Introduction

One of the keystones of a democracy is the proper representation of voters’ preferences by their representatives in parliament. Pitkin (1967) called this “substantive representation”, in which representatives make present the policy positions of the public. When voters and their representatives share the same positions on public policy, it increases the chance of those policy positions to become policy realities. Agreement between voters and representatives is seen as an indicator of a healthy democracy (Diamond and Morlino, 2005) and has therefore received much attention in the field of political science, usually under the label of “opinion congruence”. However, studies, which usually focus on the United States, find that legislatures often favor the policy preferences of the privileged social groups, typically consisting of the higher-educated or higher-income strata (Giger et al, 2012; Winters and Page, 2009). Amidst growing societal inequality, there is increasing concern that the preferences of affluent citizens receive more weight than those of other voters, and there is indeed growing evidence for a representational bias (see, for instance, Flavin, 2012; Gilens, 2005; Jacobs and Page, 2005).

Recently, scholars have begun to develop a theoretical model of inequality in representation (see, for instance, Winters and Page, 2009). Three factors stand out: a lobbying environment dominated by business organizations, political donations that only allow those approved by the wealthy to run for office, and the (non-)compulsory nature of voting, which determines whether turnout is biased toward the privileged citizens. However, all three characteristics are situated at the country level and can therefore only explain differences between countries and political systems. At the same time, studies have found differences between issues in the degree to which the preferences of different social groups are represented (Gilens, 2005, 2012; Lesschaeve, 2016). Yet, a model that can explain these differences is still missing. This study seeks to fill this void in the literature by developing and testing an issue-level model of opinion congruence inequality.

To do so, we focus on Belgium, a small consociational country in Western Europe. Using an integrated dataset containing the positions of voters and parties on 229 policy statements, our results show that, despite the favorable country context, opinion congruence inequality between the privileged and the underprivileged groups (in terms of education and income) is present in Belgium. However, we find much variation between policy issues in terms of the degree to which the preferences of social groups are represented in parliament. The country-level model thus needs to be complemented by an issue-level model of opinion congruence inequality. The issue-level model presented in this paper consists of three factors: the gap in policy preferences between social groups, policy domains of key interest to the privileged and the underprivileged groups, and the extent to which voters of a social group agree with the party they vote for on an issue, or “correct voting”.

We find that when the privileged and the underprivileged groups differ in policy positions, parliament is likely to side with the preferences of the former. In addition, the overrepresentation of the privileged groups’ policy positions is larger on issues that are important to their key interests, such as issues related to economic and tax policies, but smaller on issues of vital importance to the underprivileged groups, such as social welfare. Finally, congruence inequality is to an important degree self-inflicted. When voters of a group (either the privileged or the underprivileged) disagree with their actual party choice on an issue, their preferences are less well represented in the legislature. We discuss the normative implications of these findings in the conclusion.

Issue-Level Determinants of Congruence Inequality

Research in the United States has identified three factors of a political system through which the privileged social groups exert their influence on the decision-making process: one-sided business lobbying, financial contributions, and social bias in voter turnout. Through lobbying, representatives of specific social interests seek to establish shared perspectives with politicians on policy issues, for instance through socializing and friendship networks (Winters and Page, 2009). The second factor, political donations and spending on behalf of parties or candidates, primarily serves a selection purpose in what has been labeled the “ideological sorting” hypothesis (Ferguson, 1995). Campaign donations rig the game in favor of the privileged groups by preventing those who are unable to garner sufficient funds from competing in an election. The third factor is compulsory voting or the lack thereof. In almost every country, lower-income or lower-educated voters are consistently less likely to vote in an election (Steinbrecher and Seeber, 2011). To our knowledge, only political systems with compulsory voting have been able to completely neutralize this bias by making turnout mandatory for everyone (Hooghe and Pelleriaux, 1998).

However, lobbying, financial contributions, and voter turnout are all factors that are situated at the country level. While they undoubtedly play a role in congruence inequality, they are less able to explain variation in congruence inequality within a specific country and between issues. Some studies have indeed pointed to this variation (Gilens, 2012; Lesschaeve, 2016), but an issue-level theory is missing. This paper seeks to explore which issue-level factors can explain inequality in collective opinion congruence. We focus here on three variables: differences in policy preferences, policy domains, and correct voting.

