Differential Responsiveness in Europe: The Effects of Preference Difference and Electoral Participation

Abstract

As a core principle of contemporary democracy, political representation has been the subject of numerous studies. In particular, responsiveness has been thoroughly examined and research suggests that policies tend to reflect citizens’ preferences. However, it has also been argued that, in some instances, responsiveness systematically reflects the preferences of the rich better than those of the poor, hence violating a second democratic principle, that of political equality. While much research has focused on the United States, this study tests whether differential responsiveness also occurs in European democracies and enquires about the structural factors that may cause such inequality. The article examines to what extent the preference gap between the rich and the poor as well as the level of electoral participation can account for variation in differential responsiveness. To do this, a dataset including information on 25 European countries from 2002 to 2010 is constructed and analysed using time-series cross-sectional methods. The findings suggest that European democracies experience differential responsiveness and that the preference gap and level of turnout partly account for this.

Representation is a core feature of contemporary democracy, and has consequently been the topic of numerous studies. Following democratic norms (Dahl 1971, 1998), one important element of democracy is that people have a substantial influence over which policies a government passes and which it does not pass. Such congruence between people’s preferences and policy output has been established in a number of contexts (e.g. Erikson et al. 2002; Hobolt and Klemmensen 2008; Kang and Powell 2010; Soroka and Wlezien 2010). A second democratic principle, however, underlines the importance of political equality. This principle suggests that governments should not just be responsive overall, but they should also be equally responsive to their citizens. While this cannot mean that the government follows each citizen’s wish all the time, it does imply that any inequality in terms of representation should not be structural.

Previous research, however, has shown that such differential responsiveness exists. Much of this literature has focused on the United States (e.g. Bartels 2008; Gilens 2005, 2012). Although recent studies broaden this focus to Germany (Fortin-Rittberger and Eder 2013), Switzerland (Rosset 2013), and larger sets of countries (Reynolds 2013; Rosset et al. 2013), unequal responsiveness has scarcely been investigated in the European context. Moreover, for now, there is only little known about the possible causes for differential responsiveness. With this study, we aim to contribute to this field of research, and aim to expand our understanding of inequality in representation.

Our objective is to link people’s preferences to policy output to explore whether there is a systematic bias in responsiveness according to different income groups. We examine the effects of popular support for government welfare effort on social expenditure (policy output), and thus focus on social policy as a political issue. Moreover, we attempt to find whether this potential bias can be understood through the preference gap between income groups, as well as through the level of voter turnout. The article proceeds as follows. First, we discuss the concept of representation, its link to democracy, and review some of the relevant literature. Second, we elaborate on the reasons why we may expect to observe differential responsiveness, and in what way we can understand such a finding better. This section also formulates the expectations. Third, the data and variables are discussed, and the method used to analyse the data is introduced. We then present the results and end with some concluding remarks.

Representative Democracy and Inequality

The principle of representation is central to contemporary democracies, essentially structuring the organisation of policy formation. Although the concept of representation can be viewed in different ways (e.g. Mansbridge 2003; Pitkin 1967; Rehfeld 2009), it consists of two crucial elements: responsiveness and responsibility (Mair 2009). As a core ideal of democracy, citizens’ preferences should be deliberately incorporated into the rules, laws, and policies that govern them (Dahl 1971, 1998) (responsiveness). It is, however, also important to keep in mind that such congruence should not violate other principles of democracy such as political equality or citizenship rights, for which representatives would also be responsible (responsibility). Although mainly focusing on responsiveness, our research aims to touch on the two elements of representation by examining both responsiveness and the equality of people’s influence.

Furthermore, we specifically focus on ‘anticipatory representation’ (Mansbridge 2003), which suggests that because representatives want to win the next elections they aim to represent people in a dynamic way. They would be able to accomplish this through strengthening connections with citizens using communication and deliberation throughout (Disch 2011; Mansbridge 2003), leading to a voter/representative relationship that ‘is best conceived as one of reciprocal power and continuing mutual influence’ (Mansbridge 2003: 518). Following these theoretical characteristics, we empirically expect to see a rather dynamic and continuous correspondence of citizens’ preferences and policies. The second form of responsiveness that we examine, albeit to a lesser extent, is ‘promissory representation’ (Mansbridge 2003). Here, the idea is that citizens can be represented through a government that incorporates people’s preferences as expressed through their cast votes. Following this rationale for responsiveness, we expect that the left–right position of the government is reflected in the policies that they make.

As a crucial characteristic of representative democracy, the topic of responsiveness has received much attention. Already in the 1960s, Miller and Stokes (1963) investigated to what extent the views of citizens and elected officials align, and whether people’s preferences are also reflected in the roll-call behaviour of congressmen. More recently, scholars have looked at how public policy follows public opinion, hence focusing on the type of dynamic responsiveness implied by anticipatory representation (Erikson et al. 2002; Stimson et al. 1995). Moreover, in developing the idea of dynamic responsiveness, studies have shown that the influence of public opinion on policy output is actually reciprocal: policies and citizen preferences respond in fact to each other (Soroka and Wlezien 2010; Wlezien 1995). Overall there seems to be support for a general correspondence between what people want and what representatives do.

