Introduction
That elected representatives’ socio-economic status (SES) does not correspond with that of the average citizen has been extensively documented. In almost all countriehs where we have data from, politics remains predominantly the realm of higher-educated men from the higher social classes (Gerring et al. Reference Gerring2019). In fact, while the descriptive representation of women and ethnic minorities has improved over time (Bailer et al. Reference Bailer2022), the presence of lower-educated and lower-class individuals in politics has fared much less well. In a world in which billionaires are vastly overrepresented in politics (Krcmaric, Nelson, and Roberts Reference Krcmaric, Nelson and Roberts2023), the number of representatives from these economically disadvantaged groups has even declined over time (Best Reference Best2007; Bovens and Wille Reference Bovens and Wille2017; Carnes and Lupu Reference Carnes and Lupu2023a).
Research indicates that the limited presence of individuals with lower SES backgrounds in politics results in deficits regarding substantive representation and symbolic representation (Elsässer and Schäfer Reference Elsässer and Schäfer2022). Extant studies found that representatives with lower SES backgrounds tend to prioritize the substantive needs of their socio-economic peers, while those with higher SES backgrounds are more aligned with the broader population (Carnes and Lupu Reference Carnes and Lupu2015; Hakhverdian Reference Hakhverdian2015; Barnes, Beall, and Holman Reference Barnes, Beall and Holman2021; Lupu and Warner Reference Lupu and Warner2021; Warburton et al. Reference Warburton2021; Hemingway Reference Hemingway2022; Sevenans et al. Reference Sevenans, Awenig, Braunig, Walgrave, Soontjens and Vliegenthart2024). Consequently, the absence of diverse representatives likely contributes to policy outcomes that disproportionately favor higher SES groups (Gilens Reference Gilens2012; Bartels Reference Bartels2016; Rosset and Stecker Reference Rosset and Stecker2019; Elsässer, Hense, and Schäfer Reference Elsässer, Hense and Schäfer2021; Lupu and Tirado Castro Reference Lupu and Tirado Castro2023). Furthermore, Barnes and Saxton (Reference Barnes and Saxton2019) demonstrate that the presence of lower SES representatives enhances symbolic representation by increasing the perceived legitimacy of the political system among lower SES groups.
It is through shared experiences between representatives and their constituents that descriptive representation brings about these effects (Allen Reference Allen2022). Such shared experiences enable representatives to better comprehend and politically advocate for the preferences of their constituents (Mansbridge Reference Mansbridge1999). However, it is important to note that these shared experiences need not solely stem from the representatives’ own demographic traits. In fact, merely sharing socio-demographic characteristics with constituents is often deemed inadequate for fostering either symbolic or substantive representation, urging scholars to measure shared experiences in a more comprehensive and tangible way (Celis and Erzeel Reference Celis, Erzeel, Rohrschneider and Thomassen2020; Xydias Reference Xydias2023; Funk and Hinojosa Reference Funk and Hinojosa2023).
This study builds on these arguments by measuring shared experiences beyond representatives’ own socio-demographic attributes. We contend that these experiences can also stem from the socio-demographic profile of those closest to them. As Phillips (Reference Phillips, Rohrschneider and Thomassen2020, 177) argues, the key argument in favor of descriptive representation is that the capacity to substantively represent constituents’ grievances depends ‘… on exposure to them, whether this be direct or mediated through others’ (authors’ emphasis).
Do elected representatives count lower SES citizens in their intimate network? At present, we know close to nothing about representatives’ intimate ties. The few studies that exist focused on the characteristics of representatives’ parents (Carnes and Sadin Reference Carnes and Sadin2015; Dal Bó et al. Reference Dal Bó2017; Thompson et al. Reference Thompson2019; Dahlgaard and Pedersen Reference Dahlgaard and Pedersen2024), and comparative research is scarce (for an exception, see Han and Han Reference Han and Han2021). Data on representatives’ other intimate ties, their partner and close friends, are almost entirely lacking (see Kokkonen and Karlsson Reference Kokkonen and Karlsson2017 for a lone exception). This void is what this paper aims to tackle.
We set out to broaden the scope of the literature on descriptive representation by exploring an important way, according to the normative theory, for how representatives can develop shared experiences: the socio-economic status of their intimate social ties. Drawing on a unique face-to-face survey, we asked 1,185 representatives in 13 Western countries (Australia, Belgium (Flanders and Wallonia), Canada, Czechia, Denmark, Germany, Israel, Luxembourg, the Netherlands, Norway, Portugal, Sweden, and Switzerland) regarding (1) their own educational attainment and social class. We asked them the same questions about (2) their two parents, (3) their partner, and (4) their two closest friends.
We find that the intimate ties of representatives are starkly ‘descriptively nonrepresentative’. Although their strong ties are a little more descriptively representative than their own educational and class backgrounds, elected representatives count very few lower-educated and lower-class individuals among their intimates. In none of the 13 countries we look at do representatives’ ties mirror the share of lower SES citizens in the population. When we lower the bar and examine how many representatives have a single lower SES tie, the results are a little better regarding education, but remain bleak for social class. We find little ideological variation. Politicians identifying as more left-wing are more likely to have lower-class parents, but there are no differences for other intimate social ties or for education. To make things more cumbersome from a descriptive representation point of view, we find strong evidence for homophily. Representatives with a high own SES or a high parental SES – that is, the large bulk of the representatives in all 13 countries – tend to have partners and close friends with a similarly high SES too. Moreover, this effect is much stronger among higher SES politicians than among higher SES citizens. This means that those representatives who could profit most from gaining an understanding of lower SES perspectives through their friends or partner hardly ever have the opportunity to do so, and to an even lesser extent than higher SES citizens. In all, politicians’ networks do not compensate for their own higher SES. If anything, representatives’ ties seem to make their insulation from lower SES perspectives even worse.
