Varieties of Affluence: How Political Attitudes of the Rich Are Shaped by Income or Wealth

Sociological research often uses income as the only indicator to describe or proxy the group of the rich. This article develops an alternative framework in order to describe varieties of affluence as three-dimensional: depending on income, wealth, and origin of wealth. The relevance of such a multidimensional perspective for social outcomes is demonstrated by analysing the heterogeneity in political attitudes between different varieties of affluence. For this purpose, ordinary least squares regressions are applied to a sample from 2005, 2009, and 2014 German Socio-Economic Panel (GSOEP). The main results are, first, that the perspective of varieties of affluence reveals significant differences in social outcomes as demonstrated by political attitudes. Especially wealth possession is related to significantly more right political attitudes. Second, there is strong explorative evidence that the rich in Germany should be regarded as a heterogeneous group. These findings are robust to influential data, multiple imputations of wealth data, and endogeneity due to pooled data. The article concludes, among other things, that more data are required to make more certain assertions. Who Are ‘the Rich’ and Why Does It Matter? In recent years, there has been a shift in research on economic inequality from a primary interest in income to a focus on wealth. French economist Thomas Piketty famously predicted that wealth and its concentration might generally become increasingly important because the forecasted low economic growth might lead to decreasing chances of significant wealth accumulation through labour (Piketty, 2014; Piketty and Saez, 2014). His claim also adds relevance to the multigenerational persistence of wealth. For Germany, a recent study projected a yearly inheritance volume of up to EUR 400 billion per year until 2024—equivalent to 12.7 per cent of the GDP in 2016 (Tiefensee and Grabka, 2017). The share of pre-tax income received by the top 10 per cent has grown since the 1970s and in 2008 was at its highest level since 1917 (WID.world, 2017). In 2010, the share of wealth owned by the top 10 per cent of the wealth distribution in Germany was the highest in the Eurozone (Bundesbank, 2013: p. 30). These numbers are a reason to devote attention to the top income and wealth holders. The purpose of this article is to demonstrate to what extent social phenomena can be explained by variation among three forms of affluence: position in the income distribution, position in the wealth distribution, and origin of wealth—self-earned versus not self-earned. Reasons to make such distinctions are numerous: possible social-structural differences between these groups include ambition, family background, ability, migration background, social class, saving behaviour, and age. VC The Author(s) 2019. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. European Sociological Review, 2020, Vol. 36, No. 1, 136–158 doi: 10.1093/esr/jcz051 Advance Access Publication Date: 11 October 2019 Original Article D ow naded rom http/academ ic.p.com /esr/article-at/36/1/136/5585947 by M ax-Pck-Institut fer G esellsctsforschung user on 21 Feruary 2020 The question in how far varieties of affluence only signal such group differences or exert independent direct effects needs to be answered empirically. Political attitudes are chosen here as one example for differing outcomes because, first, they can be expected to be related in many ways to individual economic circumstances. Second, insights into political attitudes among those with different varieties of affluence might add to existing research from other disciplines, such as a democratic responsiveness bias as an indicator of political inequality (Gilens, 2012: pp. 70–96; Grimes and Esaiasson, 2014; Bartels, 2016: pp. 233–268; Elsässer, Hense and Schäfer, 2016, 2017). The increased interest in wealth and the traditional conceptualization of the affluent as those with high incomes raise the question whether ‘the affluent’ can be seen as a homogenous group in terms of their political attitudes. Is it sufficient to define them as only those with high incomes? To shed light on this question, heterogeneity among the affluent is analysed by asking two related questions: Do varieties of affluence matter? Are the rich in Germany a homogenous group in terms of their political attitudes? The analysis is based on survey data from the German Socio-Economic Panel (GSOEP), mainly for the years 2005, 2009, and 2014. While the available data are not the ideal source with which to answer the questions posed, it is the best available so far, providing initial explorative insights into social outcomes among varieties of affluence in Germany. The article is structured as follows. Varieties of Affluence section reviews previous research on the affluent in Germany, including how they are defined and assessed and develops a differentiated framework to account for varieties of affluence as a foundation for the empirical examination. Affluence and political attitudes section again consults existing literature in order to derive hypotheses regarding political attitudes among different types of affluence. Methodology section presents the data and methodology used to test the framework, the results of which are given in Results section. Finally, Discussion section discusses the results, followed by a brief conclusion. Varieties of Affluence The affluent has been analysed in sociology at least seminal since the works of Veblen (1899) and Simmel (1900). Nevertheless, a general lack of research is constantly noted and criticized (e.g. Imbusch, 2003; Lauterbach and Ströing, 2008; Page, Bartels and Seawright, 2013). In German sociology, relevant results mainly come from the intersection of two streams of literature: the traditional sociology of elites (e.g. Hradil and Imbusch, 2003; Hartmann, 2013) and a new sociology of wealth (e.g. Druyen, Lauterbach and Grundmann, 2008; Böwing-Schmalenbrock, 2012; Spannagel, 2013; Waitkus and Groh-Samberg, 2018). Lauterbach and Ströing (2008) provide a systematic definition of the affluent as a synthesis of the heterogeneous existing international literature (Figure 1). The concept starts off with the assumption of a diversified stratification of income affluence and, in the next step, differentiates it further by including wealth. According to this definition, being wealthy in terms of income starts at an income level of at least twice the mean or median. The relevance of a combined perspective of income and wealth—and sometimes other dimensions—is suggested in several recent studies (e.g. Becker, 2003; Grabka et al., 2007; Druyen, Lauterbach and Grundmann, 2008; Lauterbach, Druyen and Grundmann, 2011; Peichl and Pestel, 2011; BöwingSchmalenbrock, 2012; Rowlingson and McKay, 2012; Spannagel, 2013; Skopek, 2015; Keister and Lee, 2017; Killewald, Pfeffer and Schachner, 2017). To add to these contributions, a framework is developed in the following that includes the origin of wealth in order to analyse how individuals with different varieties of affluence vary in terms of social outcomes. Being ‘rich’ can more accurately be understood as heterogeneous combinations of different forms of affluence. Being rich shall be defined as based on three