Introduction
Political discussions occur frequently at workplaces. Unlike most other social contexts, in which individuals self-select into ideologically homogeneous environments, we typically do not choose whom we work with. Employees are therefore likely to be exposed to different political perspectives when interacting with their colleagues in the lunchroom, during meetings, and by the coffee machine, raising workplaces as a modern equivalent to ‘town squares’ (Mutz and Mondak Reference Mutz and Mondak2006). Yet, due to previous limitations in data collection practices, little is known about how such workplace socialization affects political attitudes. Previous research lacks data either on people’s colleagues, or on their political attitudes. This empirical gap is growing in importance as workplaces in Western democracies are increasingly sorted by education, which reduces differences in degrees among colleagues (Consiglio and Kurer Reference Consiglio and Kurer2025). In this article, I seek to address the social dimension at work by examining how the educational composition at workplaces shapes political attitudes.
Education represents the central political cleavage in contemporary politics (Ford and Jennings Reference Ford and Jennings2020; Garritzmann Reference Garritzmann2025; Marks et al. Reference Marks, Attewell, Hooghe, Rovny and Steenbergen2023), where individuals with lower levels of education tend to express culturally conservative views, while those with higher levels of education endorse culturally liberal values. I propose two pathways through which educational networks at the workplace may shape political attitudes. The first centers on processes of social identity and political mobilization. In this view, the clustering of individuals with similar backgrounds increases their sense of group belonging and perception of shared political interests. This hypothesis expects that an individual’s own educational background interacts with the educational environment of their workplace and polarizes political attitudes. Alternatively, the workplace is an environment of local social integration in which individuals are receptive to subtle cues, the ‘background noise’, about politically accepted views among their colleagues and conform to these in a desire to belong. From this perspective, educational homogeneity at work limits individuals’ exposure to alternative political perspectives and makes everyone align with the dominant workplace norms, regardless of their own background.
This study contributes to a wider debate on the role of education in shaping political attitudes on the socio-cultural dimension of political competition (Häusermann and Kriesi Reference Häusermann and Kriesi2015; Stubager Reference Stubager2008). To understand this relationship, much research focuses on the individual-level effects of educational attainment on political attitudes (for example, Marshall Reference Marshall2019; Scott Reference Scott2022). Far less is known about education’s social dimensions. A bourgeoning literature suggests that educational network homogeneity among close ties fosters ‘social closure’ (Attewell and Zollinger Reference Attewell and Zollinger2025; de Jong and Kamphorst Reference de Jong and Kamphorst2025), in which political attitudes amplify through similar mobilization processes that took place when unions clustered workers or the church religious people. However, the possibility of such a mobilization at the workplace may be weaker than in private contexts, as it is a setting where individuals have incentives to be less vocal when discussing politics and are more likely to encounter political and social differences. This could particularly be the case with the education divide, which is characterized by relatively weak educational identifications at the individual level, and fewer direct political appeals from external actors (Zollinger and Attewell Reference Zollinger and Attewell2026).
To study these dynamics, I leverage a unique empirical setting with population-wide administrative data linked to a representative individual-level election survey from the Swedish National Election Study (SNES) in 2022. The dataset tracks individuals’ attitudes and labor market trajectories while identifying their colleagues through official records. The set-up further enables a first-difference panel analysis on the subset of respondents who also participated in the 2018 election survey to estimate the effect of changes in workplace composition on changes in attitudes. The empirical design provides a rare opportunity to study the connection between workplace composition and political attitudes – a theoretically important relationship that is difficult to test in practice (Godechot et al. Reference Godechot, Tomaskovic-Devey, Boza, Henriksen, Hermansen, Hou, Jung, Kodama, Křížková, Lippényi, Melzer, Mun, Sabanci, Thaning, Apascaritei, Avent-Holt, Bandelj, Baudour, Cort, Elvira, Hajdu, Kanjuo-Mrčela, King, Penner, Petersen, Poje, Rainey, Safi and Soener2024). The paper thereby adds to the debate on socialization versus selection effects by examining how the workplace composition shapes political attitudes beyond individual education and labor market sorting. Although it is possible to study social homogeneity at the workplace across various social groups and attitudinal dimensions, this study focuses on divisions based on educational level, as the dominance of the ‘second dimension’ of political competition makes it the most influential determinant of voting behavior today (Häusermann and Kriesi Reference Häusermann and Kriesi2015; Hooghe and Marks Reference Hooghe and Marks2025).
The findings lend robust support to the view that the workplace functions as a meaningful site of political preference formation. A higher proportion of university-educated colleagues is independently associated with more liberal positions, suggesting that all individuals are responsive to the dominant norm at their workplace. Further analyses show that the effect seems to develop during the time spent in the labor market, is robust under a variety of specifications, and is contingent on the colleagues’ educational field background.