Representatives’ intimate ties
Our focus on parents, partners, and close friends arguably taps into the heart of representatives’ intimate network. Evidence from social psychology shows that intimate ties are uniquely relevant in shaping one’s attitudes and behavior (Dunbar Reference Dunbar2021). While weak ties or more distant acquaintances can be remarkable resources in terms of what we know (Granovetter Reference Granovetter1973), intimate ties, the people at the core of our social life, are much more relevant in shaping what we feel (Binder, Roberts, and Sutcliffe Reference Binder, Roberts and Sutcliffe2012). On average, only five people reside at the most intimate core of one’s social network (Dunbar Reference Dunbar2021). These are the people who are willing to provide us with financial, emotional, and physical support, and they are the ones we spend about 40% of all our social time with (Dunbar Reference Dunbar2020). Hence, research from social psychology shows that these intimate ties matter tremendously. But what do we know from the extant literature on representatives’ intimate ties?
In contrast to the large literature showing that representatives themselves are highly educated and belong to higher social classes (Bovens and Wille Reference Bovens and Wille2017; Carnes and Lupu Reference Carnes and Lupu2023a), only a handful of studies examines the socio-economic status of their social ties. Thompson et al. (Reference Thompson2019) use old (1940) U.S. census data to examine the education, occupation, and earnings of the parents of future members of Congress who were seated around the mid-20th century. They show that the parents of future Congressmen were hugely unrepresentative, their average earnings were twice that of the population average, and they were six times as likely as the general population to hold college degrees. Dal Bó and colleagues (Reference Dal Bó2017) present individual-level registry data on parents as well, but this time from Sweden. What they find goes against the U.S. findings. Swedish representatives’ parents, class, and earnings seem to approximate a perfect replica of the full Swedish population. Dahlgaard and Pedersen (Reference Dahlgaard and Pedersen2024) have recently replicated these findings from Sweden regarding income in Denmark. In the only comparative study we know of, Han and Han (Reference Han and Han2021) present evidence from 74 democracies between 1980 and 2011. Only focusing on the heads of government, they find that 31% had a disadvantaged family background. Kokkonen and Karlsson (Reference Kokkonen and Karlsson2017) have conducted the only study on representatives’ friends and acquaintances to date. Focusing on Swedish representatives at the municipal level, they find that almost half of the local representatives in Sweden have a blue-collar worker acquaintance or friend.
In summary, the very limited work about representatives’ social ties has produced mixed results, mostly related to representatives’ parents, and has either dealt with single countries or with heads of government only. And what we know about representatives’ friends is limited to Swedish local representatives and regards acquaintances as much as friends. In sum, there is a need for systematic comparative work drawing on more recent data that examines not only the SES of representatives’ parents but also of their other intimate ties, such as partners and close friends.
Parents’ SES is particularly important as it comes early in the socialization process and is causally prior to one’s own SES and to one’s partner and friends. Parents are an influential agent of political socialization for children at a young age, when the plasticity regarding political attitudes is at its peak (Bandura Reference Bandura1986), shaping these attitudes through both their SES and their role model behavior (Neundorf and Smets Reference Neundorf and Smets2017). Therefore, parents’ SES reflects not only the influence of intimate social ties but also the lived personal experiences of representatives when growing up.
Previous research conducted in the United States shows that parents’ SES affects representatives’ roll-call votes and political attitudes (Burden Reference Burden2007; Grumbach Reference Grumbach2015). The SES of representatives’ parents may also impact symbolic representation when these ties are visible and public. Parents may symbolize, for instance, whether a closed caste of political dynasties rules the country or whether social mobility for lower SES groups is feasible (Smith Reference Smith2018). Representatives’ parents and their upbringing also shape symbolic representation indirectly by affecting their modes of speech, mannerisms, or even attire in a way that resonates with their constituents (see Funk and Hinojosa Reference Funk and Hinojosa2023 for an overview).
However, understanding the lived experiences of individuals from lower SES backgrounds is not only a matter of one’s parental background but also of one’s partner and friends acquired later in life. The partner one chooses or the friends one has can be the consequence of one’s pre-existing attitudes and behavior. When this is the case, these ties likely reinforce the values one had before, as they get validated through one’s intimates (McPherson et al. Reference McPherson, Smith-Lovin and Cook2001). But friends and partners can also provide insights into the experience of people with a different SES (Newman Reference Newman2014). And such ties may matter for symbolic representation because they may signal to lower SES citizens that representatives are choosing to intimately connect with people like them. Conversely, Coleman and Moss (Reference Coleman and Moss2023) draw on qualitative evidence from the U.K. to show that when representatives are perceived to be insulated from friendship ties with common citizens, the political system is less likely to be evaluated as legitimate. Concerning substantive representation, there is evidence on citizens suggesting that interacting with minority groups can increase one’s tolerance and support for these groups (e.g., Pettigrew Reference Pettigrew1998). Intense and recurrent contact, as seen in close friendships or romantic relationships (Davies et al. Reference Davies2011; Newman Reference Newman2014; Paskov and Weisstanner Reference Paskov and Weisstanner2022), is also likely to generate effects through socialization mechanisms like imitation and alignment to avoid conflict over political issues (Davis and Rusbult Reference Davis and Rusbult2001).
Previous research shows that citizens with (at least) one economically struggling friend in their network are more likely to support redistributive policies (Newman Reference Newman2014). Others argue that partners are particularly potent in shaping people’s political preferences because these interactions are often more intimate and frequent than those with friends, and because they share a household (Becker Reference Becker1991; Stoker and Jennings Reference Stoker, Jennings and Zuckerman2005; Zuckerman, Fitzgerald, and Dasovic Reference Zuckerman, Fitzgerald, Dasovic and Zuckerman2005). In Sweden, Kokkonen and Karisson (Reference Kokkonen and Karlsson2017) find a correlation between local representatives with blue-collar worker acquaintances or friends and their commitment to represent blue-collar workers. In sum, the finding that social ties are crucial in explaining ordinary citizens’ political attitudes and behavior (Mutz and Mondak Reference Mutz and Mondak2006; Lazer et al. Reference Lazer2010) is very likely to apply to representatives as well. Whom politicians know matters for how they represent.
