Defending Audience Costs in the Trump Era

Scholars widely agree that citizens care about their country’s international reputation.^{Footnote 22} Research on audience costs and reputation costs consistently observes that the public disapproves of leaders who act inconsistently in foreign affairs specifically because it views inconsistency as damaging the country’s reputation for credibility and resolve.^{Footnote 23} A longstanding insight from the “cottage industry” of scholarship on audience costs is that democratic leaders’ sensitivity to public opinion ought to deter them from backing down on international threats, since voters see these threats as putting the country’s reputation on the line.^{Footnote 24} Of course, the public’s concern for national “credibility, face, and honor” has impacts that extend far beyond the audience costs literature,^{Footnote 25} underpinning theories of credibility in crisis bargaining;^{Footnote 26} the democratic peace;^{Footnote 27} effectiveness in fighting wars;^{Footnote 28} peaceful conflict resolution;^{Footnote 29} crisis escalation;^{Footnote 30} the reliability of allies;^{Footnote 31} use-of-force decisions;^{Footnote 32} and the perpetuation of international rivalries.^{Footnote 33}

Yet many observers are calling for further testing into whether existing theories of public opinion in foreign policy continue to apply to the Trump presidency.^{Footnote 34} When it comes to domestic costs for foreign policy inconsistency, the main reason for this doubt is partisan polarization. In the current political environment, Americans are sorted into more homogenous parties and less likely to take positions at odds with their own party.^{Footnote 35} Rising levels of “negative partisanship”—general disdain for members of the opposing party—further reinforce partisan tribalism among the public.^{Footnote 36} A number of past studies suggest that partisan polarization can unravel the “bipartisan liberal internationalist consensus” in U.S. foreign policy.^{Footnote 37} And some scholars specifically view partisan polarization as a moderator of audience costs, as these costs may be harder to generate when “citizens are biased for or against the incumbent relative to the opposition.”^{Footnote 38} As Daniel Drezner posits, the current degree of polarization in the United States could “shrink the audience that might change its mind about the President if he backs down in a crisis.”^{Footnote 39} If Donald Trump is indeed the most polarizing leader in American history,^{Footnote 40} then it is reasonable to presume that he is less capable of generating audience costs.^{Footnote 41}

Severe partisan polarization is also speculated to impede the ability of individuals to learn from—and react to—foreign policy failures, such as the issuance of non-credible threats.^{Footnote 42} This is because partisanship influences how people acquire, interpret, and use factual information when evaluating leaders and their foreign policy decisions.^{Footnote 43} Research on public opinion in the Iraq War shows just how powerful this effect can be: Republicans clung to false beliefs that Iraq had weapons of mass destruction, while Democrats were resistant to information showing that the 2007 surge helped reduce violence in Iraq.^{Footnote 44} When applied to the logic of reputation and audience costs, these findings suggest that if “Trump’s base will swallow whatever he tells them about foreign policy” while his detractors will unconditionally reject it, the costs of foreign policy inconsistency should be minimal—washed out by individuals’ preexisting attitudes toward the president.^{Footnote 45}

We do not doubt that the atypical nature of the Trump administration and the severity of partisan polarization in the Trump era profoundly affect the conduct of U.S. foreign policy in general. Nevertheless, we argue that the partisan polarization surrounding Donald Trump’s presidency should not affect the magnitude of audience costs that President Trump incurs—nor his ability to deflect audience costs. The empirical implications of our argument are twofold: (1) Republicans and Democrats will equally disapprove of Donald Trump if he backs down in a crisis; and (2) Republicans and Democrats will equally accept Donald Trump’s justifications for backing down in a crisis.

First, President Trump should incur equal levels of audience costs among Republicans and Democrats because the cleavage between the views of Democrats and Republicans on foreign policy issues may not be as extreme as is often characterized.^{Footnote 46} Chaudoin, Milner, and Tingley observe that contrary to critics' fears, bipartisanship in U.S. foreign policy has remained strong since the 1970s, even as partisan divisions have increased domestically.^{Footnote 47} Kertzer, Brooks, and Brooks argue that this is because partisan affiliation has little effect on how individuals evaluate foreign policy issues, and that people do not view the foreign policy positions associated with Democrats much differently from those of Republicans. Supporting that position, Milner and Tingley show that the “contrasts across the ideologies” of Democrats and Republicans “seem more muted” on use of force decisions: “At certain times practically everyone will support maintaining military might and favor military intervention.”^{Footnote 48} Much of the international relations literature suggests that the American public has regular features when it comes to the use of force: it values international reputation,^{Footnote 49} and will support military force when it is likely to succeed, is multilateral, fits with their values, and expected casualties are low.^{Footnote 50} Although some evidence shows that conservatives care more about the reputational consequences of inconsistency, and liberals more about the reputational consequences of belligerency, both care about their country’s reputation and will disapprove of leaders they perceive as damaging it; the gap between the two views is not substantial enough to impact the magnitude of audience costs.^{Footnote 51} As Michael Levendusky and Matthew Horowitz have noted, “voters are inclined to see audience costs style scenarios in terms of national interests, rather than partisan ones.”^{Footnote 52} This leads to our first hypothesis:

