Affluence and Influence in a Social Democracy

Abstract

Research from the United States and Europe suggests that affluent citizens enjoy considerably more policy influence than do average citizens and the poor. I examine the extent of unequal policy responsiveness in one of the countries that have gone farthest in reducing economic inequality and restricting money in politics: Norway. I use an original dataset on public opinion and public policy containing 603 specific issues over five decades (1966–2014). The results show that although policy is certainly skewed toward the preferences of the privileged, Norway stands out among previously studied cases for two reasons: (1) The preferences of the poor seem to have some sway on economic issues and (2) not all affluent citizens get their way: educational attainment appears to be the more important determinant. The Norwegian case suggests that influence need not be as dependent upon affluence as in the United States.

“[C]ountries in Scandinavia, like Denmark, Norway, Sweden, they are very democratic countries, obviously.” —Bernie Sanders ¹

Introduction

A defining feature of democracy is that citizens, considered as political equals, have influence over policy making. In contrast to this ideal, there is mounting evidence from the United States suggesting that the rich enjoy disproportionately large policy influence at the expense of the average citizen (Bartels 2016; Erikson 2015; Gilens 2012). Research on this topic for the rest of the world remains sparse. There are some studies that have looked at differential responsiveness in Western Europe (Elsässer, Hense, and Schäfer 2017; Peters and Ensink 2015; Rosset 2016). They tend to point in the same direction as the American studies, but they are usually not directly comparable to them. If we were to measure political inequality in exactly the same way as has been done in the most prominent U.S. studies (Gilens 2012; Gilens and Page 2014), but in a social democracy that has come far in equalizing opportunities, economic differences, and limiting the role of money in politics, what would we find? Would the poor and middle class have more influence than in the US and Western Europe?

Academics disagree about the roots of political inequality and, consequently, how it can be tackled. Some argue that institutional reform such as tightening the rules on electoral campaign finance (Page and Gilens 2020), more descriptive representation in parliament (Carnes 2013), and redistributing income and wealth (Piketty 2020), should lead to a more even dispersion of political power. Others are not that optimistic. Unequal responsiveness could be a systemic problem of capitalism or, worse, a direct consequence of any distribution of economic resources that is not equal (Przeworski 2012). The debate is not new. In fact it was salient in the early twentieth century, when the socialist movement split into reformists who believed that ”the people” could successfully voice their opinions through electoral politics, and revolutionaries who believed it to be a lost cause. The reformists developed social democracy as a compromise intended to ensure workers’ control over politics, within a capitalist system (Przeworski 1986). Whether it succeeded is an open question.

In this research letter, I apply the methodology from one of the flagship studies of political inequality in the US, Affluence and Influence (Gilens 2012), to a very different case—namely Norway. Norway is a prime example of social democracy, with low levels of income inequality, strict regulations of campaign spending, strong unions, and a generous welfare state. If social democracy is capable of curtailing the disproportionate influence of the affluent, then we should see it in this case. I constructed an original dataset of Norwegian public opinion on 603 specific policy proposals at the national level from representative surveys between 1966 and 2014. Then, for each proposal, I estimated the level of support among different income percentiles and matched those data with information on which of the proposals were subsequently adopted by government.

I find that although public policy in Norway is clearly tilted toward the preferences of high-income citizens (Figure 1), the affluent do not appear to enjoy the kind of exclusive influence that characterizes the American case. First, within economic issues the preferences of both the poor and the affluent seem to matter (Figure 2). Second, the opinions of the highly educated are strongly related to policy regardless of their income (Figure 3). These results suggest a weaker link between money and politics in Norway than in the US.

Figure 1. Policy Responsiveness to High Income and Median Income Respondents When Their Preferences Diverge by More Than 10 Percentage Points

Note: Predicted probabilities for each whole percentage between 5% and 95% support, based on the logistic regressions reported in Table 2 (rows 3 and 4).

Figure 2. Unequal Responsiveness When Preferences Diverge, by Policy Area

Note: Full results reported in Appendix Table A3. Standard errors in parentheses.

Figure 3. Income versus Education, Norway and the US Compared

Note: “High” is the 90th income/education percentile; “low” is the 10th income/education percentile. Standard errors in parentheses. Full results reported in Appendix Table E1.

Case Study: Norway

Norway is commonly perceived to be a particularly well-functioning democracy, with little room for the rich to exert disproportionate political influence (EIU 2020). There are several plausible reasons for this.

