A need for control? Political trust and public preferences for asylum and refugee policy

Political trust matters for citizens ’ policy preferences but existing research has not fully understood how this effect depends on policy design. To advance this research area, we theorise that policy controls that limit or condition policy provision can function as safeguards against uncertainty, thereby compensating for a person ’ s lack of trust in generating support. Focusing on public preferences for asylum and refugee policy, we conduct an original conjoint experiment in eight European countries. We find that individuals with lower levels of trust in European political institutions are less supportive of policies providing unlimited or unconditional protection and more supportive of restrictive policies. We also show that policy design features such as limits and conditions can mitigate perceived uncertainty for individuals who are less trusting in European political institutions. These findings have important implications for the theoretical understanding of how political trust pertains to citizens ’ preferences.


Introduction
The trust that people have in political institutions is an important ingredient in the formation of their preferences on a wide range of public policy issues. It acts as a lens through which citizens see their political institutions (Rudolph, 2017). Through this function, the degree of people's political trust influences the extent of government action they support. This is particularly important for policies that mostly benefit political minorities as previous research has shown (Hetherington, 2004;Rudolph and Evans, 2005;Popp and Rudolph, 2011;Paxton and Knack, 2012). Yet, the current academic understanding of how political trust conditions policy preference formation is still rather limited and incomplete, as Citrin and Stoker (2018) remark in their recent review essay. While previous research has demonstrated that political trust matters for policy preferences (see Rudolph, 2017 for an overview), including immigration policy (Macdonald and Cornacchione, 2021), we argue that it is limited by its dichotomous conception of public preferences (i.e., 'supporting' vs 'opposing' a policy). This dichotomous approach does not sufficiently reflect the complexity of citizen preferences as it does not explore the role of policy design.
Policy design has been shown to have its own separate influence on policy preferences (Ackert et al., 2007;Bechtel et al., 2017;Jeannet et al., 2021;Vrânceanu et al., 2022), so it is important to explore the potential interactions between policy design and political trust. A widespread lack of trust among citizens hinders the government from pursuing liberal policies of government spending because they then lack the support of conservative voters (Citrin and Stoker, 2018). At the same time, a growing body of research demonstrates that policy design mitigates this by affecting citizen orientations towards a given policy across various fields, including welfare state policy (Gabriel and Trüdinger, 2011), climate change policy (Fairbrother, 2019), government surveillance policies (Ziller and Helbling, 2021), and land-taking compensation policies in China (Cai et al., 2020).
The aim of this study is to examine how policy design and political trust interact in shaping policy preferences. We test this framework empirically by studying public preferences for asylum and refugee policy in a cross-national experimental setting. There are several reasons why asylum and refugee policy is a particularly suitable policy area for our analysis. Political trust is especially important for certain types of policies. It is particularly pertinent to policies under which the majority of citizens are not primary beneficiaries; in other words, they do not receive any direct material benefit from this policy, yet they may (or perceive to) incur the costs of these policies as taxpayers (Hetherington, 2004). Asylum and refugee policies are good examples of such policies. Moreover, the specific nature of asylum-seeking and refugee protection, in particular the complexity of the policy processes and the volatility of migrant arrivals, generates considerable uncertainty and makes citizens' confidence in the functioning of political institutions especially pertinent.
This article makes several contributions. First, we propose a new theoretical framework to explain how policy design can mitigate the role of low political trust in conditioning public policy preferences. Second, we build on earlier observational approaches and analyse the relationship between political trust and policy preferences in an experimental setting. Our conjoint methodology allows us to demonstrate empirically how certain policy instruments enable distrusting individuals to nevertheless generate support for policy areas that are not directly materially beneficial to them. Our results thus imply that political trust does not simply have a binary effect on policy preferences, as the impact of political trust critically depends on the design features of the policies. We show that specific policy design features, such as limits and conditions, are able to mitigate perceived risk and uncertainty, which we expect to be crucial for individuals who are less trusting in political institutions.
Political trust, policy preferences and the role of policy design Hetherington (2004) proposes a theory of political trust that helps us understand when political trust is relevant in preference formationin other words, why and how it is important in some policy areas and not in others. According to Hetherington, political trust plays an important role in preference formation when a policy involves sacrifice and risk (p. 6) and when it concentrates its 'benefits on a minority while imposing the real or perceived costs on a political majority' (Hetherington, 2004: 106). In other words, trust can be expected to matter most when the majority is asked to make a perceived sacrifice without receiving tangible benefits in return.