dimensions of economic resources: income, wealth, and the origin of wealth. Income can be defined as a flow of economic resources that may be received from various sources such as labour, return on capital, or government transfers. Another less-studied and rarer type of income is windfall income obtained through exogenous sources such as inheritances or lottery wins. Both of these processes are not random. There is a good reason to believe that those who are better off are more likely to receive inheritances (e.g. Hansen, 2014; Bönke, Corneo and Westermeier, 2015: pp. 11–13). On the other hand, there is evidence that lottery participation is skewed towards lower-income groups and the working class (Beckert and Lutter, 2009, 2012). The reason windfall income and inheritances are discussed combined is that they are not distinguishable in the GSOEP over time. The amount that can be spent or saved regularly grows with increasing income. Therefore, one could expect high correlations between income and wealth, but, especially in (East) Germany, they are far from perfect (Peichl and Pestel, 2011). Killewald, Pfeffer and Schachner (2017: pp. 388–390) also show this for the European Sociological Review, 2020, Vol. 36, No. 1 137 D ow naded rom http/academ ic.p.com /esr/article-at/36/1/136/5585947 by M ax-Pck-Institut fer G esellsctsforschung user on 21 Feruary 2020 United States and find that wealth is also not completely related to income in the longer run. The amount of disposable income is highly dependent on household constellation. Therefore, income is usually measured as the household equivalent income: the overall household income standardized in relation to the number of adults and children living in it according to the new OECD equivalence scale (Hagenaars, Vos and Zaidi, 1996). The second dimension of affluence is wealth that can be defined as an accumulated stock of different assets such as real estate, valuable objects, or financial assets. Wealth is built up from the different sources of income—depending on individual behaviour such as consumption and saving. Analyses of income and wealth distributions show that wealth is significantly less equally distributed than income (Frick, Grabka and Hauser, 2010: pp. 122–124). Depending on the way it is used, wealth can generate various kinds of income, e.g. dividends from stocks, rent from properties, or increasing values of valuable objects such as art.

Who Are 'the Rich' and Why Does It Matter?
In recent years, there has been a shift in research on economic inequality from a primary interest in income to a focus on wealth. French economist Thomas Piketty famously predicted that wealth and its concentration might generally become increasingly important because the forecasted low economic growth might lead to decreasing chances of significant wealth accumulation through labour 1 (Piketty, 2014;Piketty and Saez, 2014). His claim also adds relevance to the multigenerational persistence of wealth. For Germany, a recent study projected a yearly inheritance volume of up to EUR 400 billion per year until 2024-equivalent to 12.7 per cent of the GDP in 2016 (Tiefensee and Grabka, 2017). The share of pre-tax income received by the top 10 per cent has grown since the 1970s and in 2008 was at its highest level since 1917 (WID.world, 2017). In 2010, the share of wealth owned by the top 10 per cent of the wealth distribution in Germany was the highest in the Eurozone (Bundesbank, 2013: p. 30). These numbers are a reason to devote attention to the top income and wealth holders.
The purpose of this article is to demonstrate to what extent social phenomena can be explained by variation among three forms of affluence: position in the income distribution, position in the wealth distribution, and origin of wealth-self-earned versus not self-earned. Reasons to make such distinctions are numerous: possible social-structural differences between these groups include ambition, family background, ability, migration background, social class, saving behaviour, and age.
The question in how far varieties of affluence only signal such group differences or exert independent direct effects needs to be answered empirically. 2 Political attitudes are chosen here as one example for differing outcomes because, first, they can be expected to be related in many ways to individual economic circumstances. Second, insights into political attitudes among those with different varieties of affluence might add to existing research from other disciplines, such as a democratic responsiveness bias as an indicator of political inequality (Gilens, 2012: pp. 70-96;Grimes and Esaiasson, 2014;Bartels, 2016: pp. 233-268;Elsässer, Hense andSchä fer, 2016, 2017).
The increased interest in wealth and the traditional conceptualization of the affluent as those with high incomes raise the question whether 'the affluent' can be seen as a homogenous group in terms of their political attitudes. Is it sufficient to define them as only those with high incomes? To shed light on this question, heterogeneity among the affluent is analysed by asking two related questions: Do varieties of affluence matter? Are the rich in Germany a homogenous group in terms of their political attitudes? The analysis is based on survey data from the German Socio-Economic Panel (GSOEP), mainly for the years 2005, 2009, and 2014. While the available data are not the ideal source with which to answer the questions posed, it is the best available so far, providing initial explorative insights into social outcomes among varieties of affluence in Germany.
The article is structured as follows. Varieties of Affluence section reviews previous research on the affluent in Germany, including how they are defined and assessed and develops a differentiated framework to account for varieties of affluence as a foundation for the empirical examination. Affluence and political attitudes section again consults existing literature in order to derive hypotheses regarding political attitudes among different types of affluence. Methodology section presents the data and methodology used to test the framework, the results of which are given in Results section. Finally, Discussion section discusses the results, followed by a brief conclusion.

Varieties of Affluence
The affluent 3 has been analysed in sociology at least seminal since the works of Veblen (1899) and Simmel (1900). Nevertheless, a general lack of research is constantly noted and criticized (e.g. Imbusch, 2003;Lauterbach and Strö ing, 2008;Page, Bartels and Seawright, 2013). In German sociology, relevant results mainly come from the intersection of two streams of literature: the traditional sociology of elites (e.g. Hradil and Imbusch, 2003;Hartmann, 2013) and a new sociology of wealth (e.g. Druyen, Lauterbach and Grundmann, 2008;Bö wing-Schmalenbrock, 2012;Spannagel, 2013;Waitkus and Groh-Samberg, 2018). Lauterbach and Strö ing (2008) provide a systematic definition of the affluent as a synthesis of the heterogeneous existing international literature (Figure 1). The concept starts off with the assumption of a diversified stratification of income affluence and, in the next step, differentiates it further by including wealth. According to this definition, being wealthy in terms of income starts at an income level of at least twice the mean or median.