By bringing together conclusions from research on educational networks and applying them in the social context of the workplace, the paper’s contributions are threefold. First, it addresses the potential limitations and observable implications of a political mobilization of education groups through workplace interactions (Stubager Reference Stubager2010). Second, the study provides a new perspective on the influence of an individual’s labor market position: we are shaped not only by our specific work tasks (Kitschelt and Rehm Reference Kitschelt and Rehm2014) or employment structure (Rovny and Rovny Reference Rovny and Rovny2017) but also by the people around us. Third, the paper shows that while workplaces may serve as sites for countervailing processes to educational polarization, structural patterns of educational sorting in the labor market contribute to a political polarization of education groups at the aggregate level.
In the remainder of this paper, I introduce the theoretical framework that distinguishes between interactive and additive effects of the educational environment before situating the paper in the wider literature. The methods section addresses the empirical challenges of studying social networks in the labor market. The following section presents the main cross-sectional results and thereafter proceeds with a first-difference regression analysis, testing the linearity of the results, and varying the model specifications. Finally, I place the findings within the broader social science literature of the education divide and consider the implications of the results in light of structural labor market trends.
Attitudinal Implications of Workplace Interactions
This paper grapples with the question of how the educational composition of the workplace shapes political attitudes. In the absence of a contextual influence, and assuming no systematic selection into workplaces, the attitudes of university graduates and non-graduates will not vary with the composition of their colleagues. However, I propose two channels through which attitudes may change depending on the social dimension at work. First, the ‘polarization hypothesis’ expects that educational homogeneity at a workplace strengthens political group consciousness among likeminded people and interacts with an individual’s own educational background to polarize attitudes. Second, a complementary explanatory model – the ‘liberalizing hypothesis’ – proposes that all employees attune to the dominant norms, drawing on theories of intrapersonal socialization and the human desire to belong.
The Polarization Hypothesis
The polarization hypothesis proposes that the educational environment interacts with individuals’ own educational background in shaping political attitudes. It builds on the notion that homogeneous social environments intensify in-group identification and boundaries towards an out-group, by activating education-based identities. Being surrounded by others with similar educational experiences and preferences would increase individuals’ awareness about their own group belonging and of political conflicts along educational lines, which can strengthen their group-specific attitudes. A university graduate in a highly educated workplace thereby increases their political distance from the less educated through an ‘us vs. them’ dynamic, and vice versa.
Political identities are central to this perspective of network effects and the mobilizing role of colleagues. Recent research suggests that low-educated individuals often construct a moral self-image as ‘hard-working’ in explicit contrast to ‘left-wing, urban people’ (Gidron and Hall Reference Gidron and Hall2017; Stubager Reference Stubager2009; Zollinger Reference Zollinger2024, 10). Accordingly, such sentiments would intensify when individuals are surrounded by others who confirm their worldviews, and contribute to articulating shared political interests and group conflicts. Conversely, highly educated individuals embedded in homogeneous networks with other university educated would reinforce a self-conception of being liberal and open-minded, while simultaneously expressing dismissive attitudes towards the ‘xenophobic’ less educated (Zollinger Reference Zollinger2024, 10). The central argument in the polarization hypothesis, therefore, is that work environments stratified by education confirm perceptions of intergroup difference between education groups, which reinforce their conservative or liberal attitudes.
The polarization hypothesis relates to how a bourgeoning literature views social network dynamics among close ties, where homogeneous friendship groups reinforce existing educational divides (Attewell and Zollinger Reference Attewell and Zollinger2025; de Jong and Kamphorst Reference de Jong and Kamphorst2025). It further reflects the historical mobilizing role of organizations such as the church and trade unions, which clustered social groups to foster group awareness and unify their political agenda (Bartolini and Mair Reference Bartolini and Mair1990). The hypothesis posits an interaction effect between an individual’s educational background and the educational composition of their workplace (Education × Educational environment). Attitudinal differences strengthen when individuals are embedded in environments with a large share of others with similar backgrounds and habitus, and weaken in heterogeneous environments, as illustrated in Figure 1. While the literature offers several perspectives on how polarization may develop, including dynamics related to majority–minority positions (Collier Reference Collier2001), I thus focus on a conservative specification that yields the most moderate prediction of polarization in the main paper (refer to Appendix C1 for alternative theoretical specifications and tests of the polarization hypothesis). This provides a stringent test: if polarization does not emerge under this minimal formulation, it is unlikely to emerge under more demanding theoretical assumptions.
Polarization hypothesis: Predicted pattern by coworker composition (Education x Educational environment).
Note: the figure illustrates the predicted attitudes of university graduates and non-graduates across varying workplace compositions. Attitudinal differences between the groups are greatest when the workplace is dominated by low-educated or high-educated employees. The labels ‘low’, ‘mixed’, and ‘high’ approximate increasing shares of university-educated employees and reflect the cut-offs used in the main analysis rather than strict theoretical thresholds.