Touching more upon the democratic requirement of political equality, some scholars have examined congruence between preferences and policies more closely. Especially Larry Bartels and Martin Gilens have been concerned with inequality in the United States, and the possibility that some groups are better represented than others. They (Bartels 2008; Gilens 2005, 2012) find such a bias in responsiveness: the preferences of the rich appear to be systematically better represented than those of the poor. Furthermore, Enns and Wlezien (2011) disentangle differential responsiveness by focusing on various groups. They examine responsiveness and people’s preferences, and find that the middle-income group is important, too, and that differences in preferences are often more notable for different groups according to, for example, race, education, and region than to income. Although much research has focused on the US, Rosset examines whether there is a difference in ideology congruence according to income in Switzerland. He, too, finds that representatives are better aligned with the ideology of the higher-income than with the lower-income group (Rosset 2013). Furthermore, a recent study (Rosset et al. 2013) has shifted its focus to a more comparative perspective of differential responsiveness, including 24 democracies to demonstrate that under-representation of the poor is a more widespread phenomenon. By looking at European democracies, and trying to find what can help to explain the unequal patterns, our study aims to extend and contribute to this line of research.

Understanding Differential Responsiveness

From an ideological point of view, we would expect that democracies more or less fulfil the democratic conditions that Dahl (1998) posed. This means that we expect that there is at least some congruence between public opinion and policy output, and that congruence is more or less equal across income groups.

On the other hand, based on the ‘median voter theorem’ we might expect that the median citizen would determine policy output. Political parties can be expected to converge towards the median voter, or a government would include the median party (linked to the median voter), so that the governing majority is defined by that median voter (Downs 1957). Thus, in either system for translating votes into governments (plural and proportional) it has been argued that the median voter is crucial, and would be specifically influential in determining policies (see Huber and Powell 1994; Kang and Powell 2010). Considering that the issue we look at concerns income redistribution, this median citizen is most likely placed within the middle-income group. If citizens’ preferences are considered equally, we would expect that responsiveness to the poor and to the rich would be equally (ir)relevant if the distance in preference to the middle-income group is equal, and if the fact that these groups have different incomes is irrelevant.¹

However, there exist practical reasons to suggest that the latter is not the case. First, the composition of the representative body does not fully mirror society in its socio-economic characteristics, largely leaving out the lower-income group (e.g. Rosset et al. 2013). While it does not mean that these preferences cannot be represented, descriptive representation could improve substantive representation through ‘shared experience’ (Mansbridge 1999). A lack of such descriptive representation could accordingly lead to differential responsiveness. Second, the fact that some people indeed have higher incomes may imply that these people are more influential due to that economic advantage. For example, financial contributions to candidates or political parties (or ‘checkbook participation’; see Verba et al. 1995) help to strengthen these actors politically (for example in campaigns or recruitment), and could therefore make them dependent on the source of those big contributions (Scarrow 2007). Furthermore, lobbyists can influence representatives (Campos and Giovannoni 2007; Gilens 2012: ch. 5). Research has shown that this influence can be systematically biased towards the wealthier business interests, depending on the institutional characteristics of the political system (Mahoney 2007). It seems that the ‘haves’ may also have more practical political influence (see also Rosset et al. 2013).

In sum, we expect that the preferences of the rich weigh disproportionately in policy output compared to other citizens, and especially than those of the lower-income group. This leads to our first expectation:

Proposition 1: The preferences of lower-income and higher-income groups will have a different effect on policy output, in such a way that poorer people are under-represented and/or richer people are over-represented.

There are, however, also other explanations for the hypothesised differential responsiveness, which are rather indirectly related to the economic status of citizens. The first explanation of the (non-)finding of differential responsiveness has to do with the extent to which the rich and the poor differ in their preferences. When different groups in society all want the same thing, we will not observe any difference in the congruence between preferences and policies of different groups (Soroka and Wlezien 2008; Ura and Ellis 2008).² Gilens (2011, 2012) argues that a preference overlap between the rich and the poor makes policies more responsive to the poor by chance: ‘any association between preferences and policy outcomes is likely to reflect the extent to which their preferences coincide with those of the affluent’, and finds that a bigger preference gap is associated with a bigger gap in responsiveness to the extent that ‘responsiveness to the less well-off Americans is virtually nonexistent’ (Gilens 2011: 254). Thus if the government systematically follows one group over the other, any responsiveness to the disadvantaged group is mostly due to coincidental correspondence in preferences. This would only be true, however, if the difference in income indeed matters in levels of responsiveness – if income does not matter, we expect a responsiveness gap between the rich and the poor to increase only when one of these groups is closer to the preferences of the middle-income group.

However, as argued above, we theorise that income matters and therefore expect that with greater preference heterogeneity, the gap in responsiveness also becomes greater. This leads to the following (conditional) expectation:

Proposition 2: An increased preference gap between the rich and the poor diminishes responsiveness to the poor, increasing the responsiveness gap.