First, Norway is widely considered a social democracy in the comparative welfare state literature (Esping-Andersen 1990; Powell, Yörük, and Bargu 2020). As such, it has generous and universal welfare schemes and high levels of redistribution (Hicks 1999). In combination with strong unions and centralized wage bargaining that contribute to a compressed wage structure (Allern, Aylott, and Christiansen 2007; Pontusson, Rueda, and Way 2002), the resulting level of income inequality is among the lowest in the world. ² Low inequality implies less of a resource advantage for the affluent to be used (in whichever way possible) to influence politics. Additionally, it has been argued that the universality of the social democratic welfare schemes generates legitimacy and support across classes (Rothstein 2005). That means that such policies, which are usually strongly favored by the poor, have a better chance at being maintained or even expanded by government (Brooks and Manza 2008). Thus, the process might have a self-reinforcing component by which new policy gains for the poor are more easily achieved over time.

Second, in contrast to the US, where political candidates depend on large donations from private individuals and organizations to run effective election campaigns (Ferguson 1995), parties in Norway get about two-thirds of their financing from public subsidies. ³ The country also maintains a general ban on political advertisement on television, which is the major campaign expenditure in the US (Ridout et al. 2012).

Third, Norway has historically had strong trade unions. In the literature, unions have been found to be pivotal in shaping social welfare policy in the interest of organized workers (Esping-Andersen 1990). Norwegian unions have influenced economic and social policy through pressure in the corporate channel but also through their close ties with the Norwegian Labor Party (Allern, Aylott, and Christiansen 2007). In the context of policy responsiveness, this could serve as a countervailing force to the influence of the wealthy.

Fourth, Norway’s exceptional oil and gas resources have widened the spectrum of policy options available to government by allowing extra spending, usually amounting to around 5% of mainland GDP every year (Holden 2013). Potentially, this makes it easier to respond to a diverse set of popular preferences. To illustrate, Norway has been able to maintain generous welfare transfers while imposing lower tax rates than would otherwise be necessary (Holden 2013), presumably catering to the preferences of both the poor and the wealthy. Also, during economic crises, such as the Great Recession in 2008, the government has used oil money, rather than spending cuts or tax increases, to finance stimulus packages (Mjøset and Cappelen 2011).

Finally, the country’s political class is not particularly rich. Although MP salaries surely are in the top third of the income distribution, the median wealth among MPs is in fact zero, according to tax data. ⁴ Therefore, it is unlikely that policy is biased toward the rich because politicians are rich themselves (Carnes 2013).

To be sure though, relatively small differences in political influence between rich and poor in Norway could have more indirect reasons if, for instance, there were little political inequality along—or simply an absence of—other dimensions that usually intersect with economic affluence, such as gender or ethnicity (Crenshaw 1989). Previous studies in the US and Europe have suggested that the interests of both women and ethnic minorities tend to be politically underrepresented (Costa 2017; Reher 2018; Whitby 2000). However, in Norway, policy makers have actively sought to increase the political influence of women—for example, by adopting laws requiring gender balancing in government institutions and company boards, an approach sometimes called “state feminism” (Hernes 1987; Siim and Skjeie 2008). Because women on average have lower income than men do, ⁵ these measures might have indirectly increased the influence of low-income citizens. Furthermore, for most of the period under investigation here, both ethnic fractionalization in Norway and the share of foreign-born citizens have been comparatively low (Dražanová 2020; see Appendix H). The absence of a prominent ethnic cleavage overlapping with class suggests that this type of double underrepresentation might not be as common among the Norwegian poor as for example among the poor in the US (e.g., Strolovitch 2008).

Measuring Policy Responsiveness

To examine the possibility of unequal responsiveness in Norway, I constructed an original dataset, containing Norwegian public opinion on 603 concrete policy proposals at the national level as well as information on which of these proposals were subsequently adopted by government. The dataset was constructed following the same procedure as in Gilens (2012). Policy questions posed to representative samples of the Norwegian population were extracted from preexisting surveys. ⁶ Data were available for the period 1966–2014. To be included in the dataset, questions had to ask respondents whether they support or oppose some change in national government policy. This can be anything from “should the right to abortion be extended to week 16?” to “should Norway send troops to Afghanistan?” The proposed change had to be specific enough so that it could be reliably determined whether or not the change was subsequently adopted.

I managed to obtain 431 survey items from commercial surveys and 172 from academic ones (total of 603). I then imputed, for each policy proposal, the percentage that favored the proposed change among respondents at different income percentiles, using Gilens’ approach. Finally, each proposal was coded, using government and media sources, as either adopted or not adopted in the subsequent four years (for details about the data collection and imputation method, see Appendix F).

Three studies inspired by Gilens (2012) have examined unequal policy responsiveness outside the US. All find that policy is skewed toward the preferences of the affluent. Although all provide interesting results, only one of the studies, of Germany (Elsässer, Hense, and Schäfer 2017), is directly comparable to Gilens’s results for the US. For the other two, the Netherlands (Schakel 2021) and Sweden (Persson and Gilljam 2017), according to the authors, the results are not strictly comparable because their survey data partially or fully come from academic sources, whereas Gilens (2012, 54–6) used commercial survey data (the distinction is discussed in Appendix D).

