Public response to unilateral action

Unilateral action theory has dramatically changed the way scholars understand the presidency (Moe and Howell Reference Moe and Howell1999; Howell Reference Howell2003). In contrast to the “textbook” understanding of policymaking in American politics, unilateralism contends that presidents are first-movers comparatively free of collective action and agency problems faced by other policymakers. The primary obstacle is the potential for countervailing action from Congress or the Judiciary. In both theory and practice, however, these checks appear relatively weak. ^{Footnote 2}

Most critical responses to this approach contend that the president’s power has been overestimated. Christenson and Kriner (Reference Christenson and Kriner2020) pose the relative infrequency of unilateral action as a puzzle. If the separation of powers is the primary limiting mechanism, and it is frail, then unilateral action should be far more regular and successful than it is in practice. So, perhaps, other veto players in American politics challenge the president and reduce the utility of this way of policymaking. Some argue that political parties in Congress (Chiou and Rothenberg Reference Chiou and Rothenberg2017), bureaucratic actors (e.g. Lowande Reference Lowande2018; Rudalevige Reference Rudalevige2021), or successors (Thrower Reference Thrower2017) moderate the president’s actions.

Other work though emphasizes the role of the public. In a series of studies, Reeves and Rogowski (Reference Reeves and Rogowski2016, Reference Reeves and Rogowski2018, Reference Reeves and Rogowski2022a, Reference Reeves and Rogowski2022b) and Christenson and Kriner (Reference Christenson and Kriner2017a, Reference Christenson and Kriner2017b, Reference Christenson and Kriner2019, Reference Christenson and Kriner2020) investigate individual-level determinants of public support for unilateralism. There is an emerging consensus that the public checks presidential unilateralism. Christenson and Kriner argue that “legacy-minded presidents […] rationally defer taking executive action […] if they believe that the long-term political costs of pursuing an unpopular policy exceed the benefits of doing so” (2020: 7). Reeves and Rogowski write that “public support is no blank check on unilateral presidential powers” and that “Americans hold presidents accountable not only for what they accomplish but also for how they wield power” (2022a: 21). The primary dispute is not over whether the public checks the president, but why. For Christenson and Kriner, partisan loyalties and policy preferences determine whether the public punishes the president. Reeves and Rogowski argue that in addition to these factors, the public penalizes presidents for taking unilateral action because of underlying beliefs about the rule of law and the separation of powers.

These stories are complicated, we argue, by a large theoretical literature that probes the implications of the information asymmetries that mark the relationship between leaders and citizens (Ashworth Reference Ashworth2012). Put simply, presidents have private information. They know the types of actions that would shore up support with key constituencies and gain favor among the voting public. Presidents also possess better information about the likely implementation and impact of various policies.

In contrast, the public is at an informational disadvantage. It is well-known that the American public does not pay much attention to politics (Carpini and Keeter Reference Carpini and Keeter1997); and in the case of unilateral policymaking, the news media does not provide much help. The New York Times, for instance, only mentioned 15 unilateral actions per year from 2001 to 2018 (Christenson and Kriner Reference Christenson and Kriner2020). Moreover, partisan biases among the public and the news media provide filters on the quantity and quality of information on unilateral policymaking, skewing the public’s views on the president’s activities and their impacts on real-world policies.

These asymmetries have important ramifications for presidents’ incentives for unilateral action. If the public is unlikely to learn the true policy impact of a unilateral action, presidents might be incentivized to issue politically popular actions with no intention of actually changing policy (Canes-Wrone, Herron, and Shotts Reference Canes-Wrone, Herron and Shotts2001). For example, presidents might publicly sign a directive to enforce limits on foreign aid for abortions, and actually increase the number of abortions in recipient countries (Brooks, Bendavid, and Miller Reference Brooks, Bendavid and Miller2019). Alternatively, they might form a blue ribbon commission that studies the issue and circulates the president’s position in media coverage but does not change the policy. In these cases, presidents are using their ability to act unilaterally to “show off” or “pander” rather than do what is in voters’ best interests (Canes-Wrone, Herron, and Shotts Reference Canes-Wrone, Herron and Shotts2001; Judd Reference Judd2017).