One of the most important reasons why scholars argue that there is a bias in parliament toward the preferences of the privileged social groups has to do with the backgrounds of political elites. Politicians and candidates often come from the privileged groups. For instance, in Belgium, higher-educated individuals often make up more than 70 per cent of the MPs in a legislature (Bovens and Wille, 2011), and in the United States, the median individual net worth of members of Congress is six times larger than the median net worth of average Americans (Carnes, 2012). Due to their background, politicians view the world through the eyes of the privileged individuals and are thus more likely to have similar policy preferences. This link between descriptive and substantive representation is backed by a substantial literature (see, for instance Bühlmann and Schädel, 2012; Carnes, 2012)

While more and more studies are finding differences in preference representation between the privileged and the underprivileged groups that are biased in favor of the former, some studies find that there is no inequality in representation (Ura and Ellis, 2008; Wlezien and Soroka, 2012). However, Soroka and Wlezien (2008) provide a possible explanation for these conflicting results. They argue that a precondition for opinion congruence inequality is a difference in policy position. If there is a large consensus among voters regarding the future direction of public policy, then the underprivileged voters are as likely as the privileged voters to have their preferences represented. In fact, if policy preferences do not differ between social groups, opinion congruence inequality between those groups is mathematically impossible. Underprivileged voters’ opinion congruence might be poor or great, but it will at least be equal to that of the privileged voters. There has been some debate in the literature on whether or not social groups have different views on policies (see Gilens, 2009; Soroka and Wlezien, 2008). However, a more fruitful way to consider differences in policy positions would be to use it as a substantive explanatory variable of opinion congruence inequality. Opinion congruence inequality between groups on an issue depends on a difference in opinion between those groups. In sum, the descriptive bias in parliament toward the privileged groups is expected to lead to a bias in preference representation. However, this bias can only manifest itself when there is a difference in policy preference between the privileged and the underprivileged groups (also see Gilens, 2005). In addition, it is important to include the differences in policy positions in an issue-level model of congruence inequality. Otherwise, the results are likely to understate the difference in congruence between social groups. As opinion congruence inequality can only occur when voters differ in opinion, a true test of this inequality would be to study how equally or unequally the preferences of the privileged and the underprivileged groups are represented when their preferences diverge. We expect that the tendency to side with the policy preferences of the privileged groups becomes stronger as opinions diverge (H1).

The literature on inequality in substantive representation has also recently begun to study differences between policy domains. For instance, Jacobs and Page (2005) find that, in the area of foreign (economic) policy, the preferences of U.S. members of Congress are much in line with those of U.S. business elites. However, it is uncertain to what extent their findings can be generalized to other countries. The importance of U.S. foreign policy in the world and its effect on the global economy make it vital to the interests of the privileged groups, unlike the foreign policy of much smaller countries, including the case studied in this paper, Belgium. In fact, given that much of the foreign economic policy is currently being conducted at the European level, there are far fewer reasons to expect congruence inequality between the privileged and the underprivileged in the area of foreign policy because there is simply less at stake. The material interests of the society’s privileged are less likely to be threatened on the Belgian level due to the transfer of foreign (economic) policy competences to the EU level. As a result, there is less need for the privileged groups to try and exert influence in this area.

Thomassen (2012) argues that scholars should distinguish between an economic and a cultural dimension. Given the economic meaning of the left–right scale (Van Der Brug and Van Spanje, 2009), congruence measures based on it could underestimate congruence on cultural issues, in what Thomassen labeled as a “blind corner” in congruence studies. However, this blind corner affects both the privileged and the underprivileged groups. In addition, it is unclear, for instance, why on all issues related to the economic dimension there would be a bias in opinion congruence in favor of the privileged groups. The study of Winters and Page (2009) is more instructive in this regard. They claim that inequality in preference representation is most likely to occur in a number of specific economic areas, such as tax policy, that touch upon the key (material) interests of society’s “well-off’” population. Gilens (2012) put these claims to the test and found that inequality in policy responsiveness in the United States is indeed higher on issues related to economic policy. However, on issues related to social welfare, the differences between higher- and lower-income groups are smaller. This emphasizes the need to distinguish between various issues in the economic dimension. One could argue that social welfare issues are of vital importance to the underprivileged groups. Indeed, studies have shown that the underprivileged are more likely to protest when they experience deprivation or when something is taken away from them (Kern et al, 2015; Stekelenburg and Klandermans, 2013), and this is exactly what could happen when policy goes against the preferences of the underprivileged on social welfare issues such as pensions or unemployment benefits. The same can be expected when policy goes against the preferences of the privileged groups (be it more in the form of lobbying), on tax and economic issues, such as a wealth tax, that are crucial to their interests (Winters and Page, 2009). When the policy preferences of political elites go against the key material interests of a group, that group is much more likely to mobilize. This gives political parties less leeway to deviate from that group’s policy preferences on those issues. We therefore hypothesize that congruence inequality between the privileged and the underprivileged groups is higher and more in favor of the former on the issues of tax and economic policy (H2), but lower on issues related to social welfare (H3).