To make my data maximally comparable to Gilens’s U.S. data, I do three things: (1) I primarily analyze the sample of proposals from the commercial surveys, (2) I filter out policy proposals that would require constitutional change (n = 43), and (3) I remove policy proposals that were “half-adopted” (n = 3; Gilens 2012, 60). After this, 397 proposals are left for the analysis below. ⁷

Findings

I begin by looking at the overall relationship between public opinion and policy. ⁸ A simple bivariate logistic regression of policy outcome on policy support (Table 1, first row) suggests a moderate relationship between what the public wants and what it gets (b = 0.44, p < 0.001). ⁹ Predicted probabilities allow us to better understand the size of the effect: If 20% of Norwegians support a policy proposal, it has a predicted 17% chance of being adopted within the subsequent four years. If 80% favor it, the probability increases to 41% (i.e., it increases by a factor of 2.4). Public opinion matters, although with a status quo bias. This is quite similar to the results from the US (see Gilens 2012, 76).

Table 1. Policy Responsiveness by Income

Note: Bivariate logistic regression models (rows). The dependent variable is a dichotomous measure of whether or not the policy change was adopted within four years of the time of the survey question. *p < 0.10, ^** p < 0.05, ^*** p < 0.01.

Next, Table 1 presents results from bivariate logistic regression models where policy outcome is regressed on the policy support of five income percentiles. Although the effect sizes vary (from a factor 1.8 for the 10th percentile, to 2.3 for the 50th, to 2.9 for the 90th), all income groups have a positive, statistically significant effect on policy.

However, the preferences of the income groups are highly correlated. Consequently, the moderate responsiveness to the preferences of the middle-class and the poor that we observe in Table 1 could arise merely because these groups often want the same from government as do the affluent. In order to disentangle the preferences of the income groups and estimate their independent effects, one would normally use multivariate regression. But as Gilens (2012) points out, this will produce biased results due to the correlated measurement error of the opinion estimates for the income percentiles (Achen 1985). As an alternative, I subset the data, as has become standard in this type of study, keeping only policy proposals where preferences diverge by at least 10 percentage points across income groups (Elsässer, Hense, and Schäfer 2017; Gilens 2012; Schakel 2021). ¹⁰

Having filtered out the policy proposals where income groups more or less agree, Table 2 paints a very different picture of whose opinions matter. Although the preferences of the affluent still have a substantial, statistically significant effect on policy, there is no detectable effect for any of the other income groups. On the issues where the preferences of the affluent and the poor diverge, responsiveness to the preferences of the affluent is even greater than before: From low to high popularity among affluent respondents, the probability of policy adoption changes from 13% to 59% (factor of 4.7). In contrast, popularity among the poor only marginally affects the probability (28% vs. 33%, factor of 1.2), and the effect is statistically indistinguishable from zero (p = 0.66). When comparing the affluent and the middle class in the same way, I find the same: responsiveness to the affluent is substantial (factor of 3.9), whereas responsiveness to the middle class is weak and insignificant (factor of 1.1, p = 0.9). The magnitude of this unequal responsiveness is illustrated in Figure 1, which plots predicted probabilities of policy change at different levels of support among the middle class and the affluent.

Table 2. Policy Responsiveness When Preferences Diverge by More Than 10 Points across Income Percentiles

Note: Bivariate logistic regression models (rows). The dependent variable is a dichotomous measure of whether or not the policy change was adopted within four years of the time of the survey question. Poor = P10, Middle = P50, and Affluent = P90. *p < 0.10, ^** p < 0.05, ^*** p < 0.01.

To rule out that the results above are caused by the more or less arbitrary choice of a 10-point threshold for preference divergence, I also ran the regression models in Table 2 using multiple different thresholds. This showed a clear pattern of increasing unequal responsiveness the higher the threshold (see Appendix B). I also checked whether unequal responsiveness has increased over time and whether it is lower on highly salient issues. The results suggest that unequal responsiveness has been quite stable over time and that it does not decrease on the more salient proposals (see Appendix C).

Is there any policy area in which the preferences of the poor matter? Yes. Figure 2 shows estimates of responsiveness for the poor and the affluent, by policy area. ¹¹ Within moral issues, foreign policy and national security, and other (uncategorized) issues, the familiar pattern is observed: strong effect for the affluent, near zero effect for the poor. The economic area is somewhat different. Here as well, responsiveness to the affluent is substantial, but so is responsiveness to the poor. Given that the proposals of relative agreement have been excluded, it seems that on economic policy, the government sometimes follows the will of the affluent and other times the will of the poor, when the two are opposed.

