More than self-interest: Why different classes have different attitudes to income inequality

The connection between social class and political preferences is among the most well established in the social sciences. This association is typically taken as prima facie evidence of economic self-interest: Classes hold different attitudes, values, and party preferences because they have different economic interests. However, this assumption has rarely been tested empirically. In this article, we use survey data from 18 West European countries to examine why classes differ on a central aspect of political preferences, namely their views on the desirability of income inequal -ity. We find that only a moderate proportion of differences between employee classes in support for redistribution can be accounted for by contemporary differences in resources and risks; differences in economic interests to some degree account for the anti-redistributive preferences of the pro fessional middle classes compared with the working class. However, the preferences of the self-employed have a dif ferent explanation; autonomy is a better explanation of the right-wing preferences of the self-employed compared with the working class. supporting of voting derive through Lipset’s (1960) influential assertion every the mainly for parties of the


. | INTRODUC TI ON
Discussions of class differences in political preferences have been extensive since the early days of social science.
Class voting has received most attention, although studies of class differences in political attitudes, particularly those concerning inequality and redistribution, have also accumulated an extensive bibliography. This is no accident, as concern with inequality and redistribution typically lies at the heart of class politics, and conflict over inequality and its amelioration via redistribution has historically provided the axis of many Western party systems. Underpinning and perpetuating these party systems have been divisions between social classes supporting parties of the left and right. This pattern of class voting is typically assumed to derive from simple self-interest: as expressed through Lipset's (1960) influential assertion that "in virtually every economically developed country the lower income groups vote mainly for parties of the left." A view echoed recently in an extensive comparative analysis of class voting (Evans & de Graaf, 2013, p. viii), where it is assumed to result from "the rational expression of individual differences in (primarily) interests that cluster as a result of shared characteristics deriving from similar conditions of employment." In short, working-class people tend to be poorer; left-wing parties espouse redistribution from rich to poor; therefore rational self-interest leads working-class people to vote for left-wing parties, and vice versa for the middle classes. However, despite the frequency with which this implicit chain of causation linking class position with differences in resources, attitudes towards redistribution and party choice is invoked, rigorous attempts to test it are noticeable for their absence. In this paper we examine the mechanisms that link class position to support for redistribution, using one of the recently developed mediation methodologies that allows us to disentangle their relative importance.

. | WHY DOE S CL A SS AFFEC T RED IS TRIBUTIVE PREFEREN CE S?
The assumption that class voting occurs due to the classes' different economic ideologies or inequality preferences relies on there being a strong link between class and redistributive preferences. Such an association has indeed been extensively documented across Western democracies (e.g., Kalmijn & Kraaykamp, 2007, Knutsen, 2018, ch. 3, Langsaether & Stubager, 2019, such as Denmark (Stubager, 2006), the Netherlands ( Van de Werfhorst & de Graaf, 2014), Sweden (Svallfors, 2006), the UK (Evans, 1993;Evans & Tilley, 2017), and the US (Shingles, 1989). However, much less is known about why this is the case: What explains the association between class and preferences for inequality?
As with class voting, the association between class and preferences regarding inequality has been assumed to follow neatly from material interests. As Sears and Funk (1991, p. 1) put it, "[s]elf-interest is never far from front stage when we consider the ordinary individual's social and political attitudes." Thus Svallfors (2006, p. 21) claims that the classes "obviously have an interest in increasing their resources and limiting their risks …. Different class interests are then one obvious source of variation in social attitudes." This is particularly likely to be true with respect to income inequality: people in classes who have higher incomes are likely to be less supportive of income redistribution than those on lower incomes. Accordingly, from the earliest research into class differences in attitudes onwards it has been assumed that, as Berelson, Lazarsfeld, and McPhee (1954, p. 184) assert, classes are typically opposed to each other on position issues related to economic interests, such as taxation, labor management, and tariffs, and this is motivated by "[s]elf-interest of a relatively direct kind." More recently, scholars such as Kalmijn and Kraaykamp (2007, p. 550) affirm that "the attitudes class members hold are a reflection of those [economic] interests," while Curtis and Andersen (2015, p. 5) claim that it is "widely accepted that self-interest plays an important role in determining preferences for inequality." That material interests and perceptions of risk create class differences in attitudes and behavior is the assumption of "virtually all established theorizing on class economic relations" (Brooks & Svallfors, 2010, p. 200).
