Introduction
In democracies, legislators face incentives to consider the preferences of both business actors and citizens when making trade policy decisions. Business actors matter not only because their activities contribute to economic growth but also because they can exert influence through lobbying. Citizens matter because they ultimately hold legislators accountable at the ballot box and, in many cases, because legislators genuinely value being responsive to the public. These dual incentives create a dilemma whenever business and citizen majorities take opposing positions on a trade policy issue. In such situations, why do legislators prioritize one side or the other?
We argue that – in the ideal-typical case where business and citizens exhibit equal levels of political engagement – legislators are more responsive to citizens. This predisposition may stem from both instrumental considerations (securing re-election) and normative commitments (a genuine concern for public responsiveness). Two factors, however, moderate this baseline effect. First, we expect left-wing legislators to be more inclined to prioritize citizens’ preferences than their right-wing counterparts. Second, we argue that legislators who conceive of their role as delegates should place greater emphasis on citizens’ views than those who see themselves as trustees.
Three experiments included in a survey with more than 1000 legislators from 47 countries across the globe and qualitative evidence from 30 interviews in nine countries and open survey responses allow us to test these theoretical expectations.Footnote 1 For the core evidence, we relied on experiments rather than directly asking legislators in a survey because legislators would likely find it difficult to respond to an abstract question about the weight they give to business and citizen preferences in trade policy. Moreover, social desirability bias would likely influence the responses that we get. The experimental approach we take lowers concerns about both of these problems with survey questions. The qualitative evidence from interviews with legislators from a series of countries across the globe and open survey responses allows us to back up the assumptions underlying our argument and help us interpret the results from the experiments.
The results offer strong support for our expectations. Ceteris paribus, legislators are more responsive to citizen preferences, and this effect is both robust across different experimental wordings and substantively large (ranging from 13% to 21% of the scale). Moreover, the effect is particularly pronounced among left-wing legislators and among those who view themselves as delegates.
A considerable number of studies already analyze legislator preferences or voting behavior with respect to trade policy (Gartzke and Wrighton, Reference Gartzke and Wrighton1998; Hiscox, Reference Hiscox2002; Schonhardt-Bailey, Reference Schonhardt-Bailey2003; Milner and Tingley, Reference Milner and Tingley2011; Feigenbaum and Hall, Reference Feigenbaum and Hall2015; Owen, Reference Owen2017; Miler and Allee, Reference Miler and Allee2018; Campello and Urdinez, Reference Campello and Urdinez2021; Murillo and Pinto, Reference Murillo and Pinto2022; Lee et al., Reference Lee, Pomirchy and Schonfeld2023; Osgood, Reference Osgood2023; Stiller, Reference Stiller2023; Dür et al., Reference Dür, Huber and Stiller2024). Our study makes several contributions to this literature. First, existing research gives importance to either business actors or citizens, without analyzing the factors that make legislators side with one or the other type of actor. Second, whereas much existing research relies on observational evidence, our results derive from experiments with legislators.Footnote 2 Our experimental evidence complements existing observational evidence in that it allows us to investigate a specific aspect of responsiveness in isolation, aiding more credible causal identification. Third, whereas existing research nearly exclusively studies the U.S. (among the exceptions being Campello and Urdinez, Reference Campello and Urdinez2021; Murillo and Pinto, Reference Murillo and Pinto2022), our results are based on responses from legislators worldwide.
Going beyond this contribution to studies on the determinants of legislators’ trade policy positions, our research also speaks to the broader trade policy literature. So far, this literature is characterized by a divide between studies that stress the influence of business actors on policy outcomes (for example, Chase, Reference Chase2005; Ehrlich, Reference Ehrlich2008; Kim, Reference Kim2017; Bearce and Roosevelt, Reference Bearce and Roosevelt2023) and studies that focus on the role of citizens (Kono, Reference Kono2008; Baccini, Reference Baccini2012; Mansfield and Milner, Reference Mansfield and Milner2012; Grossman and Helpman, Reference Grossman and Helpman2021). Our study can help integrate these two strands of research by specifying the conditions under which legislators are responsive to business actors or citizens.
Furthermore, the paper contributes to broader debates about the interaction between elites and the public in international relations (De Vries et al., Reference De Vries, Hobolt and Walter2021; Lake et al., Reference Lake, Martin and Risse2021; Martin and Risse, Reference Lake, Martin and Risse2021). It connects to research on how rising inequality (Flaherty and Rogowski, Reference Flaherty and Rogowski2021) or increasing populist attitudes (Pevehouse, Reference Pevehouse2020) shape international outcomes. Whereas many studies examine whether elites drive public opinion or whether public opinion drives elite behavior in foreign policy (Guisinger and Saunders, Reference Guisinger and Saunders2017; Tomz et al., Reference Tomz, Weeks and Yarhi-Milo2020), there is still little research that considers public opinion and business influence simultaneously. Our focus on trade policy is particularly valuable in this regard, as trade policy involves relatively clear-cut distributional effects that can set business interests against those of citizens.
Finally, by not only providing an empirical study of responsiveness in the field of trade but also developing and testing expectations about factors that make legislators more or less responsive to business interests and citizens, we also contribute to research on representation and responsiveness more generally (e.g., Ezrow et al., Reference Ezrow2011; Burstein, Reference Burstein2014; Gilens and Page, Reference Gilens and Page2014; Costa, Reference Costa2017). In that sense, we speak to a few studies that have already analyzed the relative importance of interest groups or lobbyists and the public for policy-making (Burstein, Reference Burstein2014; Giger and Klüver, Reference Giger and Klüver2016; Rasmussen and Otjes, Reference Rasmussen and Otjes2025). What we add to this research is the focus on the moderators that make politicians focus on public opinion or business interests. The specific focus on trade policy also is a contribution to this literature, as this policy area is relatively technical, possibly making legislators less responsive to the public.
