Public support for withdrawal from international organizations: Experimental evidence from the US

Abstract

The United States has helped create and lead many international organizations (IOs). Yet in the last six years, the US announced its withdrawal from several IOs including the World Health Organization, UNESCO, and the Universal Postal Union. Do Americans care about US withdrawals from IOs? When do Americans support withdrawing from IOs and support candidates who propose this? We argue that Americans’ support for multilateralism tends to divide along party lines, and that IO withdrawal can activate those preferences. We also argue that framing an IO withdrawal as benefiting US national interests can make Americans more likely to favor IO exit. Data from four US survey experiments during the 2016–2020 Trump administration support these arguments. Democrats tend to oppose IO withdrawals while Republicans tend to support them. Further, results show that IO withdrawal (and how it is framed) affects candidate choice and policy support. This suggests that announcing IO withdrawal can be used to rally domestic electoral support. Still, the data also show that a large proportion of the US public values remaining in IOs, even when IOs are imperfect or challenging. In these cases, we note that sunk cost fallacies, status quo bias, and loss aversion may pose friction points for supporting withdrawal. Our findings have important implications for research on public opinion about international cooperation, backlash against IOs, and their life cycles.

Since the global order was established after World War II, the US has been an important founder and supporter of many international organizations (IOs). US leadership in IOs has been important – yet the US has also been the most frequent withdrawer from IOs (von Borzyskowski & Vabulas, 2019). Successive US governments have raised the possibility of withdrawing from IOs they created or joined (Birdsall & Sanders, 2020). For example, the Carter administration withdrew from the International Labor Organization (ILO) in 1977 (ILO, 1975) and the Reagan administration withdrew from UNESCO in 1984 (Washington Post, 1984). During his 2016–2020 administration, President Trump announced that the US would withdraw from several IOs including the World Health Organization (WHO), UNESCO, and the Universal Postal Union (UPU), and publicly threatened to withdraw from NATO, NAFTA, and the World Trade Organization (WTO) (Davis, 2021; von Borzyskowski & Vabulas, 2023).

It is therefore important to understand Americans’ preferences for IO withdrawal at a time when the US government was announcing it would remove itself from major IOs. While US politicians announce and justify IO membership withdrawals to the American public, underscoring their political salience, we know little about the conditions under which the US public supports or opposes IO withdrawals. Current research provides limited insights for two reasons. First, much of the existing IO public opinion scholarship focuses on what drives support for joining and participating in IOs (Bearce & Jolliff Scott, 2019; Edwards, 2009; Inglehart, 1970). Nonetheless, public opinion related to IO withdrawal may not simply manifest as the opposite of supporting IO membership because withdrawal needs to overcome friction points such as sunk cost fallacies, status quo bias, and loss aversion. We thus contend that IO withdrawal is a separate analytical category to opposing IO membership. Second, previous scholarship is largely limited to Brexit and Euro-skepticism (De Vries, 2018; Walter, 2021a). While these previous studies have examined European and British public opinion related to Brexit, we do not know whether findings generalize to US public opinion or to other IO withdrawals beyond Brexit. Understanding the preconditions for hegemonic withdrawal can also contribute to our understanding of the life cycles of IOs (Gray, 2024) and of IO vitality more broadly (Gray, 2018). Because withdrawals can impact the system of international law (Schmidt, 2024, this issue; Eilstrup & Verdier, 2024, this issue), it is important to understand support for them among mass publics.

This paper asks: under what conditions does the US public support or oppose IO withdrawals and what factors shape those opinions? Moreover, does IO withdrawal policy matter for choosing political candidates? First, we argue that US public opinion about IO withdrawal is based on Americans’ support for – or opposition to – multilateralism, which divides along partisan lines (Milner & Tingley, 2013, 2015). Announcing an IO withdrawal can activate partisan preferences on multilateralism even when voters know little about the IO itself. Thus, Republicans (rather than Democrats) should be more likely to favor withdrawing from IOs, all else equal.

Second, we argue that voters can become more supportive of an IO withdrawal when exit is framed as an action that supports US national interests. This includes justifying an IO withdrawal as supporting US national security, minimizing US spending abroad, or reducing the harms of globalization. Framing an IO withdrawal as affecting US national interests may be influential (Chaudoin et al., 2010) because it can raise withdrawal to the level of a national crisis where voters may rally ‘round the flag (Evers et al., 2019; Oneal & Bryan, 1995). This framing may also help counterbalance the friction points/perceived costs associated with withdrawal (including sunk cost fallacies, status quo bias/loss aversion, and a sense of loyalty) and instead emphasize that gains are possible from an IO withdrawal when it directly benefits the US. A national interest frame has been shown to affect EU voters’ assessments of EU disintegration because voters know and care more about national topics than EU topics (Anderson, 1998; De Vries, 2018; Hobolt & de Vries, 2016; Kritzinger, 2003; Rohrschneider, 2002; Sánchez-Cuenca, 2000), lending credibility to a similar mechanism affecting US voters. Moreover, the US national interest frame can affect voters’ choice of candidates because IO withdrawal can act as a heuristic that retrospective voters¹ recall when they are evaluating politicians at the ballot box; IO withdrawal can be touted as a foreign policy accomplishment, which politicians can invoke during the next election campaign.

To test our two arguments – that (1) preferences for IO withdrawal generally divide on partisan lines but that (2) politicians can move the needle on support for IO withdrawal by framing it related to US national interests – we fielded four pre-registered survey experiments. Two conjoint experiments assess whether Americans care about candidates’ IO withdrawal announcements during election campaigns and how much IO withdrawal policy matters for their choice of candidates compared to other candidate policies and characteristics. Two vignette experiments test whether framing IO withdrawals as addressing US national interests changes public opinion on US withdrawals.

The results support our arguments and provide the first experimental evidence on what drives US public opinion about withdrawing from IOs. The conjoint experiments show that voters care about IO withdrawals. Moreover, IO withdrawal policy can substantively influence respondents’ choice of political candidates. As expected, these effects are distinctly partisan: candidates who advocate withdrawing from imperfect IOs are likely to garner support from Republicans but lose support among Democrats. Second, the vignette experiments show how support for IO withdrawal can change: framing IO withdrawals as advancing US national interests can help increase Americans’ support for IO exit. This effect is mainly driven by independents. We also find some support that it is the US national interest frame and not just any frame by comparing to a placebo. Together, we show that Americans’ preferences regarding multilateralism can be mobilized by invoking IO withdrawal and candidates can increase support for IO withdrawal by framing exit as helping US national interests. Still, our findings also show another important point: even when IOs create challenging environments for the US, a large proportion of the US public still values remaining in IOs. Even if the 2016–2020 Trump administration advocated a “withdrawal doctrine” (Haass, 2020), the US public was not capricious about IO withdrawal.