Second, we expect that President Trump is equally capable of deflecting audience costs from Democrats and Republicans because the general public tends to be “rationally ignorant” in foreign affairs, largely deferring to the president in crisis situations.^{Footnote 53} While the public cares about credibility and resolve, the information asymmetry between leaders and populations limits its ability to independently assess the merits of leaders’ crisis decisions. By virtue of their unique access to intel and foreign policy expertise, presidents are able to mold the public’s reaction in the immediate aftermath of a crisis-decision—when citizens lack alternative sources of information, partisan lines have yet to be drawn, and public opinion is most malleable.^{Footnote 54} In this period, the information contained in the message itself should outweigh the identity of the messenger.^{Footnote 55} Levendusky and Horowitz, for instance, show that presidents are capable of reducing audience costs from co-partisans and opposition-partisans if they justify backing down as being in the “national interest.”^{Footnote 56} Given the information asymmetry favoring the president in foreign policy, mass publics may accept the executive’s justifications, or at least give them the benefit of the doubt if the logic of the justification seems plausible, before alternative sources of information on that issue become available.^{Footnote 57} Consequently, we don’t just expect Republicans to accept a reasonable justification he offers for backing down in a crisis; we expect Democrats to do so as well—at least in the short term. This gives rise to our second hypothesis:

While our hypotheses are strongly supported by past findings in the audience costs literature, we believe that this literature must be revisited as it is ill suited to the Trump era. Because prior audience cost experiments prime respondents with unnamed, hypothetical leaders (“The President”, but sometimes the “Republican President” or “Democratic President”), they provide us with little guidance when it comes to specific leaders—especially those as unique and individually divisive as Donald Trump.^{Footnote 58} Moreover, because domestic audience costs underpin many of our theories on international conflict and cooperation, as well as the domestic accountability of leaders,^{Footnote 59} it is worth revisiting whether the logic of audience costs still holds in the Trump era.

One final point is warranted before moving on. The dominant trend in political science is to hypothesize statistically significant effects. In contrast, each of our hypotheses is deliberately framed in terms of the null. We expect our main findings to be null results, with partisanship failing to achieve statistical significance as a moderator of audience costs for Donald Trump. We firmly believe that statistically insignificant results can be both theoretically interesting and substantively important.^{Footnote 60} This is especially true when it comes to research on the uniqueness of the Trump presidency, as scholars must be open to—and aware of—the possibility that in some areas of policy, Donald Trump is not significantly different than his predecessors. This is a point that we return to in the conclusion.

Experimental Design

To test our theories, we turn to an online experiment in which we recruited 1,991 individuals through Amazon’s Mechanical Turk in January 2017—after the election of Donald Trump but before he took office.^{Footnote 61} Of these, approximately 15% (292) did not pass the manipulation checks, failing to correctly identify the identity or actions of the president in the treatments to which they were assigned.^{Footnote 62} We omit these individuals from the analysis, but all results are robust to their inclusion.^{Footnote 63}

Because MTurk samples often include more women, are more liberal, younger, whiter and more educated, we collected information on the respondents’ gender, age, race, education, and partisanship.^{Footnote 64} While MTurk samples may not be representative of the general population, the differences can be reduced when controlling for measurable sample features and “MTurk respondents do not appear to differ fundamentally from population-based respondents in unmeasurable ways.”^{Footnote 65} While many have reservations about the generalizability of crowd workers to the general population, MTurk studies are perfectly valid and are “no worse than convenience samples used by other researchers in political science.”^{Footnote 66} Most importantly for our study, “liberals and conservatives in our MTurk sample closely mirror the psychological divisions of liberals and conservatives in the mass public.”^{Footnote 67} We include demographic information about the sample in online appendix A.