Since work on representatives’ intimate ties is limited, we refrain from formulating precise hypotheses about how the intimate network of representatives in the 13 countries under examination will look like. Instead, we formulate two broader expectations. First, there are reasons to expect that representatives’ intimate networks, in general, would contain few lower SES ties. First, as explained earlier, there is strong empirical support that, in most countries, educational attainment and social class are strongly correlated across generations from parents to children (Erikson and Goldthorpe Reference Erikson and Goldthorpe1992; Jackson Reference Jackson2013; Jackson and Evans Reference Jackson and Evans2018). Hence, if representatives belong to the higher SES group themselves, we can generally expect their parents to also belong to higher SES groups. Additionally, homophily – the tendency for individuals to connect with others who are like them – has been shown to be an almost ubiquitous mechanism in network formation (McPherson et al. Reference McPherson, Smith-Lovin and Cook2001). Due to these factors – both transmission (parents) and homophily (partner and friends) – we expect that representatives’ intimate ties would diverge in an upward SES direction from the population in general.
Second, the positioning of representatives on the left–right ideological spectrum can be expected to correlate with the SES of their intimate ties. Left-wing representatives explicitly aim to represent the worse-off, whereas right-wing parties place less emphasis on this goal. Additionally, working-class voters are also shown to care more about their representatives resembling them in terms of SES (Evans and Tilley Reference Evans and Tilley2017; Heath Reference Heath2018). Hence, to adequately represent them, one might expect left-wing politicians to have more disadvantaged people in their intimate networks. Empirical evidence supports this notion. Bukodi et al. (Reference Bukodi2024) document a clear, long-standing difference in parental class backgrounds between ministers from Labour and Conservative parties over the period 1945–2023, even though ministers’ adult occupational experiences have become less class-differentiated in recent decades. Similarly, Carnes and Lupu (Reference Carnes and Lupu2015) report that working-class legislators in Latin America tend to be more left-wing. However, the picture for education seems quite different; earlier research by Bovens and Wille (Reference Bovens and Wille2017) suggests that, regarding their own educational backgrounds, left-wing politicians no longer differ significantly from their right-wing counterparts.
Methods and data
Study setting
We draw on a unique survey with representatives. A total of 1,185 representatives in 13 countries were queried in 2022–2023 about their own education level and social class, as well as those of five individuals within their intimate network. Contact with members of parliament was initiated in each country through a formal e-mail. If a politician did not respond to the first e-mail, reminder e-mails were sent, follow-up phone calls were made, and text messages were sent. On average, it took around three contact attempts to convince politicians to participate in our research. At the start of the interviews, researchers introduced the study’s purpose and reassured participants that their responses would remain confidential and unidentifiable. Politicians then completed the survey on a laptop provided by the interviewer, who was present but did not intervene or observe their responses.
The countries included in the study are Australia, Belgium, Canada, Czechia, Denmark, Germany, Israel, Luxembourg, the Netherlands, Norway, Portugal, Sweden, and Switzerland. Since Belgium is strongly federalized with a separate party system and media system in the Dutch-speaking and the French-speaking parts of the country, we will treat the two Belgian language groups as separate cases (Flanders and Francophone Belgium). Therefore, we cover 13 countries but 14 political systems. In all countries, except Germany, all national members of parliament were contacted. In some federal countries, regional representatives were also included in the target population. All politicians in our sample are professional politicians active at a higher political level. The overall average response rate is 30%, with considerable variation across countries. Sample size and response rates by country are available in Online Appendix A. Online Appendix B shows that the sampled representatives resemble the population of representatives in terms of gender, age, seniority, and party ideology. The data in the Appendices are pseudonymized for ethical purposes.
Survey data were collected from February 7, 2022 to May 4, 2023 as part of the POLPOP projectFootnote 1 . Notwithstanding all being Western countries, our country sample encompasses a broad range of institutional variation, including diverse electoral systems (proportional, majority/plurality, and mixed-member systems), and different state structures (both unitary and federal systems). More importantly, our countries can be considered conservative cases when examining how unrepresentative politicians’ intimate social ties are. Despite a worldwide underrepresentation of lower SES groups, the countries in our sample perform relatively well regarding the presence of representatives with a working-class background and have relatively few billionaires in politics (Carnes and Lupu Reference Carnes and Lupu2023a; Krcmaric, Nelson, and Roberts Reference Krcmaric, Nelson and Roberts2023). They are among the most egalitarian, socially mobile, wealthy, and democratic in the world (World Economic Forum 2020; Chancel et al. Reference Chancel2022; Nord et al. Reference Nord2024). Carnes and Lupu (Reference Carnes, Lupu, Lupu and Pontusson2023b) found that working-class citizens are more likely to hold office in more equal countries. Hence, if representatives’ ties in the countries we examine here were skewed in the higher SES direction, they are likely to be so in other countries as well.
Measuring education and class
We measured the educational attainment and social class of the representatives themselves and of their five intimate ties using the following survey question: ‘We are trying to get a better understanding of both your own and your close contacts’ social background. Would you be willing to provide us with the following information about yourself, your parents, your partner, and your two closest friends? If certain categories do not apply, for example, because you currently do not have a partner, please indicate “not applicable.” Please fill in the grid below using the drop-down menus’. The questionnaire provided a grid that asked for the education level and social class of themselves and the five intimate ties – mother, father, partner, closest friend A, and closest friend B. Education level is measured for each respondent on a country-specific ordinal scale (see Online Appendix C for the country-specific questions). We harmonized these scales according to the International Standard Classification of Education (ISCED) 2011 classification (OECD, EU, and UNESCO 2015). More precisely, we recoded whether representatives themselves and their ties received (1) no or primary education (ISCED 0-2), (2) secondary and vocational education (ISCED 3–4), or (3) higher education (ISCED 5–8). In online Appendix D, we compare our self-reported education measure of the population sample in our 13 countries in 2022–2023 with population data on politicians’ education level as coded from publicly available data sources in the same countries in 2010–2014. Notwithstanding an almost 10-year time difference, our total sample yields an almost identical average share of low-educated representatives compared to the earlier recorded share in the population.
Similarly, for social class, respondents were asked to indicate for each tie whether they are (1) lower class, (2) working class, (3) lower middle class, (4) upper-middle class, or (5) upper class; we recoded the first two categories as ‘lower class’ and the latter three as ‘higher class’. Note that while someone’s SES might change dynamically, we ask representatives about their intimate social ties’ SES at the time of interview. We do not know at what point in time representatives got to know their reported close friends and partners. But based on the literature regarding citizens, we can expect most people’s close friendships to develop during formative years such as one’s adolescence and early adulthood (Dunbar Reference Dunbar2021).