Political trust can affect individuals' policy preferences by offering a way of coping with the complexities of the world today 'by structuring views about specific ( : : : ) policies according to their more general and abstract beliefs' (Hurwitz andPeffley, 1987: 1114). In particular, when it comes to highly complex policy issues, individuals are likely to rely on cognitive simplification strategies that minimise the time and cognitive effort in the formation of their judgements. In such cases, political trust functions as a heuristic device, or mental shortcut, that allows people to expend less effort in gathering information for their decision-making. This would mean that individuals who have little (or a lot of) confidence in political institutions are using their negative (or positive) evaluations as a heuristic in the formation of their policy preferences. Therefore, individuals who are distrusting of governmental institutions are inclined to restrict the scope of the state's activities and spending (Hetherington, 2004), while more trusting individuals are more open to cooperating or supporting government initiatives (also see Putnam, 1993Putnam, , 2000 such as welfare state reforms (Garritzman et al., 2023). A lack of trust in an institution also makes people less willing to accept its decisions in general (Tyler and Degoey, 1996).
However, this current understanding of how political trust affects public support for government policies is limited by its consideration of public preferences in terms of binary policy choices. It suggests that political trust determines whether individuals support or oppose government activity in a certain policy area at all. Hetherington (2004: 139) argues that 'when the public does not trust that the government will implement the policies efficiently or fairly, people will prefer that the government not be involved'. This binary approach obscures a more nuanced role that political trust can play in the formation of policy preferences. Individuals not only decide whether they support policy provisions or not, but they are also influenced by the specific design of the policy. Our aim is to go beyond Hetherington's theory by considering the role of political trust in the formation of policy preferences involving non-binary policy choices. More specifically, we theorise how certain policy designs can encourage individuals who distrust 1 political institutions to nevertheless be supportive of policies that require sacrifice but for which they do not receive tangible benefits.
From a theoretical perspective, certain designs of a policy can mitigate low political trust amongst citizens by conferring legitimacy through perceived procedural fairness (Grimes, 2006;Doherty and Wolak, 2012). Policy design features are procedural as they represent 'the accurate anticipation of the consequences of government actions and the articulation of specific courses of action to be followed' (Howlett and Lejano, 2013: 358). The collective features of a policy capture the multidimensionality of a policy rather than simply a dichotomous decision of the citizen to be 'pro' or 'con' (for instance, pro-immigration). Perceptions of procedural propriety build confidence in authorities and institutions, making citizens more confident to delegate decision-making to the state (Grimes, 2017: 257). Such procedures include formal and informal rules that determine how the state handles a particular aspect of governing, such as a policy area. This conveys a sense of 'fair rules, fairly implemented' (Pearce, 2007: 11), even if this is subjective and what is considered to be fair varies across individuals. Procedures render the process of decision-making, from the very top of the state down to the front-line bureaucrats, more transparent and responsive in the eyes of citizens (Thibaut and Walker, 1975).
The area of asylum and refugee policy is a multi-dimensional 'policy field' that involves a range of procedures and requires decision-making on multiple bundles of potential policy features. For instance, what are the criteria for granting asylum or refugee status? How, if at all, should the number of people receiving protection be regulated? What are the rights and obligations associated with refugee status? What should happen to those who are not granted asylum? To what extent, if at all, should the policy include support for other countries that host refugees? The answers to these questions, and the associated policy features, are very much linked to each other and together constitute a multi-dimensional refugee policy or 'system'. The reform of asylum and refugee protection policiesat national, regional, and global levelsis typically discussed in this multi-dimensional way. This is exemplified in the 'new vision' for asylum and refugee protection presented by the Austrian Ministry of Interior and Danish Ministry of Immigration and Integration in 2018, the European Commission's 'New Pact on Asylum and Migration' in 2020, or the United Nations' 'Global Compact on Refugees' in 2018.
Public policy debates about these questions are typically characterised by considerable uncertainty about the scale of the issue, the conditions under which it occurs, and its effects on host country citizens. While refugees and asylum-seekers remain a relatively minor component of immigration in wealthy democracies, they are nonetheless disproportionately present in the public's perception of who the immigrants are (Blinder, 2015;Blinder and Jeannet, 2018). The mass media often frames asylum and refugee issues as uncertain to the public (Esses et al., 2013), particularly after the global refugee 'crisis' in 2015-2016. The number of asylum 1 We use the terms low trust and distrust interchangeably throughout this article.
Political trust and preferences for asylum and refugee policy seekers asking for protection in a particular host country can be highly volatile and is, at least according to the letter and spirit of the 'Geneva Convention', 'unlimited'. Depending on the prevailing rules and regulations, not all asylum seekers will receive protection. Those that do are likely to acquire some rights to family reunification which can lead to more immigration. Those that are refused may be asked to return home, although return will not be legal under international law if it violates the principle of 'non-refoulement' enshrined in the Geneva Convention, and it may also not be possible for other reasons (e.g., if the migrants' origin country does not cooperate with return and readmission). All of the above further adds to uncertainty about scale and effects on the host country. Debates about asylum and refugee protection are, therefore, often about whether and how to regulate the scale of policy provision and the conditions under which it occursquestions about fundamental principles that are also at the centre of debates in other policy areas.
