State of the research field

Research on political representation has traditionally assumed that a relatively strong relationship exists between citizens' opinions and the opinions and legislative behaviour of their elected representatives (Erikson, MacKuen, and Stimson Reference Erikson, MacKuen and Stimson2002; Holmberg Reference Holmberg1997; Miller and Stokes Reference Miller and Stokes1963; Monroe Reference Monroe1979; Monroe Reference Monroe1998; Page and Shapiro Reference Page and Shapiro1983; Stimson, MacKuen, and Erikson Reference Stimson, MacKuen and Erikson1995). In addition, studies have shown that opinions affect policies and that public opinion reacts to policy changes (Soroka and Wlezien Reference Soroka and Wlezien2005; Soroka and Wlezien Reference Soroka and Wlezien2010; Wlezien Reference Wlezien1995; Wlezien and Soroka Reference Wlezien, Soroka, Dalton and Klingemann2007). While much of the literature focuses on the US, several studies from Europe also show that political representatives are (at least) fairly responsive to the public (Binzer Hobolt and Klemmemsen Reference Binzer Hobolt and Klemmemsen2005; Binzer Hobolt and Klemmensen Reference Binzer Hobolt and Klemmensen2008; Hakhverdian Reference Hakhverdian2010; Rasmussen, Reher, and Toshkov Reference Rasmussen, Reher and Toshkov2019). These early studies focused on the responsiveness of the public in general, often with a theoretical focus on the median voter. The focus on the middle class does of course make sense from a theoretical point of view – if politicians want to win elections they need to appeal to the median voter but to what extent do they do so? Given the importance of the median voter in winning elections, one could theoretically expect politicians to be even more responsive to middle-income citizens than the high- and low-income citizens.

In the last decade, however, a new strand of research has emerged that focuses on how well the preferences of different subgroups are reflected in either the behaviour of political representatives or in actual policies. Bartels (Reference Bartels2009), for example, has shown that the votes of US senators better reflect the views of high-income citizens. Moreover, Gilens (Reference Gilens2005; Reference Gilens2012) and Gilens and Page (Reference Gilens and Page2014) have shown that policies have a stronger relationship with the preferences of high-income citizens than with the preferences of low- and middle-income citizens. Gilens and Page conclude that ‘ economic elites and organized groups representing business interests have substantial independent impacts on U.S. government policy, while average citizens and mass-based interest groups have little or no independent influence’ (p. 564). Other researchers have offered alternative interpretations, suggesting that group differences in preferences are generally small and that the response gap is not substantial (Bashir Reference Bashir2015; Branham, Soroka, and Wlezien Reference Branham, Soroka and Wlezien2017; Enns Reference Enns2015; Enns and Wlezien Reference Enns and Wlezien2011; Soroka and Wlezien Reference Soroka and Wlezien2008; Ura and Ellis Reference Ura and Ellis2008).Footnote ¹

In an attempt to settle the debate in the literature, Bowman (Reference Bowman2020) evaluates a variety of model specifications and definitions of disagreement and policy change used in these studies and finds that the preferences of the wealthy significantly affect the likelihood of implementation. In contrast, the preferences of the middle class and the poor do not have the same impact. The fact that there is substantial disagreement about the extent of unequal responsiveness in the US means that comparative studies like this can help us to put these findings into perspective.

From a comparative perspective, and especially in comparison with Western European countries, the US is a special case with relatively high levels of income inequality and, presumably, strong ties between the economic and political elites. So, do we find unequal responsiveness elsewhere? Individual country studies from Germany (Elsässer, Hense, and Schäfer Reference Elsässer, Hense and Schäfer2021); Denmark (Elkjær Reference Elkjær2020); the Netherlands (Schakel Reference Schakel2021; Schakel and van der Pas Reference Schakel and van der Pas2020); Norway (Mathisen, Reference Mathisen2022); Sweden (Guntermann and Persson Reference Persson2021; Persson Reference Persson2021; Persson Reference Perssonforthcoming); Spain (Lupu and Tirado Castro, Reference Lupu and Tirado Castro2022); and EU-level institutions (Lefkofridi and Giger Reference Lefkofridi and Giger2020) point to similar patterns of unequal responsiveness. But, as said earlier, this research area lacks cross-country studies that map the terrain in many countries across a range of issues.