So far, we have focused on top-down mechanisms of inequality in opinion congruence. However, the literature has neglected the possibility that voters themselves can be responsible for the lack of preference representation. The responsible party model argues that proper congruence between voters and political elites can be achieved through a mechanism called “party–voter” congruence, in which voters vote for a congruent party (Schmitt and Thomassen, 1999). Voters need to have developed policy positions, be informed about the policy positions of political parties, and compare their own positions with those offered by the parties. Finally, the voter’s choice of party needs to be based on the congruence between his or her own policy positions and the policy positions of the chosen party. Conversely, a lack of congruence can result from voters voting for parties with which they disagree. The third factor at the issue level relates to the extent that voters agree with the party for which they voted (Lau and Redlawsk, 1997). Previous studies have found that the underprivileged voters are less likely to vote for a party with which they agree on policy issues (Lesschaeve and Meulewaeter, 2015). Many studies have indeed found that lower-educated voters are less knowledgeable about politics (see, for example, Grönlund and Milner, 2006) and are less able to process political information from the media (Eveland and Scheufele, 2000). As education and income are related, this also applies to lower-income voters. For instance, if lower-educated or lower-income voters want tougher immigration policies but vote for parties that opt to make those policies softer, then those groups give support to policies that contradict their own positions. In such cases, voters “self-inflict” opinion congruence inequality on themselves by voting for a party with which they disagree. On an aggregate level, this might lead to a situation in which the preferences of lower-educated voters are less well represented in parliament on immigration issues. However, even the privileged groups often have substantial disagreements with their party choice, their higher levels of party–voter congruence notwithstanding. This might be a deliberate choice: voters vote for a party that shares their position on the issues that matter most to them. Therefore, we can assume that on some issues, the underprivileged groups agree more with their party of choice, while on others, the privileged groups agree more. In sum, we expect that when a social group (either the privileged or the underprivileged) votes more “correctly” on an issue, its policy preferences are better represented in the legislature (H4). Our model is summarized in Figure 1, which presents the country- and the issue-level model of opinion congruence inequality. This paper will focus on the issue-level factors only.

Figure 1
figure 1

A country- and issue-level model of opinion congruence inequality

Full size image

Finding an Appropriate Case: Belgium

To study issue-level determinants of opinion congruence inequality, it is ideal test our model in a case where congruence inequality is least affected by the country-level determinants discussed above. We believe that Belgium provides us with such a case. It is a small consociational nation in Western Europe and possesses characteristics that arguably reduce the likelihood of opinion congruence inequality between social groups.

First, instead of lobbying activities being dominated by business interests, large and binding agreements regarding pensions, wages, and labor standards are reached in Belgium through the so-called “Group of 10”, a joint committee of key representatives from the most important labor and business organizations (van Gerven and Beckers, 2009). As a result, lobbying is a less effective strategy for influencing policy outcomes. In addition, contrary to any other industrialized country (including the United States), union membership in Belgium has increased rather than declined (Liagre, 2012). The increasing strength of trade unions in Belgium makes it more likely that they will continue to play a key role in socioeconomic policy making in Belgium (Naedenoen, 2008), thus countervailing the lobbying activities of business interests, and producing less opinion congruence inequality.

Second, Belgium has strict party finance laws (Weekers, Maddens, and Noppe, 2009). Donations from both corporations and trade unions have been banned, and since 1999 only individuals may donate money to political parties, and only up to a specified limit. These restrictions reduced the financial means available to parties, which were then compensated by subsidies from the state. In addition, there are strong limitations on what parties are allowed to do during a campaign (e.g., TV ads, billboards, etc.). In other words, parties and candidates need less money than in the United States because they cannot spend it on expensive ads, and the money they do need to run a campaign they get primarily from the government instead of a selective donor class.

Third, Lijphart (1997) argued that compulsory voting could help equalize representation by equalizing turnout. Due to compulsory voting in Belgium, turnout has been around 90 per cent or more since World War II.1 Not only is turnout larger than in other comparable countries, but it has also been far more equal as well. Studies have found no relation between education level and likelihood to vote in Belgium (De Winter and Johan Ackaert, 1994). However, abolishing compulsory voting would decrease turnout to about 60 per cent and would lead to an overrepresentation of higher-educated citizens among the voting public (Hooghe and Pelleriaux, 1998).

In sum, the three factors that the classic country-level theory holds as causing a representational bias in favor of the policy preferences of the privileged social groups are absent in Belgium. Instead of an overrepresentation of business interests in lobbying activities, Belgium has institutionalized negotiations between labor and business; instead of party dependence on a donor class for funds, Belgium has banned corporate sponsorship and strongly regulated the conduct of parties and candidates during campaigns; instead of voluntary voting, Belgium has compulsory voting, effectively eliminating the socioeconomic skew in turnout. As a result, when studying issue-level determinants of congruence inequality in Belgium, as we do in this paper, congruence inequality is less likely to be contaminated by country-level causes.