Our findings are important because while previous studies have shown that the public sometimes evaluates realist foreign policy actions (i.e. the use of force) from a US national interest perspective (Drezner, 2008), withdrawal from IOs is a different foreign policy action that has not previously been studied in the US context. Indeed, a key takeaway is that the public can rethink the costs and benefits of withdrawing from IOs when it is framed as being in the US national interest. This is a powerful finding since much of the IO literature emphasizes the cooperation gains from membership. Our paper is also important for understanding domestic electoral politics. While surveys have shown that many US voters do not know much about IOs,² that “few in the public have opinions on IOs,” (Guisinger & Saunders, 2017: 430) and that US public opinion plays little constraining role in foreign policy,³ other scholars see public opinion as playing an important role because it can limit foreign policy actions (Aldrich et al., 1989; Baum, 2004; Campbell et al., 2003).⁴ We therefore bridge international relations scholarship—on when and why countries may not want to be a part of IOs—and American politics, showing the interconnectedness of framing, US public opinion, candidate choice, and IO withdrawal.

Our study is particularly important as many are trying to understand how US retreats from IOs may affect international cooperation (Cooley & Nexon, 2020; Copelovitch et al., 2020; Drezner, 2017; Eilstrup-Sangiovanni & Hofmann, 2020; Lake et al., 2021; Walter, 2021c). It aligns with research showing that voters consider IOs important (Jurado et al., 2022; Dellmuth & Tallberg, 2015; Brutger & Clark, 2023; Council on Foreign Relations, 2019) and that how the public legitimizes IOs is crucial for explaining public sentiment (Buchanan & Keohane, 2006; Dellmuth, 2018; Dellmuth & Tallberg, 2021; Dellmuth et al., 2022; Rauh & Zürn, 2020; Tallberg & Zürn, 2019). While studies show that elite framing can help voters assess IO legitimacy, we show that framing related to US national interests can also shape public support for IO withdrawal. Our research also echoes recent scholarship showing that elites can blame IOs for crises in order to gain domestic support (Schlipphak, Meiners & Kiratli, 2022).

1 Withdrawal, public opinion, and international organizations

IO withdrawal occurs when a member state voluntarily removes itself from all contractual obligations and legally terminates its membership. Withdrawal is not simply the opposite of joining. Previous work has shown that only a few drivers of IO accession also influence IO withdrawal; these include a state’s preference alignment with other members and democracy levels (von Borzyskowski & Vabulas, 2019). States withdraw by providing notice to the secretariat, then waiting an IO-specified timespan to execute the exit.⁵ While IO exits happen across time and states in different world regions, the US has been the most frequent IO withdrawer since 1945.⁶ Figure 1 shows that Democratic administrations in the US have withdrawn almost as often as Republican administrations (seven vs. six times) in the post-WWII era.⁷ Historically, leaving IOs is not a new phenomenon.

Fig. 1

Withdrawal Frequency across US Administrations, 1945–2022

We examine US public opinion toward IO withdrawals, focusing on the 2016–2020 Trump administration, given the increased salience and discussion of IO exit during that time (Cooley & Nexon, 2020; Drezner, 2020, 2022 (384), 2017; Haass, 2020; Krieger, 2019). Some observers have noted that this Trump-era of IO exits and threats to withdraw signified a sea-change in commitment to multilateral institutions after a steady march toward globalization, integration, and cooperation (Shukla, 2020), and contrasts with long-term patterns of increasing membership in IOs since World War II (Drezner, 2020, p. 397; Wright, 2016). While the US is only one country, it has a unique role, status, and power in IOs, and has used this to reinforce its hegemony.

The 2016–2020 Trump administration arguably pursued a “withdrawal doctrine” in foreign policy (Haass, 2020). During this time, the US withdrew from two formal IOs and threatened to leave many others; it also raised the issue of withdrawal in speeches to domestic audiences. In the autumn of 2016 alone, Presidential candidate Trump mentioned withdrawing from IOs in at least eight campaign speeches. For example, on 25 October 2016, Trump said “We will renegotiate NAFTA or withdraw from the deal to get a much better one for our workers and our companies…”⁸ At the end of his first term, in his “Farewell Address to the Nation” on 19 January 2021 (and what some would term his first campaign speech for 2024), Trump also discussed withdrawing from IOs, saying "We reclaimed our sovereignty by standing up for America at the United Nations and withdrawing from the one-sided global deals that never served our interests. And NATO countries are now paying hundreds of billions of dollars more than when I arrived just a few years ago.”⁹

Despite the regular pattern of IO withdrawals over time and mentions of IO withdrawal in candidates’ speeches, we know little about US public opinion toward IO withdrawals. What affects public opinion on IO withdrawals, and do voters punish or support leaders who announce they will leave IOs? Thus far, many studies have focused on a single case: the United Kingdom withdrawing from the European Union (EU) (Walter, 2019, 2021a, b; De Vries, 2017; De Vries et al., 2021; Jurado et al., 2022). Studies show that Euro-skepticism (Anderson, 1998; De Vries, 2018; Kritzinger, 2003; Rohrschneider, 2002; Sánchez-Cuenca, 2000) was a key driver of Brexit which can also affect public opinion in other EU countries through encouragement and deterrence effects (Hobolt et al., 2022; Walter, 2021a, b). Several studies have also used survey experiments to examine how framing the costs and benefits of EU membership can affect support for EU membership (De Blok & De Vries, 2023; Goodwin et al., 2020; Hobolt et al., 2022).

Beyond Brexit, Bearce and Jolliff Scott (2019) show that “non-support” for IOs has increased and that regardless of education level, less-skilled US citizens express more negative views about IOs . Further, less than one third of the US public trusts IOs like the World Bank and World Trade Organization (WTO) (Mutz, 2020). Surveys by think tanks (Smeltz et al., 2019, 2018, 2017, 2016; Council Foreign Relations, 2019; Pew Research Center, 2016) and scholars (Kiratli, 2022) assess attitudes towards NATO, NAFTA, the UN, and the EU, but they do not ask respondents’ when and if they would support leaving these IOs nor whether certain justifications would make withdrawal more palatable.

We contend that non-support for IOs and withdrawing from IOs are conceptually different. People (and governments) may dislike IOs but nevertheless favor staying in them because IOs generate some benefits once states are in. Moreover, the public may perceive additional costs of IO withdrawal (beyond not joining in the first place) due to friction-points they associate with leaving. These include sunk cost fallacies (North, 1990; Williamson, 1979), status quo bias/loss aversion (Alesina & Passarelli, 2019), and a sense of institutional loyalty (Hirschman, 1972). Sunk cost fallacies may arise because the public may know that their country invested significant resources before and after joining the IO and thus they may remain tied to the IO (Copelovitch & Ohls, 2012; Epstein & Sedelmeier, 2008; Gowa & Kim, 2005; Holzinger & Schimmelfennig, 2012; Pelc, 2011; Schneider & Urpelainen, 2012; Tversky & Kahneman, 1991). Status quo bias may lead the public to be skeptical of withdrawal because institutions carry a sense of inertia once they are set in motion. Relatedly, theories of loss aversion (Tversky & Kahneman, 1991) mean that once IO membership is the status quo, voters may be reluctant to leave (Jervis, 1992). Loyalty, which relates to Hirschman’s seminal work on “exit, voice, and loyalty” (Hirschman, 1972), underscores that members may want to remain in an IO to make improvements from within (Hirschman, 1972, p. 79). As a result, “loyalty acts as a break on the decision to exit” (Hirschman, 1972, p. 88).