While educational attainment is a classic question that queries people about a factual state of affairs, social class is trickier to measure as it is a multidimensional concept related to income, occupation, social standing, parental background, and education. Scholars agree that there is no gold standard for measuring social class (Lareau Reference Lareau, Lareau and Conley2008; Carnes and Lupu Reference Carnes and Lupu2023a). Asking people about the income of their friends, for instance, would no doubt produce skewed data, as one often does not know what a friend earns. Asking them about friends’ occupations requires a lot of survey space since an exact occupation cannot be established with one simple survey question. Studies on the social class background of representatives themselves often used occupation as their prime, or only, social class indicator, but they seldom relied on survey data (e.g., Carnes Reference Carnes2013; Carnes and Lupu Reference Carnes and Lupu2015, Reference Carnes and Lupu2023a). Therefore, to be able to generate a broad comparison across representatives and citizens, across parties, across countries, and especially to be able to measure the social class background of representatives’ ties and not merely of themselves, we decided to resort to a self-reported social class measure. Apart from the practical reasons for using a subjective class measure, there are theoretical reasons as well. We are interested in representatives’ intimate ties with SES because these intense contacts can potentially shape their understanding of lower SES groups. Such sensitization probably requires the explicit recognition by the observer that the tie at hand is, in fact, a lower SES tie (Dovi Reference Dovi2002). Lower-class perspectives in representatives’ networks that go unrecognized probably do not exert a similar effect.
Nevertheless, the results for parents should be carefully interpreted since we know from sociological research that elites are prone to downplaying their higher-class origins (Friedman et al. Reference Friedman, O’Brien and McDonald2021). Yet, we did not find such a social desirability bias after benchmarking our own data. Concretely, in Online Appendix E, we compared our own data on the subjective social class of Swedish politicians’ parents with work done by Dal Bó and colleagues (Reference Dal Bó2017), who used Swedish individual-level registry data. We further discuss the implications and comparability of our subjective class measure in-depth in Online Appendix E.
Most representatives were able and prepared to provide answers to our, perhaps, intrusive social tie questions. Non-response on these items varied between 9.9% (own education level) and 25.7% (social class of partner; note that some representatives simply do not have a partner). The response rate for education level was higher than that for social class, which suggests that representatives may have found it easier to provide information about the education level of their intimate ties than about their social class. Note that the representatives’ data were pseudonymized – the goal being to avoid that individual representatives, for example, when knowing their age, sex, year of first election, and electoral district, could be individually identified. We handle missing values in this paper through listwise deletion. Descriptives of all variables and non-response for politicians can be found in Online Appendix F.
Measuring descriptive representation: comparing representatives’ ties with citizens
In order to assess descriptive representation, we compare representatives’ ties with their respective country populations. We draw on two sources. First, we use data from the Organisation for Economic Co-operation and Development collected in 2022 to benchmark education level (OECD 2023). We rely on indicators measuring citizens attaining below upper secondary education, attaining upper secondary or post-secondary non-tertiary education, and tertiary education. These data are representative of all citizens 25–64 years-old and correspond to the ISCED levels described above. Even though the data are restricted to citizens that are 64 and younger, for our countries, the OECD data are the best available at the time of interviewing. Moreover, we know that older generations are less educated than younger ones. As a result, the OECD data provide a conservative test to examine whether representatives’ social ties are representative of the population or not.Footnote 2
Second, we draw on the POLPOP project’s own citizen survey to measure social class. This data collection took place between February 7, 2022 and May 4, 2022 in the same 13 countries (N = 31,294). Weights are employed to ensure that the sample is representative of the general population in each country in terms of gender, age, education, and party choice. The exact same social tie question module was applied to citizens as well. Descriptives of variables and non-response for citizens can be found in Online Appendix G.
We compare representatives’ education level, their social class, and that of their intimate ties, with that of citizens’ in two ways. We employ a maximum proportional measure of descriptive representation and a minimum threshold measure. The ‘proportional’ benchmark implies that, within a political system, the overall share of lower SES ties in representatives’ networks should reflect the share of lower SES citizens in the entire population. This maximum benchmark is inspired by calls for a proportional descriptive representation of disadvantaged groups in parliament (e.g., Mansbridge Reference Mansbridge1999). The less demanding ‘threshold’ benchmark is based on the idea that representatives can better understand the perspectives of lower SES people when at least one of their intimate ties is lower SES. Once such a minimum presence in a representative’s network is assured, additional lower SES ties lose their marginal utility (Kymlicka Reference Kymlicka and Siedle1993; Mansbridge Reference Mansbridge1999).
For the minimum threshold approach, we use a dummified version of the intimate tie measures. For parental SES, we apply a dominance rule whereby higher dominates lower. For example, regarding education, we code representatives as having a higher educational family background when a single parent is higher educated. The reason is that a family is one unit, and that the one educated parent determines the SES of the entire family (Lee Reference Lee2023). Therefore, for the parental part of the intimate network, only the situation wherein both parents experience lower education and/or lower class truly characterizes a lower SES upbringing. We follow the opposite rule for the friends SES, whereby lower trumps higher (see also Dal Bó et al. Reference Dal Bó2017). Having one lower education or lower social class actor among representatives’ intimate friends suffices for representatives to be coded as having a lower SES actor among their close friends. For the SES of the partner, we do not use a dominance rule. Although their partners probably are part of the same household as the representatives themselves, we treat the partner as a separate tie that may give the representative access to the experience of another education or social class group. The reason for this is that while children who later become politicians are born into households shaped by both parents, they are not ‘born’ into their own relationships. Instead, the role of a partner in socialization is much more individualized, as politicians typically meet their partners later in life.