Existing work typically treats the issuance and outcome of an action as one and the same. However, the successful implementation of an action is not inevitable. President Truman, for example, nationalized the steel industry due to an imminent strike, but the order was overturned by the Supreme Court. More recently, President Obama directed the Bureau of Alcohol, Tobacco, and Firearms to conduct more extensive background checks and issued a directive closing the detention camp at Guantanamo Bay – neither were carried out.

What is the upshot of this for studying the public check? In our view, without understanding both whether presidents, in general, have something to gain from unilateral action (relative to no action) – and if they pay a cost for failed actions – we have an incomplete picture of how their incentives impact citizens’ welfare. Consequently, studying the public check requires understanding how the public reacts to two distinct, latent concepts: unilateral action and policy success or failure. This is what we designed our experiment to do.

Measuring public evaluations of unilateralism

We examine public responses to presidential power by fielding a two-wave panel survey prior to the 2020 presidential election. ^{Footnote 3} In the first wave, subjects were randomly assigned to view a president taking a policy position, lobbying for passage of a bill in Congress, or signing an executive order. In the second wave, subjects were told whether the stated policy objective of the president succeeded or failed.

Presidential accountability depends both on whether the public punishes the president for (1) policies that it disagrees with and (2) dissonant outcomes. Our setup examines several potential manifestations of these dynamics. First, if the public holds general skepticism toward unilateralism relative to other policymaking means, per Reeves and Rogowski (Reference Reeves and Rogowski2022a), they should punish presidents who go it alone over working with Congress or taking a position. Second, when the action then leads to a failure, the public could punish the president. Either would discourage a president from taking action they did not believe would benefit constituents.

The panel design helps us distinguish voters’ beliefs about executive action with and without uncertainty about its outcome. As we have argued, the vignette experiments in past work typically approximate a news release at the announcement of some presidential initiative. But like any government action, presidential unilateralism takes place over time, often in discernible stages. Policies are formulated, announced, implemented, and evaluated. By recontacting respondents to reveal the outcome of the policymaking activity, we can gauge how the public evaluates the announcement of actions with an uncertain outcome, relative to the outcome itself. We summarize our approach in Figure 1 but leave most technical details for the SI.

Figure 1. Unilateral action and policy outcomes survey design. For informational conditions and associated images by topic, president, and outcome, see Tables 1, 2, A.1, and A.2.

There are several important features of this setup that distinguish it from past work. First and foremost, we designed the experiment to examine the effects of unilateralism and policy success, in general. Our informational vignettes vary in the topic addressed, along with the president addressing it. We see presidential unilateralism as a latent treatment, with the potential for many unmeasured confounders. So, following Fong and Grimmer (Reference Fong and Grimmer2019), we randomly varied the vignettes presented to respondents. Specifically, we selected 15 topics (e.g., national monuments on public lands, foreign trade barriers, and public funding of abortion). These policy areas include foreign and domestic policy, and those the president has some legal discretion to change. ^{Footnote 4}

Because of the likelihood of moderating effects by co-partisanship of respondent, the action or position could have been taken by President Trump or Obama. This ensures that the effects of co-partisanship and party identification can be distinguished. For example, for the endangered species condition, the vignettes read

with the working with Congress condition adding

and the executive order condition instead adding

To ensure our treatments were read and received, we also included companion images for each vignette treatment (Table 1). When the public learns about executive action, it is usually accompanied by an image of the president. Imagery appears to play an important role in the public’s understanding of the presidency, especially in media reports (Howell, Porter and Wood Reference Howell, Porter and Wood2020). Howell, Porter, and Wood (Reference Howell, Porter and Wood2017), for example, write their findings “attest to the importance of visual images in public presentations. Not only were the photo and text conditions far more powerful than the text-only conditions […] the photo condition […] was consistently able to overcome the tone of the text” (28). More generally, online survey respondents are known to rely on visual cues. We expect that the topic-specific images will limit the outcome variation attributable to the passage of time between waves, and reduce the chances of null findings attributable to a weak treatment.