Although my data do not include public opinion estimates by race/ethnicity, they do contain estimates by gender. Regression results (Appendix G) suggest that on economic issues, women’s preferences are better represented than are men’s (despite being less represented overall). Because women on average have lower income than men do, this could possibly account for some of the responsiveness to low-income citizens in this area.

Is the overall policy bias toward the preferences of the affluent simply explained by the fact that they are more educated than the rest? To investigate this, I imputed the policy preferences of different combinations of income and education percentile (see Appendix E). Then, I estimated policy responsiveness to income percentiles at the same level of education. The results, presented in Appendix Table E1, show that education explains some but not all of the unequal responsiveness across economic groups. Still, what is also evident from the results is that education has a stronger effect than income. For example, the preferences of respondents with high education and low income (i.e., at the 90th education percentile and 10th income percentile) have a strong effect on policy, whereas respondents with high income and low education have a weak, insignificant effect. ¹² This suggests that education is by itself enough to see ones preferences reflected in policy. Income, on the other hand, must be coupled with a certain level of education. This is interesting because it is the exact opposite from what Gilens found in the US. In the US, “high income alone seems sufficient to ensure a strong association between preferences and outcome, while education alone does not” (Gilens 2012, 95). This contrast is visualized in Figure 3, where the results from Norway are juxtaposed to the equivalent ones for the US. ¹³

Implications

The findings of this study are twofold. On the one hand, there are surprising similarities between my results for Norway and Gilens’s (2012) results for the United States. The most important is that public policy tends to be heavily skewed toward the preferences of the privileged, violating the democratic principle of political equality. However, the Norwegian case deviates from the American in two rather consequential ways, suggesting that influence is not quite as dependent on affluence as in the US.

First, the data show responsiveness to both poor and rich citizens on economic issues. It is somewhat striking that the exception lies in this particular area, given that it is here that class interests are most clearly opposed and where one usually expects the wealthy to wield the strongest influence. One plausible explanation is the focus on economic welfare and redistribution in Norway’s social democratic project as well as the powerful allies that the poor have in this domain through the unions and the left (Allern, Aylott, and Christiansen 2007). In that case, it suggests that social democracy can be expected to empower the have-nots in one of the policy areas that affects them the most as a class. It is also possible that Norway’s gender equality policies have indirectly increased the voice of the less well-off. To be sure though, Norway’s oil wealth might play a special role within the economic sphere by granting government extra leeway to satisfy the interests of both affluent and poor (Holden 2013). If this is indeed an important part of the explanation, Norway might represent a sort of upper bound of what can be expected from social democracies in terms of equality of responsiveness.

Second, education is a stronger predictor of responsiveness than income; the opposite of what Gilens (2012) found in the US. In Norway, public policy generally reflects the preferences of low-income citizens, so long as they are higher educated. A possible explanation is the fact that government members are predominantly recruited from the highly educated, which causes their descriptive overrepresentation (Bovens and Wille 2017; Lie Andersen 2014). Although this is certainly also the case in the US (Carnes 2013), there might be more room for it to have an effect on policy in Norway. The importance of political finance in U.S. elections constrains the behavior of politicians as they try to attract private funding (Ferguson 1995). The absence of a comparable mechanism in Norway would imply that other factors, such as an education-based value system, could potentially play a larger role in determining their behavior.

It is important to emphasize that the type of data employed here, and in Gilens (2012), cannot demonstrate that popular opinions are in fact causing policy change. Even though this is the usual interpretation of opinion–policy links in the literature, there is evidence of reversed causality as well: cues from political elites can shape public opinion, and such effects might be stronger and occur more quickly for the rich and highly educated (Zaller 1992), possibly explaining part of the unequal responsiveness. Although Gilens (2012, 93–6) argued that this was unlikely in the American case, it cannot be ruled out as an explanation for my results. However, it is unlikely to be the whole story. For example, it does not explain why there would be less unequal responsiveness on economic issues, as shown above.

The main conclusion from this study is that although social democracy, of the kind that exists in Norway, probably cannot be counted on to eliminate political inequality, the link between money and politics does appear to be weaker than in the American case. Public policy, at least partially, reflects the economic preferences of the poor and is robustly associated with the preferences of the highly educated, suggesting that political influence need not be reserved for the affluent.

Achieving results similar to those of Norway in other contexts could prove difficult and perhaps most difficult precisely where reforms are most needed—that is, where the affluent already hold dominant influence. Still, reducing the initial level of economic inequality, restricting how money can be used to influence elections, and strengthening the countervailing forces (such as unions), are some steps that might change the balance of power in favor of the less well-off.