Despite the prevalence of this perspective, however, very few studies have examined whether these economic interests actually do account for differences between classes, and the few that do so have been geographically limited to Scandinavia. Moreover, even this evidence is inconclusive. In his analysis of Swedish data, Svallfors (2006, pp. 94-95) finds that working conditions, income, financial problems, and unemployment account for only a small proportion of class differences in welfare spending preferences and that "almost all class differences remain after controlling for the other variables." Brooks and Svallfors's (2010) study of mechanisms connecting class and inequality attitudes in the Nordic countries finds that self-interest factors such as household income, unemployment experience, and household economic situation account for 26%-34% of the class differences.
Finally, in their detailed study of class mechanisms in Sweden, Bengtsson, Berglund, and Oskarson (2013) find that class differences in left-right orientations are mediated to some extent by self-interest variables such as income and job security.
The relative scarcity and geographical specificity of studies testing this assumed causal chain is unfortunate, given that a model wherein class position shapes (primarily material) interests which then affect political preferences is implicit in a very large number of studies of the relationship between class and politics (examples beyond those discussed above include Campbell, Converse, Miller, & Stokes, 1960, ch. 13;Chan & Goldthorpe, 2007;Gallagher, Laver, & Mair, 2011, p. 284;Evans & Tilley, 2017, ch. 4). Our first aim in this paper, then, is to redress this omission in the understanding of class politics. We do this by examining the importance of economic selfinterest for all Western European countries with respect to the area of political preference that is most likely to be affected by material interests, namely income inequality. We focus on the key area of programmatic and policy implementation addressing this concern: support for redistribution. We examine these attitudes because of their centrality to politics and their link to class differences in material interests and because unlike voting, say, for parties of the left or right, they are not restricted by the choice-set of options presented by the political parties.
When parties do not offer significantly different choices to voters with respect to redistribution and inequality we would not expect classes to differ strongly in their vote. When inequality is depoliticized, class voting is less likely to be observed even when class differences in attitudes towards inequality and redistribution are robust (Evans & Tilley, 2012). The recent rise of radical right parties in Europe is also likely to have weakened the link between economic self-interest and party choice in some instances. Radical right parties typically mobilize support with anti-immigration stances rather than explicitly redistributive programs and as such appeal to a different set of class-related attitudes (Rydgren, 2013).
Our measure of class position is the Erikson and Goldthorpe occupational class schema (e.g. Erikson & Goldthorpe, 1992;Goldthorpe, 2000;Goldthorpe & McKnight, 2006). This is not only somewhat ubiquitous in stratification research but also benefits from detailed validation studies in which measures of occupational attributes such as employment conditions, promotion prospects, job uncertainty, and control over work have been shown to cluster into latent classes that correspond closely to the class categories in the schema (Evans & Mills 1998). The schema also has a clearly specified conceptual underpinning that evolved from a focus on work and market situations (Goldthorpe, 1980, p. 39) to the employment contracts that underlie the distribution of these attributes (Erikson & Goldthorpe, 1992;Goldthorpe, 2000). "Classes differ in sources and levels of income, their degree of economic security and chances of economic advancement … their location within systems of authority and control governing the process of production in which they are engaged, and their degree of autonomy in performing their work-tasks and roles" (Goldthorpe, 1980, p. 39) Goldthorpe went on to identify the reason for this differentiation, which lies in the nature of the employment contract. Employers prefer labor contracts for jobs that are easily monitored, and which do not require specialized knowledge or expertise, while in the opposite case they rely on a service contract or a mixed relationship. In other words, this has to do with the way in which commitment is obtained from the work-force.