To our knowledge, there is no study of the role of policy features in conditioning the relationship between political trust and policy preferences for asylum and refugee policy. Asylum and refugee policies are a clear example of policies that benefit a political minority in the host country (i.e., non-citizens seeking protection) while requiring the majority of citizens to make a sacrifice in the sense that these policies do not generate immediate and tangible material benefits for them.
There are, however, a handful of studies in other policy areas that do provide relevant insights for our analysis. Firstly, there is research showing that the uncertainty that surrounds a policy area influences the public's policy preferences. For instance, an experimental study of preferences demonstrates the importance of policy uncertainty and the level of government at which the decision is taken across policy areas (Christensen and Rapeli, 2021). Then, there is a group of studies across various policy fields showing that policy design matters for public preferences, particularly in tax and fiscal policy (Ackert et al., 2007;Ballard-Rosa et al., 2017;Bechtel et al., 2017) but also in asylum and refugee policy (Jeannet et al., 2021). Moreover, political trust appears to be an important consideration when understanding policy preferences. For example, using a conjoint experiment to investigate public preferences for technocratic expertise, Bertsou (2022) finds that less trusting individuals have distinct patterns regarding the delegation of decision-making authority. Conditionality in policy may allow low-trusting individuals to have more confidence in supporting a policy as has been demonstrated in the area of health spending (Busemeyer, 2021). While the existing body of research seems to suggest that policy features and design might be particularly pertinent to low-trusting individuals, much is still to be learned about how and why this may be the case.
So, what policy features might allow distrusting individuals to form supportive policy preferences? We know from existing research that, in addition to acting as a heuristic that helps reduce complexity, trust is an important resource for coping with uncertainty (Kollock, 1994;Yamagishi et al., 1998;Ellinas and Lamprianou, 2014), which is also relevant to policy preference formation. We argue that certain policy instruments can offer an alternative mechanism for mitigating uncertainty amongst individuals who lack trust in political institutions. Given that a sense of distrust tends to accompany 'a course of action based on suspicion, monitoring, and activation of institutional safeguards' (Lewis and Weigert, 1985: 969), we reason that distrusting individuals can support more expansive government policies if they employ explicit means of control. Such means of policy control can function as safeguards against uncertainty, compensating for a person's lack of trust in political institutions during preference formation. We identify two potential instruments of policy control: limits, which ration the policy provision and conditions, which regulate the policy's effective provision according to well-defined rules (see Spicker, 2005). 2 Recent research finds that European citizens do not only consider the personal or national material benefits when forming their preferences for asylum and refugee policy but also reconcile these with their ideological beliefs by preferring policies that include instruments of control such as limits or conditions (Jeannet et al., 2021).
Based on the idea that policy controls can act as safeguards against uncertainty, we reason further that the use of limits and conditions carries greater importance in preference formation of distrusting individuals relative to trusting individuals. This is because, if they are to support some form of policy provision, low-trusting individuals have a greater need to rely on policy safeguards to compensate for the perceived uncertainty generated by their lack of political trust. In other words, we expect that the difference between the extent of individuals' support for policies that feature limits and conditions, and their support for policies with unconditional or unlimited features, will be accentuated for individuals who are less politically trusting. Based on these expectations, we formulate the following testable hypotheses: Hypothesis 1: Individuals with lower levels of political trust are more supportive of policies that provide protection and assistance to refugees if these policies utilise controls by applying limits on admission.
Hypothesis 2: Individuals with lower levels of political trust are more supportive of policies that provide protection and assistance to refugees if these policies utilise controls through conditionality.

Empirical approach
We implemented an original choice-based conjoint survey experiment to examine if the relationship between political trust and public preferences for asylum and refugee policy is contingent upon how the policy is designed. 3 In our conjoint experiment, respondents were shown pairs of randomly generated policies and asked which of the two policies they would prefer their country to adopt. This randomised design allows researchers to isolate the separate causal effects of particular policy features in garnering public support (see Hainmueller et al., 2014). A conjoint experiment has some notable advantages over observational survey designs that makes it well-suited for this study. Most importantly, it allows us to assess the influence of policy design features on citizen support for asylum and refugee policy and how this varies across individuals who differ in their extent of political trust. Unlike previous research on political trust and policy preferences which is predominantly observational, the conjoint design helps us to minimise the possibility of social desirability bias which is crucial in policy areas that are strongly subject to ethical and humanitarian considerations. It does so by minimising the likelihood that respondents provide a response they believe to be politically correct or 'expected' by the researchers, since the different policy options vary across several dimensions (Hainmueller et al., 2014).