Results
The socio-economic status of representatives themselves
We argued that representatives have at least three opportunities to access lower SES perspectives. They can be lower SES themselves, they can be raised in a lower SES family, and they can have lower SES partners or friends. We start by examining whether representatives in our dataset from 13 countries are representative of citizens in terms of their education and class. Figure 1 displays the cross-system averageFootnote 3 as a summary overview; the first pane shows education level, and the second shows social class. Figure 2 (education) and Figure 3 (class) show the countries individually, where we each time compare representatives from a specific political system and their intimate ties with their country’s citizens. The bars each time show the distribution of educational attainment and social class of three groups: (1) citizens (red bars); (2) the total intimate network of representatives, adding up the parents and partner/friends (green bars); and (3) representatives themselves (blue bars). Comparing the red and blue bars in Figure 1 shows that representatives (blue) possess much higher levels of education and belong to a much higher social class than the people they represent (red). This replicates existing work, finding that representatives predominantly come from the more affluent and higher SES population segments (Bovens and Wille Reference Bovens and Wille2017; Carnes and Lupu Reference Carnes and Lupu2023a). In the typical country of our sample, the lower SES segments of the population are hardly represented by representatives: 1% of representatives define themselves as lower class, while 12% of citizens do, and 1% of representatives have no or primary education, compared to 17% of citizens. Among the more than thousand politician respondents, four representatives identify as lower class and eighteen as working class. Obviously, strictly speaking, this is a ‘wrong’ answer since their income and population clearly imply a higher class status. We do not know why these twenty-two politicians define themselves as belonging to the lower classes. Most likely, they consider the class they previously originated from and not their current social class. In Figure 1, the well-known picture emerges, representatives are very privileged in terms of education and class.
Figure 1. Distribution of education level and social class for citizens, representatives’ total ties, and representatives themselves (average of all 13 countries)(in %).
Figure 2. Distribution of education level for citizens, for representatives’ total intimate social ties, and for representatives per country (in %).
Figure 3. Distribution of social class for citizens, for representatives’ total intimate social ties, and for representatives per country (in %).
Figures 2 and 3 show that these findings are robust across all 13 countries. In every system, we observe the same pattern. Wherever we look, the red bars (citizens) exceed the blue bars (representatives) on the lower SES side. It is important to observe that the pattern also applies to social class since we use a subjective social class measure that may vary across countries (Evans, Stubager, and Langsæther Reference Evans, Stubager and Langsæther2022). Hence, our findings are entirely in line with what we already knew from a host of previous work. Representatives are not at all representative of citizens. This reinforces confidence in the measurement strategy we use here.
Comparing the SES of representatives’ intimate ties with the population (proportional benchmark)
Our aim with this paper is to go beyond existing knowledge by mapping the SES of representatives’ intimate ties and comparing these with the population. In this section, we show that politicians’ ties are indeed more descriptively representative than they themselves are. However, representatives’ intimate contacts still miss our maximal benchmark by a large margin; their ties strongly deviate from the population at large in the higher SES direction. We employ the data presented in the same Figures 1–3, using the maximum, proportional benchmark to compare between representatives’ intimate ties (green bars) with citizens’ SES (red bars).
Let us start by showing how politicians’ intimate ties are not descriptively representative of the population at large. First, considering education and revisiting the average country in the first pane of Figure 1, we see that contacts in representatives’ networks (green bars) in a typical country are clearly more privileged than the general population (red bars). For instance, the difference between the share of higher-educated social ties (66%) and the share of higher-educated citizens (44%) has a magnitude of factor 1.5. Examining the lower SES categories shows that our proportional benchmark is clearly not met. The number of ties with no/primary education (9%) or secondary/vocational education (25%) is much lower than in a typical population (with respectively 17% and 39%). In Figure 2, we observe that these patterns are largely consistent across countries. Of all 28 comparisons (14 political systems × 2 lower SES education levels), only 2 reveal another pattern.
Second, considering social class in the second pane of Figure 1 shows that ties are not representative in terms of social class either. Higher social class contacts are strongly overrepresented among representatives’ intimate ties. For instance, representatives consider 7% of their contacts to belong to the upper class, compared to 2% in the population, a difference with a magnitude of a factor of 3.5. The opposite is obvious as well, as shown by comparing the shares of ties with lower (3%) or working class individuals (13%) with the presence of those groups in the population (12% and 30%, respectively). Again, our maximum benchmark is far from being met. The skewed class pattern in ties is consistent in all individual countries except for the small and rich country of Luxembourg, where we surveyed the lowest number (n = 21) of representatives.
Still, although politicians’ intimate ties are not descriptively representative of the population, they are more representative of the population in terms of SES than politicians themselves. Examining Figure 1, both for education and class, we observe that the green bar depicting all intimate ties is consistently situated between the red bar representing the citizens and the blue bar representing the representatives themselves. The figure suggests that representative contacts are situated about halfway between the higher representative SES and the lower citizen SES. Wilcoxon Rank Sum tests show significant differences between politicians’ social ties and politicians with substantively large effect sizes for education (W = 3241416, P < 0.001, r = 0.61) and social class (W = 3092998, P < 0.001, r = 0.68). Hence, politicians’ intimate social ties are more likely to be lower educated and lower social class than politicians themselves are.
Figures 2 and 3 show that these findings are remarkably consistent across countries. The green tie bars in almost all countries are situated between the red citizen and the blue representative bars. For educational attainment, we observe this pattern to materialize in 39 out of 42 comparisons (14 systems × 3 education levels) – and for social class in 64 out of 70 comparisons (14 systems × 5 class levels). Differences between representatives and social ties are significant in all countries for class (all effect sizes are substantially large, ranging from 0.60 in Israel to 0.78 in Czechia), and in 11 out of 13 countries for education, with the notable exceptions of the Netherlands and Norway (all effect sizes are substantially large, ranging from 0.56 in Norway to 0.72 in Luxembourg). A full breakdown of these results by country can be found in Online Appendix H. In sum, while representatives’ ties do not at all meet our maximum benchmark and are very much situated in the higher SES segments of the population, we do see that the higher SES skew is less pronounced for representatives’ ties in comparison to their own SES.
Disaggregating representatives’ intimate networks (proportional benchmark)
The SES of representatives’ entire intimate network, being situated between their own (higher) and the population’s (lower) SES, leaves open the question of whether there are differences across the three types of ties that we grouped together in the analyses above. Do politicians’ parents, partners, and friends differ from each other in terms of SES? In this section, we will show that politicians’ parents are much more likely to be lower SES and more representative of the population than their partners and friends. Figures 4–6 below decompose representatives’ total intimate networks (the green bars in the figures above) into their three constitutive categories: parents (purple bars), close friends (gray bars), and partner (brown bars). For parents, we follow the dominance rule explained above. We code both parents together as higher educated or higher social class when a single parent is reported as such. Figure 4 displays the cross-country average as a summary overview; its first pane looks at education level, the second at social class. Figure 5 (education) and Figure 6 (class) show evidence from each country individually.