Data and Method

We use two sets of data. The first is an online voter survey of 2080 eligible Belgian voters, taken in March 2014, in the run-up to the elections for the Flemish and Walloon regional parliaments and the national parliament on May 25, 2014. The survey was conducted by TNS Dimarso. In the voter survey, respondents were asked to react to 106 (Flanders) or 113 (Wallonia) policy statements. These statements are our unit of analysis. Voters could either “agree” or “disagree” with a policy statement. Though one might argue that this leaves little room for nuance, it does represent a clearer measurement of a voter’s issue position. Rabinowitz and Macdonalds (1989) argue that additional answering categories, such as “strongly agree” or “somewhat agree”, are more indicative of the intensity of an issue position than its direction and reflect how important an issue is to a voter. Expanding the answering format would thus have conflated issue position with issue salience. In order to avoid respondent fatigue, the survey was split up into two waves. Studies have shown that when online surveys take more than 20 min to complete, the quality of the responses decreases (Galesic and Bosnjak, 2009). Due to the two-wave strategy, the average length of a survey wave was only 15 min. The survey also contained social background factors such as education and income.

In total, 12,241 individual were contacted, resulting in an average response rate of 17 per cent across both waves.2 However, due to quota sampling and the use of sampling weights, the composition of the sample accurately reflects the Belgian population.3 The most common problem of an online survey is the overrepresentation of higher-educated voters (Strabac and Aalberg, 2011), which is related to inequality in opinion congruence. Consequently, an overall average policy preference would likely reflect what higher-educated voters think. However, when calculating opinion congruence, as is explained below, we compare the policy preferences of lower- and higher-educated voters separately with the policy preferences of the parliament. This approach arguably reduces the remaining bias in the survey toward the privileged groups. Another common problem is the overrepresentation of politically interested voters, specifically among the underprivileged voters. Consequently, the underprivileged voters in the survey might be more politically interested and knowledgeable than the underprivileged voters in the population. However, if the hypothesis 4 is correct and political interest and knowledge are positively related to preference representation through correct voting, then it follows that the underprivileged voters in our sample vote more correctly and have their policy positions better represented than the underprivileged voters in general. The bias in our sample of the underprivileged voters thus leads to an overestimation of their preference representation, and is likely to make the representation gap with the privileged groups smaller. An increase in the political interest of the underprivileged groups thus makes it less likely that differences between the privileged and the underprivileged voters will be found and thus constitutes a more conservative test of our hypotheses. In sum, our approach reduces sampling biases, and any bias remaining works against the confirmation of our hypotheses.

The second dataset is a party survey. The same policy statements presented to voters in the online survey were also presented to the leaders of all political parties in Belgium who had at least one representative in either the regional or national parliament before the elections of May 25, 2014 (n = 11).4 They were given 2 weeks to confer with other members of the leadership and to develop a party position. The Belgian party landscape is split along the Flemish/Francophone linguistic divide (De Winter et al, 2006). There are six Flemish and five Francophone parties in our sample. As was the case for voters, party leaders could only react to the statements with “agree” or “disagree”. Does the position of the party leadership always match that of their rank and file MPs? Arguably, one can expect a high level of opinion congruence between the party leadership and party MPs: aspiring candidates are unlikely to join a party with which they have stark disagreements, and parties are unlikely to allow an aspiring candidate on their list if he or she does not endorse the party leadership’s positions. In addition, even in the case of disagreement, there are still important reasons to assume that MPs will vote in line with the party leadership such as anticipated sanctions or adherence to the norm to express loyalty to the party leadership (Andeweg and Thomassen, 2011). In sum, MPs and the party leadership are most likely agree on the vast majority of issues, but even when they do not, the latter’s position is the one that matters. This is shown in the Belgian case by the almost perfect degree of party cohesion during votes in parliament (Depauw, 2003).

The policy statements touched upon concrete regional and national policy issues. The statements on regional policies, however, were different for the Flemish (50 statements) and Francophone (56 statements) voters and parties and were tailored to reflect the regional differences in relevant policy issues. The statements regarding national policy issues (61 statements) were identical for both language groups. However, Belgium has separate Flemish and Francophone parties and media systems, even at the national level. Party competition for seats in the national parliament happens in each region separately. Consequently, the national parliament is subdivided into Flemish and Francophone language groups, with a fixed seat distribution for each group. In other words, the Belgian national parliament can be considered to consist of a “Flemish national parliament” and a “Francophone national parliament”. Logically, each language group in the national parliament should, collectively, represent its language community. Therefore, for national policy statements, we measure how congruent each language group is with various social groups within its own language community. Furthermore, the institutional arrangements that make Belgium a least likely case to find opinion congruence inequality apply to both the national and regional government levels. Finally, elections for both the national parliament and the regional parliaments can be considered first-order elections (Deschouwer, 2012). As a result, though the statements touch upon the same national policy issues, the separate party and media landscape make them independent cases in which to study opinion congruence inequality. This brings the total number of policy statements to 229.5