The relevance of a combined perspective of income and wealth-and sometimes other dimensions-is suggested in several recent studies (e.g. Becker, 2003;Grabka et al., 2007;Druyen, Lauterbach and Grundmann, 2008;Lauterbach, Druyen and Grundmann, 2011;Peichl and Pestel, 2011;Bö wing-Schmalenbrock, 2012;Rowlingson and McKay, 2012;Spannagel, 2013;Skopek, 2015;Keister and Lee, 2017;Killewald, Pfeffer and Schachner, 2017). To add to these contributions, a framework is developed in the following that includes the origin of wealth in order to analyse how individuals with different varieties of affluence vary in terms of social outcomes.
Being 'rich' can more accurately be understood as heterogeneous combinations of different forms of affluence. Being rich shall be defined as based on three dimensions of economic resources: income, wealth, and the origin of wealth. Income can be defined as a flow of economic resources that may be received from various sources such as labour, return on capital, or government transfers. Another less-studied and rarer type of income is windfall income obtained through exogenous sources such as inheritances or lottery wins. Both of these processes are not random. There is a good reason to believe that those who are better off are more likely to receive inheritances (e.g. Hansen, 2014;Bö nke, Corneo and Westermeier, 2015: pp. 11-13). On the other hand, there is evidence that lottery participation is skewed towards lower-income groups and the working class Lutter, 2009, 2012). The reason windfall income and inheritances are discussed combined is that they are not distinguishable in the GSOEP over time.
The amount that can be spent or saved regularly grows with increasing income. Therefore, one could expect high correlations between income and wealth, but, especially in (East) Germany, they are far from perfect (Peichl and Pestel, 2011). Killewald, Pfeffer and Schachner (2017: pp. 388-390) also show this for the United States and find that wealth is also not completely related to income in the longer run. The amount of disposable income is highly dependent on household constellation. Therefore, income is usually measured as the household equivalent income: the overall household income standardized in relation to the number of adults and children living in it according to the new OECD equivalence scale (Hagenaars, Vos and Zaidi, 1996).
The second dimension of affluence is wealth that can be defined as an accumulated stock of different assets such as real estate, valuable objects, or financial assets. 4 Wealth is built up from the different sources of income-depending on individual behaviour such as consumption and saving. Analyses of income and wealth distributions show that wealth is significantly less equally distributed than income (Frick, Grabka and Hauser, 2010: pp. 122-124). Depending on the way it is used, wealth can generate various kinds of income, e.g. dividends from stocks, rent from properties, or increasing values of valuable objects such as art. Wealth can also be directly transmitted through gifts or inheritance, whereas income usually cannot.
Finally, the third dimension is the origin of wealth. The main underlying idea is whether the accumulated wealth was mainly self-made or inherited. In economics, the problem of how to define the share of self-earned and inherited wealth has long been a matter of discussion (Modigliani, 1986(Modigliani, , 1988Hansen, 2014;Piketty, Postel-Vinay and Rosenthal, 2014;Bö nke, Corneo and Westermeier, 2015). Modigliani (1986) suggested a straightforward way to implement the share inherited in terms of data requirements. The amount of wealth ever received from inheritances is simply divided by current net overall wealth. 5 This way, the source of wealth can be analysed as a distinct dimension of wealth.
Based on these three dimensions, an overall framework of varieties of affluence can be constructed. In what follows, the question of the extent to which these varieties explain variance in political attitudes will be examined. For this purpose, the next section derives hypotheses on political attitudes among different types of affluence.

Affluence and Political Attitudes
In the article 'The Political Attitudes of Wealth ' published in 1945' published in , Almond (1945) already criticized a simplified generalization in science of those with high incomes. Today, studies still most often concentrate on only one or very few dimensions of economic affluence. For this reason, elements from heterogeneous sources must be collected to derive hypotheses for groups possessing different varieties of affluence. In what follows, empirical results for political attitudes, interests, partisanship and similar are collected, which could help us to understand political attitudes of the rich. 6 The focus is on results that might help to derive hypotheses on the main dependent variable used by the GSOEP: a left-right scale (LRS) of political attitudes. 7 Political science suggests that high income should be related to more right-wing attitudes and less desire for redistribution (Lipset, 1960: pp. 223-229). When argued in this way, more income and wealth should lead to more right-wing views, because people should be opposed to redistributive policies such as higher income taxes, wealth taxes, or inheritance taxes (Romer, 1974;Meltzer and Richard, 1981). This assumption is supported by empirical research which shows a relationship between left-right self-placement and attitudes towards redistribution (Jaeger, 2008). Despite a relevant degree of noise in this relationship, it still seems plausible to assume that both, income and wealth, should be related to more right-wing views on average.
However, another relevant question is whether, how, and why the relationship of wealth on political attitudes should differ from that of income. There are different ways in which wealth could affect political attitudes independent of income. On the one hand, it could be a spurious relationship reflecting that the same characteristics, experiences, or attitudes which lead to more wealth accumulation, conditional on receiving high income, are also related to political attitudes. 8 On the other hand, there could be a direct effect of self-earned wealth which might lead to a stronger disapproval of redistribution simply because more own, and self-earned stakes are at risk. As Killewald, Pfeffer and Schachner (2017: p. 380) put it: 'Whereas income measures the flow of financial resources at a particular time, wealth is a cumulative stock that reflects years of prior circumstances and decisions' (cf. also the discussion in Skopek, Buchholz and Blossfeld, 2014: pp. 466-470). However, the scope of this article is not to test such mechanisms. Instead, the focus here is on one step before that: first, to examine how different varieties of affluence are related to political attitudes in the first place. Second, whether heterogeneous groups among the rich can be identified empirically in terms of this outcome, and whether one of the dimensions of affluence alone is sufficient to capture one (apparent) group of 'the rich'.