The second explanation concerns political participation. Representatives would need to have some information about what people want in order to act on their preferences. Without active participation in politics, or a lack of communication between representatives and the represented, it is difficult to implement policies reflecting citizens’ preferences. Griffin and Newman (2005) underline this point with their finding that voters are better represented than non-voters. Theoretically, then, we might expect that higher turnout levels lead to less inequality in representation. Further, with many studies focusing on the determinants of political participation, it has become apparent that there are certain structural inequalities in who participates (e.g. Marien et al. 2010; Verba and Nie 1972; Verba et al. 1995). Among the structural factors for unequal participation, income is often found to be a good predictor of whether or not someone participates politically, with others typically being education, gender, and age (Marien et al. 2010; Verba et al. 1995). These inequalities are thus particularly pronounced when turnout is low, and may translate into differential responsiveness. Indeed, the possibility that unequal participation in elections is likely to lead to the representational exclusion of people has been one of the main concerns of scholars investigating it (Lijphart 1997; Verba et al. 1995), though it has often been assumed rather than empirically studied. We therefore test whether levels of turnout condition differential responsiveness, and anticipate that higher turnout promotes more equal responsiveness.

Yet we do not expect this effect to be linear. First of all, inequality in participation according to income is probabilistic in the sense that the first 33 per cent of people turning out are not also the 33 per cent richest people, and the last 33 per cent of voters are not the 33 per cent poorest. There is much more overlap than that. Second, an increase in turnout specifically implies a trade-off in responsiveness to the rich and poor. When turnout increases initially, the positive responsiveness to the rich will initially decrease as more and more people with different incomes also start participating. At the same time, the poor would be less under-represented according to the notion that an increasing number of poorer people participate and are represented. But once turnout levels are substantial, these developments will reach a minimum or maximum: when turnout is high, more people from different income groups are represented and any responsiveness advantage disappears. The two groups will end up being roughly equally responded to, as no one group systematically gets what it wants when representation is increasingly equal. We thus expect that the effect of turnout on responsiveness largely follows a quadratic function, with the responsiveness of the rich and the poor gradually balancing each other out with increasing turnout.

This leads to the following proposition:

Proposition 3: Higher levels of turnout will initially lead to less under-responsiveness to the poor and over-responsiveness to the rich, and will gradually lead to more balance in responsiveness once turnout is sufficiently high.

Variables and Data

In order to find whether there exists differential responsiveness and to what extent this is affected by preference polarisation and voter turnout, we combine data from different sources. Information concerning redistributive policy outcomes is provided by the Organisation for Economic Co-operation and Development (OECD), and is available for many European countries (including Israel), for a substantial period of time. Because we link the social spending data to data from the European Social Survey (ESS, rounds 1–5), the time period is limited to 2002–2010. The 25 countries on which data is available for at least one time-point are: Austria, Belgium, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Luxembourg, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, and the United Kingdom. The unit of analysis in our study is thus country-year, and the effective number of cases is 106.

Dependent Variable

In this study, we aim to link support for redistribution to social spending as an indication of responsiveness. Our measure for the dependent variable, economic redistributive policy output, is public social expenditure as a percentage of GDP. This measure of ‘welfare state effort’ (Brooks and Manza 2006) indicates the extent to which the government is responsible for the redistribution of wealth within a country. The OECD provides data on this measure and includes old age, survivors, incapacity-related benefits, health, family, active labour market programmes, unemployment, housing, and ‘other’ social policy areas.³ Available data on public social expenditure covers OECD countries for a period between 1980 and 2010, and extends the information to 2011.⁴ The indicator has a theoretical range of 0 to 100 since it reflects a percentage of GDP, though practically it ranges from 12.8 per cent (Estonia 2001) to 31 per cent (France 2010). Table 1 gives some descriptive information on the indicator.

Table 1 Overview of the variables and description of the data

	Mean	Median	Std. Deviation	Minimum	Maximum
Redistributive spending	22.62	22.50	4.31	12.80	31.00
Support general population	72.73	74.58	12.02	38.01	93.23
Low-income support	79.48	81.08	10.46	43.44	96.50
High-income support	65.14	64.71	14.10	31.79	95.25
Difference in mean preference	0.38	0.39	0.17	0.02	0.85
50% vote threshold	0.84	1.00	0.37	0.00	1.00
Government ideology	−2.40	−3.83	14.38	−37.81	48.19
Annual % growth GDP	2.27	2.50	3.87	−18.00	12.20

Independent Variables

The main independent variables in this study are support for more redistribution (aggregate and per income group), the difference in preferences on redistribution between higher- and lower-income groups, and turnout. Concerning the support for redistribution, we use data from the ESS (rounds 1 to 5) to create an aggregate measure of support (country-year). Thus, for each country and each two years,⁵ a percentage of people in favour of more redistribution is calculated, combining respondents that either ‘agree’ or ‘strongly agree’ with the statement that ‘the government should take measures to reduce differences in income levels’. This measure can be interpreted as a general indicator of support for the welfare state, and refers to the role that the government should play in the redistribution of wealth within a country (see e.g. Brady and Finnigan 2014; Brooks and Manza 2006). As such, the percentage of people supporting the government’s role in redistribution indicates a level of welfare state support that relates to the level of welfare state effort (social spending). Our focus here is thus on connecting preferences for welfare to the ‘amount’ of welfare that is provided, both over time and across countries.