Argument
The main effect: responsiveness to citizens
In trade policy, it is not uncommon for a majority of business interests to oppose a majority of citizens.Footnote 3 Other types of actors, such as labor unions and citizen groups, may also influence trade policy debates, but we abstract from their influence to reduce the complexity of the theoretical argument. Introducing them would not change the theoretical expectations that we derive below. The Stolper-Samuelson theorem provides one possible explanation for why, under certain circumstances, citizens (who are mainly endowed with labor) and business actors (capital owners) may have different trade preferences. Diverging preferences, however, may also arise if business actors focus on the economic gains from trade and the public on trade’s non-economic effects (e.g., for the environment). Examples of such a preference constellation abound. British business interests overwhelmingly supported continued British membership in the EU, whereas a majority of citizens voted to leave the EU. Similarly, the vast majority of U.S. business actors supported the Transpacific Partnership (TPP, 2008–2017) and European business actors backed the Transatlantic Trade and Investment Partnership (TTIP, 2013–2017), but both agreements did not materialize because of popular opposition. Beyond these high-profile cases, opposing views between business actors and citizens can also be present at the level of legislators’ electoral districts, for example, when citizens as consumers expect to gain from a trade agreement, but a few import-competing firms dominate the business side of the debate.
In all such cases, legislators must decide whether to align with the preferences of the majority of citizens or the majority of business actors. Business actors are important for legislators because of their control over investment decisions, which have a large influence on the state of the economy. To avoid an economic downturn, politicians may feel the need to keep business actors happy. Legislators may also adopt the business position because they are used to interacting with business interests and receiving resources from them, such as money and expertise. Business actors may also use their resources to get legislators elected that are business friendly. Finally, business actors may be able to offer personal awards to legislators who adopt a business-friendly position, such as the promise of a job after the end of the political career.
Given these reasons to expect business interests to matter for legislators’ decisions, it is no wonder that many authors focus on lobbying by business actors when explaining trade policy outcomes (Fordham and McKeown, Reference Fordham and McKeown2003; Drope and Hansen, Reference Drope and Hansen2004; Chase, Reference Chase2005; Tavares, Reference Tavares2006; Ehrlich, Reference Ehrlich2008; Gawande et al., Reference Gawande, Krishna and Olarreaga2012; Milner and Tingley, Reference Milner and Tingley2015; Kim, Reference Kim2017; Bearce and Roosevelt, Reference Bearce and Roosevelt2023). These studies are in line with a broader strand of literature that sees politics dominated by elites, and in capitalist economies especially by business actors (Mills, Reference Mills1956; Schattschneider, Reference Schattschneider1960; Domhoff, Reference Domhoff2013; Gilens and Page, Reference Gilens and Page2014; Bartels, Reference Bartels2016). Indeed, all legislators that we interviewed mentioned the important role of business lobbying in trade policy-making. One, for example, argued that trade policy is about “the coffee grower who goes out to defend his own interests, the textile manufacturer, the shoe producer who goes out to do the same” (Interview #8).
Nevertheless, our baseline expectation is that, ceteris paribus, legislators are more responsive to citizens than to business. This is so because in democratic political systems, they need to compete for votes in elections (Burstein, Reference Burstein2014). Politicians either offer policy outcomes that satisfy citizens or face “punishment” in the next elections. For them, avoiding voter discontent is thus the best way of staying in office. Although not all legislators strive for re-election, many do, meaning that on average legislators should try to be responsive to citizens.Footnote 4 Next to this instrumental rationale for considering citizens’ preferences, politicians may also be normatively committed to representing the public. Attesting to the importance of citizens’ preferences, a legislator we interviewed argued that if there is “discontent on the part of citizens, it is difficult for [a trade agreement] to be politically viable” (Interview #1).
It is no wonder, then, that several authors have attributed an important role to citizens (as voters or via public opinion) in their explanations of trade policy outcomes (Dutt and Mitra, Reference Dutt and Mitra2002; Kono, Reference Kono2008; Baccini, Reference Baccini2012; Mansfield and Milner, Reference Mansfield and Milner2012; Grossman and Helpman, Reference Grossman and Helpman2021). Moreover, the broader literature on representation has found congruence between public opinion and policy outcomes or politicians’ positions, even if partly contingent on other factors (Manza and Cook, Reference Manza and Cook2002; Clinton, Reference Clinton2006; Butler and Nickerson, Reference Butler and Nickerson2011; Burstein, Reference Burstein2014; Rasmussen et al., Reference Rasmussen, Reher and Toshkov2019; Sevenans, Reference Sevenans2021; Soontjens and Sevenans, Reference Soontjens and Sevenans2022).
Several objections may be raised to challenge the idea that legislators are more responsive to citizens than to business interests. First, citizens may not have much knowledge about either the issues on the political agenda or decisions taken concerning them (Converse, Reference Converse and Tufte1970; Guisinger, Reference Guisinger2009). Citizens may even lack a clear and stable preference on issues that politicians could consider. Indeed, Rho and Tomz (Reference Rho and Tomz2017) show that most U.S. citizens have little understanding of the economic consequences of trade. Citizens, however, do not have to be highly sophisticated to form a mood about a policy (Enns and Kellstedt, Reference Enns and Kellstedt2008). Unsurprisingly, therefore, politicians often collect information on public attitudes and even commission polls, meaning that they are concerned about public opinion after all (Druckman and Jacobs, Reference Druckman and Jacobs2006; Hillygus Reference Hillygus2011; Walgrave and Soontjens, Reference Walgrave and Soontjens2023). All our interviewees also told us how important it was for them to be informed about citizens’ preferences. This not only concerned citizens in their electoral districts (so, narrowly speaking, their own voters) but also citizens across the country. Legislators we interviewed also stressed that citizens regularly share their views with them, for example during public hearings or through e-mail and social media (e.g., Interview #26).