For inheritances and windfall income (IWI), theoretical explanations are not easy to give because they cannot be distinguished in this work from other types of income. The presented results must, therefore, be seen as explorative and further research is needed to understand how this dimension affects political attitudes. In general, theoretical predictions of theories of self-interest are supported by previous empirical studies. Table 1 summarizes the relevant literature.
Studies analysing political attitudes related to wealth show strong support for more right-wing views-although relevant studies are only a few. For Germany, the support for the CDU is strongly related to higher positions in the distribution of net overall wealth (Bach and Grabka, 2013). Analysing views on different policies, Page, Bartels and Seawright (2013) report more conservative views of the wealthy in the United States. They also find evidence that the relationship might be very strong for extremely large wealth. Therefore, the following hypotheses are drawn: H1: The possession of wealth is related to an identification as right-wing.
H1a: The possession of more wealth is associated with more right-wing attitudes among the non-rich.
H1b: The possession of more wealth is associated with more right-wing attitudes among the rich.
H1c: Those with very large wealth are more to the right than other groups among the rich.
The picture for income is a bit more ambivalent. Bach and Grabka (2013) report that support for the CDU and FDP increases with increasing household equivalent income. The relationship is weaker compared with wealth, however. Different studies for the United States found that those with high incomes are more conservative in economic terms but more liberal in social terms; that is, in topics such as gay rights or abortion (Gilens, 2009;Page and Hennessy, 2010;Gilens, 2012). Others find only slight differences (Ura and Ellis, 2008), or even none at all (Soroka and Wlezien, 2008). The geographical region also seems to matter and might mediate the opinions of the wealthy (Flavin, 2011). One could, therefore, expect that high income is related to more 'right-wing' political attitudes at least in terms of economic and redistribution topics. Because of the reported socially liberal orientation of highincome earners, the effect is expected to be smaller than that of wealth. This is also in line with recent findings by Piketty (2018) for the United States, France, and United Kingdom. The following hypotheses shall be tested: H2: High income is related to an identification as right-wing.
H2a: High income is related to an identification as right-wing for the non-rich.
H2b: High income is related to an identification as right-wing among the rich.
H2c: The effect of income on being right-wing is weaker than the effect of wealth.
European Sociological Review, 2020, Vol. 36, No. 1 As another frequently mentioned group among the rich, an 'economic elite' is of interest to a long tradition of elite sociology. In traditional elite studies, wealth and economic resources are often related to a group of top executive managers as wealthy group, and there is a general interest and previous research on their political attitudes (e.g. Galonska, 2012;Allmendinger et al., 2013;Hartmann, 2013;Hecht, 2017). Therefore, an economic elite is included in the analysis. Results from the sociology of elites can be drawn on here when assuming that the 'economic elite' are top-managers with especially high incomes. They could be conceived as a special subgroup of those with high incomes: high-performing, topincome earners with high-responsibility positions. Their political attitudes might be distinct to other high-income earners or wealth holders. Galonska (2012) finds strong support for the CDU and FDP among this group and far right-wing political self-classification in comparison to other groups. Hartmann (2013) shows findings from elite interviews in which economic elites present arguably conservative views on social inequality. When looking at the LRS of the underlying data (Allmendinger et al., 2013;cf . Table 1), a self-classification of 5.9 on the LRS is found among this group. In sum, elite groups can be expected to deviate to the right from other groups possessing varieties of affluence: H3: Being a member of the economic elite is related to an identification as right-wing among the rich.
The last relevant dimension is origin of wealth, or IWI. For Germany, a study on behalf of Deutsche Bank interviewed different heirs and future bequeathers as part of an overall representative sample (Blumenthal and Hö rter, 2015). Those who had already inherited agreed most with opinions paraphrasing that inheritances above EUR 1 million should be taxed higher and that current inheritance law reproduces social inequality. One could interpret these findings as an indicator for more pro-redistributive and, therefore, left opinions among this group.
The literature analysing the effect of lottery wins in the US and the UK points in a different direction. There is some evidence that the amount won in a lottery is related to more conservative views and partisanship in the United Kingdom and support for the Republican party in the United States (Powdthavee and Oswald, 2014;Peterson, 2016). Doherty, Gerber and Green (2006) find no effect of winning the lottery on opinions on social inequality, but opinions on inheritance taxes change after winning. These results again give evidence that inheritance and other windfall income might not be easily comparable. Not only that the groups of heirs and lottery winners are structurally very different but also the direct effects of receiving an inheritance or a lottery win might be incomparable. For example, inheritances are often planned by bequeathers and expected by prospective heirs which might influence attitudes, e.g. towards inheritance taxes before and after inheriting. However-over time-the two sources of income are not distinguished in the GSOEP (cf. Methodology section). To account for this, three different hypotheses are derived here. First, the absolute amount of windfall income is assumed to be related to more conservative views, following the lottery studies.
H4a: Having received larger amounts of windfall income is related to identifying as right-wing among the non-rich.
H4b: Having received larger amounts of windfall income is related to identifying as right-wing among the rich.
The second follows the indication from the Deutsche Bank inheritance survey: H4c: Those who received any amount of inheritances or other windfall income are more left compared to the overall average.
These hypotheses, derived from various pieces of heterogeneous extant literature, will be analysed mainly using three waves of the GSOEP in the following.

Methodology
This section presents the data and variables, the sample, and the research design. The analysis is based on the GSOEP, an annual panel assessment with more than 12,000 German households and 20,000 respondents. Three features of the GSOEP concerning economic affluence are especially advantageous for this work. First, in addition to income, wealth is also assessed in some years (Frick, Grabka and Marcus, 2013). Second, high-income households were oversampled from 2002 onwards, which increases statistical power for this group . Finally, GSOEP assesses inheritances in 2001 (Rosenblatt, 2002: p. 10) and windfall income after that. On top of the detailed assessment of economic affluence, there are also relevant questions on political views. Among others, the LRS was assessed in some years, which is used as a dependent variable here.