However, since the survey question that we use asks whether the respondent wants the government to reduce income differences, it can also be interpreted more literally as a question about policy change. For this reason, we additionally analyse whether preferences for change in redistribution also translate into a change in policies. Thus, we regress the change in social spending on the mean preference for redistribution, and report the results for the full models in addition to the models analysing the relationship in levels.

In order to calculate support within each income group, the sample for each country and year needs to be split in three different income groups of more or less equal size. This is important because we aim to compare the relationships of support within different income groups to policy output, and need to exclude the possibility that a group is better represented simply because it is larger in size. Using this relative measure of income also implies that being poor according to our standard may imply something different in, for example, Norway and Spain. Since a lot of such difference is based on economic inequality (or the extent of redistribution) these differences will also form the variation in levels of responsiveness in our analyses – in countries where poor is poorer, the role of money will be more pronounced and the preferences of the income groups most likely more contrasted, leading to a bigger responsiveness gap.

We systematised the split-up according to the distribution of income for each country-year. The low-income group is defined as people with an income below the mean minus half a standard deviation (equation (1)); the high-income group as people with an income above the mean plus half a standard deviation (equation (2)); and the middle-income group is defined as people with an income between these two cut-off points (equation (3)):(1) (2) (3)

where μ is the mean value for a country-year and σ is the standard deviation of household income for each country-year. This results in three groups of more or less equal size,⁶ with the level of income as the only systematic difference between them. For each of these groups, the level of support for redistribution is calculated as indicated above. The theoretical range of the variables is between 0 and 100, since they involve a percentage of people that express support for the welfare state. The practical range for the support measure, however, is between 43.44 per cent (Denmark 2010) and 96.5 per cent (Greece 2004) for the low-income group and between 31.79 per cent (Denmark 2010) and 95.25 per cent (Greece 2004) for the high-income group.

In order to explain differential responsiveness (the relation between popular support for welfare and welfare effort), we first look at the differences between preferences of the higher- and the lower-income groups. This difference is based on the mean score on the statement ‘the government should take measures to reduce differences in income levels’, where respondents could indicate they ‘strongly agree’ (1), ‘agree’ (2), ‘neither agree nor disagree’ (3), ‘disagree’ (4), or ‘strongly disagree’ (5). The difference between the mean in support of lower- and higher-income groups constitutes the preference difference. This indicator ranges from 0.02 (Israel 2010) to 0.85 (Czech Republic 2008).

To test our expectations concerning the effect of voter turnout on responsiveness, two approaches are possible. First, we can measure a level of electoral participation within each income group based on the data from the ESS.⁷ This would allow us to differentiate participation between groups and isolate the effects of different levels of turnout on responsiveness for each group. The disadvantages of this approach are (1) that self-reported turnout is less reliable than officially reported turnout levels, and (2) that group turnout is kept separate – i.e. the measures do not incorporate a balance or proportion of group turnout. The second approach possible is to include a national level of turnout, with the disadvantage that we cannot monitor explicitly whether the lower-income group indeed turns out less than the higher-income group. While the two possibilities each have advantages and disadvantages, they are largely equivalent. We opt here for the second option because it provides us with a relative measure, and allows us to test whether contexts with higher turnout also have less inequality in responsiveness. Data are provided by the International Institute for Democracy and Electoral Assistance (International IDEA), and are based on the voter age population. However, as the former possibility offers a valuable addition to test our expectation, we also include it in our analyses, and report the results in the online appendix. Results are highly similar. Table 1 provides an overview of some descriptive information on the indicators.

Control Variables

Although we may consider several control variables to account for the variation of social spending, just two control variables are included. The reason for this is that our aim is not to explain levels of social spending as such, but rather to find whether there is a relation between what people want and what they get regardless of other explanations. It is nonetheless important to account for some minimal variation due to country and time differences. We therefore include two important factors in the analyses: government ideology and economic growth. First, the ideology of the government in office may affect social policies (e.g. Bartels 2008; Brooks and Manza 2006; Hobolt and Klemmensen 2008; Huber and Stephens 2001). Generally, it can be argued that the policy output of a government is strongly influenced by the specific parties that are in government, and government ideology can be seen as indirect responsiveness or promissory representation (Mansbridge 2003). Here a government’s actions are a reflection of people’s expressed preferences during the previous election. More left-wing governments are expected to have higher levels of social spending. To measure the left–right placement of political parties, the coding of party manifesto data was used; negative scores are associated with more left-wing parties, whereas positive scores indicate more right-wing parties. Since many of the included governments are coalitions, we weighted the manifesto ideology score according to their number of seats in parliament (using data from ‘parlgov’⁸) (see also Kim and Fording 2003). Our measure is summarised in equation 4:(4)

Where is the ideology of a government party, the number of seats of that government party, and the number of seats that the government holds.

Since our dependent variable is a relative measure to GDP, the change in GDP needs to be taken into account as a control variable. Social spending relative to GDP can change due to an actual change in social spending, but it also fluctuates according to a change in GDP – when policies linked to social spending remain stable but GDP increases (decreases), the share of social benefits will automatically be smaller (bigger). We control for this type of mechanical fluctuation of social benefit spending by including growth in GDP as a control variable (see also Kang and Powell 2010; Wlezien and Soroka 2012). The World Bank provides data on economic growth for a wide range of countries and years.⁹ Table 1 again reflects some descriptive information on the data for these two variables.