There are at least four common but rarely tested hypotheses in the literature which aim to explain the extent of unequal responsiveness: that is, economic inequality, campaign finance regulations, union density, and voter turnout.

First, the literature often points to economic inequality as a possible cause of unequal policy responsiveness. There are several reasons why this might be the case. In economically unequal societies, low-income citizens may be more politically marginalized and lack the means to voice their opinions in politics. Similarly, high-income citizens are likely to be in a better position to influence policy because of their stronger relative position of power in society. On the other hand, however, elites in societies with greater economic equality do not possess the same economic power to exert their influence over low-income citizens. Therefore, the first hypothesis we test is whether higher income inequality is associated with higher levels of unequal responsiveness.

Second, Gilens suggests that unequal responsiveness in the US may be due to high-income citizens making large campaign contributions. This could create a link between the economic and political elites, which compels politicians to grant more influence to high-income citizens. Therefore, regulating campaign contributions could be a way to weaken the undue influence of economic elites. The second hypothesis we test is whether the absence of campaign finance regulations is associated with higher levels of unequal responsiveness.

A third factor that often comes up in the discussion is union density (Becher, Stegmueller, and Käppner Reference Becher, Stegmueller and Käppner2018). The assumption is that a well-organized labour movement with high union density is advantageous in enforcing the preferences of low-income citizens. At the very least, especially when left-wing parties are in power, high union density might be expected to put pressure on those governments to implement policies that matter to the working class (Kwon and Pontusson Reference Kwon and Pontusson2010). The direct influence of union density on policy under right-wing governments may be less direct but a well-organized labour movement could help to put their preferred issues on the agenda.

The fourth hypothesis we test is whether the level of voter turnout is related to unequal levels of responsiveness. Since research shows that political representatives are more responsive to active citizens rather than inactive citizens (Griffin and Newman Reference Griffin and Newman2005; Peters and Ensink Reference Peters and Ensink2015), it is reasonable to assume that low political participation by the least affluent citizens could lead to lower consideration of low-income citizens' preferences in the political decision-making process. If low-income citizens participate less than high-income citizens, this is likely to translate into unequal influence in the policymaking process. We also know that economic inequality lowers both voter turnout and non-voter participation, which causes a greater income bias in participation (Solt Reference Solt2008; Solt Reference Solt2010). While there are many ways in which citizens might try to influence politicians, we focus here on one particular form of political participation, namely voter turnout. We do so because turnout is central to the distribution of power in society and is of direct importance to policymakers, and also because it is a form of political participation for which it is feasible and possible to use comparative country data. While it is difficult to obtain data on the relative turnout of different socio-economic groups, much research shows that lower turnout is likely to be associated with a more unequal distribution of turnout between the different groups (Persson, Solevid, and Öhrvall Reference Persson, Solevid and Öhrvall2013; Tingsten Reference Tingsten1963). The fourth hypothesis we test is, therefore, whether the level of turnout is related to unequal levels of responsiveness.Footnote ²

There are, of course, many other factors that might influence the extent of unequal responsiveness but there is no clear consensus in the literature about which factors might facilitate equal representation of citizens' preferences. The role of interest groups remains contested and studies point in different directions about whether they strengthen the link between opinion and policy (Bevan and Rasmussen Reference Bevan and Rasmussen2020; Klüver and Pickup Reference Klüver and Pickup2019). Factors that appear to have a significant ex-ante impact on how well different income groups are represented, such as a proportional voting system, are not effective in empirical studies of ideological congruence (Guntermann Reference Guntermann2021). Similarly, Rasmussen, Reher, and Toshkov (Reference Rasmussen, Reher and Toshkov2019) show that the effects of electoral systems and power distribution have a minimal impact on political responsiveness and that the number of coalition partners is not associated with lower responsiveness (see also, Toshkov, Mäder, and Rasmussen Reference Toshkov, Mäder and Rasmussen2020). Lupu and Warner (Reference Lupu and Warner2022a; Reference Lupu and Warner2022b), in two papers mainly concerned with ideological congruence, find that, while there is clear evidence that the wealthy are more often better represented than the poor, there is no evidence that factors such as political institutions, globalization, or citizens' political behaviour moderate this relationship.