Hypothesis 1: The cultural liberalism of both university graduates and non-graduates is reinforced when they work with people who share similar educational backgrounds.
The Liberalizing Hypothesis
On the other hand, the liberalizing hypothesis expects an additive effect of the workplace environment, irrespective of individuals’ educational backgrounds. Under this view, workplace interactions foster a general cohesion and attitudinal accommodation among employees but do not contribute to politicize the education divide. This hypothesis rests on the idea that the need to belong and to receive social approval is a key human motivation (Baumeister and Leary Reference Baumeister and Leary1995; Zuckerman Reference Zuckerman2005). Intrapersonal relationships convey both explicit and implicit cues about the social norms, and individuals can make attitudinal adjustments to maintain or improve their social standing (Apfeld et al. Reference Apfeld, Coman, Gerring and Jessee2023; Huckfeldt and Sprague Reference Huckfeldt and Sprague1995; Mendelberg et al. Reference Mendelberg, McCabe and Thal2017). The desire to be valued by our peers implies that attitudes will slowly change depending on what is respected and rewarded by the specific group. University graduates and non-graduates alike will thus adopt more liberal values as they move into more highly educated work environments, since they adjust to the general political climate in their workgroup. Hence, the educational composition shapes political attitudes as it changes the dominant norms at work, but it does not strengthen a political group consciousness or serve as a basis for a deeper mobilization of the education divide.
The challenges for a political mobilization through social network dynamics may be more pronounced in workplace interactions. It is more likely that individuals are exposed to divergent political perspectives at the workplace than among strong ties that are often shaped by pre-existing political similarities. Individuals may also be more cautious about drawing sharp boundaries towards others and less vocal about their preferences in work environments, making implicit cues more prevalent in this environment than among private ties. Such an interpretation is reflected in recent qualitative evidence on the socializing role of work by Kamphorst et al. (Reference Kamphorst, Hooghe, Marks and Wang2025). They highlight that working with people who have certain ideas may shape opinions, even if those ideas are conveyed through the general atmosphere and subtle signals, rather than openly talked about (Kamphorst et al. Reference Kamphorst, Hooghe, Marks and Wang2025, 35).
The lack of mobilizing effects from social interactions at the workplace could further be particularly applicable in the case of the education divide. This results from the absence of external organizations that promote education-based political mobilization, unlike the trade unions or religious institutions (Bartolini and Mair Reference Bartolini and Mair1990; Hooghe and Marks Reference Hooghe and Marks2025). Politicians also tend not to make appeals directly targeted at specific educational groups (Thau Reference Thau2019), reflecting the fact that most individuals tend to identify more strongly with other attributes than with their education (Zollinger Reference Zollinger2024). Taken together, both the bottom-up process of generating an education-based identity and the top-down process of group appeals and classifications may be weaker than for comparable social groups or in other social contexts (Westheuser and Zollinger Reference Westheuser and Zollinger2025).
Consequently, as the proportion of university-educated colleagues grows, so too does the likelihood of exposure to liberal political attitudes. Regardless of an individual’s own educational background, they will be equally influenced by the dominant workplace norms, but the gap between the education groups remains the same. Figure 2 illustrates this relationship (Education + Educational environment).
The liberalizing hypothesis: Predicted pattern by coworker composition (Education + Educational environment).
Note: the figure illustrates the predicted attitudes of university graduates and non-graduates across different workplace compositions. Attitudinal differences between the groups are the same at the different compositional levels. The labels ‘low’, ‘mixed’, and ‘high’ approximate increasing shares of university-educated employees and reflect the cut-offs used in the main analysis rather than strict theoretical thresholds.
Hypothesis 2: The share of highly educated at the workplace has an additive liberalizing effect on cultural liberalism.
The Workplace as a Site for Political Mobilization or Social Integration
Both hypotheses share the premise that heterogeneous networks moderate attitudes; non-graduates in an educationally mixed setting are more liberal than in low-educated environments, while university graduates in the same environments express comparatively more conservative views. The key divergence lies in the expectations for individuals situated in homogeneous workplaces. The polarization hypothesis anticipates stronger in-group identification and attitudinal polarization when individuals are surrounded by others with similar educational backgrounds. In contrast, the liberalizing hypothesis suggests that exposure to the dominant workplace norms exerts a local social influence that leads to gradual attitudinal adjustments for all. The employees’ educational backgrounds are therefore used as a proxy for their attitudes: if workers with different levels of education but similar attitudes work together, the social norms in the group are expected to remain the same. However, educational backgrounds are also linked to identity elements which could reinforce attitudinal transmittance (Stubager Reference Stubager2009), as highlighted in the ‘polarization hypothesis’.