The aim in this study is to find whether different income groups have different relations to policy output, and to what extent this differential responsiveness can be explained. Pooled cross-section time-series models are useful for this purpose because the variation in the dependent variable is both cross-sectional and dynamic, and it allows for testing the propositions over time and across space. For all models, the assumptions of ordinary least squares were tested, and formed the basis of the specific choice for the models. First, to deal with persistent panel heteroscedasticity and serial correlation, the models are based on Prais–Winsten (AR1) regression analyses with panel corrected standard errors (PCSE; see Beck and Katz 1995). Second, some models are not able to explain all cases equally well (outliers), and some of these also had too much influence over the results of the analyses (too high leverage) – these cases were removed from the analyses and this is specified where the results are reported. Although these adjustments only aim to correct the models methodologically, they should allow the estimator to be the best linear unbiased estimator (BLUE). There were no problems with a unit root or specific time or country heterogeneity. Furthermore, in line with most research on responsiveness, popular preferences are lagged by one year since it takes some time to implement policy changes. All other independent variables are also lagged by one year.

Robustness of the Results

In order to illustrate the robustness of our findings, we include the results of some further analyses. First, as discussed above, the results section below will discuss the differenced analyses dealing purely with changes in policies and preferences, as opposed to levels. Furthermore, the online appendix reports the results of a fixed effects model, controlling for country-specific effects.¹⁰ Moreover, models including the alternative measure for turnout (specified to income group) are included. And lastly, the online appendix reports a replication of the two full models, substituting the lower-income group with the middle-income group. These last analyses are to underpin the disproportionate effect of the rich on policies, even when compared to the middle-income group.

Results

If income does not play a role in responsiveness, we might expect the median citizen to have a pivotal role in determining policy output. From this perspective, we would expect a difference in responsiveness only if one group is closer in the level of support to the middle-income group. Therefore, the first step in the analysis is to examine the differences in support between the groups. Table 2 shows the results of a t-test, and suggests that the levels of support for redistribution between the three groups differ significantly. Moreover, it shows that overall the level of support for the low-income groups is closer to the level of the middle-income groups than that of the high-income groups with differences of 4.4 per cent and 9.9 per cent respectively. Figure 1 further illustrates the difference in support between the income groups. If the median citizen is indeed pivotal and if income is irrelevant, the support of the lower-income groups should be more closely related to output compared to that of the high-income groups. Our analyses compare responsiveness of the rich and poor and are discussed next. The results of the comparison of responsiveness between the higher- and middle-income groups are reported in the online appendix, and further illustrate the disproportionate role of the rich in determining welfare state efforts.

Table 2 Differences in levels of support for redistribution between income groups

	Mean (standard error)	Difference (standard error)
Low-income support	79.481 (0.971)
		4.406*** (0.370)
Middle-income support	75.075 (1.078)
		9.938*** (0.556)
High-income support	65.137 (1.310)

Note:

Results are based on a t-test.

* p value ≤ 0.10

** p value ≤ 0.05

*** p value ≤ 0.01.

Figure 1 Differences in support for redistribution between income groups

We expected to find that people with a lower income are under-represented, and/or people with a higher income are over-represented based on the specific idea that a higher income helps to get better representation. The results of these analyses are reported in Table 3. It shows that general support for redistribution is negatively and significantly related to social spending, in both model (1) and (2). This suggests that higher levels of support for redistribution are related to lower levels of actual social spending, and seems to indicate a general under-responsiveness of preferences in terms of anticipatory representation. People seem to get less redistribution than they would like. Furthermore, models (2) and (4) show that the indicator for government ideology has a negative and significant relation to social spending. More right-wing governments thus tend to have lower levels of social spending, supporting our expectation. It further underpins the idea of promissory representation, where elected governments aim to fulfil the promises made before the election. Furthermore, Table 3 indicates that economic growth is negatively related to social spending (models (2) and (4)), and highlights that a higher percentage growth is related to lower proportional levels of social spending. We suggested that economic growth needed to be taken into account since the percentage of GDP would partly be a function of the growth in GDP, where spending would vary mechanically according to the extent that GDP increases or decreases. The results support this idea.

Table 3 General and differential responsiveness on redistributive spending

	Total social expenditure in % GDP
	(1)	(2)	(3)	(4)
Support general population	–0.053*	–0.051**
	(0.030)	(0.026)
Low-income support			–0.196***	–0.088*
			(0.059)	(0.048)
High-income support			0.087*	0.010
			(0.049)	(0.038)
Government ideology		–0.055**		–0.086***
		(0.024)		(0.028)
Annual % growth GDP		–0.692***		–0.700***
		(0.094)		(0.108)
Intercept	26.722***	28.423***	32.967***	31.211***
	(2.208)	(1.983)	(2.119)	(2.383)
N	104	104	104	101
N of countries	25	25	25	25
Adjusted R-square	0.734	0.825	0.662	0.820

Notes:

Models report results from Prais–Winsten regression analyses. Jackknifed panel-corrected standard errors are reported in parentheses. All independent variables are lagged with one year. Model (1) excludes Denmark 2009 and 2011; model (2) excludes Estonia 2009 and Greece 2011; model (3) excludes Ireland 2011 and Israel 2011; model (4) excludes Estonia 2009, Greece 2011, Norway 2007, 2009, and 2011, because these cases could not be predicted by the models and have too high leverage on the general results.