The key features of employment contracts are their degree of "asset-specificity" and the degree of monitoring difficulty (Goldthorpe, 2000, p. 213). Employees with a "service" contract-typically in professional and managerial jobs-score highly in both these respects and exercise delegated authority with specialized knowledge and expertise in the interest of their employing organization. Their commitment is bought via higher salaries and better prospects (Goldthorpe & McKnight, 2006). Their asset specificity also endows them with higher levels of autonomy in their work. Those on "labor" contracts-typically in working-class jobs-are more heavily monitored with consequently less autonomy and tend to provide discrete amounts of labor under closer supervision, as well as receive lower incomes and less job security. Class positions are therefore quite directly related to income, autonomy, and security. It is the influence of these mechanisms that we intend to examine in the analysis that follows. Working-class occupations have poorer working conditions, lower incomes and fewer opportunities and the working class are expected to find redistributive policies more appealing than those in more advantaged situations.

. | UNDER S TAND ING WHY THE PREFEREN CE S OF THE S ELF-EMPLOYED D IFFER FROM THE WORK ING CL A SS
We further propose that the story of class differences in interests is not likely to be as simple as hitherto assumed, as it may vary according to the classes being considered. A key distinction is in how we might account for differences in attitudes between the middle class and the working class and the self-employed, or petite bourgeoisie, and the working class.
While professional and managerial classes have much higher and more stable incomes than the working class, the self-employed do not. In their study of the economic interests of classes, Goldthorpe and McKnight (2006, p. 21) claim that fluctuations in earnings are much higher among self-employed than among employees in general "as a direct consequence of the relations that characterize their class position." They also cite qualitative evidence that this economic uncertainty creates problems in both business planning and family budgeting.
Moreover, the incomes of the self-employed are closer to those of the working class than to those of the service classes (Goldthorpe & McKnight, 2006, pp. 127-128), even when correcting for potential under-reporting of their income. Finally, while the self-employed have the lowest risk of unemployment (due to being their own employer), underemployment is another source of economic insecurity for this group (Goldthorpe & McKnight, 2006, p. 114). Despite these economic disadvantages the self-employed have consistently been found to be among the most economically right-wing citizens (e.g., Evans, 1993, p. 455;Jansen, 2011, pp. 104-105;Knutsen, 2018, p. 91;Svallfors, 2006, ch. 3-4;Van de Werfhorst & de Graaf, 2014) who tend to vote for right-wing parties (e.g., Bengtsson, Hansen, Hardarson, Narud, & Oscarsson, 2014, p. 159;Jansen, Evans, & De Graaf, 2013;Langsaether, 2019b, ch. 2). While this could be related to their opposition to labor regulations or employee rights, it is hard to find an economic interest basis for their support for income inequality. Previous studies have not considered this issue and have also tended to use composite measures of class differences (i.e., the kappa index-see Brooks & Svallfors, 2010) that do not distinguish how mechanisms might work differently for different classes. This leads then to our first hypothesis: Hypothesis 1 Economic interests can account for differences in support for redistribution between the middle and working classes, but are less likely to do so for differences between the self-employed and the workers.
The similarity in economic situation between the self-employed and the working class suggests that we need to look at class differences outside of the realm of economic interests when trying to account for this axis of class political division. Key amongst these is work autonomy and control. According to Kohn (2001), people learn on the job and extend the lessons learnt to realities outside of work. Kohn focused on authoritarianism, showing that the degree to which someone is monitored does affect their level of authoritarianism (Kohn, 1969), a line of inquiry also considered by Bengtsson et al. (2013, p. 698). However, the argument can be generalized. We argue that people who experience autonomy in the workplace may have a more acute sense of the relationship between decisions and outcomes than employees with low levels of autonomy. Generalizing from this experience to political attitudes could lead to a belief that people are responsible for their own outcomes, thus justifying a higher level of inequality, a smaller welfare state, and lower taxes. A similar argument has been made by Kitschelt and Rehm (2014, pp. 1674-1675, who argue that people who enjoy discretion and autonomy in their jobs generalize these experiences to other spheres of life, inducing a tendency towards right-wing ideology in the economic domain. Jansen (2019, p. 11) also argues that job autonomy can be associated with opposition to redistribution and other government interventions. However, while professional and managerial classes generally have more autonomy than the working class (e.g., Erikson & Goldthorpe, 1992;Evans, 1992), work autonomy is a particularly important source of difference between the self-employed and the working class. We therefore predict: Hypothesis 2 Job autonomy will be substantially more important for differentiating between support for redistribution between the self-employed and the working class than between the middle and working classes.