Related Literature
This article combines insights from the broader literature on education and social networks, workplaces, and preference formation. Scholars suggest that the workplace is the most central context for everyday conversations for adults (Alt et al. Reference Alt, Jensen, Larreguy, Lassen and Marshall2022; Mutz and Mondak Reference Mutz and Mondak2006), which involves more cross-cutting political discourse than other social contexts and includes a larger number of individuals. For instance, Neuberger (Reference Neuberger, Auhagen and von Salisch1996, 272) observes that the workplace is a unique setting compared to families, voluntary associations or neighborhoods, as ‘Encounters and relationships are unavoidable: it is impossible to simply “keep out of the way” of certain people even if they have not selected each other’. Yet empirically we know more of the attitudinal effects of an individual’s occupation and employment structure than of the social dimension at work (Kitschelt and Rehm Reference Kitschelt and Rehm2014; Rovny and Rovny Reference Rovny and Rovny2017).
The seminal paper by Kitschelt and Rehm (Reference Kitschelt and Rehm2014), for example, studies the task structure of different occupations, in which individuals internalize the norms and logics associated with their work roles. They also consider collegial relationships through the lens of workplace hierarchies. However, attitudinal change stems from the individuals themselves – through the tasks they perform and their possible placement in a leadership position. More recent work shows that the characteristics of the people one meets in the workplace also matter. Andersson and Dehdari (Reference Andersson and Dehdari2021) leverage administrative data of ethnic workplace composition and find that contact with immigrants reduces support for the radical right at the precinct level, while Aggeborn and Andersson (Reference Aggeborn and Andersson2025) show an increase in the likelihood of becoming a politician based on their colleagues. This should come as no surprise since ample research in the intersection of political science and sociology shows that voters rely on cues from their surroundings when shaping their political attitudes and voting decisions.
As with the attitudinal effects of occupations, much work focuses on the individual-level processes underlying the link between higher education and liberal political attitudes. These relate to the development of cognitive skills, or socializing experiences, from university education (for example, Apfeld et al. Reference Apfeld, Coman, Gerring and Jessee2023; Stubager Reference Stubager2008). Critically, they also involve the influence of self-selection mechanisms prompting certain individuals to pursue university education as well as hold more liberal attitudes (Lancee and Sarrasin Reference Lancee and Sarrasin2015; Lindskog et al. Reference Lindskog, Gustafsson and Voytiv2025). However, the educational background of people’s close ties such as one’s friends also matters politically, beyond an individual’s personal university experiences (Attewell and Zollinger Reference Attewell and Zollinger2025; de Jong and Kamphorst Reference de Jong and Kamphorst2025; for a review, see Lindskog and Persson Reference Lindskog, Persson, Janmaat and Dijkstra2025). Even if education lacks a causal effect on attitudes during the time spent in education (Scott Reference Scott2022), education-based networks may thereby shape attitudes and have electoral consequences later in life (see also Persson Reference Persson2015 on the role of educational networks for political participation).
Taken together, this paper advances research on the social dimension of the education divide by focusing on workplace interactions. Individuals spend a large share of their hours awake at work, where they are exposed to structural shifts in collegial composition which shape the local social norms. The frequency of these everyday interactions, the potential for social mixing, and the subtle cues they convey are particularly characteristic of the workplace compared to other social ties. Still, there is no reason to expect that the influence of colleagues would not generalize to other social contexts involving relatively weak ties, to which individuals are similarly exposed and where they seek social belonging. In the next section, I proceed with the empirical strategy to study these dynamics.
Methodology and Data
It is methodologically challenging to estimate the attitudinal effects of social networks. There are uncertainties in identifying individuals’ actual networks as well as risks of confounding factors that shape both individuals’ social ties and their attitudes. An ideal research design would randomly assign individuals to different coworkers for an extensive period to imitate a real-world scenario. Obviously, such a design could not realistically be implemented, and we are therefore left with observational designs.
Past research on the link between education networks and attitudes let respondents indicate their four to five closest social contacts and their backgrounds (Attewell and Zollinger Reference Attewell and Zollinger2025; Mutz and Mondak Reference Mutz and Mondak2006; de Jong and Kamphorst Reference de Jong and Kamphorst2025). This set-up raises the possibility that it is the politically convinced and attitudinally extreme individuals that are drawn to individuals from similar backgrounds rather than the other way around. Relying on self-assessed ties further assumes that the respondents’ recollections about their social connections are correct, despite well-known limitations in human memory (Lim Reference Lim2008). What is at the top of respondents’ minds at the time of the survey – for instance, whether they recently attended a family gathering or a trip with friends – may further influence which social ties they consider their most important discussion partners.Footnote 1 On the other hand, the limited number of papers that examine the political effects of workplace composition using official records to identify coworkers focus on political participation (Aggeborn and Andersson Reference Aggeborn and Andersson2025), or use aggregated data at the precinct level to assess consequences in vote choices (Andersson and Dehdari Reference Andersson and Dehdari2021). Still, macro data obscure individuals and risk introducing ecological biases into the interpretations (Healy et al. Reference Healy, Persson and Snowberg2017).