Our analysis focuses on Europe for several reasons. Just as in the USA, Europe has also experienced a decline in public trust in political institutions in the last decade (Brechenmacher, 2018). Several recent studies have shown that political trust is relevant for policy preferences in various European countries (Trüdinger and Steckermeier, 2017;Fairbrother et al., 2019;Fairbrother et al., 2021), and for Europeans' immigration policy preferences in particular (Macdonald and Cornacchione, 2021). Europe also offers a rich environment for empirical study since it includes different national settings with varying levels of political trust.
Our survey experiment was conducted online in May 2019 across eight European countries: Austria, France, Germany, Hungary, Italy, Poland, Spain, and Sweden. The countries have been selected on the basis that they represent a variety of experiences with refugees and asylum seekers, cover several geographic areas of the European Union, and include a wide variety of labour market conditions, welfare systems, and cultural institutions. These countries are also among the most populous countries in the European Union, also making our sample more representative of European public preferences overall. 4 The total sample size was 12,000 adults, comprising a nationally representative sample of 1,500 in each country. 5 Respondents first read the instructions of the survey and were then shown an introductory page that briefly explained the key terms used in the experiment (such as 'asylumseeker', 'refugee', and 'resettlement'). 6 Each respondent was asked to make five binary policy comparisons, meaning that after completing the survey each respondent had considered and assessed ten randomly generated policies. All asylum and refugee policies shown to respondents included five policy dimensions, with two to three possible policy features selected randomly within each dimension. As part of our experimental design, we also include a sixth dimension, 'decision-making,' that does not intend to test our hypothesis regarding policy conditionality but, rather, is included as a validity-checking exercise to exclude the possible framing effects of having either the national government or the EU government, as the decision-making bodies of the asylum system, influencing our findings.
For each asylum and refugee policy that a respondent considered, we constructed a variable policy support and coded it as 1 if an individual chose this policy and 0 if it was not chosen. To investigate a more nuanced portrayal of policy support, we also ask respondents to rank each policy on a scale from 1 to 7 after choosing one of the two policies. 7 After completing the five conjoint tasks, the survey asked respondents a series of questions about their age, gender, education, political orientation, preferred scale of immigration, and political trust. 8 We randomised the order of the questions about immigration policy and political trust across the respondents. 9 Table 1 shows the randomly allocated policy features for each of the dimensions of asylum and refugee policy. We identify five core dimensions of asylum and refugee policy, drawing on recent research (Jeannet et al., 2021). These relate to: the right to apply for asylum; the resettlement of recognised refugees; the return of asylum seekers whose applications for protection have been unsuccessful to countries where they might face harm; the right to family reunification for recognised refugees; and the provision of financial assistance to first countries of asylum, i.e., lower income countries outside Europe that host large numbers of refugees near conflict regions. A sixth dimension regarding the level of decision-making for asylum claims (EU vs national) is included as a validity check, as described above. As shown in Table 1, within each dimension, we randomise policy design features that include or exclude limits or conditions. An example of a conjoint task as it appeared in our survey can be found in online Appendix D. The order in which the dimensions were listed was randomised for each respondent.
We used two survey items to measure a person's political trust. 10 Given the multi-level governance of asylum and refugee issues in the European Union, we measure both trust in EU institutions and national government institutions. EU trust and national political trust are 4 We have intentionally excluded the UK as our study occurred after the UK's referendum on EU membership and during the Brexit negotiations. 5 The survey company that implemented the experiment, Respondi, uses matched sampling procedures which has been shown to be a highly accurate technique for approximating a random sample (see Ansolabehere and Schnaffer, 2014). 6 The text of this introduction can be found in the online Appendix A. To be sure that these definitions did not prime the respondents' conjoint tasks, a group of respondents (n = 1015) was not shown this introduction page. To rule out a priming effect, we do not find significant differences between the preferences of individuals who were shown this introductory page and individuals who were not (see online Appendix N for estimates). 7 We have used these ratings as a robustness check for our dependent variable measurement (see Appendix M). It also allows us to validate the measurement of policy choice as well as identify individuals who were inattentive (e.g., because they gave inconsistent answers) and whose choices may thus decrease data quality. conceptually distinct (see De Vries, 2018), and we expect both to matter in the formation of Europeans' asylum and refugee policy preferences. While national political institutions govern many dimensions of asylum and refugee policy, asylum and refugee policy is widely perceived as a European issue due to the way it has been framed in the public discourse (d' Haenens and de Lange, 2001;Horsti, 2007;Slominski and Trauner, 2018). For these reasons, we expect political trust in European institutions to play a particularly important and potentially dominant role.