* p value ≤ 0.10

** p value ≤ 0.05

*** p value ≤ 0.01, two-tailed.

Regarding the existence of differential responsiveness in Europe, Table 3 provides support for the first proposition. Model (3) suggests that support for redistribution of the lower-income group is negatively and significantly related to social spending. Higher levels of welfare support within the lower-income group are thus accompanied with lower levels of social spending. This relation remains negative and significant after introducing the indicators for government ideology and economic growth, and supports proposition 1 with the finding that poorer people seem under-represented. Moreover, model (3) reflects a positive effect of support for redistribution within the higher-income group. Richer people thus appear over-represented compared to those on lower incomes. In model (4), this relation is no longer significant. However, the results reported in Table 3 provide support for proposition 1 in that the two groups have a different relation to policy output, where the poor seem to have a disadvantage compared to the rich. These findings corroborate those reported in the study by Rosset et al. (2013).

The second proposition concerned the size of the difference in preferences between the two different income groups. We expected that when these differences are relatively big, we would also see more pronounced differential responsiveness. Table 4 reports the results of the analyses that test proposition 2. The models in this table now also include the interaction effects between support for redistribution and the difference in preferences between the two income groups. It shows that when this difference is non-existent neither of the income groups has a significant influence on levels of social spending. In this situation, we thus do not observe differential responsiveness. The interaction effects suggest, however, that when the preference gap between the two groups increases, the effects of both groups increase: the effect of support of the lower-income group increases negatively, while the effect of support within the higher-income group increases positively. This finding therefore supports our second proposition in that larger differences in preference do indeed make differential influence more pronounced. Moreover, the marginal effects graph of the relationship between support and social spending for model (5) (presented in Figure 2) illustrates that the poor have a negative and significant effect on policy output when the preference gap is larger than about 0.45. So, while the rich do not seem to have a significant effect on spending, the poor are increasingly under-represented the larger the preference gap is. This finding holds when government ideology and economic growth are included in the analysis, although the effect of support for redistribution on social spending is no longer significant for either income group, even when the preference gap is large. While the preferences of the rich remain insignificant, it is not clear why the size of the effect increases with the size of the preference gap. Comparing the effects of the preference gap between the middle- and higher-income groups (reported in the online appendix), the results are less clear. Here, even when the gap in preferences is relatively large, responsiveness for neither higher- nor middle-income groups is significant, even though the (smaller) gap between the higher- and middle-income groups is similar to that between the rich and the poor.

Table 4 Differential responsiveness and preference difference between income groups

	Total social expenditure in % GDP
	(5)	(6)
Low-income support	–0.012	–0.017
	(0.169)	(0.222)
High-income support	–0.002	–0.025
	(0.138)	(0.161)
Difference in mean preference	26.580	8.785
	(16.401)	(22.984)
Low-income supp.*difference	–0.541*	–0.280
	(0.287)	(0.407)
High-income supp.*difference	0.285	0.238
	(0.194)	(0.205)
Government ideology		–0.050**
		(0.025)
Annual % growth GDP		–0.561***
		(0.117)
Intercept	23.542***	26.789***
	(6.077)	(8.871)
N	105	106
N of countries	25	25
Adjusted R-square	0.614	0.771

Notes:

Models report results from Prais–Winsten regression analyses. Jackknifed panel-corrected standard errors are reported in parentheses. All independent variables are lagged with one year. Model (5) excludes Czech Republic 2003, because this case could not be predicted by the model and has too high leverage on the general results.

* p value ≤ 0.10

** p value ≤ 0.05

*** p value ≤ 0.01, two-tailed.

Figure 2 Marginal effects of preferences, conditioned on preference difference (model 5)

Furthermore, we expected that the level of turnout would affect differential responsiveness in a quadratic way. In order to test our proposition 3, we therefore included interaction effects between turnout and responsiveness as well as between squared turnout and responsiveness. We expected that levels of responsiveness for the poor and the rich would increase/decrease initially, and would present a responsiveness trade-off between the rich and the poor when turnout would increase.

Table 5 reports the results of the analyses to test this proposition. It shows that support of the low-income group is negatively related to social spending when turnout is zero, while this relation is positive for the support of the high-income group. The latter is, however, no longer significant when government ideology and growth in GDP are introduced in model (8). Figure 3 illustrates the quadratic marginal effects of turnout on levels of responsiveness, based on model (8). It shows that the relation between support and spending for the rich is positive (and partly significant, when turnout is around 30 per cent) but remains about zero once turnout passes 50 per cent. The relation between support and spending for the poor is negative and significant when turnout is relatively low (<40 per cent), and remains almost zero when turnout is higher than 50 per cent. However, the effects of lower-income preferences is negative again when turnout is very high, suggesting that the disadvantage of the poor does not dissolve completely when turnout is high. These results suggest that higher levels of turnout may indeed help to reflect preferences of different groups more equally although the difference does not disappear completely. The findings thus largely support our proposition 3.