The analysis below circumvents these problems by combining population-wide administrative data covering the full Swedish population with a representative election survey from the Swedish National Election Survey (SNES) in 2022. Information about each respondent’s workplace and labor market trajectory is provided through register data from Statistics Sweden. The governmental agency merged administrative data using the respondent’s unique workplace ID, allowing for verified socio-economic information of all other individuals working at the same location as the respondent. All employers are legally required to provide employer reports and information about their workplaces and employees. The informal sector in Sweden is very small by international standards, accounting for around 3 per cent of the workforce (Engzell and Wilmers Reference Engzell and Wilmers2025).
The linkage of the large-scale administrative micro data to the election surveys thereby generates close to complete information on the respondents’ workplace networks and their career trajectories over time. A subset of the respondents who participated in the 2022 election survey also participated in the 2018 election survey and answered identical attitudinal questions, thereby enabling a two-wave panel analysis. The panel analysis allows for the estimation of the effect of changes in workplace composition on within-individual changes in attitudes. The fielding of the surveys during election periods increases the likelihood that politics is a conversational topic at the workplace. Hence, the empirical approach relies on the use of granular administrative data on coworkers and workplaces, combined with high-quality data on respondents’ attitudes over time. The research design provides the strongest evidence to date on the effects of workplace networks on political attitudes.
Still, while the register data include clear advantages in identifying individuals’ complete workplace networks, they have limitations in testing the precise mechanisms through which such effects operate. The register data do not provide information on the relationships between the colleagues and the respondent, their colleagues’ attitudes, or other non-observable factors that are correlated with the educational composition at the workplace. While the workplace composition thus could reflect developments that are not possible to test directly, such as improvements in a firm’s economic conditions or changes in the organizational structure at the workplace that may exert an independent effect on the respondents’ attitudes, the cross-sectional analysis is complemented with a panel analysis. This analysis provides additional evidence of contextual workplace effects on attitudes using a harder empirical test.
The main analysis in the paper examines the conditional mean differences in cultural liberalism in various steps, including a closer investigation of the thresholds for when the educational composition influences attitudes. In a second step, I proceed with a first-difference regression analysis and test the linearity of the findings and the robustness of the results under various specifications.
The Composition of Workplaces
The workplace is identified at the most granular postal level. A workplace refers to each address, property, or group of properties where the company conducts its operations. Every active company has one workplace for each address. All individuals who have performed work and received any kind of financial compensation from the employer are connected to a workplace. For example, individuals working at a university are distinguishable at the department level and teachers by their school. The general rule is that an employee’s workplace is where they perform the majority of their work. This means that for employees working in an office, factory, store, or similar setting, the employer’s location is typically their workplace.Footnote 2 It is possible to identify the specific industry and similar workplace information, but the exact workplace ID is anonymized due to the low level of aggregation.
One challenge when studying cross-sectional data on the effects of the share of university-educated individuals in an individual’s context is that respondents’ own level of education is included in the aggregated measure of workplace composition (Persson Reference Persson2015). I therefore exclude each respondent’s own educational background before calculating the workplace’s composition in these analyses. This approach reduces the correlation between the two measures and helps equalize measurement error across workplaces of varying sizes, as the respondent’s own education would otherwise constitute a substantially larger share of the compositional measure in small workplaces than in large ones. However, it only has minor implications for the compositional measure (Figure A1).Footnote 3 , Footnote 4
As in most studies of social composition, variations within the units of analysis create a gap between the predicted and actual social contacts. The frequency and intensity of social ties depend on the different working groups within the companies, the extent to which there is freedom to choose your own work team, and the hierarchies between different roles. Although the data do not identify different working groups within the workplaces, it does capture the respondents’ complete set of colleagues. As 50 per cent of all workplaces have fewer than 72 employees, and 25 per cent have fewer than 21 employees, the unit of analysis is thereby on average smaller than the usual network sizes (Lim Reference Lim2008).Footnote 5 As highlighted by Godechot et al. (Reference Godechot, Tomaskovic-Devey, Boza, Henriksen, Hermansen, Hou, Jung, Kodama, Křížková, Lippényi, Melzer, Mun, Sabanci, Thaning, Apascaritei, Avent-Holt, Bandelj, Baudour, Cort, Elvira, Hajdu, Kanjuo-Mrčela, King, Penner, Petersen, Poje, Rainey, Safi and Soener2024), most employees know each other on at least a basic level and share a lunchroom.