In addition, studies show that the public cares about IOs and suggest that public opinion toward IOs can be malleable. For example, information about institutional performance shapes confidence in the UN (Dellmuth & Tallberg, 2015; Tallberg & Zürn, 2019), information about procedural and outcome quality affects citizens’ evaluations of international cooperation (Bernauer et al., 2020), information about financial benefits of US leadership can increase public support for IOs (Brutger & Clark, 2023), and information about IMF conditionality can reduce public support for the IMF (Handlin et al., 2023), emphasizing that citizens care about what IOs do for their nation. While these studies show that context and framing shape how and when the public cares about IOs, we still do not know what shapes public opinion about leaving IOs.

Perhaps the best evidence underscoring that the public may care about IOs is that political elites regularly communicate IO strengths and weaknesses to the public (Davis & Pratt, 2021; Dellmuth & Tallberg, 2021). Politicians frequently blame IOs for bad policies (Heinkelmann-Wild et al., 2020) and endorse IOs if this can help them achieve policy objectives (Chapman & Reiter, 2004).¹⁰ For example, politicians seek UN Security Council authorization to increase the perceived legitimacy of US military action (Voeten, 2005; Chapman, 2009; Grieco et al., 2011; Thompson, 2006).¹¹ Some governments have even posed IO withdrawal questions as part of referendums, emphasizing the importance of public opinion in this foreign policy realm (Walter, 2021a, b). We do not argue that policymakers make IO decisions based on public opinion, but publicizing their foreign policies regarding IOs can help politicians gain (or lose) public support among some groups.

2 IO withdrawal policy and framing

When does the US public support IO withdrawals and candidates who propose this? We present our theoretical expectations in several steps starting with partisan baselines, moving to whether IO withdrawal policy matters, whether the framing of that policy matters, and finally looking at effects on independents. All hypotheses we discuss and test are pre-registered; Table A1 in the Appendix A provides an overview.¹²

First, we argue that Americans’ attitudes about IO withdrawal are largely driven by their preference towards multilateralism and that this divides along partisan lines: Republicans nowadays should be more supportive of IO withdrawal than Democrats. We base this argument on post-1995 surveys that show Republicans are less supportive of multilateralism whereas Democrats are more likely to support multilateralism and international organizations (Smeltz et al., 2022b, p. 6 & 21; Milner & Tingley, 2013, 2015; Broz, 2011; Friedhoff, 2021). Republicans tend to be less favorable toward multilateralism because they fear losing control over foreign policy issues (Milner & Tingley, 2013). They prioritize preserving the country’s freedom of action and are generally opposed to multilateral initiatives that they believe would undermine sovereignty without delivering sound policy outcomes (Busby et al., 2012). Democrats, on the other hand, tend to be more favorable toward multilateralism because they believe that the US needs to enlist other nations and cannot solve most problems alone (Busby et al., 2012).¹³

Since IOs institutionalize multilateralism, we contend that individuals may extrapolate their general attitudes about multilateralism to their support for (or opposition to) IO withdrawal. IO withdrawal backs the US away from multilateralism. We therefore expect that baseline preferences for IO withdrawal differ between Republicans and Democrats. Republicans should be more likely to favor withdrawing from IOs (to denounce multilateralism) while Democrats should be more likely to favor remaining in IOs (as an institution to promote multilateralism):

• Hypothesis 1 (Divergent partisan baselines): Republicans should be more supportive of IO withdrawal than Democrats.

Moreover, we argue that IO withdrawal policy should matter to citizens’ choice of political candidates, not just in isolation (which is not a reasonable assumption in the real world) but when considered in a multidimensional policy space—that is, when potential voters weigh several attributes of political candidates. The reason IO withdrawal might matter at the ballot box and not just as a stand-alone question is that research shows that the public is more likely to hold their government accountable for actions at the international level when the decisions are politicized at the national level (Schneider, 2018). We surmise – but do not test – that IO withdrawal might be salient for voters because its nature (a country is either in or out) makes it easily retrievable from memory (Aldrich et al., 1989; Ottati & Wyer, 1990). Voters retrospectively consider politicians’ past actions (Aldrich et al., 1989; Ottati & Wyer, 1990), making this foreign policy decision easy for voters to remember, and thus ripe for later priming (Iyengar & Kinder, 1987; Zaller & Feldman, 1992). For example, during an election campaign, voters often want to know “what have you done for me lately” (Shepsle & Weingast, 1981). In contrast, other foreign policy actions may be more nuanced and thus less readily retrievable when voters try to recall politicians’ “achievements”.¹⁴

If Hypothesis 1 is true and partisan baselines indeed diverge, then IO withdrawal policy may not affect candidate choice on average because the diverging opinions of partisan groups may cancel each other out. Thus, the average effect across groups may be indistinguishable from zero (as we note in the pre-registration).

Second, beyond partisan baselines, we argue that framing IO withdrawals as supporting US national interests boosts public support for IO withdrawal and for the politicians supporting it. This argument builds on research in American and comparative politics¹⁵ as well as international relations (Guisinger & Saunders, 2017) about how framing can shape people’s preferences (Druckman, 2011). Elites can strategically use cues to activate certain frames that individuals hold in memory. This provides heuristics that influence people’s preferences, aiding decision-making by simplifying information overload (or information shortfalls). For example, research shows that elite framing can help voters assess the legitimacy of IOs (Dellmuth, 2018; Dellmuth & Tallberg, 2021; Nielson et al., 2019). These findings provide the foundation for understanding how political leaders justify why IO membership or its termination might benefit the country.

We argue that framing an IO withdrawal as helping US national interests should appeal to respondents more than no framing (pure control) or a non-US national interests frame (such as bureaucratic reasons). This argument that a national interest frame may nudge more support for IO withdrawal builds in part from research on public support for European integration and disintegration. These surveys show that heuristics such as perceptions of national interest drive public support for EU membership (Anderson, 1998; De Vries, 2018; Hobolt & de Vries, 2016; Kritzinger, 2003; Rohrschneider, 2002; Sánchez-Cuenca, 2000).