It is important to mention that, as common in quantitative survey research, measurement error is also a problem in the GSOEP. Specifically, the most important independent variables of income and wealth are a matter of this problem. Respondents may not answer, or misreport, e.g. due to matters of cognitive limitations, privacy concerns or social desirability (Schräpler, 2002). Because of these problems, a careful inquiry as well as elaborate editing and imputation was performed and validation with corresponding information from national balance sheets gives confidence in the data (e.g. Frick, Grabka and Hauser, 2010;Grabka, Marcus and Sierminska, 2013: pp. 4-5;Grabka and Westermeier, 2015). There is some evidence indicating that measurement error in the GSOEP might be higher for income than for wealth, possibly because it is considered as more private than wealth as studied and discussed in Riphahn and Serfling (2005: pp. 530-534). Measurement error among the independent variables might attenuate regression coefficients which should be kept in mind for the analysis (Fox, 2011: pp. 112-115).

Operationalization and Sample
The classical LRS is assessed as an 11-point scale.
Respondents are asked to position themselves from 0 (far left) to 10 (far right) (TNS Infratest Sozialforschung, 2005: p. 32). Despite criticism of this instrument (Weber and Saris, 2015;Bauer et al., 2017;Otjes, 2018), it can be useful to make international comparisons (e.g. Giger, Rosset and Bernauer, 2012) and offers a well-known one-dimensional measure of political attitudes. Varieties of affluence are relevant in other national contexts as well. It, therefore, makes sense not to rely on a measure such as partisanship that is presumably even more dependent on the national context.  (Grabka, 2014: p. 41) and weighted with the OECD equivalence scale (as proposed by Hagenaars, Vos and Zaidi, 1996; cf. Varieties of Affluence section). 9 The measure is standardized for the respective year. 10 Measures for level of income richness correspond to double and triple the sample mean. The double mean income in 2014, for example, is EUR 53,252 per year. The central measure chosen for wealth is net overall household wealth. This measure is the sum of all kinds of wealth in a household net of debts. In contrast to income, wealth is not standardized per capita. 11 Wealth is assessed as self-estimated market value of different assets such as real estate, valuables, or financial assets. Cars and tangible and financial assets of less than EUR 2,500, and consumer debts of less than EUR 2,500 are excluded (Frick, Grabka and Sierminska, 2007: pp. 2-3). Assets are totalled and all debts subtracted to calculate net overall wealth. The value is calculated as the mean of the provided multiple imputations for every case (Frick, Grabka and Marcus, 2013). Wealth is used standardized to the mean for each year. Level of wealthiness is introduced as a categorical variable distinguishing between comparable categories suggested in the literature: the affluent with wealth of more than EUR 500,000, and HNWI with wealth of more than EUR 1,000,000. 12 There are few cases involving more than EUR 30,000,000 of wealth in the SOEP, therefore, only the first two categories are used and the HNWI category is top-coded.
Finally, the origin of wealth is measured by referring to IWI, which includes inheritances, gifts, and other windfall income such as lottery wins. The reason for this combined perspective is solely data restriction (cf. Varieties of Affluence section). Two measures are used for IWI. The first is the standardized absolute amount of IWI received. It is composed of two items: inheritances until 2001 and yearly windfall income received after that. Although types of inheritances were assessed, the exact values per type are unknown. This makes it hard to capitalize them in relation to the real value in the year they were received. Therefore, the total inherited value is used without accounting for value changes, for example, through inflation or capital returns. 13 From 2001 onwards, all windfall income is assessed annually. Both amounts are added up until the year of analysis. Like the other wealth and income measures, also IWI is standardized for the respective year. The second measure is categories of the absolute amount of IWI received.
In analogy to the wealth categories, thresholds are below and above 500,000e of IWI.
The economic elite is defined as follows: Those who are employees with comprehensive executive tasks are taken as a basis. From this group, those who are additionally in the top 1 per cent of individual labour income in each year are coded as economic elite. As an example, the top 1 per cent of labour income in the sample begins at EUR 112,713 in 2014. Only 12 per cent of this subgroup is female.
Relevant confounding variables are chosen informed by the literature on political attitudes and partisanship in Germany (e.g. Zuckerman and Kroh, 2006). The main interest is in the direct relationship between varieties of affluence and political attitudes, net of other relevant influences. Therefore, other confounding variables relevant for political attitudes are controlled for. A variation of the standard Erikson -Goldthorpe -Portocarero (EGP) class scheme (Erikson and Goldthorpe, 1993) is used in its four-class version (Breen, 2009) and a category for the unemployed is added. 14 Education is based on the 1997 International Standard Classification of Education (ISCED) scheme, aggregated to low, medium, and high education (Eurostat, 2011). Religion was assessed in the years 2003, 2007, 2011, and 2015. Confession for the years of analysis is taken from the closest assessments. Migration background is an indicator that includes both direct and indirect migration background. Other control variables are interested in politics, age, age 2 , self-employment, federal state, gender, marriage status, and wave as a time dummy. 15 The analyses are based on two pooled unbalanced samples with a total of 47,978 cases from the GSOEP, covering the years 2005, 2009, and 2014. The composition of the samples is illustrated in Table 2. The analyses are, therefore, conditional on being rich or not being rich. A joint analysis is presented in the Supplementary material.
The dependent and all independent variables and their descriptive summary statistics for the sample are displayed in Table 3. The descriptive statistics show that some of the relevant categories are small. This is especially the case for large IWI, and to a lesser extent for wealth. This needs to be kept in mind when evaluating the results. Correlations displayed in Table 4 demonstrate that interpretation of the continuous variables should not be affected by strong multicollinearity, and the correlation between income and wealth is also relatively low.