Second, to the extent that citizens manage to express a preference on an issue, that preference may not be genuine but may have been shaped by either politicians or other political entrepreneurs (for the general argument, see Matsubayashi, Reference Matsubayashi2013; specifically on trade, see Guisinger, Reference Guisinger2017; Dür, Reference Dür2019; Dür and Schlipphak, Reference Dür and Schlipphak2021). While elite influence on public opinion is likely, the fact that political elites regularly attempt to shape public opinion on trade via campaigns is indicative of the influence that political elites assign to public opinion. For example, across many European countries opponents of the proposed TTIP relied on mobilizing public opinion. If public opinion were irrelevant for legislators, such campaigns would only make sense in the context of referendums.
Third, for legislators’ re-election, citizens overall may be less relevant than the subgroup of citizens who are likely to vote for them or their party. In the absence of information on that subgroup, however, the majority opinion should serve as a cue for legislators. Indeed, some studies offer experimental evidence that legislators try to be responsive to the opinion of the public in general rather than only a subset (e.g., Butler and Nickerson, Reference Butler and Nickerson2011; Ezrow et al., Reference Ezrow2011).
Finally, business actors possess resources that give them channels of influence that are not equally available to citizens. Our expectation of greater responsiveness to citizens thus refers to a baseline scenario in which citizens and business actors exhibit the same level of political engagement. When business actors devote substantial resources to shaping a policy while citizens remain passive, legislators are likely to be responsive to business.Footnote 5 Conversely, when citizens mobilize (for example, by organizing protests or making their views otherwise visible) while business actors remain inactive, legislators should be even more inclined to prioritize citizen preferences. Our baseline expectation of greater citizen responsiveness is therefore net of such asymmetries in engagement. Overall, therefore, we expect:
Hypothesis 1: Legislators are more responsive to citizens’ trade preferences than to business preferences. Going beyond this baseline expectation, legislators’ relative responsiveness to citizens and business should vary depending on legislators’ characteristics. In the following, we discuss two such characteristics, namely political ideology and legislators’ conceptions of their role.
Moderator I: political ideology
Left-wing legislators may be opposed to and right-wing legislators may be supportive of trade agreements as a matter of principle. Indeed, earlier research has shown that left-wing and right-wing parties adopt different positions toward trade in their party manifestos (Milner and Judkins, Reference Milner and Judkins2004), which could indicate deeply held views on this issue. Principled support or principled opposition could lead legislators to adopt firm positions toward trade agreements that would not change as a result of information about the winners or losers of a trade agreement.
Contrary to this baseline expectation, we conjecture that legislators’ ideology pushes them to support the kind of trade agreement that favors the constituency they stand for. Left-wing politicians, qua their ideology, prioritize the needs of workers and employees over those of business owners. Left-wing ideology emphasizes reducing social hierarchies, which aligns with advocating for workers and employees who typically have less economic power compared to business owners. Left-wing legislators should also exhibit a willingness to support state intervention in the economy to address perceived market failures or redistribute income, which locates them further away from business preferences. Economically right-wing politicians, by contrast, should be more willing to support the position that business profits will eventually trickle down to workers. What is more, they might highlight that workers will lose their jobs if firms fail (and several interviewees made exactly this point). They hence might mainly support trade agreements that favor business interests. Legislators’ ideology hence is likely to predispose them to exhibit greater responsiveness with either business actors or citizens.
Existing research also shows that right-wing political parties have stronger links with business groups than left-wing parties (Otjes and Rasmussen, Reference Otjes and Rasmussen2017). Indeed, a right-wing legislator that we interviewed indicated: “Sometimes I initiate meetings [with business actors] myself, as I’m interested to know [their opinions] prior to ordaining or legislating. […] I am in contact with businessmen, I am interested in their opinion, they are the engine of the economy” (Interview #5). By contrast, a legislator from a far-left party that we interviewed stated that he had very few contacts with business actors, because “we are not part of this lobby” (Interview #2). Instead, left-wing parties traditionally have had stronger links with labor unions (especially social democratic parties) or broader social movements (e.g., green parties). Even if the strength of such links have weakened over time (Allern and Bale, Reference Allern and Bale2012), they may still make economically left-wing legislators relatively more eager to bring their positions in line with citizens’ positions on trade and right-wing legislators to be relatively more concerned about business demands. In the form of a hypothesis, we expect:
Hypothesis 2: Left-wing legislators are relatively more responsive to citizens’ trade preferences than right-wing legislators.
Moderator II: trustees vs. delegates
Legislators differ not only in their political ideology but also in how they conceive of their role: some see themselves as trustees, while others emphasize their function as delegates.Footnote 6 Trustees insist on their greater expertise than citizens; they are hence motivated to lead based on their judgment of what is good for the country rather than to listen to citizen demands (e.g., Pitkin, Reference Pitkin1967). For example, a Chilean Senator explained his vote in favor of Chile’s accession to CPTPP adopting a trustee position, arguing that the “public is being misled with false claims” (Senado de Chile, Reference Senado2022). One interviewee clearly stated that his task was to lead the public and create support for the position he advocates (Interview #21). Thus, depending on their judgment of what serves the greater good, trustees may at times align with citizens and at other times with business interests.
Legislators who conceive of themselves as delegates, by contrast, put more emphasis on responsiveness to voters and the representation of citizen preferences (Miller and Stokes, Reference Miller and Stokes1963). In the words of Mansbridge (Reference Mansbridge2003, 516), such a representative “promises to follow the constituents’ instructions or expressed desires.” Business actors are also constituents; but delegates are less likely to assign them the same role as voters. Exhibiting such a delegate role conception, one legislator we surveyed stated that “Parliamentarians are the voice and the hands of those who voted for them” (open response in our survey). Similarly, an interviewee stated that it was a principle of his party “to be with majorities, to be with the citizens” (Interview #3), whereas another indicated that he “cannot be legislating behind the backs of the people who vote for [him]” (Interview #17).