| DATA AND ME THODOLOGY
To test the above hypotheses, we rely on the European Values Study (EVS, 2008-10), which covers all the West European countries and includes a measure of redistribution preferences, suitable class measures for both current class and class origins (see below), and a range of relevant variables capturing aspects of the different mechanisms for explaining class differences in preferences. 1 These include perceived job autonomy, income, unemployment experience, and social security dependence. We utilize data on 9,570 respondents from Austria, Belgium, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the UK. We limit our analyses to Western Europe as the recency of the post-communist transition suggests there are features of class politics that need consideration in their own right: the class structure is rather different and class differences in income as well as redistributive preferences are smaller.

| The dependent variable: Income inequality preferences
We rely on an item in the European Values Study asking respondents to place their views on a scale from 1-10, where 1 denotes "incomes should be made more equal," while 10 denotes "there should be greater incentives for individual effort." 2 We have reversed the scale so that higher values indicate that the respondent is more in favor of redistribution.

| The independent variables
To measure class, we utilize the EGP class schema (Erikson & Goldthorpe, 1992;Erikson, Goldthorpe, & Portocarero, 1979), which is "the only systematically validated measure of class position" for use in large-scale comparative projects (Evans & De Graaf, 2013, p. 13). We use a modified version of this schema. We include the higher service class and the lower service class, which correspond to class I (higher controllers) and class II (lower controllers) in the original schema. Furthermore, we include the self-employed (sometimes referred to as the petite bourgeoisie), which consists of class IVa, IVb, and IVc in the original schema, that is, self-employed with and without employees, and self-employed farmers. We also include the routine non-manual employees, which consist of class IIIa (routine non-manual employees) and IIIb (lower sales-service employees). Some scholars include class IIIb among the workers. In Section 5 of the Online Appendix, we re-estimate all analyses with class IIIb categorized as working class, and the results are robust to this change. Finally, we have the working class, consisting of manual supervisors (class V), skilled and unskilled workers (class VI and VIIa), and farm labor (class VIIb). 3 The EVS contains pre-coded variables for the respondents' EGP class location as well as that of the parents when the respondent was aged 14.
We use the former to measure class, and the latter to measure class origins.
We include class origins to identify background effects. Early life experiences are generally considered to affect political attitudes, values, and identities (Neundorf & Smets, 2017;van Deth, Jan, Abendschön, & Vollmar, 2011). The formative years between childhood and adulthood are considered especially important in these respects (Neundorf & Smets, 2017, p. 4). A recent British Household Panel Study (BHPS) analysis has found intergenerational transmission of both parents' ideologies, as well as persistent effects of growing up in poverty (O'Grady, 2019). As current class is associated with class origins, class differences in socialization and early experiences have a potentially confounding impact on contemporary attitude differences between classes. We thus present analyses with controls for background here, while we present analyses without such controls in Section 9 of the Online Appendix, and results are highly similar.
We also include age, gender, and education as control variables. We distinguish between respondents with low, medium, and higher education. The first group consists of those who have no or only primary education (ISCED one digit 0 and 1), the second group consists of respondents who completed lower or upper secondary education or post-secondary but non-tertiary education (that is, ISCED one digit 2-4) and finally, the third group consists of respondents who have completed first or second stage of tertiary education (ISCED one digit 5 and 6).