Research Design
The research strategy is made up of two steps. The first is to describe how different dimensions of affluence are European Sociological Review, 2020, Vol. 36, No. 1 related to political attitudes in the two samples: the nonrich and the rich. Second, heterogeneity of political attitudes among those who can be considered 'rich' in terms of income or wealth will be analysed. To analyse heterogeneity, ordinary least squares (OLS) regression models with robust standard errors are used to estimate differences between the dimensions of affluence for the whole sample, while controlling for confounding factors. For wealth, the mean of the multiple imputed wealth values provided by the DIW is used, without adjusting models for multiple imputations in the main analysis but as a robustness check in the Supplementary material.

Results
The hypotheses drawn in Affluence and political attitudes section are tested with the results of the OLS (Table 5). To avoid multicollinearity, continuous and categorical independent variables are analysed in different models. Results for the non-rich sample (M1-M6) are discussed first, and for the rich (M7-M13) second. Killewald, Pfeffer and Schachner (2017: p. 382) suggest that the consequences of wealth may vary across its distribution. To account for this, interaction terms were included for wealth, income, and between to the two, to describe the relationships in the samples as exhaustive as possible. Marginal effect plots are presented as suggested by Berry, Golder and Milton (2012) to allow their appropriate interpretation.
Among the clearest results is support for Hypotheses H1a. Wealth seems to be an important predictor for more right self-classification in both samples. This result persists with or without controlling for confounding variables. Model M2 suggests that the possession of one standard deviation (SD) additional wealth is related to an increase of self-classification on the LRS of 0.62 for the non-rich.
However, there is evidence for non-linearity of the effect of wealth in both samples. For the non-rich, the conditional effect of wealth on wealth increases (Figure 2b). That is, the relationship becomes even stronger for high levels of wealth. In contrast to this, the effect of wealth seems to decrease with increased wealth in the rich sample (Figure 3b). For both samples, the coefficient of wealth is estimated to be larger than zero with 95 per cent for a wide range of observed wealth values. There is one exception, though, in both cases: the effect of wealth conditional on income is not distinguishable from zero anymore if income is far above the mean. That is, if income is ca. 1 SD above the mean for the non-rich (Figure 2c), or ca. 15 for the rich (Figure 3c), wealth does not show a significant effect anymore. However, this finding could also be due to the low number of cases with very large income or wealth. Overall, wealth is strongly related to right-wing self-classification but the relationship is non-linear and is conditional on being rich or not, and on the position in the wealth and income distribution.
For income, results are more ambivalent. For the non-rich sample, the effect of income without controlling for confounding variables seems to be related to slightly, though significantly, more left-wing self-classification. However, controlling for confounding variables, neither a significant unconditional effect of income can be found (M2), nor significant conditional effects for any observed position in the income distribution ( Figure 2a). Therefore, hypothesis H2a cannot be supported because income alone does not seem to have an effect for the non-rich. For the rich sample this is different. First of all, income seems to be related to right-wing self-classification (M2). There is evidence for a decreasing marginal effect for higher levels of income ( Figure 3a). Further support is given by the finding that those with at least 300 per cent mean income are significantly more right than those who are not income rich. Hypothesis H2b, therefore, also finds support-though only from 300 per cent of the mean upwards.
When it comes to IWI, results are somewhat ambivalent again. The first finding is that, for the non-rich, there seems to be a significant negative effect of the absolute amount of IWI received. But, on closer inspection, this can easily be challenged. First of all, there is no such effect for the rich. Second, when excluding influential cases (Supplementary material A4) this effect does not persist and it can, furthermore, only be reproduced for two of the three waves (Supplementary material A5). Hypothesis H4a must, therefore, be rejected. There is some indication, however, that receiving windfall income or inheritances is related to more left selfclassification. Compared with those who did not receive any IWI, those who received between zero and EUR 500,000 IWI are slightly less conservative. This result seems more robust, at least in terms of influential data, and supports hypothesis H4c. Taking into account the discussed problems of the measure, there is some indication that IWI might matter but better data are necessary to make more robust claims.
The second step of the analysis is to look closer for heterogeneity among the rich only: those with either an income of at least 200 per cent of mean income or at least EUR 500,000 of wealth. For this group as well, 1 SD of increase in wealth is related to a significant increase in right-wing self-classification in all models, M7 to M12. Again, the effect is conditional on the position in the wealth and income distribution. When income is equal to the mean, the effect of wealth is estimated to be larger than zero for the whole observed range (Figure 3b). However, with increased income above the mean, the conditional effect of wealth decreases to zero and even significantly negative-although the number of cases is too low in the range of the negative effect to make any robust claims (Figure 3c). Wealth still seems to be among the most important and persistent predictors for right LRS identification and H1b can, therefore, be supported. Though, the effect is more equal to that of income for the rich, and its strength also depends on income. Compared with those with only high incomes, respondents with wealth of more than EUR 500,000 and EUR 1,000,000 seem to be significantly more conservative than the average rich person in the sample, and hypothesis H1c thus finds good support.
Total income shows a significant effect in all models for the rich. In addition, those with at least 300 per cent of the mean are significantly more conservative than those with income of 200 per cent of the mean or less. This result contributes to the image that among the rich, those with only 200 per cent of mean income or less are the least conservative group. Other evidence suggests that higher income of at least 300 per cent of the mean is associated with more conservative views. Therefore, hypothesis H2b can be supported.

146
European Sociological Review, 2020, Vol. 36, No. 1 Downloaded from https://academic.oup.com/esr/article-abstract/36/1/136/5585947 by Max-Planck-Institut fuer Gesellschaftsforschung user on 21 February 2020 The rich The rich The rich The rich The rich The rich  is very high, wealth does not show a significant effect anymore. Yet, the non-continuous measures indicate that wealth might overall be more strongly correlated to right self-placement. H2c, therefore, finds support especially for the non-rich, where wealth is clearly more important than income. For the rich, coefficients of wealth seem more equal to those of income-but together with the evidence of the high wealth categories, H2c finds support. This finding is also discussed further in the next section. The effect of IWI shows a similar pattern for the rich as for the non-rich. Receiving IWI at all seems to make a difference, whereas its amount does not. The results for IWI, therefore, only allow to make one claim which is that the origin of wealth seems to matter. However, the directions, mechanisms, and strengths of effects need to be analysed with more elaborated data sources on this matter. Especially a differentiation between inheritances and other windfall income would enable to understand better whether only one of the two or both matter and in what way exactly.