Our dependent variable is the difference in opinion congruence between the privileged and the underprivileged groups on a single policy statement. The first step is to calculate opinion congruence. While many studies focus on “dyadic” opinion congruence, the congruence between voters and a specific party or representative (see, for instance Giger et al, 2012; Walgrave and Lefevere, 2013), Pitkin (1967) herself emphasized the normative ideal of having a legislature that reflects the will of all people. For this “collective” opinion congruence, elections are seen as the mechanisms through which voters ensure that a legislature as a whole is a proper reflection of the public in terms of policy preferences (Andeweg, 2011; Weissberg, 1978). Instead of measuring how congruent voters are with a party, this collective perspective focuses on how the distribution of preferences in a legislature matches the distribution of preferences within a population. In addition, as inequality is usually defined as the extent to which political elites or political institutions as a whole favor the preferences of certain groups above those of others (Lefkofridi et al, 2012), a collective approach fits our research question best. Inequality in (collective) opinion congruence occurs when the distribution of preferences in a legislature has a better match with the distribution of preferences in one social group than the distribution of preferences in another. Some studies focus on governments, as this comes closer to actual policy (Giger et al, 2012). However, government formation is ruled by its own dynamics (Martin and Stevenson, 2001), and a preference bias in a legislature as a whole makes it more likely that any government supported by a legislative majority will be biased toward those same preferences as well. To study opinion congruence inequality, we thus take the collective approach and base our operationalization on Golder and Stramski (2010). Proper collective opinion congruence is achieved when the distribution of preferences in a legislature matches the distribution of preferences in specific social group:

$$ {\text{collective}}\;{\text{congruence}} = 1 - \left[ {\sum\limits_{i = 1}^{k - 1} {\left| {F_{i}^{*} ({\text{parliament}}) - F_{i}^{*}({\text{social}}\,{\text{group}})} \right|} } \right] $$

where k stands for the number of categories in an opinion measure and F* for the relative cumulative frequencies.6 If, for instance, 50 per cent of parliament agrees with a statement and 60 per cent of a social group agree, then that social group is 1 − |50 − 60 per cent| = 90 per cent congruent with parliament. The result of the above formula thus indicates absolute levels of opinion congruence. Opinion congruence inequality, however, refers to a relative difference between the opinion congruence of various social groups. Therefore, we require an additional calculation to indicate whether one social group is more congruent with the legislature than the other. To do so, we subtract the collective opinion congruence of a privileged social group from the collective opinion congruence of an underprivileged social group:

$$ {\text{inequality}}\,{\text{in}}\,{\text{c}} . {\text{congruence}} = {\text{c}} . {\text{congruence}}\,({\text{priviliged}}\,{\text{group}}) - {\text{c}} . {\text{congruence}}\,({\text{unpriviliged}}\,{\text{group}}) $$

The above two formulas were applied to all policy statements, and the result of this subtraction is the dependent variable in this study, opinion congruence inequality. When comparing the collective opinion congruence of two social groups, there are three possibilities: (1) the privileged group is more congruent with the legislature than the underprivileged group, (2) the underprivileged group is more congruent with the legislature than the privileged group, and (3) the privileged and the underprivileged groups are equally congruent with the legislature. By subtracting the opinion congruence of the underprivileged group from the opinion congruence of the privileged group, our measure of opinion congruence inequality is positive when the privileged groups are more congruent with the legislature than the underprivileged groups, negative when the underprivileged groups are more congruent with the legislature than the privileged groups, and zero when collective opinion congruence is equal.

The distinction between the privileged and the underprivileged voters follows the division of society in various social strata. Social stratification refers to the distribution of resources in a society (Beeghley, 2015), and the determinants of one’s stratum can be largely traced back to two factors: education and income. Education has a strong impact on the occupation a person is able to attain (Jerit, 2009) and relates strongly to voters’ ability to vote for a congruent party in elections (Lesschaeve and Meulewaeter, 2015). Income is related to an individual’s wealth and material well-being. In addition, income is the most-used variable in congruence inequality research (see, for instance Flavin, 2012; Giger et al, 2012) and has been found to be related to certain material interests with regard to social welfare and economic policy (Winters and Page, 2009). However, instead of selecting one approach, we choose to include both. Doing so will allow us to test the robustness of our results, and it also constitutes a more thorough test of our hypotheses. Therefore, we calculate (1) the opinion congruence inequality between lower-educated (voters who have no degree or only an elementary school degree) and higher-educated voters (voters who have a university degree or higher) and (2) the opinion congruence inequality between the lowest two income deciles and the highest two income deciles. For instance, if lower-educated voters are 40 per cent congruent with the legislature on an issue and higher-educated voters are 60 per cent congruent with the legislature, we subtract the congruence of the former from the congruence of the latter: 60 − 40 per cent = 20 per cent. We conclude that there is a 20 per cent opinion congruence inequality in favor of higher-educated voters.