As a precaution, we take several steps to validate our measures of political trust to check for any potential bias. First, we must consider the possibility of measurement error in the wording of the question. This might be particularly delicate when it comes to trust in national institutions (see footnote 8), which, if worded in a certain way, could capture satisfaction in the current government. Given its cross-national design, our survey was conducted in seven different languages (respondents in both Germany and Austria were surveyed in German), and we used 'back-translation' to make sure that the wording of the question was translated in a way that conveyed our intended question.
Another potential limitation of our experimental design is that the measurement of political trust might be biased by the ordering of the survey questions. Respondents were first asked to complete the conjoint tasks and then shown a series of attitudinal questions, including the question about political trust. In this sense, simply thinking about different aspects of asylum and refugee policy might have had a priming effect. To validate our measures and to rule out these possible sources of bias, we calculate the national means in national trust for each of the countries institutions." Respondents were asked to choose from the following response categories: entirely trusting, somewhat trusting, a little bit trusting, a little bit distrusting, somewhat distrusting, entirely distrusting.
in our study and find that they are highly correlated with the means for these same groups of countries surveyed in 2018 in the European Social Survey. 11 To analyse the results of our experiment across sub-groups of respondents with different levels of political trust, we follow the approach by Leeper et al. (2020) and compute the conditional marginal means, as this allows us to compare the effects of different levels of political trust on individuals' policy preferences in a more intuitive manner than other approaches. When computing the marginal means, we follow standard practice and apply cluster-robust standard errors at the respondent level to correct for possible within-respondent clustering. In all our analyses, we use entropybalancing survey weights to correct for sampling error. For comparison, we show average marginal component effects in Appendix K. As shown in Appendix L, the cross-country estimates are well powered (1-β = 0.8 for α = 0.05) to detect an average marginal component effect of 0.01, while the single-country estimates in the Appendix are powered to detect an AMCE of 0.03.

Experimental results
We find strong evidence that policy preferences for asylum and refugee policies are conditional on a person's trust in EU institutions. The results are displayed in Figure 1 below in the form of marginal means. We distinguish between six sub-groups of respondents who differ in their degrees of political trust in European institutions. 12 The marginal means can be interpreted as an indication of how favourably a policy is viewed. In a forced-choice design such as ours, where respondents need to choose exactly one of the two policies they are shown, a person randomising their choice would select each policy feature with a probability of 50%. A marginal mean of, for example, 55% indicates that policies that include this particular feature are selected with a probability of 55%.
The findings in Figure 1 strongly support our expectation that individuals who are more distrusting tend to be less supportive of policies that include expansive, unlimited, and unconditional features and more supportive of policies that eliminate protection/assistance in some policy dimensions. For example, in the asylum dimension, distrusting respondents are significantly less likely to support policies that feature unlimited asylum applications than the most trusting respondents. 13 Similarly, considering unconditional family reunification, the most distrusting people are considerably less likely to support policies that allow for unconditional family reunification than the most trusting people. 14 The same patterns of lower support for unconditional policies amongst individuals with less political trust can be observed for never returning refused asylum seekers to places where they could face harm, unconditional financial assistance to non-EU countries hosting refugees, and high levels of refugee resettlement.
At the same time, less trusting respondents are significantly more likely than trusting persons to support policies that eliminate protection and assistance. As can be seen in Figure 1, this holds for all our policy dimensions that feature the elimination of protection, rights, or assistance altogether: distrusting individuals are significantly more likely to support policies that do not provide any financial assistance to non-EU countries (difference in marginal means between low-trust and high-trust respondents is 5.1% points) and that do not allow for any refugee resettlement (difference in marginal means is 3.8% points on the three-point scale trust variable). Distrusting individuals are also considerably less likely than trusting individuals to oppose policies 11 We use the 2018 European Social Survey and construct an index for political trust using three items: trust in political parties, trust in parliament, and trust in politicians. The correlation coefficient between the mean national political trust in our survey and the mean political trust in the European Social Survey for these eight countries is 0.82. 12 Results for all respondents without distinguishing by level of political trust are available in Appendix E. 13 The difference in marginal means between the high-and low-trusting sub-groups of respondents is 3.7% points and even 7% points if not aggregating trust from the six-point scale into a three-point scale, cf. Figure K5 in the Appendix. 14 The difference between the marginal means of the most and least trusting sub-groups is ten percentage points ( Figure K5). In the aggregated form in Figure 1 it is nine percentage points. that never provide family reunification for recognised refugees (marginal means 49.5 15 and 40.1%, respectively).