In our survey, all respondents received the following vignette at the introduction:

This vignette is derived from existing audience costs experiments, which enables us to firmly situate our results within the literature.^{Footnote 68} We deliberately refrained from identifying specific countries in the vignette because, as Levendusky and Horowitz observe, specifying the countries involved in the crisis runs the risk of introducing country-specific confounding.^{Footnote 69} We did, however, ask respondents which specific countries they believed were involved in the scenario to assess whether our results change if respondents were thinking of any particular crisis.^{Footnote 70}

After reading the vignette, respondents were randomly assigned information about the partisanship of the president and the president’s actions towards the conflict, producing a 3x3 factorial design. To realistically manipulate the identity and partisanship of the president, we told respondents that the leader was either “The President,” “President Barack Obama,” or “President Donald Trump.” Then, respondents were told that this president: (1) decided not to take any action; (2) promised to intervene and then backed down; or (3) promised to intervene and backed down, but justified his decision to back down on the basis of new information indicating that involvement was not in America’s interests.^{Footnote 71} For the full wording of the vignettes, refer to online appendix E.

For the main outcome of interest, we asked respondents whether they approved of the president’s handling of the situation in the vignette on a scale ranging from 1 (strongly disapprove) to 5 (strongly approve). We collected information about respondents’ partisanship based on a 7-point scale which we condense to three groups (Democrat, Independent, and Republican).^{Footnote 72} Our main analysis excludes independents to specifically assess how partisans (including leaners) impose audience costs, leaving us 1,454 Democrats and Republicans. However, the results are robust to the inclusion of independents and a continuous measure of partisanship.^{Footnote 73} We also asked respondents who they voted for in the 2012 and 2016 presidential election to examine whether there are any leader-specific (as opposed to party-specific) effects, which recent work on leader-specific reputation in international relations suggest as a possibility.^{Footnote 74} This question further allows us to assess whether the results change when excluding non-voters.^{Footnote 75} Finally, we included a series of foreign policy knowledge questions, adopted from Chen, to assess whether—and how—political awareness might condition audience costs.^{Footnote 76} Previous research has argued that people may need a certain level of political awareness to impose audience costs on presidents who back down during crises.^{Footnote 77}

This experimental design is tailored to address how Donald Trump’s foreign policy behavior is evaluated compared to past presidents. We focused our attention on audience costs because voters’ disapproval for leaders who renege on military threats is one of the most consistent findings on public attitudes towards foreign policy, and therefore, an ideal place to test whether Donald Trump is unique from past presidents.

Our design complements and improves upon existing studies in a number of ways. In traditional audience costs studies, an unnamed U.S. president (“The President”) takes one of two actions: (1) he stays out of the conflict or (2) he issues a threat but subsequently backs down.^{Footnote 78} By their nature, the presidents included in these vignettes are never specified, which prevents researchers from knowing with certainty whether respondents are thinking of the current president, a past president, or are genuinely entertaining the hypothetical and not thinking of a specific president at all. Moreover, the public image and reputation of these hypothetical leaders is never filtered through—and shaped by—the gauntlet of partisan elites, the media, election campaigns, and the mass public, as would certainly happen in practice. If, in the real world, partisanship affects the way in which citizens perceive and react to leaders’ policy decisions, experiments that exclude contextual information on partisan cues might artificially amplify the effect of audience costs.^{Footnote 79}

Our design enhances the external-validity of the treatment (varying the partisanship of the president) in two key ways. First, by using the names of presidents, we more closely approximate the American context in which presidents are viewed simultaneously as individuals and partisans. In our experiment, we specifically used the names of presidents Barack Obama and Donald Trump because, in the United States, presidents vary in the extent to which they are perceived as representatives of a party, as well as in the extent to which they personally evoke partisan polarization among the public.

Second, the timing of our experiment allows us to vary the president’s identity and partisanship without introducing an implicit confounding treatment: real versus hypothetical—or current versus former—presidents. In 2012, Levendusky and Horowitz varied partisanship in their survey vignette by saying either “the Democratic President” or “the Republican President” issued a threat. But at the time of their survey, one of those treatments was factual, while the other was nonfactual and it is possible that respondents evaluate factual and nonfactual crisis scenarios differently.^{Footnote 80} By fielding our experiment during the 2016 presidential transition—after Donald Trump won the election but before he took office—we take a modest step toward disentangling the effect of partisanship from the effect of factual versus nonfactual information. Because respondents could feasibly imagine either Barack Obama or Donald Trump as the incumbent president at the time, we maximize the real-world plausibility of priming respondents with the names of two polarizing presidents, from two different parties.