The distribution of policy positions of a social group is derived directly from the voter survey. Per social group and per policy statement, we calculate the collective public opinion: the percentage of voters who agree or disagree with a statement. This is important, as many studies have cast doubt on whether voters hold “true” preferences on issues (Converse, 2006; Zaller, 1992), and the lack of a neutral category forced these voters to choose a side on the issue. Consequently, several of the positions of voters on the statements have to be considered random and therefore not reflective of an actual position. However, Page and Shapiro (1992) argue that the presence of such non-attitudes is not problematic if one wants to measure collective public opinion. When aggregating public opinion across all voters or a subset of voters, these random answers cancel each other out. Consequently, the measurement of collective public opinion is “largely free of the random error associated with individual attitudes” (p. 16).

The distribution in parliament is calculated using the party leaderships’ positions and the seats parties received after the elections of May 25, 2014. The proportion of the legislature that “agrees” or “disagrees” with a policy statement equals the sum of the seats of all parties that agree or disagree with that policy statement. For example, in a parliament with 100 seats, if three parties with 10, 15, and 20 seats, respectively, agree with a policy statement, then the proportion of the parliament that agrees with that policy statement is (10 + 15 + 20)/100 = 45 per cent. As there are two possible answers to a policy statement, it follows that 55 per cent of the parliament disagrees with that policy statement.

Our independent variables are the differences in policy position, the policy domains of economic and tax policy and social welfare, and correct voting. The difference in policy position is the absolute difference between the percentage of voters who agree with a policy statement within a privileged and an underprivileged social group. Its value is 0 when both groups have an equal percentage of voters who agree and disagree with a policy statement, and 100 when all voters of one group agree and all the voters of the other group disagree with a policy statement. For the two policy domain variables, economic and tax policy and social welfare, we create two dichotomous variables indicating whether a statements belongs to a certain policy domain (1) or not (0). To measure differences in correct voting, within each social group and for each policy statement, we calculate how often voters are congruent with their preferred party. This gives us a percentage of correct voting for each statement for each social group. For each policy statement, we then deduct the percentage of correct voting in the underprivileged group from the percentage of correct voting in the privileged group. The result is a measure of inequality in correct voting. This variable is positive when higher-educated or higher-income voters vote more correctly than lower-educated or lower-income voters, and it is negative when lower-educated or lower-income voters vote more correctly than higher-educated or higher-income voters. Finally, as our policy statements come from different linguistic regions and parliaments, we control for the language group (Flemish or Francophone) and the legislature (regional or national) in our analyses. Table 1 gives an overview of all of the variables.

Table 1 Descriptives of all the variables
Full size table

Results

Based on a large sample of 229 policy statements, we analyze how often and why the privileged groups (higher-educated and/or higher-income voters) are more congruent with the legislature in Belgium than the underprivileged groups (lower-educated and/or lower-income voters). When we look at the average values of congruence inequality (Table 1), we find them to be significantly different from each other, both for education (t[228] = 5.23, p < 0.001) and income groups (t[228] = 3.38, p < 0.001), indicating a bias toward the policy preferences of higher-educated and higher-income voters. Figures 2 and 3 visualize the opinion congruence inequality for each individual policy statement for education and income, respectively. Every bar in the figures represents one policy statement, and the height of each bar shows the opinion congruence inequality. The statements are placed in order of decreasing difference in collective opinion congruence, and the x-axis displays the relative rank order of a policy statement (rank order divided by the total number of policy statements [229]).

Figure 2
figure 2

Opinion congruence inequality between higher- and lower-educated voters

Full size image
Figure 3
figure 3

Opinion congruence inequality between higher and lower incomes

Full size image

It is clear that there is a lot of variation in the degree to which the privileged and the underprivileged groups’ preferences are represented. For instance, regarding the statement “The ban on smoking should be relaxed in the hotel and catering industry”, higher-educated voters are 31 per cent more congruent with parliament than lower-educated voters, and higher-income voters are 41 per cent more congruent than lower-income voters. With respect to the statement “The headscarf should be banned for students in formal education”, there are almost no differences between the different groups. Finally, concerning the statement “The living wage should rise”, opinion congruence inequality is reversed in favor of unprivileged groups: higher-educated voters are 22 per cent less congruent with parliament than lower-educated voters, and higher-income voters are 36 per cent less congruent than lower-income voters.7 The aim of this paper is to explain why inequality is higher or lower on some policy statements than on others.