Our results suggest that distrusting respondents prefer policies that abandon protection to policies that provide protection without limits or conditions. The inverse applies to the most trusting individuals. For example, with regard to family reunification, the most distrusting individuals are more supportive of a policy that abolishes the right to family reunification than a policy that provides this right unconditionally (marginal means difference = 8.9% points). In stark contrast, among the highest trusting individuals, a policy that abandons family reunification is 14.2% points less likely to be supported than a policy that allows for unconditional family reunification. A very similar pattern can be observed for financial assistance to non-EU countries. In the case of resettlement, distrusting people, unlike those individuals with high amounts of political trust, prefer policies that do not allow for any resettlement to policies with high levels of resettlement.

15
It is striking how large the differences within a broader trust category are, i.e., when people move from somewhat distrusting to highly distrusting. The average of 49.5 here hides that the former have a marginal mean of 47.4 while the latter have one of 51.7. Please refer to Appendix K for the more fine-grained respective graphs. Therefore, if we consider respondents' preferences for policies that include 'extreme' policy features only, i.e., 'no protection/assistance' and 'protection without limits and conditions', we find support for Hetherington's (2004) argument that distrusting individuals on average prefer no intervention by the government over government intervention. However, our analysis of the role of political trust goes beyond this binary understanding and also considers policy preferences when policy controls such as limits and conditions are employed. Our results show an important nuance and new insight, namely, that even distrusting individuals can support policies if they include limits or conditions. Figure 1 shows clearly that individuals with lower levels of political trust are more supportive of policies that provide protection and assistance to refugees if these policies utilise controls. This provides empirical support for both Hypothesis 1 (which pertains to limits) and Hypothesis 2 (which relates to conditions). It holds across all five policy dimensions that include values with features of limits or conditions. For example, individuals with low levels of trust show greater support for asylum and refugee policies that include limits on annual asylum applications (MM = 0.54) than for policies that do not include such limits (MM = 0.46). Similarly, people with low trust show considerably more support for asylum and refugee policies that condition family reunification on the refugee's ability to cover the costs of living of their family members (MM = 0.56) than for policies that facilitate family reunification without this condition (MM = 0.44). Similar preference structures can be observed in the policy dimensions relating to return, financial assistance to non-EU countries hosting refugees, and refugee resettlement.
Across all policy dimensions except for the 'return dimension', a low level of political trust accentuates the relative difference between individuals' support for asylum and refugee policies that feature limits and conditions and their support for policies that do not. 16 In other words, policies that feature limits and conditions are more relevant to the formation of supportive policy preferences of low-trusting individuals. For example, considering financial assistance to non-EU countries hosting refugees, the difference between policies that include conditions on financial assistance and unconditional financial assistance is much larger for people with low trust than for individuals with high trust. In fact, people with the highest degree of trust in our sample do not differ in their support for policies that provide conditional or unconditional financial assistance. As we expected, low levels of political trust amplify the positive role of limits and conditions in generating support for asylum and refugee policies that provide protection and assistance.
Overall, our results support our theoretical argument that policy controls can compensate, partially or even fully, for a lack of trust in generating support for asylum and refugee policies. They also show that, in some cases, distrusting people can prefer policies that utilise policy controls to policies that provide no protection or assistance. For example, with regard to family reunification, distrusting people are more likely to support the conditional policy (MM = 0.56) than a policy of no family reunification at all (MM = 0.49). Appendix F, Panels 1-8, replicates these results separately by country. This general pattern, can be found in all countries. 17 The results discussed above all relate to individuals' trust in EU institutions which we find plays a much larger role in conditioning public asylum and refugee policy preferences than people's trust in their national government institutions. Still, our results for national political trust indicate that the use of limits and controls can generate support among people with low trust for asylum and refugee policies that provide protection to refugees.
As shown in Figure 2, people with low degrees of trust in their national government institutions show significantly greater support for asylum and refugee policies that provide protection/ 16 There is one exception to this statement: In the 'governance' dimension, the experimental design does not allow for a possible policy condition or limit. 17 Due to the lower sample sizes in the national subsamples, the error bars are larger, especially for those with the least trust in the EU, the category with the fewest observations. assistance to refugees if these policies include limits and/or conditions. For example, low-trusting respondents prefer policies that include conditional rather than unconditional family reunification policies (MMs = 0.57 and 0.47, respectively), conditional rather than unconditional financial assistance to non-EU countries (MMs = 0.53 and 0.45, respectively), limited rather than unlimited numbers of asylum applications each year (MMs = 0.53 and 0.47, respectively), and restrictions on protections for failed asylum seekers. These results are consistent with our hypothesised claim that policy controls (limits and conditions) can compensate for low political trust and generate policy support even from distrusting people.