Framing IO withdrawals around US national interests may increase public support for the policy and the candidate proposing it because it is popular, invokes a sense of crisis, and/or because it signifies domestic benefits. We unpack each of these arguments. First, reasoning related to US national interests is broadly popular among the public. Surveys show that a supermajority of Americans supports policies on behalf of “US national interests” (Evers et al., 2019, p. 438) and that broad support for topics related to US national interests, including globalization and US national security, is stable over time (Smeltz et al., 2016, 2017, 2018a, b; 2019; Pew Research Center, 2016). Second, the US national interest frame may invoke a sense of crisis that can create a rally ‘round the flag effect.¹⁶ In crisis scenarios presented to voters, Evers et al., (2019, p. 438) show that when the president “justifies his decision to back down as being ‘in America’s interest,’ he no longer incurs an approval hit among Democrats or Republicans.” A further reason why the US national interest frame might nudge more support for IO withdrawal and the candidates advocating it is that the frame might highlight how IO withdrawal may be domestically beneficial to voters. Without this frame, the public may think more about the friction-points they associate with leaving that we mentioned above (sunk cost fallacies, status quo bias/loss aversion, and institutional loyalty).

Indeed, President Trump justified his withdrawal from the UN Human Rights Council as acting “in the national interest.”¹⁷ Trump also linked withdrawal from the Universal Postal Union (UPU) to US national interests amidst the rising power of China (Guardian, 2018; New York Times, 2019), which garnered widespread attention. As one indication of public interest, online search popularity for the UPU peaked shortly after Trump threatened to exit.¹⁸ President Trump also cited a national crisis due to China’s increasing influence and the negative ramifications on US national interests when he announced the US would withdraw from the World Health Organization in 2019.¹⁹

In sum, US citizens should be more likely to support IO withdrawal if it is justified as being in the US national interest. A politician may explicitly reference the phrase “US national interests” or invoke this idea in other ways. Three manifestations of US national interests stand out as previous framings:²⁰ justifying IO exit with the domestic costs of multilateralism, addressing national security, or curbing domestic problems related to globalization. Figure 2 illustrates that over the last hundred years, more than 90 percent of the justifications for US withdrawal decisions and threats have centered on US national interests: 44 percent are tied to the costliness of international cooperation, 41 percent relate to concerns over national security or sovereignty, and 12 percent tied withdrawal to the negative effects of globalization.²¹ Nonetheless, in some cases, US politicians have announced IO withdrawals without any justification (3 percent of cases).²² In addition, politicians can use alternative framings not related to US national interests including focusing on the IO not achieving its mandate or mission, but such alternatives are rarely used.

Fig. 2

Framings of US Withdrawals and Withdrawal Threats (1919–2022)

Therefore, we expect that these three manifestations of a US national interest frame increase public support for IO withdrawal. Because we view each of these three frames—reducing the costs of multilateralism, increasing national security, and reducing the challenges of globalization—as manifestations of US national interests, we have no strong expectation that any one of these three frames would work better (as noted in the pre-registration).

• Hypothesis 2a (Average effects of framing vs control): Framing IO withdrawal as related to US national interests (versus no framing) should generate higher public support for IO withdrawal, the candidate’s achievements, and a higher likelihood of voting for the candidate.

We expect that the effect is due to the “US national interest framing” rather than offering any explanation for IO withdrawal. Research in political psychology notes that offering any explanation can be more persuasive than no explanation (Langer et al., 1978), so we aim to distinguish the effect of our hypothesized framing not just from no framing (the pure control) but also from a placebo framing (Porter & Velez, 2021). While the pure control allows comparing the effect of a treatment text versus no treatment text, the placebo condition holds the treatment mode constant (added text) but provides information not relevant to the outcome (like a bureaucratic reason instead of US national interests). Nonetheless, comparing a treatment to a placebo instead of (or in addition to) a pure control usually leads to conservative effect estimates because the treatment mode is identical, statistical power is lower (due to including a third group), and because selecting a non-relevant placebo can be challenging.

• Hypothesis 2b (Average effects of framing vs placebo): Framing IO withdrawal as related to US national interests (versus a placebo framing) should generate higher public support for IO withdrawal, the candidate’s achievements, and a higher likelihood of voting for the candidate.²³

The counterfactual is that framing IO withdrawal around US national interests may reduce (instead of increase) support because IOs are supposed to be venues of international cooperation; justifying withdrawal as helping national interests may therefore strike some as being self-centered. A US national interest framing is also not costless. It may generate audience costs for the leader if withdrawal is justified by national interests but then the leader does not follow through with withdrawal. Levendusky and Horowitz (2012, p. 326–7) note that voters assess audience costs – whether they will punish a President for backing out of a foreign policy commitment – in terms of national interests. This logic is enhanced if framing makes withdrawal sound important for the country’s survival/prosperity.

Finally, we hypothesize that independents²⁴ should be more likely to be influenced by IO withdrawal policy than partisans. Individuals with strong partisan affiliations should be less likely to be swayed because their attitudes are more entrenched and they are also more informed. A candidate’s policy on IO withdrawal may therefore matter little for partisan voters beyond the candidate’s party.²⁵ For example, Republican voters may mostly support Republican candidates, regardless of IO withdrawal policy, and Democratic voters may mostly support Democratic candidates (Green et al., 2004). Voters may infer the candidate’s party directly from the party label or indirectly from positions on domestic policies. In contrast to Republican or Democratic respondents, independent or other respondents (whom we group together under the label "independents") are less affected by dominant party cues and we argue that they might thus be more affected by the IO withdrawal policy of the candidate. Moreover, polls show that non-partisans have the least knowledge of international affairs (Pew Research Center, 2022, p. 13–14). A frame that provides justification for IO withdrawal may provide independents with more information about international affairs and this may be what nudges them more than partisans. Consequently, we expect the following:

• Hypothesis 3a (Independents—policy): The effect of IO withdrawal policy should be stronger for independent respondents than Republican and Democratic respondents.

• Hypothesis 3b (Independents—framing): The effect of framing related to US national interests (versus no framing) should be stronger for independent respondents than Republican and Democratic respondents.

We recognize that voters form preferences and make decisions about alternatives in a multidimensional way. Below, we thus outline how we lean on conjoint survey experiments to randomize information about different candidate policies and demographic characteristics, which helps us assess whether/how much IO withdrawal policy matters relative to other candidate characteristics, before testing the framing effects.²⁶

3 Testing whether/how IO withdrawal policy matters

To assess whether and how (much) the US public cares about candidates’ IO withdrawal policies, we conducted two conjoint survey experiments using a nationally representative sample (with quotas on age, gender, race,²⁷ region, and partisanship on Lucid) in the US just before the 2020 US national election. The conjoint design explicitly mirrors previous designs related to US public opinion, candidate choice, and foreign policy (Tomz & Weeks, 2020; Tomz et al., 2020). It allows us to disentangle whether respondents care about issues like IO withdrawal or solely make candidate decisions based on other characteristics like domestic policy or demographics (e.g. gender). This is useful given our theoretical interest in comparing how and if IO withdrawal policy matters among the multiple other factors that contribute to voters’ choices. Moreover, the conjoint design allows us to test how IO withdrawal policy matters for candidate choice: whether it hurts or helps candidates on average, and whether the effect depends on the partisanship of voters.