The Swedish Exception: Labor Market Trends in a Comparative Perspective
Western democracies are experiencing structural shifts in their labor markets. Two trends are especially salient. First, major developments in automation contribute to a decline in routine middle-skilled jobs, and labor market opportunities concentrate in either low- or highly skilled jobs (Autor and Dorn Reference Autor and Dorn2009; Mijs and Roe Reference Mijs and Roe2021). Second, the proportion of highly educated individuals continues to rise, which increases the likelihood that university graduates are embedded in educationally homogeneous workplaces.
Sweden differs from several advanced democracies by following a trajectory of a general occupational upgrading, with a stable job growth in low-skilled occupations and a very strong job growth in high-skilled occupations (Oesch and Piccitto Reference Oesch and Piccitto2019). Likewise, Sweden maintains a highly compressed wage structure, making low-skilled labor relatively expensive and high-skilled labor comparatively cheap. The economic incentives to hire highly educated workers contribute to the country’s extensive educational upgrading and university-educated employees are today working in all parts of the occupational structure. Sweden has also been characterized by relatively high unemployment figures in recent decades, primarily for individuals with lower levels of education, following the supply of more educated workers (Berglund and Ulfsdotter Eriksson Reference Berglund and Ulfsdotter Eriksson2024). Figure 3 shows the percentage of the labor force working alongside a majority of university-educated coworkers (>66 per cent) across four decades. Both non-graduates and graduates are increasingly likely to have highly educated colleagues over time, but the trend is especially clear for university graduates themselves.Footnote 6
Percentage of the Swedish labor force working with a clear majority of university-educated colleagues (>66 per cent) between 1992 and 2022.
Note: the figure shows the share of individuals working in workplaces with a clear majority (<66 per cent) of university-educated colleagues, by university graduates and non-graduates. The figure is based upon Swedish population-wide administrative record and covers the full labor force working at workplaces with at least four employees. N varies between 3,620,107 and 4,496,728.
Main Variables
Respondents’ main variables are studied using SNES data. As an attitudinal measure of cultural liberalism, I combine two items: one on immigration ‘Allow more refugees to enter the country’, and one on authoritarianism ‘Impose stricter prison sentences’. These are dominant issues that define the ‘socio-cultural’ dimension in Western Europe today (Lancaster Reference Lancaster2022), and are items commonly used to study education divides (Attewell and Zollinger Reference Attewell and Zollinger2025; Lindskog and Oskarson Reference Lindskog and Oskarson2023; Stubager Reference Stubager2008). Both items range from 1 to 5, and the index is calculated as their mean (range: 1–5), where higher values indicate more liberal attitudes.Footnote 7 The combined index shows a high correlation between the items (α = 0.77).
A respondent is classified as highly educated (graduate) if she has completed at least three years of university education. Those who have completed vocational training or shorter periods of university education are classified as medium educated, while respondents whose highest level of education is upper secondary school or equivalent are classified as low educated. In the main analyses, I focus on a binary distinction between respondents who have a university degree or not, while presenting detailed variations in educational attainment at the individual level in robustness analyses (see Figure 7).
Mean cultural liberalism (1–5 liberal direction) by educational attainment and workplace composition
Table 1. Long description
The table presents mean cultural liberalism scores categorized by educational attainment and workplace composition. It has four rows and four columns. The columns are labeled Low share (0-33%), Mixed share (33-66%), High share (66%), and Total. The rows are labeled Non-graduates, Graduates, and Total. Each cell contains a mean score and a percentage value in parentheses. Row 1: Non-graduates, Low share (0-33%), 1.88 (58%); Mixed share (33-66%), 2.14 (28%); High share (66%), 2.33 (14%); Total, 1.96 (35%). Row 2: Graduates, Low share (0-33%), 2.30 (13%); Mixed share (33-66%), 2.60 (35%); High share (66%), 2.81 (52%); Total, 2.40 (32%). Row 3: Total, Low share (0-33%), 1.96 (35%); Mixed share (33-66%), 2.40 (32%); High share (66%), 2.71 (33%). The table notes significant differences at p 0.05 and p 0.1 levels, with specific markers indicating these differences.
Note: N = 4,384. *Notes a significant difference at p < 0.05 compared to the previous column within the same row (i.e., compared to the workplace category with a lower share of university educated), and † marks a significant difference at p < 0.1 compared to the row above within the same column (i.e., compared to the group with a lower level of individual education). Estimates are mean levels with share of the sample in parentheses.
The share of highly educated at the workplace is independently associated with liberal cultural attitudes
Table 2. Long description
A table with six models comparing the impact of individual and workplace education on cultural liberalism. The table has six columns labeled Model 1 to Model 6 and several rows with different categories. Row labels include Individual education, Workplace education, Individual ed. # Workplace ed., Constant, Workplace controls, Individual controls, R2, and n. Each model shows values for University graduate, Mixed (33-66%), High (>66%), Graduate # Mixed share, Graduate # High share, Constant, Workplace controls, Individual controls, R2, and n. For example, in Model 1, University graduate is 0.47 (0.04), Mixed is 0.31 (0.05), High is 0.47 (0.05), Graduate # Mixed share is 0.03 (0.11), Graduate # High share is 0.05 (0.11), Constant is 1.80 (0.03), Workplace controls are NO, Individual controls are NO, R2 is 11%, and n is 3,345.