However, we do not find that it is only, or primarily, people with low trust in national political institutions who prefer policies that include controls. As can be seen in Figure 2, most respondents, regardless of their degree of trust in national governmental institutions, prefer policies that include limits to conditions to policies that provide unlimited and unconditional protection. There is no evidence in our data that the presence of policy controls makes a larger difference in public support among the lowest trusting respondents compared to more trusting individuals. More broadly, in contrast to our analysis of the role of trust in EU institutions, we find no evidence to support the idea that individuals who are less trusting in their national government institutions tend to be less supportive of policies that include expansive, unlimited, and unconditional features and more supportive of policies that eliminate protection/assistance in some policy dimensions. In other words, our results on trust in national government institutions do not support the theoretical expectations based on Hetherington (2004). We provide potential explanations in the conclusion.
Assessing robustness and the role of moderating variables We conduct a series of robustness checks to verify our results. We investigate whether our result on the conditional role of EU trust is merely an artefact and instead might mask the influence of another individual-level characteristic that is correlated to EU trust. To investigate this possibility, we conduct a battery of robustness checks in which we interact the policy dimensions with different variables to see if trust is indeed the most relevant dimension or whether it is driven by omitted variables that are correlated with trust. The results can be found in online Appendix G. We start by comparing different model specifications with the help of a nested model in Table G1. Model 2 provides results without any further control variables and thus represents the approach taken throughout the rest of the paper. Models 3-5 then add additional interactions with age groups, education, and political ideology (liberal to conservative). These models can be thought of as horse race specifications. The results suggest that interactions between policy features and individual characteristics can affect the strength of the relationship between policy features and individuals' trust in EU institutions somewhat but that the relationship is quite robust. Readers should exert care in interpreting the individual coefficients from the latter three models because adding these control variables can create biases and even inflate type I error (see e.g., Wang et al., 2017).
These previous results suggest that factors such as immigration attitudes and political ideology may be relevant moderators, which we investigate further in Figures G1-G3. In each of the figures, we vary EU trust within a given column and another factor (immigration attitudes, political ideology, and political alignment) in the others. Figure G1 shows that the differences in support between trust levels are considerably stronger and more systematic than the differences arising between individuals who differ in their immigration attitudes. There is typically little change in support within a given trust level when varying immigration attitudes. This suggests that the immigration attitude is not an important moderator for EU trust's role in explaining asylum and refugee policy preferences.
Similarly, Figure G2 shows that political orientation, i.e., the differences between the left, right, and centre, is less pronounced than differences by trust in most dimensions. Note, though, that political orientation matters for the policy preference toward financial solidarity, as indicated by the stance on 'no financial solidarity' with third countries hosting refugees switching from opposition to support, for low compared to high trust respondents on the left and in the centre. Figure  G3 provides the same analysis for political ideology (conservative/centrist/progressive) and finds only very minor differences. Overall, our results suggest that neither immigration attitudes, political alignment, political ideology, nor perceived government effectiveness play a strong role in moderating the relationship between EU trust and the policy preferences that we study.
In Figure G4, we analyse whether there are substantial differences between countries that have more or less exposure to migration. We use the 2019 OECD figures on the stock of migrants in the country (by nationality) and population figures to calculate the per capita number of migrants for each country. Splitting the sample at the median and comparing results, we find that countries with above-median migrant numbers have a greater preference for imposing annual limits. This may be expected, but it is remarkable given that the above-median countries are Austria, Germany, Spain, and Sweden and the below-median countries are France, Hungary, Italy, and Poland. Some might have expected public attitudes towards limits in countries with governments that have been proposing limits (especially Hungary and Poland) to be more restrictive. Yet, we find larger support for limits in countries that have high numbers of migrants and experienced large inflows of asylum seekers in 2015/2016. Still, one has to be careful not to overinterpret this pattern as a causal effect of higher migrant numbers because those numbers are themselves an endogenous outcome: asylum seekers, refugees, and other migrants are more likely to move to and stay in countries that they perceive as more welcoming. In our design, with its focus on similarities and differences of individuals within and across countries, the potential for further comparative analyses of country features is limited because of the low effective sample size (eight countries). Overall, the results hint at interesting country-level differences that merit further study in future research.
Conjoint tasks are cognitively demanding and therefore require respondents to devote a certain degree of concentration. To be sure that participants were able to focus sufficiently on the conjoint tasks, we required them to complete the survey only on a computer and not allowed to complete it on a mobile device. We also took measures to reduce bias from potential survey fatigue. We restricted the number of tasks to five per individual, which is well within the number of tasks that a respondent can complete before fatigue reduces response quality (Bansak et al., 2018). In addition, we analysed whether estimated preferences depend on the number of conjoint tasks that have already been completed to ensure that any remaining form of fatigue does not affect our results strongly. Reassuringly, as is shown in Appendix H, overall patterns do not change substantially or become weaker as respondents conduct additional tasks.