Before administering the conjoint experiment, we measured respondents’ demographics and political attitudes. We first measured respondents’ preferences toward two domestic policies (economic and education)²⁸ by describing what each policy entails, what a small or large government role would look like, and then asking whether respondents preferred the government to take no, a small, medium, or large role for each policy. We took a similar approach to measure attitudes about IO withdrawal policy, explaining what international organizations are (like the media frequently does when discussing IO withdrawal), giving examples of IOs so that the notion of an IO was specific and not just hypothetical. These examples included the North Atlantic Treaty Organization (NATO), World Trade Organization (WTO), and World Meteorological Organization (WMO). This cross-section of regional and global organizations included one security, one economic, and one standards-based entity, which represent the types of IOs from which the US announced potential withdrawals during the 2016–2020 Trump administration. We then noted that sometimes IOs fall short. This balanced information about IOs – that they are a useful means to pursue US policy (positive) but that they sometimes fall short (negative) – allows a measure of baseline support without priming respondents in only one direction.

We then noted that the US can choose to withdraw from or remain in such imperfect organizations and asked respondents, “Which approach to such international organizations do you prefer?” We asked the question concerning withdrawal about IOs in general because our theory relates to voter’s broader preferences about IOs not just one specific organization. Answer options were definitely withdraw (9%), probably withdraw (26%), probably remain in (40%), and definitely remain in (25%). These findings show that in the aggregate, Americans support remaining in IOs, even when they are imperfect (65%). Only 35% of Americans leaned toward withdrawing. This is an important finding in itself, given the salience of IO withdrawals at this time.

After measuring respondents’ preferences and demographics, we introduced the context of the subsequent survey questions (a hypothetical presidential election in 2028)²⁹ and measured respondents’ attentiveness (see Appendix A). Our analyses focus on the 1,824 respondents who passed both attention checks; Appendix A shows analyses for all respondents (including inattentives and controlling for inattention; see Berinsky et al., 2021), with fairly similar results. The selection of 2028 as the election year was important as no 2020 Presidential candidate had been chosen and thus voters would not be making their own analysis of a 2021–2024 president.

3.1 Treatments – conjoint experiments

We then presented respondents with four pairs of hypothetical candidates that varied randomly on policy and demographic dimensions to inform them about candidate characteristics and mitigate potential confounding of this relationship.³⁰ We randomized all attributes independently, allowing us to estimate how much weight (if any) respondents attach to each.

In the first conjoint experiment, the policy attributes included IO withdrawal policy which we randomized to indicate whether the candidate advocated definitely withdrawing, probably withdrawing, probably remaining in, or definitely remaining in an imperfect international organization (four categories). We also included domestic economic and education policy (which plausibly varies across parties) because economic and social issues are key factors driving voters’ decisions.³¹ For economic and education policy, we randomly varied four categories: whether the candidate advocated no role, a small role, medium role, or large role of government in the economy or, respectively, in education policy (specifically related to affirmative action). The second block randomized four demographic attributes of candidates: gender (female, male), race (white, Black, Asian, Hispanic), years of experience in politics (4, 8, 12, 16, 20, 24, 28, 32), and home region (Northeast, South, Midwest, West).

We supplemented this first conjoint experiment with a second conjoint survey experiment (on a different sample of respondents) to account for candidates’ party. Party is a crucial factor for vote choice, especially in a two-party system like the US. Omitting the candidate’s party would mean that estimated effects may be limited in external validity, as they may be overcome by party effects. However, including party means that full randomization along policy dimensions (economic and education policy) would make some profiles unrealistic. For example, a Republican candidate favoring strong affirmative action and a strong government role in the economy is possible but unlikely, as is a Democratic candidate advocating no government role in economic and education policies. To reduce potential contradictions between party and domestic policy, we held domestic policy positions constant at a “medium role” for economic and education policies in the sample with the party attribute. We then independently randomized party (instead of home region),³² gender, race, years of experience, and IO withdrawal policy. We presented the candidate information in a table. Tables 1 and 2 show examples of conjoints which could result from random assignment. The attributes with all possible levels/contrasts are shown in Fig. 3 below.

Table 1 Example of Conjoint with party label

Table 2 Example of Conjoint with home region

Fig. 3

Conjoint – Aggregate Results

3.2 Outcomes – conjoint experiments

After each candidate pair, we asked respondents to make a vote choice: “If you had to choose, which candidate would you vote for?” We assigned each candidate a score of 100 if the respondent said they would definitely or probably vote for the candidate, and a score of 0 if the respondent said they would definitely or probably not vote for the candidate. The eight candidate profiles and choices for 1,824 respondents result in 14,592 total candidate choices.

We estimate the effect of each attribute on vote choice using Average Marginal Component Effects (AMCEs) via linear regression (Hainmueller et al., 2014; Bansak et al., 2020), clustering standard errors by respondent. AMCE represents the extent to which a given level of a candidate attribute increases or decreases respondents’ probability of voting for that candidate profile relative to a baseline level, averaging across all other candidate profile attributes and all respondents. This allows us to observe whether changes in preferences toward a Presidential candidate are caused by changes in candidate attributes. Each attribute is dummied out with the baseline value set as the omitted category. For assessing heterogeneous effects by respondents’ partisanship, we estimate conditional AMCEs.

3.3 Results – conjoint experiments

Results from the two conjoint experiments are presented in Figs. 3 and 4, indicating estimated effects of candidates’ policy and demographic attributes on respondents’ preference for that candidate.³³ Figure 3 shows average effects while Fig. 4 shows effects conditional on respondents’ partisanship.

Fig. 4

Conjoint – Heterogenous Results

While Fig. 3 shows that the average effect of IO withdrawal policy is indistinguishable from zero (Fig. 3a) or negative (Fig. 3b), Fig. 4 reveals that these average effects mask strong partisan differences, in line with Hypothesis 1. In both versions of the conjoint experiment, all else equal, Republicans were 5% more supportive of candidates advocating IO withdrawal policies compared to candidates advocating to remain in IOs. In contrast, Democrats were 6–10% less supportive of candidates pursuing IO withdrawal policies. Independents were 4–5% less supportive of such candidates. This indicates that partisan preferences are diametrically opposed: Republicans favor and Democrats oppose IO withdrawal policies and candidates who propose them. These results show that Americans have substantively important opinions about whether the US should withdraw from imperfect IOs, and they divide along partisan lines. Perhaps surprisingly though (and counter to Hypothesis 3a), independents are not more influenced by IO withdrawal policy than partisan groups.

The results for IO policy are substantively similar when accounting for attentiveness (Appendix A Figures A5-A8)³⁴ and they are stronger when restricting the focus to only voters (those who had already voted or said they were likely to vote).³⁵ They are also fairly robust to adjusting for multiple comparisons (Appendix Tables A3-4). In short, results indicate that IO membership policy can matter for one’s choice of political candidates, not just by itself but also when accounting for other common domestic policies and candidate attributes.