Table 5 Differential responsiveness and voter turnout

	Total social expenditure in % GDP
	(7)	(8)
Low-income support	–3.310***	–2.707***
	(1.223)	(0.968)
High-income support	1.732*	1.387
	(0.961)	(1.113)
Turnout	–5.202***	–4.638**
	(1.763)	(1.809)
Turnout²	0.043***	0.039***
	(0.013)	(0.014)
Low-income supp.* turnout	0.106***	0.092***
	(0.040)	(0.031)
High-income supp.* turnout	–0.052	–0.043
	(0.032)	(0.034)
Low-income supp.* turnout²	–0.001***	–0.001***
	(0.0003)	(0.0002)
High-income supp.* turnout²	0.0004	0.0003
	(0.0003)	(0.0003)
Government ideology		–0.076**
		(0.030)
Annual % growth GDP		–0.676***
		(0.117)
Intercept	176.026***	152.077***
	(57.0465)	(57.687)
N	106	104
N of countries	25	25
Adjusted R-square	0.714	0.800

Notes:

Models report results from Prais–Winsten regression analyses. Jackknifed panel-corrected standard errors are reported in parentheses. All independent variables are lagged with one year. Model (8) excludes Estonia 2009 and Greece 2011, because this case could not be predicted by the model and has too high leverage on the general results.

* p value ≤ 0.10

** p value ≤ 0.05

*** p value ≤ 0.01, two-tailed.

Figure 3 Marginal effects of preferences, conditioned on voter turnout (model 8)

It is not surprising that low turnout benefits the rich more than a high turnout benefits the poor. The studies on unequal participation discussed above suggest that people with a higher income are more likely to turn out than people with a lower income. This would therefore also mean that a high turnout includes both high- and low-income groups, and would therefore theoretically lead to a more widespread incorporation of preferences: representatives may be better able to represent different groups in society when they have more information about their preferences. This in turn could lead to more equal representation, which the findings reported in Table 5 and Figure 3 support. They show that higher levels of participation in elections seem to lead to reduced differential responsiveness, even though the effect of the poor and the rich on spending is not fully equalised. Low levels of voter turnout, on the other hand, appear to make differential responsiveness more pronounced in such a way that the poor are under-represented and the rich over-represented. This finding is strikingly similar when the effects of the middle- and high-income groups are compared (online appendix): the effect of middle-income group is almost identical to that of the poor in the comparison with the rich. This further highlights the exceptional position of the rich in comparison to both the poor and the middle-income groups. It suggests that representation of the rich is not based on a coincidental overlap with the middle-income group: the middle-income groups do worse in terms of representation than the rich, and are more similar to the poor.

Lastly, we more explicitly explore temporal responsiveness. Table 6 shows the effects of preferences for change in redistribution on changes in social expenditure for the two full models including the interaction terms of the preference gap and voter turnout. While these models focus solely on explaining policy changes and the number of cases has reduced to 80, the main findings align with those discussed above. Figure 4 illustrates that with a change in the size preference gap, levels of responsiveness for the different groups also change: when the difference gets larger (negative values), preferences of the poor have an increasingly negative impact on change in social spending, while those of the rich have an increasingly positive effect. On the other hand, when differences in preferences get smaller (positive values) the effects of preferences become highly similar. This supports the idea that the poor are less under-represented primarily due to coincidental overlap in preferences with the rich.

Table 6 Differential responsiveness and change in spending

	∆ social expenditure in % GDP
	(9)	(10)
Low-income preference	3.237**	–93.103**
	(1.586)	(38.522)
High-income preference	–1.521	54.877**
	(1.150)	(21.139)
∆Difference in mean preference	40.483**
	(16.522)
∆Low-income pref.*difference	–13.768**
	(6.125)
∆High-income pref.*difference	4.848
	(3.729)
Turnout		–5.764**
		(2.576)
Turnout²		–0.044**
		(0.019)
Low-income pref.* turnout		3.238***
		(1.199)
High-income pref.* turnout		–1.969***
		(0.662)
Low-income pref.* turnout²		–0.026***
		(0.009)
High-income pref.* turnout²		0.016***
		(0.005)
∆Government ideology	–0.017	–0.036*
	(0.021)	(0.020)
Annual % growth GDP	–0.556***	–0.610***
	(0.114)	(0.109)
Intercept	–5.557*	174.335**
	(3.335)	(83.055)
N	80	80
N of countries	24	24
Adjusted R square	0.442	0.512

Notes:

Models report results from Prais–Winsten regression analyses. Jackknifed panel-corrected standard errors are reported in parentheses. All independent variables are lagged with one year. Model (9) and (10) exclude Greece 2011 because this case could not be predicted by the model and has too high leverage on the general results.

* p value ≤ 0.10

** p value ≤ 0.05

*** p value ≤ 0.01, two-tailed.

Note, also, that the main independent variable is now the mean answers to the statement ‘the government should take measures to reduce differences in income levels’, with the (recoded) categories 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree.

Figure 4 Marginal effects of preferences, conditioned on changes in preference difference (model 9)

Furthermore, Figure 5 shows the effects of levels of turnout on dynamic responsiveness. The results are very similar to those concerning the levels of responsiveness, albeit more pronounced. In contexts where turnout is lower than about 35 per cent, the rich are significantly over-represented while the poor are under-represented. Beyond that threshold, the responsiveness gap is much reduced, and even shows that the poor are over-represented when turnout is relatively high.