This paper brings two main contributions to the field. First, it moves beyond single-country studies and evaluates responsiveness to specific issues using comparative data. While other studies have also used comparative data, such as the recent papers by Lupu and Warner (Reference Lupu and Warner2022a; Reference Lupu and Warner2022b), they look at representation in terms of left-right ideology. As Broockman (Reference Broockman2016) has shown, ideological scores ‘say little about citizens’ views within domains, on issues themselves’ (p. 8). Moreover, when using ideological scores it is not possible to distinguish legislators who are ideologically ‘mixed’ from those who are ‘moderate’. Moreover, several studies have convincingly shown that people have opinions on distinct issues beyond their ideological dispositions (Ansolabehere, Rodden, and Snyder Reference Ansolabehere, Rodden and Snyder2008; Tausanovitch and Warshaw Reference Tausanovitch and Warshaw2013).Footnote ³ This paper takes a country-comparative approach to studying unequal policy responsiveness by using issue-by-issue data. Secondly, it adds an empirical evaluation of the impact of four of the most commonly occurring mechanisms in the field. These mechanisms have been largely ignored in empirical studies.

As the reader may have noticed, there is a lack of conceptual clarity in this research area. Researchers often use terms like ‘representation’, ‘responsiveness’, and ‘congruence’ interchangeably. In general, scholars refer to congruence to study whether there is a match between what kind of policy people want and what kind of policy is in place (note that this is a different approach to the many studies focusing on ideological congruence (see Golder and Stramski Reference Golder and Stramski2010)). Responsiveness, on the other hand, is sometimes thought to imply a temporal dimension where some scholars define policy responsiveness as a causal reaction to an opinion, while others define it more loosely as a correlation between opinion and policy: as Wlezien (Reference Wlezien2017) puts it, ‘when most people want a lot of policy […] they would get a lot of policy’ (p. 562). Others refer to the same relationship as ‘influence’ or simply policy representation (see Bartels (Reference Bartels2021) and Wlezien (Reference Wlezien2017) for conceptual overviews). Using our data we can merely study the correlations between opinions and policy, not whether there is a causal relationship. Still, in our dataset policy, outcomes are measured years after the measurement of opinion, which mitigates the risk of reverse causality. Further methodological considerations are discussed in the next section.

To make clear what we are empirically investigating in this paper, we focus on ‘congruence’. By that term, we refer to the extent to which actual policy changes match public support for those policy proposals. We do this by looking at how large the share of citizens is that hold preferences matching the policies that are already in place. Empirically, we calculate congruence by examining the extent to which political support in different income groups is congruent with policy changes. In the Appendix, we examine responsiveness more traditionally with regression analyses of the influence of political support in different groups on policy changes.

Data

The unit of analysis in the dataset is policy proposals in specific country years. The main independent variables are the level of support in different income groups, while the dependent variables measure whether policy proposals were implemented at different points in time. To collect information on preferences for policy proposals in different income groups, we use data from the following cross-country surveys: The Comparative Study of Electoral Systems (CSES), The Eurobarometer (EB), The European Social Survey (ESS), The European Values Study (EVS), The International Social Survey Programme (ISSP), and The World Values Survey (WVS).

We searched these datasets for questions about specific policy proposals. For a question to be selected, it had to be possible to investigate, in an appropriate way, whether it had been implemented or not. Questions should ask about a preference for particular policy proposals that translate into a single, specific, measurable, and identifiable policy. Thus, we look at questions that ask about citizens' preferences for specific policy proposals such as nuclear energy development, allowing same-sex marriage, tax increases/tax cuts, and public spending in different domains. The questions should be specific enough so that coders know what information to look for to enable them to examine implementation. We did not include questions that asked respondents about priorities, importance, or conditionalities. However, questions may ask about absolute changes (such as joining NATO or closing all nuclear power plants) or relative changes (increasing or decreasing spending in various areas). The questions can also be about new proposals or about policies that have already been implemented. As is common in this field, we changed the direction of the opinion variables for the status quo questions in the empirical analyses so that all opinion variables indicate support for policy changes.