The results have to be interpreted in light of our choice to apply equal weights for each country due to the similar sample size in each of the countries included in our survey. To obtain the estimates for the preferences of an average citizen across the eight countries, we reweight the results using the size of the represented population in each country (see Appendix I). Weighting by population decreases the influence of the smaller EU member states (esp. Austria, Hungary and Sweden) on the overall results and increases the role of larger states. Since the relationship between EU trust and policy preferences holds across countries and does not differ with country size, the differences to the previous results are minor.
Despite our battery of robustness checks, our results are still subject to some important limitations. First, as a person's political trust is not randomly assigned, we introduce endogeneity into the research design. While we do control for a series of observable control variables as a check to avoid the risk of omitted variables that are highly correlated with EU trust, biasing its relationship with policy preferences, it is still possible that there might be omitted variables that nevertheless confound our estimates. For instance, in our control variables, we control for right-left orientation in our liberalism variable, but this would not specifically identify the influence of relevant political affiliations such as populist voters. It is thus important to bear in mind that from our results we can only conclude significant differences in political trust by sub-groups but not actually explore the causal effect of political trust on preferences. These differences by group and the analyses of moderating variables are thus exploratory. Moreover, our experiment was conducted at a single point in time (2019), and we are unable to verify in the context of our experiment if similar results would be found at other points in time. In this same vein, we have designed the experiment to be cross-national, but the influence of different national contexts on our findings is not statistically identified because it is not driven by exogenous variation, thus creating the potential that a third factor or reverse causality might distort the relationship between national context and our findings.

Discussion and conclusion
This article provides experimental evidence about the relationship between political trust and policy preferences and a novel analysis of how this relationship is contingent on the design of policies. To assess this question empirically, we conducted an original cross-national conjoint experiment to examine how a person's trust in political institutions conditions his or her preferences for asylum and refugee policy. Randomising the policy features, we demonstrate that individuals with lower levels of EU trust are more supportive of policies that provide protection and assistance to refugees if these policies utilise controls such as limits or conditions. Moreover, we find that there is less divergence between low-and high-trusting individuals when policies feature instruments of control such as limits and conditions. We have argued that this is the case because policies that feature policy controls can function as safeguards against uncertainty, which allow for distrusting individuals to nonetheless form supportive policy preferences.
Our finding that trust in EU institutions plays a much more significant role in conditioning public preferences for asylum and refugee policy than trust in national government institutions does not come as a surprise. Conceptually, trust in European political institutions and national institutions is different as trust in the EU has less of a rational, performance basis (Harteveld et al., 2013) but instead depends on what individuals extrapolate from their media environment (Brosius et al., 2019(Brosius et al., , 2018. In fact, the visibility of the role of Europe in asylum and refugee issues has increased since the Syrian refugee 'crisis', in the sense that the regulation of asylum and refugees is perceived to be a European issue as opposed to a national issue (European Commission, 2019). 18 Furthermore, over the past few years, the European Council, the European Commission, and individual EU Member States have made a considerable number of policy proposals on how to reform Europe's asylum and refugee policies (Geddes and Ruhs, 2018). These proposals have led not only to extensive political debates across the EU but also to extensive media coverage of these issues in EU Member States, which is likely to have strengthened Europeans' perception of asylum and refugee issues as European policy questions.
Beyond the specific analysis of public support for asylum and refugee policies, our results also have important implications for the role of political trust in the formation of policy preferences more generally. The finding that politically distrusting individuals are less supportive of asylum and refugee policies that provide expansive and unlimited protections and rights are in line with the established theory and argument put forward by Hetherington (2004). Yet our research also refines this argument and common understanding by demonstrating how even distrusting individuals can generate support for policies that are not directly beneficial to them if certain policy controls are in place. In fact, we find that individuals who lack trust in European institutions are most attracted to this alternative and more conditional way of providing protection to asylum seekers and refugees. Our results imply that certain policy controls, such as limits or well-defined conditions, have a compensatory effect in the sense that they act as safeguards that can counter-act and, in some cases, completely offset an individual's lack of political trust in his or her preference formation.
Future research is needed to refine these results. There is still much to be understood about various aspects of how policy controls, such as limits and conditions, can offset a person's distrust in political institutions in the formation of policy preferences. For instance, how strong must policy controls be to compensate fully for a person's lack of trust in political institutions? What exactly makes a policy control 'weak' or 'strong' in this context? Our experimental design tests the impact of the basic principle of using limits and conditions in asylum and refugee policy but not the required strength of the controls and conditions. These questions can and should be analysed also in the context of other public policy areas where political trust would be expected to be consequential, such as minority rights or the provision of international development aid.