Figure 4. Distribution of education level and social class for representatives’ intimate social ties split up by group (parents, close friends, and partner) (average of all 14 systems)(in %).
Figure 5. Distribution of education level for representatives’ parents, close friends, and partners per country (in %).
Figure 6. Distribution of social class for representatives’ parents, close friends, and partners per country (in %).
Figure 4 shows that representatives’ parents have a lower SES and, therefore, form a better descriptive SES match with the population than their friends or partners. The two latter categories – partners and friends – look very similar in terms of SES. For instance, in the first pane, the share of parents with none/primary education is 13%, compared to 1–2% among the partner/friends; 30% of representatives’ parents received only secondary or vocational training, as compared to 18% of their close friends, and 19% of their partners. Similarly, for social class and in the second pane, we observe that 4% of parental ties are in the lower social class category, compared to 1% of close friend ties and 1% for partners; 19% of parents are categorized as working class, as compared to only 7% of the close friends and 6% of the partners. Kruskal-Wallis tests confirm the differences between tie groups regarding class (H(2) = 140.25, P < 0.001) and education (H(2) = 243.16, P < 0.001). Post hoc tests show that partners and close friends do not significantly differ from each other regarding education, but they both significantly differ from representatives’ parents. Concerning social class, parents, partners, and close friends all significantly differ.
Unpacking differences between political systems, we turn to Figures 5–6. Again, they reveal a great deal of consistency across political systems. Wherever we look, representatives’ parents shown in purple bars are much more likely to offer a lower SES perspective to representatives than their close friends depicted in the gray or the partner in the brown bars. Of all 28 comparisons (14 political systems × 2 disadvantaged education levels), only two cases reveal the opposite pattern for education, and the pattern holds for 21 out of 28 cases for class.
Representatives’ access to lower SES individuals through their intimate network (threshold benchmark)
So far, we showed that our demanding maximum benchmark for full proportional descriptive representation is largely unmet by representatives’ social ties. In the next step, we significantly lower the bar and proceed with a much more conservative test. We examine the share of representatives that reach the minimum threshold by having at least one lower SES tie among their intimate ties. Lower SES here means those who did not follow higher education or those reported as belonging to the lower or working class. We will show that even in some of the world’s most democratic, wealthy, socially mobile, and equal countries, many representatives do not meet that benchmark regarding education and a vast majority of them do not meet it regarding social class.
Table 1 reports a country-by-country summary of the share of representatives with at least one lower SES tie among their different types of intimate contacts. To compare, the table contains the share of lower SES citizens in the respective populations. Countries are ranked from high to low based on how many representatives have at least one lower SES tie in their intimate network (politician all ties column).
Table 1. Share of representatives (in %) with both parents classified as lower SES, a lower SES partner, at least one lower SES close friend, or meeting at least one of these conditions among all their intimate ties, compared to the percentage of lower SES citizens in the population
Looking at the average across countries (last line), 55% of citizens are lower educated (see Citizens column), while 58% of representatives have at least a single lower-educated actor in their intimate network (see Politician all ties column). So more than half of the representatives have access to at least one intimate tie that can possibly provide them with an understanding of lower-educated experiences; inversely, a small half of the representatives (42%) do not have any lower-educated tie among their intimates. In Czechia (80%), the share of representatives with at least one lower SES tie is highest, while in the Netherlands (32%) it is lowest.
While one may optimistically conclude that quite a lot of representatives can be in touch with lower-educated citizens through their strong tie networks, the same does not apply to their access to lower-class citizens. Overall, only 31% of the representatives report having at least a single lower-class tie among their intimate contacts, while 42% of the citizens report belonging to the lower classes. This means that more than two-thirds of the representatives (69%) declare that they do not have a single lower-class person in their intimate network. Norway (52%) and Australia (39%) are performing best with relatively large shares of representatives reporting to know at least one lower class person very well but even in those countries the comparison with the shares of lower class individuals in the population (54% and 56%, respectively) sketches a bleak picture of lower class representation in the networks of representatives.
Comparing our findings regarding class and education, it becomes clear that politicians are much more likely to have lower educational ties than lower-class ties. In all countries, there are more representatives with at least one lower-educated tie than with at least one lower-class tie. The difference in percentage points ranges from 12 in Israel to 56 in Luxembourg, with a difference of 27 in the average country. This difference is statistically significant when considering all representatives together (X2 = 166.64, P < 0.001), and in 9 out of 14 country cases. The complete results are provided in Online Appendix I.
To summarize our assessment of the threshold benchmark, even when we look at the very minimum benchmark of having at least one lower SES tie in their network, many representatives do not meet that benchmark regarding education and a vast majority of them do not meet it regarding social class. These findings paint a grim picture, even more so, provided that the countries in our sample are among the most democratic, socially mobile, rich, and equal in the world. Even in these very likely cases of strong lower SES representation, the minimum benchmark is predominantly not met.
Which politicians have lower-SES intimate ties?
While our aggregate findings may have painted a grim picture, there may be differences between individual politicians. As argued above, based on the literature, ideological left-right self-placement of representatives can be expected to be predictive of the SES of their intimate ties (Carnes and Lupu Reference Carnes and Lupu2015; Bukodi et al. Reference Bukodi2024). Also, representatives’ own SES could predict the SES of their friends and partners later acquired in life, given how ubiquitous homophily is among citizens (McPherson et al. Reference McPherson, Smith-Lovin and Cook2001). In our analysis, we compare representatives to citizens to test whether homophily and left-right effects are more outspoken among representatives than among citizens. As Kertzer (Reference Kertzer2022) argues, comparing citizens to politicians can help us understand whether politicians’ characteristics are due to basic compositional differences or due to their own specific position in society that comes with the job. In our case, this comparison is especially relevant regarding homophily. Are higher SES politicians similarly insulated from lower SES citizens in their intimate network as higher SES citizens are?