Table 1. Photos for the position, congress, and executive order interventions in Wave 1, by President

In addition, by assigning some subjects to view only the position of the president, our study distinguishes between the signal about the president’s policy preferences sent by an action and the action itself. Most prior research compares public evaluations of backing legislation to taking unilateral action (e.g., Christenson and Kriner Reference Christenson and Kriner2020). However, separating the action from the substance of the policy is critical for determining differences in public evaluations of the means of policymaking, relative to support for the policy position itself. This also matters for presidents’ incentives. For example, presidents may be rewarded for appearing to take charge of an issue by acting. Alternatively, if acting and position-taking are treated equivalently by respondents, presidents may be able to shirk from policymaking and focus their attention on signaling their support for politically advantageous policies.

Finally, separating these actions from the results of policy allows us to address the underlying questions posed by theories of accountability. In Wave 2, the treatments inform respondents that the policy change the president supported or acted upon occurred or did not occur. Like the treatment conditions in the first wave, the second wave’s success and failure conditions include companion images (Table 2). We define failure as the status quo policy remaining unchanged, or the absence of the desired liberal or conservative movement. For example, in the endangered species condition, the failure prompt reads

Table 2. Photos for the success and failure conditions in Wave 2, by President

while the success prompt reads

To measure public evaluations, we used three primary outcomes: respondents’ assessment of handling of the policy, respondents’ overall assessment of the president, and respondent vote preference for the 2020 presidential election. For the subsequent analyses, the third outcome is whether the respondent indicated they would vote for Donald Trump. In the SI, we also report same-party vote, which is the intention to vote for the 2020 presidential candidate that shared the party of the president shown in the treatment condition. In the following section, we note when these two specifications differ substantively.

We obtained a large national survey sample from the online vendor Lucid. In general, the sample of 4,350 U.S. citizens of voting age reflects the partisan identification, region, age, gender, and ethnicity of the Unites States (Section A.4 in the SI). Wave 1 was fielded September 7–9, 2020, while Wave 2 was fielded September 16–18. About half of the respondents could be recontacted after the first wave. We found that recontact was not associated with treatment condition or political party (Section A.4 in the SI). Prior to fielding, we preregistered our hypotheses and conducted simulations to ensure the experiments were adequately powered to detect substantively meaningful effects (Section A.2 in the SI). ^{Footnote 5}

How the public responds to actions

Overall, we find that public evaluations of presidential actions were mostly predicted by the set of demographic variables that form policy-related fault lines in American politics (e.g., partisan identification, age, race, and sex). Since we chose to vary contextual factors (i.e., president and topic) along with our treatment conditions of interest, we rely on parametric, linear models that include all informational treatments and demographic covariates to predict evaluations. Because the results are not sensitive to modeling decisions, we summarize our findings with estimates based on logistic regressions and binary versions of the dependent variables in Figures 2, A.3, 3, and A.6, and Tables A.9, A.10, A.13, and A.14. We report additional tabular results with alternative model specifications in the SI. ^{Footnote 6}

Figure 2. The public is mostly indifferent to the way policy is made. Plots simulated marginal effect estimates, based on logistic regressions that include demographic controls; error bars indicate conventional and Bonferroni-adjusted CIs; see Tables A.9 and A.10 for full results.