In short, delegates can be expected to be more concerned with representing citizens’ interests than trustees. Following the same logic, Schonhardt-Bailey (Reference Schonhardt-Bailey2003) explained trade liberalization in mid-19th century Britain as a result of legislators switching from voting as trustees to voting as delegates. Also more broadly, existing research has found a greater coincidence between delegates and public opinion (e.g., Studlar and McAllister, Reference Studlar and McAllister1996; Herrick, Reference Herrick2013). We hence expect:
Hypothesis 3: Legislators who conceive of themselves as delegates are relatively more responsive to citizens’ trade preferences than legislators who conceive of themselves as trustees.
Research design
To test our three hypotheses, we rely on original survey-experimental data. Between April 2021 and March 2022, we surveyed members of national parliaments from 47 democracies across the globe. This covers all democratic countries for which we were able to find e-mail addresses for members of parliament, except some smaller countries with official languages that were not otherwise part of our study.Footnote 7 We contacted 12,578 legislators in both lower and upper chambers of national parliaments via e-mail and offered an online questionnaire in one of eight languages.Footnote 8 After sending reminder e-mails, we received responses from 1057 legislators, which amounts to a response rate of 8.4%. The experiment with the fewest valid responses was completed by 894 legislators (7.1%). Given the difficulty of surveying political elites, especially because mass e-mails are likely to be categorized as spam and may never reach the intended recipients’ mailboxes, this overall response rate is decent.
Response rates vary strongly across countries, as only about 1% of Indian legislators but 32% of Swiss parliamentarians responded (see Figure E1 in the Appendix). These different response rates appear to be not only a function of legislators’ willingness to respond to a survey but also of how many of our e-mails made it through spam filters. There are no major differences in the response rates by gender, government status, or political ideology (although overall legislators from parties to the right of the political spectrum are slightly over-represented), but those from upper chambers exhibited a clearly greater willingness to fill in the survey (see Figures E2 and E3 in the Appendix). We account for the latter by excluding all respondents from upper chambers in a robustness check.
Although we sent e-mail invitations directly to legislators, we cannot be certain that they personally completed the questionnaire. Nevertheless, several pieces of evidence suggest that this was largely the case. First, we frequently received follow-up questions or comments that clearly originated from legislators, as well as e-mails from aides noting that a legislator would soon respond or confirming the legislator just had completed the survey. Second, our follow-up interviews reinforced the impression that legislators had filled out the survey themselves. Finally, in some countries legislators have only very small teams of assistants, which makes it even more plausible that they personally replied to our inquiry.
The resulting sample is very diverse (see also the information in Section D in the Online Appendix). We received responses from 714 men (68%) and 343 women (32%, which reflects the lower share of women in many parliaments across the world). The mean age of the respondents was 53 years. The youngest respondent was only 22 years old, whereas the oldest was 96. Among the respondents, 53% came from government parties, whereas 47% were from opposition parties. 53% served their first term in parliament. Only 76 legislators (7%) served in a trade committee in parliament when responding, whereas the rest did not. Nevertheless, on average the respondents indicated that they had considerable interest in trade policy. Finally, 78% of respondents hold a university degree, whereas 22% do not. These percentages are identical for men and women. No fewer than 109 respondents hold a PhD as their highest degree.
Using a cross-national sample has both advantages and drawbacks. On the downside, heterogeneity across countries – in institutional structures, lobbying practices, and other factors – can introduce noise into the results. At the same time, this variation enhances the generalizability of the findings (spatially and temporally), as many factors are likely randomly distributed across countries and therefore cancel each other out. In contrast, single-country studies or studies with only a few countries may be systematically affected by factors that are constant within that context and hence cannot be controlled for. Overall, we consider the benefits of a large, diverse sample to outweigh the drawbacks. To account for cross-country heterogeneity, all models below include random effects at the country level, with fixed effects used in robustness checks (see Section M in the Appendix). We also implement jackknifing to ensure that results are not driven by any single country (see Section J in the Appendix).
Qualitative evidence
The survey contained an open question inviting legislators to give us feedback beyond their responses to the closed questions. About 10% of the respondents made use of this possibility. Some also sent us follow-up e-mails, commenting on aspects of the survey. Since we found many of these comments highly valuable, we decided to carry out a series of follow-up interviews with legislators (Section B in the Online Appendix contains the questions used). For these semi-structured interviews, we selected members of parliamentary committees dealing with trade policy, and tried to attain variation in terms of ideological leaning and home country of our respondents. In the end, we spoke by phone or video conference to 30 legislators from nine countries with political ideologies that range from far-left to far-right, some – but not all – of which had also filled in the survey. The open responses together with the interview evidence provide us with a cross-check of the assumptions made in developing our argument, and also help in the interpretation of the results from the experiments.
The experiments
For the core analysis, we rely on experiments because directly asking legislators to assess the relative importance they attach to business versus citizen preferences in trade policy decisions would have been problematic for two reasons. First, it is unlikely that legislators can provide an objective self-assessment of how much they cater to each actor type. Second, even with survey anonymity, responses to such a direct comparison would likely be affected by social desirability bias. Our experimental design addresses these issues by asking legislators about their attitudes toward specific trade agreements, providing citizen and business preferences only as contextual information. Moreover, each legislator saw only a single treatment, making it difficult for them to determine which response might be socially desirable. Indeed, ignoring business opposition may not be desirable either. If social desirability bias still affects responses in this specific context, it likely similarly influences actual trade policy debates.
We included three experiments in our survey, all of which address the same issue with slightly different wordings. If the main results remain the same despite these differences in wording, we can be confident that the results are not just driven by any specific formulation. We introduced experiments 1 and 3 in random order, but we always kept what we call experiment 2 to the end, as it is the only one to mention a specific country and we wanted to avoid priming respondents to think about that country in the other experiments.