*p < 0.1, **p < 0.05, ***p < 0.01. OLS regression models. The dependent variable, cultural liberalism, ranges between 1 and 5. Workplace clustered standard errors in parentheses and the models are weighted against the official election results. The table is based on data from SNES 2022 and administrative workplace records from Statistics Sweden. The dependent variable is a combined index of immigration and law and order attitudes, with higher values indicating more liberal views.
Information on employees at the workplace is drawn from the population-wide administrative data. Their educational attainment is coded using the Swedish SUN classification, which is linked to the International Standard Classification of Education (ISCED).Footnote 8 The main contextual measure captures the share of employees with university education and is divided into three key groups in the main analysis: low share of university educated (<33 per cent), mixed share (33–66 per cent), and high share (>66 per cent). The richness of the data further enables empirical testing of the thresholds for ‘the education effect’ at the workplace level, as is considered in detail (see Figure 8). Figure 4 illustrates the distribution of university-educated employees at higher levels across industries. The figure shows that average skill level varies substantially across different industries, being just above 10 per cent in construction and over 60 per cent in education, science, and law. Likewise, individuals with a doctoral degree only appear in certain industries and are almost fully absent in others.
Distribution of higher education degrees across industries.
Note: the figure shows the share of individuals with different levels of education by industries. Proportions sum to 100 per cent when including all employees. Bachelor’s degree corresponds to three years of university education and master’s degree to 4–5 years of university education. The figure is based upon Swedish population-wide administrative records linked to the Swedish National Election Survey.
Control Variables
At the individual level, gender is treated as a dummy variable where 1 equals being a woman. The respondents’ occupations are divided according to the ISCO scheme (International Standard Classification of Occupations 2008) and categorized into nine major occupational groups. Disposable income is logged to reduce earnings skewness and age ranges between 18 and 75 years. Only individuals who are actively participating in the labor market as their main daily activity are included in the sample. The analyses further control for whether the individual holds a fixed-term contract and whether they were born in Sweden. Table A3 shows descriptive statistics for the individual-level variables.
Since the distribution of highly educated employees is uneven across different types of workplaces and may coincide with other workplace factors that may shape attitudes, I control for various workplace characteristics in parts of the analysis. These include the sector, industry, geographic location, and workplace size, as well as the share of immigrants. The workplace sector is recoded into a public–private distinction.Footnote 9 The types of industries are generated based on the workplace SNI classification, an international system based on the EU’s industrial classification standard, NACE. I categorize the sixteen categories of industries into four main groups.Footnote 10 The type of municipality captures the location of the workplace, from being located in a municipality in a big city to a municipality in the countryside without a tourism industry. It is divided into three main groups: city, mid-size municipality, and countryside municipality. Similarly, I control for the share of individuals at the workplace who were not born in Sweden and include a logged measure of workplace size. Table A2 shows descriptive statistics for the workplace variables. The workplace characteristics explain 36 per cent of the variation in the educational composition, leaving 64 per cent unexplained when including the full set of controls.Footnote 11
Model Specification
To clarify the empirical specification used in full models in the main analysis, I estimate the following relationship:
$${Y_{iw}} = {b_0} + {b_1}{X_{iw}} + {b_2}{E_{w}} + {b_3}({X_{iw}}*{E_{w)}} + {b_4}{W_w} + {b_5}{C_{iw}} + {e_{iw}}$$
The dependent variable Y iw is an indicator of cultural liberalism for individual i in workplace w. The intercept is noted by b 0, and b 1 X iw represents the binary educational status of the respondent. b 2 E w denotes the categorical measure of workplace educational composition in three levels (low, mixed, and high). Thereafter follows an interactive term in which the individual’s education status is interacted with the educational composition at their workplace, b 3(X iw * E w ). I further include a vector of workplace controls b 4 W w and individual-level controls, b 5 C iw , as specified previously. Finally, e iw represents the error term. Standard errors are clustered at the workplace level.
Main Results
I start by comparing the average attitudes towards cultural liberalism for university graduates and non-graduates who work at workplaces with various shares of university-educated employees. This initial analysis does not account for any other characteristics of workplaces or individuals, but simply shows the bivariate relationship.
Table 1 shows a clear pattern that follows the expectations for the liberalization hypothesis: individuals working at workplaces with a higher share of university-educated colleagues tend to express more culturally liberal attitudes. This association holds across both levels of individual education. Notably, a non-graduate working in a highly educated work environment expresses views that are as liberal as a university graduate in a low-educated environment. This suggests that the workplace has the potential to moderate attitudinal differences between education groups. Moving forward, I introduce models that consecutively incorporate the workplace-level controls, followed by the individual-level controls.