A team of research assistants coded the policy outcomes. For each question in each survey, it was coded whether the policy was implemented at three points in time after the survey (we code the exact month and year of the survey question): that is, (1) by the end of the incumbent government's term, (2) by the end of the successor government's term, and (3) five years after the question was asked. The implementation variables are dichotomous and take the value 1 if a policy is changed and 0 if it is not changed (or if a decision is made in the opposite direction). Thus, the implementation variables are coded 1 if an official government decision was made or an outcome occurred (depending on whether the survey question referred to a decision or an outcome). If no information can be found after exhaustive research, we classify the proposal as not implemented. In cases where the de jure implementation of the policy differs from the de facto implementation, we focus on the official factual data (de jure). We always make a comparison between the three points in time and the time of the survey (for example, if a decision is made three years after the survey but reversed the following year, it is coded as not implemented after five years). For statements about relative change, we look at the differences between the level in the survey year and each of the three time points. If there has been a relative change in the direction of the policy proposal, the implementation variable is coded 1; otherwise, it is coded 0.

The instructions to coders were straightforward: code whether or not the policy proposal was implemented at the specified time points. To this end, coders had to refer to different sources depending on the type of policy. For example, some survey questions asked specifically about policy decisions; for these questions we looked at the minutes of the national parliament. Other questions focus on the implementation of proposals. For these questions, we looked at the relevant sources for the topic in question. For example, these may include budgets, administrative files, documentation on the closure of nuclear power plants, or the construction of an infrastructure project. Also, for some issues that were not implemented, there is simply no source or documentation (as these policy changes were not events).

As with the opinion data, the original response options were coded in different ways. Therefore, following the standard practice in the field (Elsässer, Hense, and Schäfer Reference Elsässer, Hense and Schäfer2021; Gilens Reference Gilens2005; Gilens Reference Gilens2012; Schakel Reference Schakel2021), we standardized each survey item so that it captured the percentage of respondents that indicated support for policy change, from 0 to 100. Because income was measured in different ways in the different surveys, we also harmonized the income scales by assigning the respondents to income values that corresponded to the middle percentile of their respective income categories, taking into account the distribution of income groups in each survey. To estimate support for change across income groups, we calculated the average support for each proposal in each country among respondents with a percentile rank in the bottom 20 per cent, the middle 60 per cent, and the top 20 per cent of the income distribution. This is our measure of support for the policy proposals among the poor, the middle class, and the wealthy.

In total, we use data on the policy preferences of about two million individuals. Overall, we cover thirty countries and thirty-eight years from 1978 to 2017 (516 country years in total). The countries included all EU countries except Malta) plus Iceland, Norway, Switzerland, and the UK. For a full list of countries and years see the Appendix.Footnote ⁴

The policy proposals cover a wide range of matters. For example, economic issues that mainly concern welfare spending, redistribution and state-market regulations; fiscal policies; and non-economic issues that cover foreign policy, immigration/multiculturalism, liberties and rights, environmental policies, and law and order. See the Appendix for further information on the number of issues in the different areas.

To test the impact of income inequality, we used data from ‘The Standardized World Income Inequality Database’ (SWIID) (Solt Reference Solt2020). We used SWIID because it is the database on inequality that covers most countries and years. The Luxembourg Income Study (LIS) is considered the standard measure of inequality in comparative research and the SWIID ‘estimates the relationships between Gini indices based on the LIS and all of the other Ginis available for the same country years and uses these relationships to estimate what the LIS Gini would be in country-years not included in the LIS but available from other sources’ (p. 1,183).

As for campaign regulations, we used data from the V-Dem Institute (Coppedge et al. Reference Coppedge2021). These data are based on expert interviews with country experts that provide information on individual years and countries. We used the question, ‘Are there disclosure requirements for donations to national election campaigns?’ with responses on the following scale: 0 (No), 1 (Not really), 2 (Ambiguous), 3 (Mostly), 4 (Yes).

We used the information on turnout in the last general election from the Comparative Political Data Set, 1960–2016 (Armingeon, Engler, and Leeman Reference Armingeon, Engler and Leeman2020). From the same dataset, we used data on union density, defined as the ‘net union membership as a proportion of wage and salary earners in employment’.Footnote ⁵