We run several logistic regressions regressing at-least-one-lower-SES-tie, dummies from Table 1 on politicians’ and citizens’ own education and on their parents’ social class, as well as on their left-right self-placement while controlling for age and gender and including country-fixed effects (the full models are reported in Online Appendix J). We operationalize our main independent variables as follows. We employ a dummy for education indicating whether a representative is lower educated or not. Our approach differs when testing social class homophily; here, we use a dummy variable measuring whether both representatives’ parents are lower social class or not. There are barely any representatives who define themselves as belonging to the lower class (only 22 do so), and, hence, there is not enough variation in representatives’ own social class SES to use it as a sensible independent variable. Left-right placement is operationalized on a scale from 0 to 10, where 0 means left and 10 means right. We control for age and gender, both variables are dummies with male vs. female and below vs. above the median age in the population of representatives – this raw categorization of age is needed to safeguard the anonymity of our representative respondents. For citizens, we draw on the survey described above. The variables are operationalized in the same way as for the politicians, except for the age control, which is based on citizens’ year of birth.
The results of our models provide strong support for the existence of a homophily mechanism. Representatives with a lower education themselves and representatives from disadvantaged social class households are much more likely to have a disadvantaged actor among their two closest friends or partner. Indeed, compared to highly educated representatives, lower-educated representatives have 4.1 times the odds of having a lower-educated partner or close friend (P < 0.001). However, the effect for citizens is much larger than this effect for representatives. Lower-educated citizens have 8.5 times the odds of having a lower-educated partner or close friend compared to higher-educated citizens (P < 0.001). The results for class are very similar. Representatives from lower social class households have 3.1 times the odds of having a lower social class partner or close friend (P < 0.001) compared to those from higher social class households. Running the same analysis for citizens, we find that they have 6.9 times the odds of having a lower-class partner or close friend (P < 0.001). Our results demonstrate that higher SES politicians are much more insulated from lower SES friends or partners than higher SES citizens. To ease the interpretation of our findings, we plot the corresponding predicted probabilitiesFootnote 4 of having a disadvantaged class or education partner/friends in Figure 7.
Figure 7. Predicted probabilities of having a disadvantaged education or class partner/friends with 95% confidence intervals.
The predicted probability of having one lower educated partner or close friend is 27% for higher-educated representatives (95% CI [0.18, 0.38]), whereas it is 60% for lower-educated representatives (95% CI [0.42, 0.76]). These probabilities are much higher for citizens. Their probability of having one lower-educated partner or close friend is 47% for higher-educated citizens (95% CI [43, 51]), and 88% for lower-educated citizens (95% CI [86, 90]). Again, the dynamics are similar for class; we find that the predicted probability of having one lower class partner or close friend is 8% for representatives originating from a lower middle class, middle class, or upper class household (95% CI [0.04, 0.16]), and 23% for representatives originating from a lower or working class household (95% CI [0.11, 0.40]). For citizens, we find that the probability of having one lower-class partner or close friend is 25% for higher-class citizens (95% CI [0.21, 0.28]), and 69% for lower-class citizens (95% CI [0.65, 0.73]). The main takeaway from Figure 7 is that higher SES citizens are much more likely to have a lower SES partner or close friend than higher SES politicians.
In contrast to our results concerning homophily, we do not find strong evidence for left-right differences regarding politicians’ intimate ties. We only find significant and robust results for the class of politicians’ parents. For each one-point increase in left-right self-placement, the odds of having lower-class parents decrease times 0.88 (P < 0.01). We find a similar but weaker effect for citizens; for each one-point increase on the left-right scale, the odds of these citizens having lower-class parents decrease times 0.98 (P < 0.05). So, more left-wing is associated with slightly lower-class parents. To ease the interpretation of these coefficients, we calculate predicted probabilities in Figure 8.Footnote 5
Figure 8. Predicted probabilities of having lower-class parents by left-right self-placement with 95% confidence intervals.
For politicians, the predicted probabilities of having lower-class parents range from 17% for those identifying as most left-wing with a score of 0 on the left-right scale (95% CI [0.07, 0.31]) to 6% for those identifying as most right-wing with a score of 10 (95% CI [0.03, 0.12]). The predicted probabilities of citizens having lower-class parents show that the effect of left-right self-placement is smaller than the one for politicians, with predicted probabilities ranging from 43% for the most left-wing citizens (95% CI [0.38, 0.47]) to 38% for the most right-wing ones (95% CI [0.34, 0.42]). Left-right self-placement does not reliably predict the education level of politicians’ intimate ties, nor does it predict the class of politicians’ close friends and partners. The significant odds ratio for having at least one lower-class intimate tie among parents, friends, and partners (OR = 0.90, P < 0.05) is not robust after dropping Wallonia from the dataset.
We conducted several robustness checks reported in Online Appendix K. First, we reran the models with a different specification of our dependent variable. Concretely, we employ a linear scale accounting for how highly educated or higher class a representative’s intimate ties are. Second, we test for the robustness of our homophily effect by excluding representatives’ partners from the dependent variable. Third, we drop radical right and green politicians from our dataset to examine whether they do not suppress left-right differences. Finally, to ensure that the findings reported above are not driven by one country, we dropped one country at a time from our models. All robustness checks are fully in line with the results reported here, except for left-right not significantly predicting lower class ties in politicians’ overall intimate networks after dropping Wallonia.
In sum, our results suggest that those representatives who could profit most from having a lower SES tie among their friends and partners are not very likely to have one, while those who already have exposure to lower SES through their own education and/or parental class background have more lower SES individuals among their friends and partners. Higher SES citizens turn out to be much more likely to be in contact with lower SES citizens than politicians who also report to be higher SES. Additionally, politicians identifying as more left-wing are only slightly more likely to have lower-class parents, but they are not more likely to have other disadvantaged SES intimate ties in terms of education and class.
Conclusion
We set out to examine the socio-economic status of representatives’ intimate networks. The underlying idea was that representatives may not only share experiences with lower SES groups by belonging to those groups themselves but also by virtue of the SES of their most intimate contacts. Intimate social ties are very likely to shape representatives’ experiences. Thus, our network approach answers scholarly calls for a more granular and comprehensive measurement of descriptive representation as driven by shared lived experiences that extend beyond the descriptive traits of representatives themselves (Celis and Erzeel Reference Celis, Erzeel, Rohrschneider and Thomassen2020; Allen Reference Allen2022; Xydias Reference Xydias2023; Funk and Hinojosa Reference Funk and Hinojosa2023). Examining the intimate ties of representatives serves as a particularly valuable complementary measure for highly underrepresented groups such as the lower educated and lower class (Bovens and Wille Reference Bovens and Wille2017; Carnes and Lupu Reference Carnes and Lupu2023a). We ventured into an uncharted domain by examining the SES of the intimate contacts of representatives – their two parents, their partner, and their two closest friends in 13 countries.