The Figures also shows that country-level determinants need to be complemented with issue-level factors. If country-level determinants were enough, we would arguably find low levels of inequality in congruence across all policy statements. While we do not include country-level variables in the multivariate models below, merely finding (substantial) variation in inequality in opinion congruence while keeping country-level factors constant shows that there is more going on than country-level variables can explain.

Table 2 shows the results of two OLS regressions with opinion congruence inequality between education and income groups as the dependent variables. The difference in opinion between social groups is the strongest predictor of the difference in collective opinion congruence, reaching statistical significance in both model 1 (education) and model 2 (income). More importantly, however, the signs of the coefficients, which are always positive, are indicating that large differences in opinions correlate with more pronounced positive differences in collective opinion congruence, which are in favor of higher-educated or higher-income voters. This supports the hypothesis 1.

Table 2 Explaining opinion congruence inequality
Full size table

When we look at the policy domain variables, we find evidence for the hypothesis 2 and partial evidence for the hypothesis 3. In model 1, for education groups, policy statements related to economic and tax policy coincide with a higher level of opinion congruence inequality, meaning that the legislatures in Belgium more often prefer the policy positions of higher-educated voters on these issues. On statements related to social welfare, on the other hand, opinion congruence inequality is lower, indicating a more equal preference representation or a reverse inequality in favor of the policy positions of lower-educated voters. In model 2, for income groups, though all coefficients are in the expected direction, we only find a marginally significant effect for policy statements related to social welfare. We are thus able to confirm the hypothesis 2, but for the hypothesis 3, we can only confirm congruence inequality between the lower- and higher-educated groups.

We find more consistent evidence for the hypothesis 4. The inequality in correct voting between lower- and higher-educated groups and lower- and higher-income groups is a highly significant predictor of opinion congruence inequality. In addition, it is one of the strongest predictors of congruence inequality. This indicates that part of the differences in opinion congruence between the privileged and the underprivileged groups is the result of their voting behaviors – specifically, the extent to which their party choices are related to their policy preferences. With this result, we are able to confirm the hypothesis 4.

Finally, the type of legislature (regional or national) plays a role in explaining the opinion congruence inequality between the lower- and higher-educated groups: the regional parliaments seem to be more biased toward higher-educated groups than the national parliament. This could be explained by the fact that the elections for both the regional and national parliaments were held on the same day, May 25, 2014. While they are both normally considered to be first-order elections (Deschouwer, 2012), it could be that, because they coincided, the election for the national parliament became more important. This may have given parties and political elites less leeway to deviate from the policy preferences of the underprivileged voters on the national level, but more so on the regional level.

The language group has an effect on the inequality between higher- and lower-income groups. In the Francophone language group, the inequality between those groups is significantly lower. This could be explained by the difference between the Flemish and Walloon social-democratic parties. The former has lost much more of its connection to its traditional underprivileged voter base, while the latter is still clearly a travaillist party (Coffé, 2008). This would suggest that inequality in opinion congruence not only depends on the choices voters make, as the hypothesis 4 predicts, but also on the choices given to the privileged and the underprivileged groups.

To get a better idea of the relations between congruence inequality, differences in policy preferences, and inequality in correct voting, we calculate the predicted values of opinion congruence inequality for each value of the two independent variables based on the models in Table 2. Figure 4 represents the relation between the difference in opinion and opinion congruence inequality for the education and income groups. It shows that when the policy preferences of the privileged and the underprivileged diverge, the opinion congruence inequality steadily increases, from no difference in collective congruence when groups have the same policy positions up to differences of 64 and 41 per cent for education and income groups, respectively, when there is a total opinion divergence between the social groups. Opinion congruence inequality thus increases in favor of the positions of higher-educated and higher-income groups as the opinions diverge more. Figure 5 shows the relation between inequality in correct voting and opinion congruence inequality. When one social group has voted more correctly than another social group, then the preferences of the former will be better represented than those of the latter. This applies to both the privileged and the underprivileged groups; both are able to turn congruence inequality in their favor by voting for parties with which they agree in terms of policy.

Figure 4
figure 4

The relation between differences in policy positions and opinion congruence inequality

Full size image
Figure 5
figure 5

The relation between differences in party-dyadic congruence and opinion congruence inequality

Full size image

Conclusion

This paper examines to what extent and why the policy preferences of the privileged groups (higher-educated and higher-income groups) are better represented in Belgium’s regional parliaments and national parliament compared to those of the underprivileged groups (lower-educated and lower-income groups). Previous studies identified three major causes of this inequality: one-sided business lobbying, political donations, and a social skew in electoral turnout. These factors, however, are situated at the country level and are unable to explain variation in congruence inequality between issues. The purpose of this paper was to develop a model capable of filling this gap. Our results show that when social groups differ in policy positions, legislatures are more likely to be in line with the preferences of the privileged groups than with those of underprivileged groups. A representational bias thus becomes more pronounced as opinions diverge.