Other candidate attributes also matter. Figure 3b shows that compared to candidates who advocated no role for government in economic policy, candidates favoring a stronger role were about 8% more likely to be preferred for President; the same for education policy was up to 4%. Unsurprisingly, party ID and demographic attributes of candidates matter for vote choice as well. All else equal, Figs. 3 and 4 show that candidates with at least some experience performed substantially better than otherwise comparable candidates with no experience (though there is not a linear increase in support with years on the job). Democratic candidates receive 3% more support than Republican candidates, in line with Biden’s 4% advantage in the actual 2020 national election a few days later. Among the other demographic attributes, the candidate’s home region does not matter for vote choice, and results for gender and race are not consistent as they differ depending on whether party or home region is included as an attribute.

While these results align with previous studies of public opinion on IOs, showing that support for IO withdrawal divides along party line, our theory goes further. It argues that framing the withdrawal as benefiting US national interests can nudge increased support for IO withdrawal. We next turn to testing this framing argument.

4 Testing whether framing matters

To test whether framing the IO withdrawal policy matters for public support and partisan baseline differences, we conducted two vignette survey experiments in the US. The vignette design also allows us to test the effect of no framing versus US national interest frames on IO withdrawal policy. As explained above, we test the overall concept of “US national interests” and more specific aspects including “strengthening US national security,” “reducing losses from globalization,” and “mitigating the costs of multilateralism.” Before each vignette experiment, we measured respondents’ political attitudes, demographics, and attentiveness.³⁶ As above, our analyses focus on respondents who correctly answered attention checks; Appendix A shows that effects are similar but somewhat smaller when including inattentive respondents or controlling for inattention (Appendix Figures A11b, A15, A17, A18).

The two vignette experiments differed in a few ways, including timing, platform, and content. The first vignette experiment was fielded just after the 2018 US midterm elections and embedded in a nationally representative Cooperative Congressional Election Study (CCES) survey fielded by YouGov (N = 862) (Ansolabehere et al., 2019). We fielded the second vignette experiment just before the 2020 US Presidential election on Lucid (N = 2,826). In terms of content, the two vignettes varied in treatment texts (see below) and the particular IOs we named. They also varied in the messenger, as we invoked an analyst in the first experiment versus a political candidate in the second.³⁷ These differences were purposeful and help extend the external validity of our study. Despite the differences in IOs, frames, messengers, timing, and platforms, our results are substantively the same.

In the first vignette, we focus on economic organizations because they are the most frequently left IOs and they make up the largest share of IOs.³⁸ By narrowing to economics-related IOs, we also rule out public opinion explanations tied to issue area variation rather than the role of framing.³⁹ To make the concept of economic organizations more specific and realistic, we named several organizations including the World Trade Organization (WTO), International Whaling Commission (IWC), and Asia Pacific Economic Cooperation (APEC). We chose these three economic organizations because they vary in prominence, specificity (broad versus specific), and geography (global versus regional).

In the second vignette, we go beyond economic organizations to understand whether findings from the first vignette are generalizable to broader issue areas. We thus list a security-based organization (the North Atlantic Treaty Organization/NATO), an economic organization (the World Trade Organization/WTO), and a standards-setting organization (the World Meteorological Organization/WMO) which also vary in prominence.

We do not assume that respondents are familiar with these IOs; instead, by including both well-known and lesser-known organizations, we emulate real withdrawal cases. The inference that we make therefore comes not from knowing or not knowing the organizations but from how framing withdrawal impacts respondents’ attitudes. Moreover, the vignettes ask about several organizations to increase the generalizability of our claims (beyond a single IO) and to mimic reality, as countries sometimes announce withdrawals from several IOs at once (for example, as the result of a budget review). Thus, withdrawals sometimes come in waves where voters think more about the implications of membership and multilateralism rather than the specifics of one particular IO.

4.1 Treatments – vignette experiment 1

The first vignette survey experiment included the following text:

“There are several hundred international organizations in the world today, such as the World Trade Organization (WTO), International Whaling Commission (IWC), and Asia Pacific Economic Cooperation (APEC). The United States has joined several international organizations whose core mission is to cooperate on economics and trade. In one of these economic organizations, the US has tried to improve the international trade situation for the US, but its efforts have come up short. [insert experimental group text] Nonetheless, the US respects international law.”

To test how framing drives citizens’ attitudes about IO withdrawal, we randomly assigned respondents to one of four conditions regarding why the US is considering withdrawal: national security (treatment 1, N = 221), globalization (treatment 2, N = 215), costs of the IO (treatment 3, N = 228), or no reason (control group, N = 198). We inserted these reasons as the next-to-last sentence into the vignette. That is, the vignette is identical for all respondents except for one sentence, which was:

[Treatment 1/security:] “Unfortunately, analysts argue that the international organization has reduced US national security.”

[Treatment 2/globalization:] “Unfortunately, analysts argue that the international organization has increased the negative effects of globalization.”

[Treatment 3/costs:] “Unfortunately, analysts argue that the international organization is too costly for what the US gets from it.”

The main independent variable, Framing: Any US national interest is coded 1 for the three treatment groups combined (national security, globalization, costs) and 0 for the control group without framing. As noted above and in the pre-registration, we do not have hypotheses about differences among national interest framings. We use this variable to test Hypothesis 2a that withdrawal from an IO should have more public support if it is framed around US national interests than if there is no framing. While we did not have strong expectations about any particular US national interest treatment being stronger than others, we also code each treatment separately to assess potential differences in exploratory analyses.

4.2 Outcome – vignette experiment 1

We then asked respondents about their support for IO withdrawal, i.e. whether “The US should…”.⁴⁰

• definitely withdraw from the international organization

• probably withdraw from the international organization

• I am undecided

• probably stay in the international organization

• definitely stay in the international organization

We recorded these responses on a five-point scale (1 to 5) with higher values indicating a stronger preference for withdrawal.

4.3 Results – vignette experiment 1

Before evaluating our main theoretical contribution about the effect of IO policy on attitudes, we evaluate Hypothesis 1 about baseline differences between Republicans and Democrats. Figure 5a shows strong differences, with 42% of Republicans but only 4% of Democrats saying the US should withdraw from an imperfect IO. This difference in proportions is highly significant (p < 0.001), providing strong support for Hypothesis 1.⁴¹ The substantive and statistically significant difference of over 30 percentage points also exists when narrowing the sample (to only the control group without framing), broadening the sample (to include inattentive respondents; Appendix A Figures A11a/b), and exists also for a similar pre-treatment question in the second vignette (Appendix A Figure A12). Figure 5b provides further detail on the distribution of partisan and non-partisan groups across response categories, showing strong baseline partisan differences between Republicans and Democrats in the “withdraw” and “stay in the IO” answers.