We should, however, note that a few experiments do include leader-specific information in their treatments, bringing in former presidents or past foreign policy crises.^{Footnote 81} Yet the results of these experiments may be confounded by the influence of partisan media or historical memory. Incumbent presidents are polarizing figures,^{Footnote 82} but research shows that past presidents tend to fade from the memory of U.S. citizens—this holds even for relatively recent presidents.^{Footnote 83} For instance, the percentage of Democrats who have a favorable view of former President George W. Bush jumped from 10% in 2009 to 54% in 2017.^{Footnote 84} Comparing Donald Trump to a former president, such as George W. Bush, could lead us to overestimate his uniqueness. The two presidents may share similar titles, but they aren’t necessarily comparable. Consequently, using contextual information that is specific to past events or presidents cannot tell us if Donald Trump’s current behavior is actually being evaluated differently from the way past presidents were evaluated while they were still in office.^{Footnote 85}

Discussion, Limitations, and Future Research

In support of our hypotheses, these results underscore that individuals’ partisan identity does not moderate the magnitude of audience costs in the Trump era. Republican and Trump voters punish Donald Trump and Barack Obama relatively equally for backing down in a foreign policy crisis when they do not provide a justification for doing so. The findings provide support for the notion that voters are inclined to evaluate audience costs scenarios in terms of national rather than partisan issues, and that even atypical leaders like Donald Trump are not immune from censure from their supporters.^{Footnote 92} Our results extend recent findings in the international reputation literature, which show that leaders develop reputations for resolve on their own, and that leaders who fail to back up assertive statements of resolve incur reputation costs among international audiences.^{Footnote 93} We find evidence of a similar effect among domestic audiences when it comes to the issuance of an international threat, whether by Donald Trump, Barack Obama, or a generic president.

We also demonstrate that presidents can decrease the magnitude of audience costs by justifying their decisions to back down in a crisis, and that individuals with low levels of political awareness may approve of the president regardless of inconsistencies in foreign policy behavior. These findings support the conclusion by Michael Tomz that the political fallout from making empty threats is concentrated among the most politically active—and arguably the most influential—segments of the population.^{Footnote 94} What this means, more concretely, is that Trump’s impulsive foreign policy behavior has the potential to hurt his approval among some of his base.

Still, our study is not without limits. The believability of our partisan treatments varying whether Donald Trump or Barack Obama was president in the scenario—required us to field this experiment during the 2016 presidential transition, before Trump had taken office and dealt with a real crisis. As a result, our experimental scenario was not grounded in any particular crisis, which raises concerns of external validity since the public may evaluate President Trump differently during genuine crisis situations. To address this concern, albeit imperfectly, Table 8 of online appendix D restricts our analysis to the 66% (n=975) of respondents who thought Russia was the attacking country in the scenario, as the vignette text largely resembles Russia’s annexation of Crimea from Ukraine. Our results are robust even when models are restricted to the subsample of respondents who specifically thought about Russia and the Crimea crisis. This post hoc check for heterogenous treatment effects reduces concerns that crisis-specific confounding may be biasing our results. Yet future research would make a valuable contribution to the literature by specifically tailoring the framing and timing of audience costs experiments to investigate whether respondents treat real (versus hypothetical) and ongoing (versus former) crisis scenarios differently.

A second limitation of our research design concerns the comparability of President Obama and President Trump. At the time of our experiment, Donald Trump had yet to assume the Oval Office. While Barack Obama had a clear record of foreign policy behavior and an established reputation on the basis of that behavior, Donald Trump did not. This was a necessary byproduct of our decision to field the experiment during the presidential transition when respondents could feasibly imagine either Donald Trump or Barack Obama as the president, maximizing the plausibility of our partisan treatments. It is, however, possible that the public updates the way in which it evaluates foreign policy decisions in light of leaders’ past behavior in international affairs. If this is the case, our finding that the public does not evaluate Donald Trump differently than Barack Obama would be specific to the period in which President Trump was a new leader, without an established reputation and track record in foreign affairs.^{Footnote 95} Still, we do not expect this research design limitation to affect our conclusion that partisanship fails to moderate audience costs for Donald Trump (a unique and new leader), Barack Obama (an incumbent), or “The President” (a generic leader). But we acknowledge the possibility that, over time and in reaction to Trump’s behavior on the international stage, partisan polarization could grow to have a greater effect on how the public evaluates President Trump’s foreign policy. Accordingly, it will be important for scholars to continue to question and reinvestigate our findings as Donald Trump’s presidency progresses.