We find that the intimate social ties of representatives cannot be considered descriptively representative of their respective populations. This overall pattern applied to all 13 countries examined in this study. We consistently observe a skew in the SES composition of elected representatives’ networks, with a notable overrepresentation of individuals with higher SES backgrounds. Almost none of the 13 countries in our dataset met the maximum proportional benchmark. Results of the minimum threshold benchmark point are in the same direction. When considering the proportion of representatives with at least one lower-educated or lower-class citizen among their intimate contacts, we again observe a pronounced skew in the SES composition of representatives’ networks. Results are a little better for education than they are for class, but many representatives are almost entirely insulated from lower SES contacts in their most intimate circle.
To make things worse from a descriptive representation perspective, we clearly identify a homophily mechanism. Representatives who personally experienced a lower socioeconomic background or upbringing are much more inclined to have lower SES close friends or partners. Conversely, the vast majority of representatives who are not experiencing lower SES or do not come from better-off families are much less likely to develop lower SES social ties. In fact, our evidence suggests that higher SES politicians are even more insulated from lower SES ties in their intimate networks than higher SES citizens at large are. Higher-class or university-educated citizens are more likely to have a lower SES partner or close friend than the large majority of politicians who come from higher-SES backgrounds. This insulation further undermines the potential of politicians’ social ties to ‘compensate’ for the absence of lower-SES perspectives. Taken together, these findings erode the potential of social ties to compensate for representatives’ own higher SES and underscore the importance of politicians who have lower SES themselves.
Our study also highlights that the different types of intimate ties are not entirely similar in terms of SES. Representatives’ family background is more representative of the population than the close friends and partners they acquire later in life. Friends and partners, we find, are extremely similar regarding their SES. These findings support the idea that representatives in our country sample might have enjoyed some social mobility but confine themselves to the most privileged members of society due to the intimate social ties they acquire later in life.
Clearly, politicians’ social ties do not reflect broader societal diversity in terms of SES. But one might have assumed that left-wing politicians, given their ideological commitments, maintain closer ties with disadvantaged groups. Surprisingly, however, we find limited evidence that left-right differences predict whether politicians form intimate ties with lower-SES individuals. While left-wing politicians are more likely to come from lower-class parental backgrounds – echoing UK data that Labour ministers are more often from working-class families than their Conservative counterparts (Bukodi et al. Reference Bukodi2024) – this pattern does not extend to the class of their partners or close friends acquired later in life. Moreover, no left-right differences emerge in educational backgrounds, aligning with earlier findings that left-party politicians no longer possess markedly lower educational credentials compared to those on the right (Bovens and Wille Reference Bovens and Wille2017).
Are these results good or bad news for substantive and symbolic representation? At first sight, they are bad news. From the perspective of substantive representation, the higher SES character of representatives’ networks comes on top of their own SES privilegedness and most likely further acerbates representatives’ tendency to cater to the interests and preferences of the better off. This effect has been observed in previous studies regarding the influence of parents in the United States (Burden Reference Burden2007; Grumbach Reference Grumbach2015). From the perspective of symbolic representation, our findings partially validate the common perception that representatives are confined to an elitist bubble (Coleman and Moss Reference Coleman and Moss2023). This is most evident when we examine their partners and closest friends.
Yet, on the other hand, we do find that representatives’ networks are considerably more diverse in terms of SES compared to their own socio-economic status. This is especially true when we compare representatives’ SES to their parents’ SES. Hence, our findings suggest that for some representatives, their intimate ties provide them with unique insights into lower SES perspectives that they lack through their own experience. How and whether these insights translate into effects on symbolic and substantive representation are beyond the scope of this paper and are left for future research to unpack.
Future research could also go beyond our measurements of subjective social class and education. Occupation, social status, and income are also relevant to examine. But since these other dimensions of SES are known to be correlated with education and class (Hout Reference Hout, Lareau and Conley2008), we contend that our two measures tell us a good deal about the socio-economic status of representatives’ intimate contacts. Moreover, our choice to focus on representatives’ two closest friends opens up avenues for future research. Although people have rarely more than five most intimate ties (almost always including the partner and very often family members, see Dunbar Reference Dunbar2021), studying only the two closest friends remains to some extent an arbitrary cut-off point. It is possible that a third close friend could be lower SES and strongly contributes to that representatives’ access to low SES people’s experiences.
We identified a consistent pattern across all 13 Western countries examined, but to what extent can we generalize our findings from 13 Western countries? The countries in our sample had varied political systems, but they all are affluent. Countries in our sample have more representatives with a working-class background and fewer billionaires in politics than most others (Carnes and Lupu Reference Carnes and Lupu2023a; Krcmaric et al. Reference Krcmaric, Nelson and Roberts2023). They are also among the most egalitarian, highly educated, socially mobile, wealthy, and democratic in the world (World Economic Forum 2020; Chancel et al. Reference Chancel2022; Nord et al. Reference Nord2024). We think that it is telling that even in these countries, we see a serious skew in the SES composition of representatives’ networks. If anything, we expect that our findings would even be more pronounced in other countries. Even in some of the countries where we expect to find relatively unbiased networks, representatives’ intimate social ties are poorly representative of lower-educated and lower-class citizens.
Wrapping up, we believe that our study made a substantive leap forward in the search for the mechanisms explaining unequal representation. Representatives are privileged themselves, which we already knew, but they are surrounded by people in their intimate network with whom they interact intensely and frequently, who are relatively privileged as well. This likely has implications for how representatives relate to low socio-economic status groups and for how they are perceived by these groups in turn.
Acknowledgments
We thank the entire POLPOP II team for their feedback and their efforts in gathering the data this paper draws upon. Many more people than just the authors of this piece have contributed to data-gathering. We also want to thank Silvia Erzeel, Julie Sevenans, and Rozemarijn van Dijk for their valuable comments on earlier versions of this manuscript.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1475676525100042.
Open access