The following experiments intentionally simplify reality to isolate a specific causal mechanism. This does not imply that other factors – such as media coverage, institutional rules, or lobbying – are unimportant. On the contrary, precisely because these factors influence real-world outcomes, experiments are necessary to examine a specific causal link. In this sense, experimental studies like ours complement observational research: experiments identify causality, while observational studies show how multiple mechanisms interact to produce observed outcomes.
Our methodological approach also needed to consider several other issues that result from our target group being political elites and from ethical considerations. For one, while some legislators only have little knowledge of trade matters, others have an excellent understanding of the subject matter. To allow the former to participate, treatments had to be relatively simple.Footnote 9 At the same time, for ethical considerations we did not want to provide treatments that were evidently incorrect or misleading. Because of this, and in line with recent methodological advice (Brutger et al., Reference Brutger2023), we kept the experiments hypothetical. This was also necessary because non-hypothetical treatments could have idiosyncratic effects in individual countries. Only in one experiment, therefore, did we use the name of a specific country as a partner in a trade agreement.
Second, given that legislators are very busy, we also needed to keep the questionnaire short. Among other things, this made repeating a conjoint experiment several times or larger sets of questions impractical. Third, we also needed to consider that because of a combination of a relatively low response rate and a limited universe of respondents, the number of responses would be limited. To maintain statistical power, this meant that we had strict upper limits to the number of treatment groups for each experiment. The power problem is aggravated by the fact that we expected treatment effects to be moderated by legislators’ ideology and role conceptions. Two of the three experiments below hence (effectively) only have two treatment groups, and one has four treatment groups.
In experiment 1, we varied whether business associations support or oppose a trade agreement with the population taking the opposite position. This exactly reflects the trade-off we referred to in the argument. The full wording of the experiment is as follows:
Experiment 1: Imagine COUNTRYFootnote 10 negotiates a new trade agreement with an important trading partner. [Business associations support this agreement, but there is opposition within the population to it. / There is opposition within the population to this agreement, but business associations support it. / Business associations oppose this agreement, but there is support within the population for it. / There is support within the population for this agreement, but business associations oppose it.] How much would you support or oppose this trade agreement? (11 point scale from “Would strongly oppose” to “Would strongly support”)
Business associations here serve as placeholders for businesses, whereas the population proxies for the citizens mentioned in the hypotheses. Since two of the four treatments just invert the order in which information is provided, without this affecting the substantive results, in the analysis we reduce this experiment to two treatments.Footnote 11
In experiment 2, we provide a specific mechanism why citizens may support or oppose a trade agreement: namely because of the impact of trade agreements on wages. In many countries, the effects of trade on jobs and wages dominate public debate over trade (see, e.g., Guisinger, Reference Guisinger2017). We again included a direct trade-off, as we stressed both the winners and losers from the agreement in the same treatment. In this experiment, we also referred to a specific trading entity, namely India in the case of most countries and the European Union in the case of India and Bangladesh, with whom the home country signs a trade agreement. We opted for India because this is a relatively large economy, meaning that trade agreements with that country can have significant economic consequences. At the same time, the aim of mentioning India (instead of, for example, China) was to reduce the influence of geopolitical considerations on responses. Finally, India is a good case because it does not yet have trade agreements with most of the countries covered by our study. Using the European Union in India and Bangladesh has the same advantages. Again, two of the four treatments provide equivalent information, just in a different order, allowing us to work with only two treatments in the analysis:
Experiment 2: Trade agreements tend to create winners and losers. Concretely, a trade agreement with IndiaFootnote 12 could [lead to lower wages for workers in COUNTRY but increase profits of firms from COUNTRY exporting to India./increase profits of firms from COUNTRY exporting to India but lead to lower wages for workers in COUNTRY./lead to higher wages for workers in COUNTRY but reduce profits of firms from COUNTRY that compete with imports from India./reduce profits of firms from COUNTRY that compete with imports from India but lead to higher wages for workers in COUNTRY.] How much would you support or oppose this trade agreement? (11 point scale from “Would strongly oppose” to “Would strongly support”)
For experiment 3, finally, we decided to have separate treatments for winners and losers, rather than include both in a single treatment. What is more, we account for likely heterogeneity for both types of actors by specifying for each actor type a percentage that benefits or loses. Again, we refer to workers (citizens) and firms (business actors). The full wording of the experiment, which had four treatment conditions, was as follows:
Experiment 3: Imagine COUNTRY negotiates a trade agreement with an important trading partner. [70 percent of workers in COUNTRY would benefit from this agreement via higher wages./30 percent of workers in COUNTRY would lose from this agreement via lower wages./70 percent of firms in COUNTRY would benefit from this agreement via higher profits./30 percent of firms in COUNTRY would lose from this agreement via lower profits.] How much would you support or oppose this trade agreement? (11 point scale from “Would strongly oppose” to “Would strongly support”)
This set-up allows us to test our hypotheses in two comparisons, namely workers winning versus firms winning and workers losing versus firms losing.
Experimental designs are strong in establishing causal inference, but this often comes at the cost of low external validity. In the real world, factors (by design) excluded from our experiments matter for legislators’ decisions to side with either business or citizens. Consequently, our results are not point predictions, but rather serve as indications of the direction of the effects of individual factors. If we find empirical support for our hypotheses, we can infer that, ceteris paribus, legislators are inclined to defend citizens’ preferences in trade and that ideas play a moderating role in shaping this inclination. Once the ceteris paribus condition is relaxed, we can consider scenarios in which citizens and business actors differ in their degree of political engagement. Support for H1 would then imply that business actors must exert comparatively greater effort merely to neutralize legislators’ underlying predisposition toward citizens. However, further research on actual trade policy decisions is necessary to bridge the gap between experimental insights and real-world complexities.