Fig. 5

Baseline Differences between Republicans and Democrats (Vignette 1)

To assess framing effects, we estimated OLS models with robust standard errors, where the unit of analysis is the individual. Our specifications are:

$${{\text{Y}}}_{i}=\alpha +{\beta }_{0}{{\text{T}}}_{i}+{\epsilon }_{i}$$

and

$${{\text{Y}}}_{i}=\alpha +{\beta }_{1}{{\text{T}}1}_{i}+{\beta }_{2}{{\text{T}}2}_{i}+{\beta }_{3}{{\text{T}}3}_{i}+{\epsilon }_{i}$$

where ${Y}_{i}$ is an outcome for individual i. In the first equation we estimate the causal effect (${\beta }_{0}$) of receiving any of the US national interest treatments ${T}_{i}$. In the second equation we separate each treatment condition. Figure 6a displays coefficient estimates for testing the US national interest framing versus no framing (the pure control condition, Hypothesis 2a), along with 95% confidence intervals.⁴²

Fig. 6

a Framing Effects on Public Support for IO Withdrawal (Vignette 1) b Substantive Effect of Framing on Withdrawal Support (Vignette 1)

Results show that framing IO withdrawal around US national interests increases public support for withdrawal. Figure 6a indicates that the coefficient on Framing: Any National Interest, indicated by a blue circle, is positive and highly significant (p < 0.003).⁴³ Compared to the control group of respondents who did not see a US national interest frame, respondents who were provided this framing were significantly more supportive of withdrawal. The substantive size of the effect is 0.35 points on a 5-point scale or 31% of a standard deviation change in the outcome. Figure 6b shows this in terms of a bivariate difference in means with 95% confidence intervals.⁴⁴ Public support for IO withdrawal doubles from 12% to 25% when justified by US national interests.

As noted above, we have no strong expectations about which of the three national interest framings is more important for public opinion. Nevertheless, in exploratory analyses, we unpack the framing effect into the three reasons to see if the average result is driven by a particular treatment or hides significant variation. The estimated coefficients are also shown in Fig. 6a and are indicated by orange squares; the excluded reference category is again the control group of no framing. The estimates all point in the same direction and are of similar size; they are statistically different from zero (cost/benefit at p < 0.06) but not different from each other. In other words, framing matters in terms of whether IO withdrawal decisions are justified in one of the US national interest justifications but the specific type of US national interest justification does not differentially affect public support for withdrawal.

In the Appendix we document that results are robust. Results are substantively similar when using an ordered logit estimator (Appendix A Figure A19)⁴⁵ or randomization inference (Keele et al., 2012); when checking for unbalanced covariates (Appendix A Tables A7-A8), checking and adjusting for inattentive respondents (Appendix A Tables A10-A11 and Figure A15), and adjusting for multiple comparisons (Appendix A Table A15; Benjamini & Hochberg, 1995).⁴⁶

While these results indicate that framing IO withdrawal policy around US national interests can increase support for the policy, they also have a few limitations. The vignette did not allow us to assess whether any frame works – since no placebo other than the pure control was included. Further, while we assume that candidates may tout IO withdrawal as an achievement on the campaign trail, the first vignette did not explicitly ask how this would affect respondents’ support for a candidate or their vote choice. Finally, it is possible that during a more polarized or high stakes context – right before a Presidential election – we may not find the same results. To address these limitations, we fielded a second vignette survey experiment to include a placebo frame (testing Hypothesis 2b), to ask whether IO withdrawal policy affects candidate assessments or voting, and to understand support during a presidential election campaign.

4.4 Treatments – vignette experiment 2

The second vignette experiment presented respondents with a hypothetical scenario where the US presidential candidate in a future year (again 2028) discusses their accomplishments on the campaign trail. We randomly assign one of three vignettes: no framing (a pure control), a framing related to US national interests (treatment text), and a non-US national interests framing (placebo text around changing IO headquarter locations). The vignette reads:

“Now we would like to get your opinions about the policy achievements of one of these Presidential candidates in 2028. Imagine in a campaign speech for the 2028 election, the Presidential candidate says, “In the last four years we got a lot of things done. On the domestic front, we created many jobs, maintained strong economic growth, and reduced the budget deficit. In foreign policy, we pulled out of several international organizations, including the North Atlantic Treaty Organization (NATO), the World Trade Organization (WTO), and the World Meteorological Organization (WMO).”

We then randomly assigned respondents to one of three groups. The control group saw the text above without any framing. Respondents in treatment groups 1 and 2 saw one additional sentence with different framing.

• 1. [Treatment/US national interests:] “These international organizations do not advance US national interests or prioritize American goals in the world.”

• 2. [Placebo/headquarters:] “These international organizations have headquarters in locations which need to be moved somewhere else.”

That is, as in the first vignette, experimental groups are identical except for one sentence. We chose this placebo frame about headquarters location because it addresses foreign policy and crosses issue areas but may be perceived by respondents as largely bureaucratic (or at least focused mostly on IO particulars), versus being tied to US national interests. We also wanted a placebo that would not invoke partisan differences, which ruled out many possible frames including spending or size because conservative voters tend to favor smaller government involvement.

4.5 Outcomes – vignette experiment 2

We then asked three questions about the candidate and one question about the withdrawal policy to measure outcomes of interest:

• “Given this information, how much would you say this candidate has achieved for you?”

• “Given this information, how much would you say this candidate has achieved for the US?” Answer options for both of these questions were a lot, a moderate amount, a little, nothing.

• “Given this information, how likely would you be to vote for this candidate?” Answer options were very unlikely, somewhat unlikely, somewhat likely, very likely.

• “How much would you approve or disapprove of these policies?” Answer options were strongly approve, somewhat approve, somewhat disapprove, strongly disapprove.

Since all outcomes are recorded on 4-point scales, we estimate OLS with robust standard errors.⁴⁷ To assess the framing effect, we estimate differences in means between the control group (no framing) and treatment group (US national interests’ frame). We use the placebo condition to assess whether differences between the control and treatment could be due to the desire to advance US national interests or because offering any framing changes attitudes (no matter how meaningful the explanation). As in the conjoint experiment, we also investigate heterogeneous effects by respondent partisanship with an indicator for Independents.

4.6 Results – vignette experiment 2

Results in Fig. 7 and Table 3 show that framing withdrawal around US national interests increases public support for withdrawal on average.⁴⁸ In the upper panel of Fig. 7, the coefficient estimates on the treatment are positive and statistically significant for all four outcomes. Figure 7 shows that the size of the effect is about 0.15 points on the 4-point scale across outcomes, which is about 16% of a standard deviation change in the outcomes. Substantively, the increase in public support is about 6% for withdrawal, 4–6% for perceptions of candidate achievements, and 6% for the likelihood of voting for the candidate.