To test the hypotheses, we include three variables measuring economic interests. The first is household income. In each country, the respondents are asked to place their household income in one of several income categories-these vary from country to country. The EVS team has then converted these categories into euros (1000), corrected for purchasing power parity, and harmonized into a cross-nationally comparable variable. 4 The second is unemployment experience. The respondent is asked "[d]uring the last five years, have you experienced a continuous period of unemployment longer than three months?" The variable is coded 1 if the answer is yes, and 0 if the answer is no. Furthermore, we use a measure of social security dependence. In this case, the respondent is asked "during the last five years, have you been dependent on social security at any time?" The variable is coded the same way as unemployment experience. These variables index core material interests related to income levels and employment security. Finally, we measure job autonomy. Respondents are asked "[h]ow free are you to make decisions in your job?" and may indicate any value from 1 (which reads "none at all") to 10 ("a great deal").

| Methodology
To test the hypotheses, we employ the KHB method of mediation analysis (Kohler, Karlson, & Holm, 2011). While this methodology was originally developed to deal with mediation analysis in logistic regression analyses, it can also be used in the ordinary least squares (OLS) setting. In this case, it is equivalent to traditional mediation analysis; however, an important advantage is that the KHB package in Stata allows us not only to assess how much of the effect of class that is due to other (preceding or intermediate) variables, but also to disentangle the mediating effect of various variables simultaneously (Breen, Karlson, & Holm, 2013). 5 In the first model, we regress income inequality preferences on current class, as well as age, gender, and education. In the second model, we also include class origins. These are all potential confounders as they may affect both current class and redistributive preferences. By including education, we control to some extent also for other potential confounders that are associated with educational attainment, such as career ambition or aspirational values. The same is the case for parental class, which effectively controls for early socialization related to the class environment one grows up in, as well as childhood experiences such as growing up in poverty (see the discussion below). However, we acknowledge that there may be confounders we have overlooked (although sensitivity analyses provide some comfort in this regard, see Online Appendix, Section 10) or reverse causality issues that we are unable to handle with cross-sectional data.
Finally, in the third model, we also add the three economic interest variables as well as job autonomy. Applying the KHB method, we then disentangle the mediating effect of these variables. Hypothesis 1 receives support if the economic interest variables reduce the coefficients of the service classes more than that of the self-employed from Model 2 to Model 3, controlling for job autonomy. Conversely, Hypothesis 2 receives support if job autonomy reduces the coefficient of the self-employed more than those of the service classes from Model 2 to Model 3, controlling for economic interests. 6

| RE SULTS
Before moving on to the hypothesis testing, we want to illustrate the importance of class. First, classes do indeed differ in terms of income and job autonomy in the way discussed above. Figure 1 plots the marginal effect of class on income (left panel) and job autonomy (right panel) in OLS regression models including country-fixed effects and cluster robust standard errors (see Table A.1.1 in the Online Appendix for the full regression table). As is clear, the workers have the smallest incomes, followed closely by the routine non-manual employees and then the selfemployed. The lower service class earns substantially more, while the higher service class by far has the highest incomes. For job autonomy, the picture is different. The workers and routine non-manual employees are clearly least autonomous. The lower service class is a full scale point more autonomous, followed by the higher service class. The self-employed report by far the highest job autonomy, a full scale point above even the higher service class. Clearly, income and job autonomy are unequally distributed in different classes.
Despite arguments about the death of class (see, e.g., Evans, 2017 for a brief overview), political preferences also follow class lines. Figure 2 compares the association between current class and redistributive preferences (left panel) with the association between left-right self-placement and redistributive preferences (right panel), from models that include country fixed effects and controls for age, gender, and education (see Online Appendix,  The results are in line with previous findings (see, e.g., Langsaether, 2019b, p. 14).
We now move on to test the hypotheses. The first row in Table 1 shows the effect of current class on income inequality preferences. All estimates are as expected. The higher service class and self-employed are approximately equally right-wing, while the workers and routine non-manual employees are equally in favor of redistribution. The lower service class is in an intermediate position. In line with previous research, women are slightly more in favor of redistribution than men, while age and education do not affect income inequality preferences when controlling for occupational class.