Moderators
To assess Hypotheses 2 and 3, we needed to operationalize two moderators. Starting with the political left-right positioning of legislators (H2), we rely on legislators’ party membership as a proxy. We used parties’ economic left-right position instead of asking legislators directly about their ideology because we tried to avoid priming respondents before the experiments. The survey also needed to be short to maximize the response rate. Finally, the expert coding of parties may be more comparable across countries and world regions than legislators’ own assessments of their ideology.
Concretely, we used the economic left-right variable from the V-Party dataset (Lindberg et al., Reference Lindberg2022). Whenever we had missing observations, we further relied on variable V4 from the Global Party Survey (Norris, Reference Norris2020), which classifies the current stance of a party on economic issues such as privatization, taxes, regulation, government spending, and the welfare state based on expert judgments (measured on a continuous scale from 0 for far-left to 10 for far-right). To make the two datasets compatible, we converted the values from the second into an ordinal variable going from far-left to far-right. We relied on manual coding to close a few remaining gaps. Our sample covers legislators from the full ideological spectrum, with the modal legislator being from the center-right. In robustness checks, we substituted this variable with one capturing parties’ cultural ideology (for which we have a larger number of missing observations), without this changing the substantive findings (see Section H in the Appendix).
The second moderator is legislators’ role conception as either delegates or trustees (H3). For this variable, we relied on the following battery of six items included in our survey:
“To what extent do you agree or disagree with the following statements?
T1 In a representative democracy, parliamentarians should act in the interest of the country even if that means making unpopular decisions.
D1 In a representative democracy, parliamentarians should act according to public opinion even if it is at odds with their own opinion.
D2 Demonstrations and letter-writing campaigns are legitimate ways to influence the decisions of parliamentarians.
T2 Trade agreements are so technical in nature that ordinary citizens cannot comprehend them.
D3 The influence of economic elites on trade negotiations is too big.
T3 Populists exploit the lack of knowledge of ordinary citizens about trade agreements.”
Respondents could answer each item on a five-point scale from “strongly disagree” to “strongly agree.” For the variable we use, we subtracted the mean of all items capturing the support for a delegate role (indicated by D in the list above) from the mean of the items capturing support for a trustee role (indicated by a T). This operationalization results in a measure of support for a trustee model relative to support for a delegate model. Because of problems with bootstrapping when using the continuous version of this variable, we recoded this variable to indicate whether legislators are delegates (–0.5 or smaller), trustees (0.5 or larger) or neutral (> –0.5 but <0.5). Right-wing legislators and men are slightly more likely to exhibit a trustee understanding of their role (see Tables D2 and D3 and Section L in the Online Appendix). In robustness checks, we a) show that results hold when including the interactions with ideology and role conceptions in the same model and b) present results for the individual items included in this battery.
Estimation
As our surveys are nested within countries, we rely on linear mixed-effects models with random intercepts at the country level. These intercepts capture heterogeneity across countries that may influence legislators’ stances toward trade agreements, such as level of economic development and electoral system. In robustness checks, we substitute the random effects with fixed effects without this changing the substantive results. Although random assignment to the treatment groups obviates the need for other control variables, we present models with and without control variables. The models with controls allow for more precise estimates (for the results of these full models, see Section I in the Appendix). The variables that we control for in the full models are legislators’ age and gender, whether they form part of government or opposition parties, whether they are elected in single or multi-member districts, whether they are first-timers, and in which chamber of parliament (lower or upper) they serve. All of these variables were coded by us using publicly available information. We include interactions with both moderators in these models. Since our observations are clustered at several levels (within countries, chambers, political parties etc.), we use bootstrapping with 1000 replicates for standard errors and confidence intervals. Descriptive statistics for all variables used in the analysis are provided in Table D1 in the Appendix.
Results
Responsiveness to citizens
Figure 1 presents the results for the test of H1. The y-axis shows the marginal effects of the citizen treatment relative to the business treatment. Here and below, we focus on marginal effects as they allow us to test our hypotheses most directly. In the Appendix (Section G), we also show predictions from the various models. As expected, legislators exhibit greater responsiveness to citizens than to business. In experiment 1, we find that legislators are more supportive of trade agreements when the population supports them despite business opposition. The marginal effect here amounts to 1.4 points on the 11-point scale (around 13%). Experiment 2 confirms this finding, showing that legislators are more supportive of a trade agreement that increases wages at the cost of lower profits. The marginal effect here is even larger than in experiment 1 (2.3 points or 21%).
Hypothesis 1 – main effects.
Note: Figure 1 shows marginal effects for the citizens-business comparison. Larger values indicate more support for a trade agreement when citizens support it or benefit from it (relative to when business supports it/benefits from it). Table F1 in the Online Appendix reports the regression coefficients. 95% confidence intervals are bootstrapped.
Finally, experiment 3 shows that legislators react more negatively to losses for the population than to losses for firms. The marginal effect is –1.3 points and hence similar to the results found with experiment 1. In the realm of gains, however, we find no difference between business and citizens (the average marginal effect is –0.01, which is not statistically significant). The lack of differences between the firm and population treatments in the realm of gains may be due to ceiling effects, namely legislators evaluating agreements with 70% winners (independent of whether they are firms or citizens) very positively. Figure D1 in the Appendix, which shows the predicted values, lends some support to this argument, as the mean evaluation of an agreement that benefits firms or workers is 8 (on a scale that ranges to 10). The fact that legislators were not told whether the remaining 30% are losers or simply unaffected may also go some way in explaining this finding. These points also apply to the findings for this experiment reported below.
It may be argued that social desirability desire affects these results. Given that we asked legislators about their position toward a trade agreement and introduced citizen and business preferences only as context information, however, social desirability bias should play less of a role. As argued before, moreover, even if social desirability bias does play a role, our finding is important. If social desirability bias makes legislators indicate a preference for citizens in an experimental setting embedded in a confidential survey, it should also matter in actual trade policy debates. In turn, this should influence the policies they support, even if in the real world other factors – such as campaign donations, the salience of the issue, party dynamics and so on – that we cannot control for here also affect outcomes. In short, this evidence is strong support for H1. In the following, we investigate to which extent these findings are moderated by political ideology and legislators’ role conceptions.