Fig. 7

Framing Effects on Public Support for IO Withdrawal (Vignette 2)

Table 3 Aggregate and Placebo Framing Results (Vignette 2)

Robustness checks are detailed in the Appendix. Results are substantively similar when using an ordered logit estimator (Appendix A Table A14) or randomization inference, and when adjusting for unbalanced covariates (Appendix A Table A9 and Figure A13), inattentive respondents (Appendix A Tables A12-A13, Figures A17-A18), and multiple comparisons (Appendix A Table A15).⁴⁹

Moving on to heterogeneity, we test whether the effect of framing related to US national interests (versus no framing) is stronger for independent respondents than Republicans and Democrats (Hypothesis 3b). As pre-specified, we test this by distinguishing partisan sub-samples. The lower panel of Fig. 7 shows estimates separately for Democrats, Independents, and Republicans. Across all four outcomes, independents are significantly affected by the US national interest framing while partisans are not. This supports Hypothesis 3b. Interestingly, this differs from the conjoint results (Hypothesis 3a), suggesting that shortly before elections and in the context of this experiment, framing is more effective on independents who otherwise care less than partisans about IO withdrawal policy itself.

Using the placebo framing (moving the IO’s headquarters), we also assess whether a different frame (placebo) works as well as the US national interest frame (treatment). Here the results are mixed and remain suggestive, as shown in Table 3. The difference in coefficients between the treatment and placebo is significant for only one of the four outcome variables: respondents’ perception of how much the candidate has achieved for them. IO headquarters may not be as strong a placebo as we had intended; studies suggest that countries can leverage IO headquarter locations for political advantage (Johnson, 2015; Sommerer & Tallberg, 2019; Clark & Dolan, 2021; Kilby, 2013). Further, while both the treatment and placebo invoke the idea of (unrealized or potential) benefits, US national interests may invoke values, while headquarters may invoke concerns about fairness, which may explain the weaker placebo effect. Given these considerations, it is challenging to identify an apolitical placebo framing.⁵⁰

The results in Table 3 show that relative to the placebo, the US national interests treatment generates stronger effects in two ways. First, the treatment has consistent effects across all four outcomes while the placebo is insignificant for two outcomes. Second, the treatment generates larger effects than the placebo in terms of standard deviation shifts in outcomes: between 14.2% and 17.6% versus the placebo’s 0.3% to 14.0% change. Again, exploratory analyses show that results are similar when we restrict the focus to only voters.⁵¹

Using observational data, we also explore whether respondent demographics influence support for IO withdrawal. Appendix Table A16 examines support for IO withdrawal using the pre-treatment question and the relevant outcome measure from the vignette experiment. We find that respondents who are older and more educated (which could also be a proxy for political sophistication) are less likely to support IO withdrawals, which aligns with previous research showing that these factors may matter for Americans’ foreign policy preferences (Balestrini, 2014; Kertzer, 2021). Gender and income flip signs across the outcome measures and are thus less consistent findings, even though some prior work has found these demographic variables to be significant in understanding some US foreign policy preferences.⁵²

5 Conclusion

Under what conditions does the US public support withdrawal from IOs and candidates who propose this? This question is significant given the US’ leadership role in IOs and international cooperation more broadly, and because the US is the most frequent withdrawer from IOs. Furthermore, during 2016–2020 observers sometimes referred to President Trump’s IO policy as a “withdrawal doctrine” (Haass, 2020). But to date, we know little about the conditions under which Americans support withdrawal from IOs and what factors shape their support. Researchers have not yet studied whether Americans care about IO withdrawal and whether it matters for choosing political candidates.

We argue that (1) announcing IO withdrawals activates Americans’ partisan preferences regarding multilateralism and that (2) the US public is more likely to favor withdrawing from an IO when it is framed as serving US national interests. We discuss why the public might change their support for a candidate based on IO withdrawal policy. The nature of IO membership may make withdrawal relatively easy for voters to recall at election time when the politician can prime them to remember the IO withdrawal. This may help politicians capture “what they have done for voters lately” as it invokes opposition toward multilateralism. This can gain support from some voters at the ballot box. The US national interests frame may also rally voters to conceive of IO withdrawal as a crisis-related achievement (Iyengar & Kinder, 1987; Zaller & Feldman, 1992). We also reason that a US national interest frame may sway public opinion by casting withdrawal as domestically beneficial, which can help overcome respondents’ sunk cost fallacies, status quo bias/loss aversion, and loyalty (all of which may otherwise steer them to remain in the institution).

The results of four survey experiments support our arguments and provide the first experimental evidence on what drives US public opinion about withdrawing from IOs. These results suggest that US politicians can raise and frame IO withdrawals relative to US national interests to gain support in future elections. These findings also show that there is an important domestic electoral component of international cooperation and withdrawal from IOs. This extends previous work that broadly connects IOs to domestic policy by drawing a clear theory regarding electoral politics and withdrawal (Putnam, 1998). It also builds on previous work on public opinion towards EU (dis)integration, extending research to the US context.

Do these findings generalize? We test our general theory on the specific context of US public opinion in 2018 and 2020 (two time points during the 2016–2020 Trump administration), and we leave it to future research to extend the generalizability of our findings to other timeframes. Politicians’ before Trump have promoted IO withdrawal as a foreign policy accomplishment. And IO withdrawal has been framed as relating to national interests/sovereignty in other states as well (von Borzyskowski & Vabulas, 2019, Fig. 4). While we cannot be sure that the empirical findings fully generalize to other timeframes, there is reason for confidence. Our findings are consistent with previous research that has been conducted in different countries. This research shows that European voters supporting Brexit focus on how membership cessation could improve national interests such as immigration (De Vries, 2017; De Vries et al., 2021; Jurado et al., 2022; Walter, 2019, 2021a, b). Moreover, our findings are in line with literature on IO legitimacy that underscores how elite framing and cueing can shape voters’ considerations on an IO’s cachet (Dellmuth & Tallberg, 2015; Tallberg & Zürn, 2019). Previous scholarship shows that voters’ preferences can enhance governments’ bargaining power in international negotiations (Caraway et al., 2012; Hug & König, 2002). Moreover, politicians incorporate public opinion into foreign policy decision-making (Hagemann et al., 2017; Schneider, 2018; Tomz et al., 2020), especially when international cooperation is salient to the public (Wratil, 2018). Future research can consider how and if politicians explicitly leverage public support if they seek to withdraw from IOs.

Our findings are also important for understanding what might drive some of the different stages of IO life cycles. Like other articles in this Special Issue, we move away from the state as the unit of analysis. We unpack the black box of the state and start to shed light on how public opinion might affect states’ membership in IOs. This introduces the possibility that various stages of IO life cycles are not just driven by the IO’s effectiveness or outputs but also by domestic electoral politics.

Our findings are also crucial in expanding our understanding of the backlash toward international institutions. Our results challenge the notion of a broad backlash against IOs (von Borzyskowski & Vabulas, 2025) and the liberal international order by highlighting the nuances of partisanship and cueing in support for IO withdrawal. In some respects, the experimental results can inspire cautious optimism about international cooperation (Copelovitch et al., 2020). The data show that even when IOs create challenging environments for the US—as presented in each of the experimental vignettes—two thirds of the public still values staying in IOs as a baseline perspective. This emphasizes the long view of IO life cycles that is underscored in this Special Issue. In other words, the US public is not nonchalant about IO withdrawal and regards international cooperation as ongoing and important.