The second row depicts the results from Model 2, adding also class origins to control for class differences in socialization and early experiences. There is a modest effect of having a background from the higher service class rather than from the working class, but otherwise class origins have very little effect when estimated simultaneously with current class. As can be seen in Row 4, controlling for class origins does not change the coefficients of current class position by much. This implies that current class differences in inequality preferences are not due to growing up in different class environments.
Finally, Model 3 includes controls for the economic interests variables as well as job autonomy. We notice that income has a substantial effect: Moving one standard deviation on the income scale (1.52) would reduce opposition to income inequality by 0.21 scale units. We note, however, that this means that one would have to move almost four standard deviations across the income scale to achieve an effect similar to the difference between the higher service class and the workers in Model 1. Occupational class is indeed important for these preferences. Having experienced unemployment in the last 5 years does not affect redistributive preferences, controlled for all the other variables, whereas having been dependent on social security does increase support for redistribution by 0.35. Job autonomy, while less important than income, has a substantial impact on redistributive preferences: People who are one standard deviation (2.26) more autonomous in their jobs are 0.16 units less opposed to income inequality.
The final row shows how the class coefficients from Model 2 change when including the economic interest variables as well as job autonomy. We see that these become substantially smaller. The difference between the higher service class and the workers is reduced by 0.25, that is, 31%. Approximately a third of the difference between the lower service class and the working class also disappears. More than a quarter of the difference between the self-employed and the working class is eliminated. All these changes are statistically significant at the 0.01 level. Routine non-manual employees were not different from the working class prior to introducing the mediators, so this coefficient hardly changes and the change is not statistically significant.
To be able to test hypotheses 1 and 2, however, we need to know to what degree this mediation is accounted for by different variables. The KHB methodology allows us to do this. Table 2 shows mediation by variable for the three classes that were more right-wing than the workers and the routine non-manual employees: The two service classes and the self-employed. Table 2 shows that of the material interest variables, only income accounts for a substantial proportion-about a fifth-of the right-wing orientation of the higher service class, and the change is about three times as large as the standard error. The same is true for the lower service class: 15.3% of the difference between this class and the workers and routine non-manual employees is due to income. Unemployment experience is neither substantially nor statistically significant for these classes' right-wing orientations, whereas social security dependence accounts for around 2% of the difference between them and the workers.
This is also the case for the self-employed. However, in contrast to the service classes, a much smaller proportion of the difference between the self-employed and the workers can be attributed to income differences-only around 6%. Put simply: Material interest does not seem to be an important reason for the right-wing orientation of this class. Hypothesis 1 receives support.
Moving on to job autonomy, we see that this variable mediates 11%-15% of the right-wing orientations of the two service classes. However, it accounts for a fifth of the right-wing orientations of the self-employed. Particularly noticeable is the difference in mediation effects between the higher service class and the working class (11.2%), and the self-employed and the working class (19.9%). The latter is almost twice the level of the former. In other words, while the job autonomy of the service class is associated with being more right-wing, this factor is far more important for the self-employed, in line with Hypothesis 2. To ensure that these results are not driven by any country in particular, we have re-estimated all the models 18 times, each time leaving out all the respondents from one country (see the Online Appendix, Section 8). The results are always very similar to those reported in Tables 1 and 2.

. | CON CLUS I ON AND D ISCUSS I ON
This article addresses an understudied yet important question within class studies. We broaden the geographical scope compared to the few previous studies on mechanisms connecting class and redistributive preferences and we disentangle the importance of individual mediators. Our results hold regardless of which mediation technique we utilize; whether we examine redistributive preferences or broader economic left-right ideology; whether we include control variables or not; whether we exclude any individual country; whether we include class IIIb among routine non-manual employees or the workers; and regardless of whether we adjust income for household composition. These robust findings yield three general conclusions. First, in Western Europe, self-interest accounts for less than a fifth of overall class differences in attitudes to income inequality, an attitude chosen as a most likely case for the expression of self-interest. Furthermore, most of the class differences accounted for by self-interest result from current household income rather than unemployment experience or social security dependence.