The moderators
Hypothesis 2 expects that left-wing legislators are relatively more concerned with being responsive to citizens (compared to business) than right-wing legislators. Figure 2, which shows the marginal effects of the citizen relative to the business treatments, suggests that this is indeed the case. Starting with the left pane (Experiment 1), we observe a substantially larger treatment effect of support by the population relative to support by business among left-wing respondents. The marginal effect of the population treatment is 2.8 (on an 11-point scale) for the most left-wing legislators, a value that declines to roughly 0 for the most right-wing legislators. For the full range of the moderator, the effect hence is 2.8, or 25% of the full scale. This is strong support for our expectation that left-wing legislators do not object in principle to trade agreements, but that their evaluations of trade agreements depend on their expected consequences for the constituency they favor. Figure G2 in the Online Appendix shows that this effect is driven by left-wing legislators’ reaction to the “business associations support but population opposes” treatment (whereas right-wing legislators support all trade agreements, independent of the treatment).
Hypothesis 2 – the moderating effect of political ideology.
Note: the interpretation is as for Figure 1. We show results for both a continuous (lines) and a categorical (points and whiskers) measure of ideology to check for possible non-linearity of the interaction. Table F2 in the Appendix reports the regression coefficients. 95% confidence intervals are bootstrapped.
We observe a similar picture when it comes to the relative importance of higher wages or higher profits (Experiment 2). Left-wing legislators exhibit a greater positive treatment effect for higher wages relative to higher profits than right-wing respondents. The difference from the most left-wing to the most right-wing legislators here amounts to 2.5 points on the 11-point scale (or around 23%). The difference to experiment 1 is that even the far-right in terms of political ideology slightly prefers higher wages for workers to increased profits for firms.
Experiment 3 further corresponds with this overall pattern. Here we show the marginal effects of the citizen versus the business treatments separately for whether these actors benefit (from higher wages or higher profits) or lose (from lower wages or lower profits). For the loss treatment, we again find that left-wing legislators are more responsive to citizens (hence the negative marginal effect) than right-wing legislators. On the 11-point scale that we used in this experiment, the difference between the most left-wing and the most right-wing legislators amounts to 1.6 points (15% of the overall scale). The effect is substantially smaller (0.7 point) and does not reach statistical significance for the benefits treatment (and there is also some evidence of non-linearity). Figure G2, which shows predictions instead of marginal effects, shows that the effect of this experiment is mainly driven by the workers losing treatment.
In line with the experimental evidence, a center-left legislator from South Africa indicated that “Agreements which only ensure higher profits are no guarantee that the larger populace would also benefit. In a country like South Africa, this may lead to heightened levels of inequality” (open survey response). By contrast, a legislator from a business-friendly party reasoned that “profits for companies will also benefit employees” (open survey response). An interviewee indicated that ideology also affects whether legislators side more with citizens or business through another channel: as he is economically right-wing, he has many more contacts with business actors, because actors self-select into contacting him depending on his ideology (Interview #23). Overall, the evidence hence strongly supports Hypothesis 2.
Moving on to Hypothesis 3, we anticipate legislators who conceive of their role as delegates to pay greater attention to the preferences of citizens relative to those of business than legislators who see themselves as trustees. The findings for this hypothesis are shown in Figure 3. In this figure, the x-axis indicates whether respondents tend to see themselves as delegates, trustees or are rather indifferent, whereas the y-axis again shows marginal effects for citizens supporting or workers gaining. The findings largely correspond with Hypothesis 3. In experiments 1 and 2, legislators that conceive of themselves as delegates (left side of the figures) attach much more weight to citizen preferences than those who have a trustee vision of their role (on the right side). The substantive effect is larger in experiment 1 (3.2 points on the 11-point scale) than in experiment 2 (1.8 points). In experiment 2, moreover, there is some evidence of non-linearity. A closer look at the results (see Figure G3 in the Appendix) indicates that these effects are mainly driven by trustees and delegates reacting very differently to the “business supports/population opposes” and “firms win/workers lose” treatments.
In experiment 3, for both benefits and losses the difference in the marginal effect of the citizen treatment between the legislators seeing themselves as delegates and those seeing themselves as trustees amounts to about 0.6 points. While both treatments go in the direction expected in Hypothesis 3, neither effect is statistically significant.
While thus only two of the three experiments clearly support Hypothesis 3, the qualitative evidence strongly backs our expectation. A legislator with a conception of her role clearly in line with the delegate model wrote in the open responses that “[p]olitical representation is granted by citizenship, not by the business elites or the lobbies that try to manage the future of the world in their interests” (open survey response, our translation, emphasis in the original). By contrast, a legislator whom our battery identifies as a trustee wrote that trade agreements “are as important as they are necessary, even if they are unpopular” (open survey response, our translation). Overall, therefore, the evidence also backs H3.
Robustness checks
To analyze the robustness of our findings, we first focus on alternative operationalizations of the core variables to account for potential sensitivity to measurement choices (see Section H). Concretely, we substituted parties’ cultural ideology for the measure of economic ideology without this changing the substantive results. The results also suggest that economically left-wing parties independent of their stance on the cultural dimension react differently from their right-wing counterparts. Splitting our relative trustee measure into its individual six items and including them separately in the model does not change the substantive interpretation of the results either. Even if in some cases weaker, the effects generally remain consistent with our expectations.
Second, to account for potential omitted variable biases, we replicate our main results with the set of control variables mentioned in the estimation subsection. These models also include both two-way interactions used to test Hypotheses 2 and 3. Section I in the Appendix presents the findings. Overall, the findings support the robustness of our findings, but the results for experiment 3 are weaker than in the lean models. This is driven by a smaller number of observations (because of missing observations for control variables) and a larger number of predictors in the models.