In addition, we found that policy domains matter: the preferences of the privileged groups are better represented on issues vital to their interests such as economic and tax policies, but on issues related to social welfare, which are key to the interests of the underprivileged voters, preference representation is far less skewed toward society’s well-off population. Finally, we find that the degree to which groups vote correctly (i.e., for parties that share their policy positions) also affects congruence inequality. When one group votes more correctly than another on an issue, the former’s policy positions will be better represented than those of the latter. This relation applies to both the privileged and the underprivileged groups and indicates that congruence inequality is, to an important degree, self-inflicted by social groups themselves.

Our results raise normative questions. Is the opinion congruence bias toward the preferences of higher-educated or higher-income groups problematic? Research has shown these groups to be the most informed and interested in politics (Hillygus, 2005). Our findings may therefore sound pessimistic to proponents of democratic theory, who emphasize an equal representation of policy preferences (Dahl, 1989; Page and Shapiro, 1992), but they could sound encouraging for those who consider large portions of the public, predominantly from the lower strata of society, to be ill-informed (Lippmann, 1955). Yet authors of the latter conviction also believe that politics should advance the general interest, and it is, however, uncertain whether an adherence to the policy preferences of the privileged groups will lead to the pursuit of the general interest rather than the pursuit of the interests of a specific social group.

The purpose of this study is largely exploratory, distinguishing between country- and issue-level factors of congruence inequality and developing a model for the latter. We believe that future research could build on and expand this model to develop it further. For one, the data on the policy preferences of voters were collected before an electoral campaign, and there are two reasons to assume that the congruence gap between the privileged and the underprivileged groups is underestimated. First, while campaigns are known to be information-dense moments (Alvarez, 1998), research has suggested that campaigns are more likely to benefit already knowledgeable voters (usually the higher-educated or higher-income ones) instead of voters who stand to benefit most from the information disseminated in a campaign (Lesschaeve and Meulewaeter, 2015). The gap in correct voting—and, by extension, the inequality in collective congruence—between the privileged and the underprivileged groups that exists before the campaign may even be larger after it. Second, we cannot exclude the possibility that, on some issues, and in anticipation of the electoral campaign and the media and public scrutiny it entails, parties may take policy positions in order to be more congruent with the underprivileged groups. Scholarly attention should therefore also focus on inequality in collective congruence between elections. We would expect the differences between the privileged and the underprivileged groups to be even higher than those found in the present study.

In addition, while we attempt to take into account the saliency of issues by identifying policy domains that touch upon the vital interests of the privileged or the underprivileged groups, they remain crude measures. Future studies should try to include more precise measures of salience. This could be done, for instance, by looking at the media attention given to the various policy issues, or by measuring the importance of each policy issue to the various groups. This could give more insight into why, on certain policy issues, parliamentary opinion favors the position of higher-educated or higher-income groups.

In conclusion, the focus of the literature on country-level characteristics threatens to underexpose differences in the congruence inequality between issues. By distinguishing between a country- and an issue-level model, we believe that we have presented a novel and complementary way to start thinking about congruence and representational inequality.

Notes

  1. 1

    http://www.ibzdgip.fgov.be/result/nl/main.html. In Belgium, voters are required to vote. However, voters can still cast a blank vote.

  2. 2

    The survey was conducted by TNS, and all respondents come from its “managed access panels”. Response rates are increased through careful panel management so as to avoid contacting the respondents too frequently and give them participation incentives.

  3. 3

    The voter survey was weighted in order to accurately reflect the Belgian population in terms of six characteristics: gender, age, occupation, education level, social class, and Nielsen region, based on the most recent population data provided by the Centre for Information on the Media. Every respondent was designated a weight between the minimum value of 0.0001 and the maximum value of 2.

  4. 4

    The parties included in the party survey are Groen and Ecolo (green parties), Sp.a and PS (social democrats), CD&V and CDH (Christian democrats), Open VLD and MR (liberals), N-VA and FDF (regionalist parties), and Vlaams Belang (extreme right/separatist party).

  5. 5

    A full list of the statements can be found in the online appendix.

  6. 6

    In addition, we subtract the formula of Golder and Stramski (2010) from 1 so as to have high values reflect high levels of congruence and low values reflect low levels of congruence.

  7. 7

    As a robustness check, we examined the difference in opinion congruence between the privileged groups and the middle-class (middle-educated voters and voters from the two middle-income deciles) and found that, while the inequality in congruence is smaller, it is also in favor of the preferences of the higher-educated or higher-income groups.