The low level of mediation via self-interest is thought provoking given how many studies in the field seem to assume that this mechanism should account for (net of measurement error) most or all class differences in redistribution preferences. We are aware that this test of the self-interest perspective is not comprehensive, and more and better measures of material interests might increase the level of mediation. However, our findings are consistent with the limited number of previous studies examining the Nordic countries discussed in the introduction, some of which do include a more extensive range of mediating variables. The implication of our findings is that class analysts may benefit from relaxing the assumption that class differences in redistributive preferences are only (or perhaps even primarily) related to differences in self-interest, and try to theorize and empirically investigate other sources of class differences in preferences. One fruitful area for such research might well lie in examining network effects on class attitudes. The more people interact within a group, the more likely they are to influence each other (Zuckerman, Dasovic, & Fitzgerald, 2007, p. 31), and people tend to interact most with people in the same class, for instance due to clustering in families, schools, workplaces, and neighborhoods (Kalmijn & Kraaykamp, 2007, p. 550). This means that there are likely to be important class differences that we can explain by social interactions, norms, and identity rather than economic self-interest per se.
Second, we find that job autonomy plays a hitherto undetected role in accounting for certain class differences, even when controlling for income. While income is more important for the service classes' right-wing orientation than job autonomy, the latter is much more important for the right-wing orientation of the self-employed. This is in line with the claims of authors such as Kohn (1969Kohn ( , 2001 and Kitschelt and Rehm (2014), who argue that people generalize experiences from their work to other aspects of life, including political attitudes. It also helps to explain the puzzle of why a group of people who are not much better off than the working class are still much more in favor of income inequality. Heterogeneity of mechanisms between classes may therefore be a promising line of inquiry in future efforts to understanding class differences in preferences.
Thirdly, we find that class origins are not important for current class differences in support for redistribution.
What people in different classes think about income inequality is more related to their current situation than to their early socialization. 7 This non-finding differs from the literature on social mobility and class voting, which tends to find that the preferences of the socially mobile lie between their origin and destination (e.g., De Graaf, Nieuwbeerta, & Heath, 1995). That party allegiance is perhaps more strongly influenced by origins than are redistributive preferences may well relate to the continuing, if weakened (Dalton & Wattenberg, 2000), presence of durable party affiliations in many Western democracies. Such party affiliations are likely to be more stable than are issue preferences (Huddy, 2013) and less likely to change following changes in resources and interests (Svallfors, 2006, p. 21). 8 It suggests that further consideration of the factors that mediate the relationship between class position and redistributive preferences is likely to make progress through further examination of aspects of the current situation rather than background factors. This might well include an expansion from considerations of narrow material/economic self-interests to a broader conception of personal self-interest. This could well include factors such as social norms, identities, and historical legacies.
To conclude, although based on a partial test, our findings imply that we cannot simply assume that class differences are only due to economic self-interest, and that the mechanisms explaining different class differences in political attitudes are general across classes. We need to recognize that different influences are likely to be at play in explaining why some classes differ from others-particularly in the case of job autonomy and the self-employed.
We also need to move beyond studying differences in material interests, and to examine interactions, norms and identities that might throw further light on class divisions.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in GESIS at http://doi.org/doi:10.4232/1.10188, reference number ZA4800.

N OTE S
1 Alternative data sources such as the ISSP inequality module or the ESS work wave unfortunately lack measures of job autonomy.
7 As noted earlier, in analyses examining generalized left-right values rather than, specifically, support for income redistribution we do find some effect of background on these more stable and enduring aspects of belief systems (on which, see Evans & Neundorf, 2018). 8 It is also possible self-selection may bias our results: It may be that people who are both opposed to redistribution and enjoy autonomy self-select into self-employment. Long-term panel data studies are needed to examine such selection processes.