Figure 5 Marginal effects of preferences, conditioned on voter turnout (model 10)

Conclusion

In this study, we aimed to link people’s preferences to policy output to see whether there is a systematic bias in responsiveness according to different income groups. Moreover, we explored whether this potential bias is partly explained by the preference gap between the income groups, and by the level of voter turnout. While responsiveness is an important condition for democracy, political equality also constitutes such a condition. Each citizen should, ideally, have an equal voice in a country’s regulations and policies. And while it is impossible for a government to incorporate each citizen’s preferences all the time, from a democratic point of view there should not be a systematic inequality in whose voices are heard. Our analyses show, however, that there is little support for structural equality. Lower-income groups tend to be under-represented while higher-income groups appear over-represented. Furthermore, we found that differential responsiveness is more pronounced in situations where the preferences of the rich and the poor diverge more. When these preferences do not align, governments tend to follow the preferences of the rich more than those of the poor. Our analyses moreover show that the level of participation within democracies is important to understand why differential responsiveness occurs. When participation is relatively high, the preferences of more people are communicated to the representatives, and can lead to a more equal representation of preferences. Especially low levels of turnout seem to emphasise over-representation of the rich and under-representation of the poor. These findings relate to levels of welfare state effort, as well as to changes over time.

What is more, the difference in responsiveness cannot be attributed to similarities in preferences between the higher- and middle-income groups. Indeed, comparing responsiveness between these groups suggests that the effects of the lower- and middle-income groups are more similar than those between the higher- and the middle-income groups. This is especially pronounced in the results for the effects of turnout, which suggest that the rich are not coincidentally represented due to responsiveness to the middle-income groups.

With this study, we have aimed to contribute to the relatively new area of research that focuses on explaining differential responsiveness cross-sectionally. Future research should explore further explanations of such structural inequalities in order to better understand this phenomenon. For example, the status of the government (minority/majority/oversized), the number of veto points and players (for a reflection of narrow/broad interests), institutions and use of direct democracy as well as levels of economic inequality may help account for variation in differential responsiveness. Another way of exploring responsiveness in more depth is to specify the link between people’s preferences and policies. More can be done, for example, in more qualified measurement of policies and policy output (see e.g. Immergut and Abou-Chadi 2014), as well as exploring responsiveness in various policy areas. Lastly, we have also found some support for the notion of promissory representation, where representatives aim to implement the policies promoted during the election campaign. Although we have not fully explored this form of representation, we do find initial support for its existence. To what extent people’s votes also express their preferences, and to what extent this form of representation encourages equality among different groups in society, should be explored in future research.

Notes on Contributors

Yvette Peters is Postdoctoral Researcher at the Department of Comparative Politics, University of Bergen and has been on leave from the Humboldt University until recently. Her research interests include political participation, the functioning of democracy, representation and responsiveness, and the impact of (changing) political institutions on citizens’ behaviour and the quality of democracy – from a comparative perspective. [[email protected]]

Sander J. Ensink received his Master’s Degree at Leiden University, the Netherlands where he compared the quality of democracy between developed states. He then took part in the PhD programme at Trinity College Dublin (not completed), where he studied the relationship between economic inequality and democratic responsiveness. He currently works as a data analyst for an international business intelligence company in Leiden, the Netherlands. [[email protected]]

Acknowledgements

We would like to thank Tarik Abou-Chadi, Cornelius Cappelen, Raj Chari, Oscar Gabriel, Evelyne Huebscher, Eddie Hyland, Michael Tatham, Till Weber, and the two anonymous reviewers for their valuable comments, suggestions, and discussions on earlier versions. We would also like to extend our thanks to the panel participants of the panel ‘Attitudes towards representatives and representative institutions in democratic systems’ at the 7th ECPR general conference in Bordeaux, August 2013. The usual disclaimers apply.

Notes

1. In this article we focus on the comparison between the rich and the poor, although we do make reference to the situation of the middle-income group. The online appendix provides analyses comparing responsiveness of the higher- and middle-income groups.

2. The suggestion that there is little difference in the preferences of richer or poorer people within the US has been challenged by Gilens (2009).

3. See http://www.oecd.org/els/soc/socialexpendituredatabasesocx.htm

4. Although our time period covers 2002–2010, we also need information for 2011 so that the independent variables can be lagged with one year without losing cases. These data are based on estimates using projections in the OECD Economic Outlook and the European Union’s Annual macro-economic database (AMECO) (see also Adema et al. 2011).

5. Each two years as the ESS rounds are only available every other year.

6. The middle-income groups are slightly bigger (38.2 per cent based on the normal distribution) than the lower- and high-income groups (30.9 per cent each). However, due to multicollinearity problems in the analyses, this middle-income group is dropped, leaving only the lower- and higher-income groups which are similar in size.

7. We would like to thank the reviewer for this idea.

8. http://www.parlgov.org/

9. http://data.worldbank.org/indicator/NY.GDP.MKTP.KD.ZG

10. Though it needs to be noted that with so few time-points the fixed effects models are inefficient, increasing the chances of both a type I and type II error.