Figure 2 plots marginal effect estimates for policymaking via Congress and unilateral action – relative to position-taking – simulated based on an observed case approach (Hanmer and Ozan Kalkan Reference Hanmer and Ozan Kalkan2013). There is little evidence of backlash against unilateralism, and perhaps some evidence that it may be beneficial for the incumbent. Specifically, neither overall approval of the president nor approval of the president’s handling of the topic was associated with the means of policymaking. Nonetheless, respondents appeared to identify the treatment. Consistent with Reeves and Rogowski (Reference Reeves and Rogowski2022a), the executive order condition was associated with reduced beliefs that the president respected the rule of law and worked with Congress (Figure A.3). Those beliefs simply did not weigh heavily into overall assessments of presidential performance. Opponents of the president, defined as those that identify with the other major party, tended to rate the president’s handling of the topic about 7 percentage points higher if they worked with Congress, relative to position-taking. But these evaluations were not distinguishable from acting alone, and did not generalize to independents or copartisans.

Most strikingly, viewing the executive order signing is associated with a roughly 3 percentage point increase in the probability of reporting a vote preference for President Trump. Curiously, this effect is driven both by those who saw Obama and Trump vignettes. Figure A.4 shows that the executive order treatment did not increase the likelihood the respondent would vote for the party of the president they saw in 2020. That is, those who saw Obama did not become more likely to say they would vote for Biden.

In short, however, there is little evidence from the initial panel that respondents systematically punish presidents for taking unilateral action. Assessments of performance were either unmoved, improved, or driven mostly by respondents’ priors. These findings also suggest that, in the absence of information about outcomes, the public tends to treat unilateral action equivalently with position-taking. ^{Footnote 7}

How the public responds to outcomes

In contrast to our initial findings, the second wave demonstrates that the public can punish presidents for failing to produce. We estimate respondent assessments as a function of our treatment conditions and controls, while conditioning on Wave 1 outcome values. We condition on Wave 1 values in order to isolate the effect of learning about the outcome of the policy from respondents’ initial assessments of the policy change. In pre-analysis simulations, we found that two other approaches – namely, predicting a change score as a function of Wave 2 treatments, or predicting Wave 2 outcome without conditioning on Wave 1 – resulted in less power. Figure 3 shows that respondents’ evaluations of the president’s handling of the topic took a significant hit when it was revealed that the stated objective of the president’s policy did not occur. Specifically, respondents were about 6 percentage points less likely to approve of the president’s handling of the topic, which is large, relative to the baseline probability of approval of 52.4%.

Figure 3. The public punishes presidents for failing. Plots simulated marginal effect of estimates of failure relative to success; see Tables A.13 and A.10 for full results. Figure A.5 shows that failure did not have a discernible effect on respondents’ support for the 2020 presidential candidate from the same party.

Interestingly, there is no evidence of heterogeneous effects by partisan identification. Members of the opposition, along with copartisans and independents, all tended to have lower approval ratings of presidential performance. This is notable because the symmetric partisan identification of the president in question. Republicans and Democrats may have been assigned to Trump or Obama, but in both instances, they tended to punish the president for failing to get a win. The additional outcomes suggest a mechanism. As Figure A.6 demonstrates, failure led the public to rate presidents’ ability to “get things done” lower by 4 percentage points. The public punished the president for failing to achieve a stated objective, regardless of their party or the means they used to do it. ^{Footnote 8}

Importantly, the finding leaves open an avenue for presidential accountability. Respondents’ assessments of presidential performance under the first wave and the second wave success condition are indistinguishable – which suggests that the initial announcement of unilateral action is treated as a “win” by default. Follow-up information on outcomes, however, can alter public assessments.

There are two important caveats. First, the penalty was limited to one question – topic handling – which may or may not be sufficient to inform the president’s thinking. The strongest argument that it is sufficient is that job approval and voting outcomes may be the aggregation of many such “handling” questions for many different topics. Repeated failure to implement policies, therefore, could mean reduced overall job approval ratings and less support at the ballot box. It is rare for any single issue to move something as entrenched as job approval or vote choice.

Second, this accountability mechanism is conditional on the informational environment the public encounters when evaluating presidential performance. In other words, it is not enough that the public punishes the president for dissonant outcomes under idealized experimental conditions, there must be a plausible case that these dissonant outcomes are observable in practice. We discuss this latter point in the next section.