Finally, the economic elite was tested here as a potential distinct group among the rich. Results suggest that this group is far more conservative than the average among the rich, and hypothesis H3 finds support for this sample. In fact, coefficients are comparably large as those for being wealthy with at least EUR 500,000.
All of the presented results, except when stated otherwise, are robust to the exclusion of influential data, separate analysis for the individual waves, as well as adjusted models for multiply imputed wealth data, as discussed and presented in the Supplementary material. 16 The next section discusses these results and puts them into context. European Sociological Review, 2020, Vol. 36, No. 1 Discussion Two related questions are discussed in this section in light of the results: Do varieties of affluence matter for the explanation of social outcomes? Are the rich homogenous in terms of their political attitudes?
The first question can be answered with a definitive yes, at least for political attitudes. This is not necessarily a surprising outcome and the only reason this differentiation has not been made so far is probably due to lack of relevant data. However, in the literature, discussions too often include implicit or explicit generalizations from income richness to wealthiness. Varieties of affluence can make a difference for social outcomes, as demonstrated here for political attitudes. It is especially remarkable that income, the dimension most often referred to, seems to have the smallest impact on political attitudes. This is even more striking when compared with the impact of wealth. Results suggest that the perspective of varieties of affluence could also add to other fields of sociology, such as the study of context effects (e.g. Brooks-Gunn et al., 1993), intergenerational transmission of advantage (e.g. Pfeffer, 2011;Pfeffer and Schoeni, 2016), health inequality (e.g. Pollack et al., 2013), or the study of political attitudes.  Are the rich homogenous in terms of their political attitudes? When looking at the data available for Germany today, a tendency can be identified, but more data are necessary to understand what is going on in the top categories. Figure 4 shows comparisons between those who are not rich and those with different combinations of wealth and income. Most of the variations on the 0-10 point LRS occur between 4.5 and 6, but there are some significant differences. Most of the relevant variation is taking place between owning wealth or not. It also seems that receiving more than 300 per cent of mean income adds to the effect of being wealthy, though not perfectly. A second finding is that, according to the analysed data, no justifiable assertion can be made about those who own more than EUR 1 million of wealth. The large level of uncertainty within this group might well be due to the relatively small sample sizes (cf .  Table 3), or it might indicate more heterogeneous views. However, looking at top wealth holders in the sample may give a preview of the potential rewards of collecting more data on those with very high wealth.
The pattern in Figure 5 is in line with similar findings that Page, Bartels and Seawright (2013: pp. 64-65) report for the United States: there is some evidence that very immense wealth might be related to far more conservative views. As in the US study, sample sizes here are very small. The GSOEP data comprise only 17 households with more than EUR 5 million. Wealth records of these respondents vary between EUR 5.1 and 50.5 million and are original records, not imputed statistically. It is remarkable that those few individuals with by far the highest wealth in the SOEP position themselves far right on the LRS-one even at 10. If heterogeneity among the rich can be described with more certainty based on more extensive data, then this is, of course, only the first step in determining the causal mechanisms for the differences. Several causes are plausible that would have to be determined by future research. In any case, they will surely be dependent on different forms of affluence.
Heterogeneity of political attitudes and opinions might be underestimated when looking at income only. This interpretation finds some support when inspecting results from existing studies in political science. In many studies comparing the opinions of those with high incomes and lower-income groups, variation within the highest income group seems to be largest, as indicated by standard deviations or standard errors (e.g. Soroka and Wlezien, 2008: p. 320;Flavin, 2011: p. 42;Peters and Ensink, 2015: p. 583;Bartels, 2016: p. 262). While some authors explicitly address this and either do not find systematically different within-income-group variation (Gilens, 2012: pp. 91-92), or provide an alternative explanation (Ura and Ellis, 2008: p. 789), this fact might indicate that there is more heterogeneity among the affluent than in the compared groups-especially when only measured as highest income tertile, quintile, or decile. However, it might also be due to smaller sample sizes. To clarify this, possible heterogeneity among the rich should be given more attention in future studies.
Overall, the results presented here indicate that there is considerable heterogeneity among different groups. This insight and the consciousness about it might be helpful in identifying causal mechanisms for how political phenomena involving the rich work. For example, if social scientific research wants to explain why 'the rich' are better represented than other social groups as suggested by responsiveness bias research (Gilens, 2012;Bartels, 2016), it has to identify mechanisms which account for this heterogeneity. Namely, which group is assumed to be better represented than those with lower incomes or wealth? Comparing political attitudes between potentially relevant groups suggests that these attitudes differ, as presented in Figure 6.
Overall, social scientific research can only benefit from more interest in varieties of affluence. Opinions of distinct groups among the affluent on more gradual measures such as opinions on individual policies could be especially beneficial. After all, as Almond (1945: p. 213) put it, 'the existence of deviational groups of this type and size plays havoc with any simple effort to account for the political attitudes of wealth, or for that matter, of any economic and social group'. With increasing availability of data accounting for this variety, his claim is more relevant than ever today.

Conclusion
This article discussed definitions of being rich in relation to varieties of affluence such as income, wealth, and origin of wealth, proxied by IWI. Varieties of Affluence section developed a framework with which to analyse differences in social outcomes. Affluence and political attitudes section derived hypotheses on political attitudes for different dimensions of affluence. Methodology section and Results section applied the varieties of affluence perspective to data from the GSOEP by analysing heterogeneity in terms of political attitudes for different groups among the rich. Finally, Discussion section discussed the results in light of relevant open questions.