Critics might also wonder whether partisanship is the only cue, or even the dominant one, that respondents receive from the “President Trump” treatment condition. Donald Trump’s populist appeal may affect individuals’ approval for him in ways unrelated to partisanship. President Trump’s rhetoric of identity politics, anti-elitism, and pro-nationalism clearly resonates with his supporters.^{Footnote 96} If anything, Trump’s brand of populism suggests that identity politics—abstract notions of who or what Donald Trump represents—should matter more to his supporters than his substantive foreign policy behavior, diluting the extent audiences costs he faces for backing down.^{Footnote 97} Yet our finding that Republicans, Democrats, Clinton voters, and Trump voters equally disapprove of Donald Trump for backing down in a crisis suggests that Trump’s populist appeal does not shield him from audience costs. We do, however, acknowledge that populism may condition the effectiveness of Donald Trump justifying his decisions to back down in a crisis. In our experimental design, the justification that Donald Trump provided appeared relatively plausible. Trump’s populist appeal, however, may mean that his supporters are more willing to accept patently implausible justifications for foreign policy inconsistency, and this possibility is an important consideration for the design of future audience costs experiments.^{Footnote 98}

Donald Trump’s authoritarian tendencies may also be a strong cue baked into the “President Trump” treatment.^{Footnote 99} Such tendencies manifest in Trump’s projection of a strong-man personality, fearmongering, demonization of the opposition, and attacks on the media.^{Footnote 100} While our design does not enable us to isolate this effect from that of partisanship, we posit that Donald Trump’s fearmongering and strong-man style of leadership should actually make him more sensitive to audience costs than other leaders because backing down and lacking resolve undermines these narratives. Moreover, Trump’s demonization of political opponents ought to increase the partisan division in audience costs imposed upon him by further polarizing the public. Our results, however, show Donald Trump is not more sensitive to audience costs, and that there is no in-party versus out-party (or Trump supporter versus Clinton supporter) difference in the magnitude of audience costs imposed on him. Hence, we do not expect that these dimensions of Donald Trump’s authoritarian style of leadership is confounding our partisan treatment effects.

At the same time, Donald Trump’s inclination to disparage the media and discredit unfavorable coverage does pose concerns of external validity. In the real world, the public’s information on crisis situations is filtered through the media, and it is likely that Donald Trump’s supporters are less willing to receive—or trust—news that depicts him as behaving inconsistently or irresolutely. Introducing a mediating role for media sources in audience costs experiments thus would be particularly enlightening given President Trump’s general hostility to the press.

Another limitation of our experiment is that it excludes counter-rhetoric from opposing partisan elites, potentially overstating Donald Trump’s ability to deflect audience costs by providing a justification for backing down. Levendusky and Horowitz, for instance, suggest that “if the President can get elites to support his actions, then the mass public will follow suit,” implying that elite opposition could potentially increase the magnitude or durability of audience costs imposed by the public.^{Footnote 101} Counter-rhetoric challenging justifications for backing down could hold leaders’ feet to the fire, reducing their ability to deflect audience costs. While our experiment only included one piece of rhetoric—the president’s own justification—we expect that conflicting rhetoric from in-party and out-party elites plays an important role in real-world crisis scenarios, and thus, warrants further investigation.^{Footnote 102}

It is also possible that Republicans and Democrats are imposing audience costs for different reasons, even if the aggregate magnitude of audience costs between the two groups is equivalent. Kertzer and Brutger find that conservatives impose audience costs due to concerns with the reputational consequences for inconsistency, while liberals do so because of the negative reputational consequences of both inconsistency and belligerency.^{Footnote 103} We acknowledge that our experimental design cannot neatly disentangle the effect of belligerence from that of inconsistency, though in figure 10 of online appendix C, we use causal mediation analysis to partially subtract out the mediating effect of belligerence costs from the total audience costs that Republicans and Democrats impose on each leader.^{Footnote 104} Most important for our study, both the main results and this robustness check indicate that there is no partisan divide in the total magnitude of audience costs that President Trump incurs for backing down in a crisis. Neither Republicans nor Democrats will let Donald Trump off the hook for backing down in a crisis, unless he provides a justification for doing so. Based on previous research, we expect that the negative reputational consequences of inconsistency play a larger role in driving the results among conservative Republicans. Nevertheless, we save this for future research to address in greater detail.

Supplemental Materials

• Appendix A. Sample Characteristics

• Appendix B. Balance across Treatments

• Appendix C. Robustness Checks

• Appendix D. Hypothetical Vignette Check

• Appendix E.: Survey

To view supplementary material for this article, please visit https://doi.org/10.1017/S1537592718003390