases. Although Internal Revenue Service (IRS) regulations require regular public disclosure of the donors to 527 groups, the IRS is expressly prohibited from releasing the identities of donors to 501(c) nonprofit groups, thereby preventing the press and voters from identifying who paid for those ads. ³¹ Largely in response to more active enforcement of 527 group donor-disclosure requirements, donations to 501(c) groups tripled in 2008 relative to 2004 levels and there is strong reason to expect that this trend will continue (Campaign Finance Institute, February 25, 2009). The degree to which Super PACs will be able to shield donors from disclosure is not yet clear at the time of this writing, but early indications are that at least some donors have been able to hide their identifies behind difficult-to-trace limited liability corporations (Confessore and Luo 2012); moreover, donors who wish to remain anonymous continue to have the option of doing so through a 501(c)4 organization.

Some members of Congress consider those kinds of donor disclosure loopholes to be a problem and have proposed bills like the “DISCLOSE Act” which, among other things, required that the biggest donors to independent groups be identified specifically within advertisements sponsored by those groups. ³² Under conditions of clear donor disclosure, corporate and union sponsors with membership, shareholder, and/or customer bases they cannot afford to alienate would presumably be more likely to shy away from the most negative and controversial ads. Yet many Republican leaders strenuously object to this legislation and its passage in a filibuster-prone Senate anytime soon seems extremely unlikely (Franz, 2010, p. 19). Moreover, even if the legislation did pass, many individuals and some groups would be largely unaffected by disclosure. The present study shows that the variation in net ad effectiveness between sponsors is due to differences in backlash rather than differences in persuasive power. As such, even increased disclosure will not change the fact that harsh attack ads sponsored by outside entities will tend to be more effective than comparable ads sponsored by candidates, which will tend to enhance the relative power of ads sponsored by independent groups on the airwaves.