The main finding is that wealth-and not incomeseems to be the dominant dimension for political attitudes among the rich. There are also indications that very large fortunes might be associated with strongly conservative views. Similarly, IWI seems to be a potentially relevant dimension, which should be analysed further with more data on the very wealthy and those who inherit large sums of wealth. Finally, qualitatively disparate groups could be identified that exhibit significantly heterogeneous political attitudes. This insight sheds light on the relevance of heterogeneity among the rich.
Several shortcomings of this study could not be overcome due to the available data. First, the sample is not representative because weights for the combination of subsamples used are so far not provided. Also, sample sizes of the very wealthy and inheritors of large wealth are still very small. This is because, in contrast to income, those with high wealth are not oversampled in the data. In addition, longitudinal wealth data could lead to more certain assertions about mechanisms and consequences related to wealth possession. The fact that only wealth data assessed 2-3 years before the dependent variable was available is unfortunate and introduces some inevitable bias to the results. In addition to that, measurement bias of wealth and income might be evident which possibly attenuated the found effects. Overall, the available data were exploited as the best source available today. However, the data basis must be extended to allow for results that are more than an initial exploration.
Today, when many indicators point towards an increasing concentration of wealth, income, and inheritances, there is a good reason to devote more attention to different kinds of economic affluence and understand the variance between individuals and groups holding such economically privileged positions in our societies. As far as this is possible, integrating them into standardized surveys assessing all aspects of life would enable us to finally analyse the rich with representative, comparable data, as it is done for almost all other members of society already. Existing studies on the very wealthy (e.g. Lauterbach et al., 2016) could be built upon and ideally integrated into standard large-scale panel surveys, as is by now planned for one of the next GSOEP releases. In addition, more creative and heterogeneous measures and methods will be necessary to understand how varieties of affluence are obtained and maintained, and how they influence individuals, their lives, and therefore, society as a whole in the long run.

Notes
1 See King (2017) for a review of the critiques of Piketty's claim. 2 In what way, and how much, varieties of affluence matter will, therefore, strongly depend on the studied outcome. Political attitudes are chosen as one example here; however, other outcomes European Sociological Review, 2020, Vol. 36, No. 1 might be related to them in different direct or indirect ways. 3 For now, the term affluent is used-in accordance with most of the existing literature-as being affluent in one of two dimensions: income or wealth. Terms such as affluent, wealthy, and rich are, therefore, used interchangeably. A more differentiated definition is given in Varieties of Affluence section. 4 The inclusion of pension claims is also often suggested and discussed. It is disregarded in this article because of its main interest in the top group in terms of different dimensions of financial wealth in Germany. However, it might be reasonable to include pension claims, e.g. for cross-country comparisons. 5 The reason capitalization is disregarded here is data restriction (cf. Operationalization and sample section). 6 Political attitudes, orientation, and right-and leftwing orientations are used interchangeably in this article. 7 For a discussion of the meaning of the LRS for Germany see Fuhse (2004). To analyse the available empirical results about the affluent, it is assumed that the Free Democrats (FDP) and Christian Democrats (CDU) are seen as right from the centre. This is also in line with robust findings from voter surveys, e.g. reported by Stö ss (1997) or Urban Pappi, Kurella and Brä uninger (2016). Results from the United States referring to liberal and conservative are regarded as roughly equivalent to the left-right distinction. In recent years, there seems to be an overall shift towards the left that is possibly due to right-wing populist movements (Infratest dimap, 2015;Piketty, 2018). Therefore, the year of inquiry is included as confounder in the analysis. 8 See, for example, Hecht (2017) for a description of different potential mechanisms relevant for political attitudes, and heterogeneous groups among the group of top income earners in London. 9 There are two reasons why income is used before government transfers and taxes. The first is that the absolute amount of pre-government income entails the amount of tax relevant income which might be relevant for political attitudes. Second, receiving high pre-government income implies that one is not dependent on government transfers. This might influence political attitudes, e.g. towards redistribution.
10 Taking the example of wealth this means: Where Z is the standardized transformation used in the analyses, X y is the originally observed value of wealth in year y, X À y is mean wealth in that respective year (i.e. the year in which the variable was assessed), and r Xy is the standard deviation of wealth in year y. Standardization was performed for the rich and the non-rich combined in each year and not separately for each sample. 11 The first reason for this is that wealth thresholds are defined in absolute terms. Second, using overall as well as non-capitalized wealth, the stakes for exceeding the thresholds are lowered which is preferable here because sample sizes are very limited in the first place. However, some cases might not be considered rich in terms of wealth when applying a household standardized measure. 12 However, in contrast to the literature, net overall wealth is used here instead of available net financial capital which means stakes for being rich in terms of wealth are lowered significantly. The used categories simply overtake descriptive definitions often used in the literature. This shall not imply, however, that they have a substantial sociological meaning. They are rather used here as a starting point to identify heterogeneous groups and to see whether they, in fact, do empirically identify distinct groups in terms of political attitudes. This is also why the thresholds in dollars are simply used in euros in the analysis. For a discussion of a more substantial definition see, for example, Fessler and Schü rz (2017). 13 A more detailed presentation of the measure and a summary of its shortcomings is provided in the Supplementary material A1. 14 Households are assigned according to the occupation of its head as suggested by Goldthorpe (1983) and the retired are assigned according to their last occupation. 15 A full table of all included variables and descriptive statistics can be found in the Supplementary material A2. 16 In addition, the main results do not change when bootstrapping percentile standard errors or when excluding confounders which could be suspect to introduce post-treatment bias (such as arguably class or political interest) are excluded (not reported).
17 Unfortunately, for most of the studies based on the GSS, it is not clear whether family income or personal income is used. Both were assessed in the GSS (Hout, 2004); most often, the articles do not specify. There are some implicit remarks that suggest individual labour income might have been used more often.

Supplementary Data
Supplementary data are available at ESR online.