Third, we adopted two alternative approaches to deal with heterogeneity across countries. On the one hand, we replaced the random effects with country fixed effects, without this changing the substantive results (see Section M). On the other hand, we used jackknifing – namely iteratively running the models while each time dropping one of the countries – to assess whether individual countries have an undue influence on the results reported. Figures J1 to J3 in Appendix Section J show that no country drives the results we report. Finally, we dropped all legislators from upper chambers of parliament, as their incentives may differ from those of members of parliaments’ lower chambers. Even with this more restricted sample, however, the results remain very close to those reported before (see Appendix Section K).
Conclusion
We have argued that in democracies, legislators have incentives to be responsive to both business actors and citizens when taking trade policy decisions. For legislators, this creates a dilemma whenever (a majority of) business and (a majority of) citizens have divergent trade policy preferences. In these situations, legislators need to decide whether to lean toward business or citizen preferences. Our argument has been that, ceteris paribus, legislators should show greater responsiveness toward citizens. We also expected this effect to be moderated by legislators’ political ideology and role conception. Relying on three survey experiments carried out with over a thousand legislators from around the globe, we have found strong support for our expectations.
Our findings are of broad relevance at a moment when trade has become politicized in many countries (Kay and Evans, Reference Kay and Evans2018; Dür et al., Reference Dür, Eckhardt and Poletti2020). Deindustrialization in developed countries has been accompanied by debates about the distributional effects of trade, which have influenced electoral outcomes (Margalit, Reference Margalit2011; Autor et al., Reference Autor2020; Baccini and Weymouth, Reference Baccini and Weymouth2021; Che et al., Reference Che2022). This has led to the observation of a globalization backlash, which seems to be also driven by a feeling that globalization reflects business interests more than citizens’ interests. In turn, this feeling can lead to people rejecting the liberal international order and looking for radical alternatives (De Vries et al., Reference De Vries, Hobolt and Walter2021; Lake et al., Reference Lake, Martin and Risse2021).
We make several contributions to these debates. For one, by focusing on legislators’ attitudes rather than policy outcomes, our research allows for a better understanding of the process linking the distributional effects of trade to outcomes. Moreover, we offer evidence from democracies across the globe, which contrasts with the large number of studies on the U.S. Finally, our experimental evidence complements the observational evidence that so far dominates research in this area.
The findings have important implications for the broad literature on trade policy-making. In line with some studies (McGillivray, Reference McGillivray2004; Kono, Reference Kono2006; Mansfield and Milner, Reference Mansfield and Milner2012), they show that an effect of citizens’ trade attitudes on trade politics should not be discarded too readily. Whereas the literature on the politicization of trade has already taken this into account (Kay and Evans, Reference Kay and Evans2018; Dür et al., Reference Dür, Eckhardt and Poletti2020), other strands within the trade policy literature still assign citizens little to no role. Legislators’ willingness to respond to citizens’ preferences as shown in this paper may explain why, on average, democracies have lower trade barriers than autocracies, as citizens qua consumers may have a preference for cheaper imports. Moreover, our findings make it plausible that politicians’ ideas and ideology have some impact on trade policy outcomes (Goldstein, Reference Goldstein1993), at least because they shape legislators’ relative responsiveness to different constituencies.
A possible limitation of our study is that the survey provides only a single snapshot in time. We cannot assess whether the effects would look different if the experiments had been conducted at a different time. For instance, the salience of trade policy has increased considerably in the wake of the tariffs imposed by the second Trump administration. However, the fact that our results are highly consistent across countries, despite trade policy debates being influenced by different factors in each, suggests that the effect we identify is robust rather than an artifact of timing.
Going beyond research on trade policy, the paper contributes to the large literature on representation and responsiveness. Some of this research finds that average citizens only have a little impact on policy outcomes (Gilens and Page, Reference Gilens and Page2014), whereas other studies uncover a coincidence between public opinion and public policies (e.g., Canes-Wrone, Reference Canes-Wrone2015; Rogowski, Reference Rogowski2023). We cannot fully resolve this difference, because our experimental evidence abstracts from the many ways in which business actors could win out over citizens even if legislators are predisposed to give more weight to citizen preferences. Nevertheless, finding such a predisposition in the trade domain is noteworthy, since trade is often considered a “hard” case to detect an influence by public opinion. In a rather technical policy field, legislators may be more likely to ignore citizens’ trade preferences. Future research could expand the present study to policies other than trade – such as immigration – that also have distributional effects and where at least sometimes business actors and citizens have opposite preferences.
Normatively, our findings make us cautiously optimistic about responsiveness in democracies. Being responsive to citizens turns out to be more important for legislators on average than being responsive to business interests. This can serve as a counterweight to the influence that business can exert via lobbying during the policy-making process. However, this only works if legislators know what citizens want or which policies would serve the broader public, which may not always be the case. To further develop these normative implications, future research should link our results to legislative votes to see to which extent the patterns we find for legislator attitudes also show up in legislator behavior.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1755773926100320.
Data availability statement
A replication archive is available at https://doi.org/10.7910/DVN/GRDYEI.
Acknowledgements
We gratefully acknowledge help with the data collection from Yannick Stiller. We also thank audiences at the annual conferences of the European Consortium of Political Research and the European Political Science Association, the Dreiländertagung, and the University of Konstanz for helpful comments on earlier versions of this paper.
Funding statement
This research received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No. 724107).
Competing interests
The authors declare none.
Ethical standards
The study received ethics clearance as part of the ethical review of the TRADEPOWER project. Respondents were asked for informed consent before starting the survey or doing the interviews (see also Section C in the Online Appendix). No deception was used either in the experiments or the interviews. Since we promised anonymity to our respondents, the interviews are cited only with a numeric ID.
Open access