Table 2 presents six OLS regression models examining the relationship between cultural liberalism and the share of university-educated individuals in the workplace, while adding controls for individual-level and workplace-level characteristics. Varying the inclusion of controls reduces the effect size of both the individual and workplace education, but makes little difference in the direction or significance of the results. In Models 1 and 2, both individual-level education and the educational composition of the workplace are significantly associated with more liberal attitudes. Compared to those without a university degree, respondents with a degree score around 0.47 points higher than those without, and individuals working in high-education workplaces score 0.47 points higher than those in low-education workplaces. In Models 3 and 4, workplace fixed effects are added, which slightly reduce the magnitude of the effects from the educational composition. In Models 5 and 6, additional individual-level controls are included, and the coefficients for university graduates and for working in a high-education workplace stay positive and statistically significant. The reduced coefficients for workplace education in Model 5 suggest that part of the association is explained by individual characteristics, such as income, occupation, and demographic traits, that are correlated with both liberal attitudes and the likelihood of working in a highly educated environment.
Notably, none of the interaction terms between individual education and workplace education are statistically significant, suggesting that the effects of contextual education do not differ across individual education levels. Overall, the results show that both individual education and the educational composition of the workplace are independently and additively associated with more liberal attitudes.Footnote 12 Table B1 in the Appendix provides additional reassurance that these results are unlikely to be driven by other observable compositional characteristics of the workplace, as controlling for the share of women, urban residents, and younger employees at the workplace does not alter the results.Footnote 13 In addition, Appendix C examines alternative specifications of the polarization hypothesis, including models emphasizing the degree of educational imbalance at the workplace and the respondent’s minority or majority status (Figure C1 and Table C1). These alternative specifications do not receive empirical support. Table C2 further replicates the main specification across three consolidated sector tiers; the results indicate that the effect is not solely confined to a particular sector.
Figure 5 shows the predicted attitudes across university graduates and non-graduates working in different educational compositions in their workplaces. The pattern clearly mirrors that of the ‘liberalizing hypothesis’. The association seems to be symmetrical for the educational groups, supporting the notion of social belonging as a shared human motivation.Footnote 14
Workplace-level and individual-level education independently predict higher levels of cultural liberalism.
Note: the figure is based on Model 6 from Table 2 and relies on data from SNES 2022 and administrative records from Statistics Sweden. Cultural liberalism ranges between 1 and 5, in a liberal direction. All control variables are held at their means. 95% confidence intervals.
Addressing Three Central Concerns
While the cross-sectional results show a linear correlation between the educational composition of the workplace and individuals’ attitudes, I now turn to a series of analyses to test the robustness of the findings. The central concern relates to the possibility that the results are solely driven by selection effects, in which individuals with certain predispositions self-select into workplaces with a low or high share of university educated. A second concern is that the grouping of the educational composition into three categories obscures important thresholds in the effect. A third concern is that the results are driven by specific model specifications and overlook important field-level differences within the highly educated group (cf. Hooghe et al. Reference Hooghe, Marks and Kamphorst2025). The following sections of the paper address each of these issues.
Self-Selection Versus a Direct Workplace Effect
Although the cross-sectional results remain robust after several fine-grained controls, individuals may not change their attitudes during their time at the workplace; rather, the collegial composition could result from self-selection mechanisms at an earlier stage in life. Despite the unique linkage between the data sources, such an opportunity cannot be ruled out when examining only a single point in time. To test the causal interpretation of the estimates using a more demanding test, I therefore run a first-difference regression on the restricted panel sample that answered the survey in 2018 and in 2022. Finding a within-individual attitudinal effect of changes in workplace composition provides additional evidence that part of the effect of the colleagues stems from the time spent in the labor market.
This first-difference design avoids many problems associated with endogenous sorting, under the assumption that the change in the collegial educational composition between the two time points is as ‘good as random’ and not correlated with other confounding workplace-level characteristics. While the empirical design cannot fully rule out such possibilities, particularly in the case of non-observable workplace factors, I run a variety of analyses to test the validity of this assumption. Table B1 shows that the general effect of the share of highly educated colleagues remains unaffected when controlling for other compositional workplace characteristics and is substantially stronger than these alternative observable measures. A placebo test further shows that changes in the share of university-educated coworkers do not correlate with attitudes in 2018 (Table D6). Table D7 further indicates that the respondents’ level of education and workplace size are not significantly associated with future changes in workplace composition, while income shows a modest negative association, likely reflecting the fact that higher-income respondents are already employed in relatively highly educated workplace environments. Furthermore, social networks are a common pathway through which individuals find new jobs (Engzell and Wilmers Reference Engzell and Wilmers2025; Gee et al. Reference Gee, Jones and Burke2017), raising the possibility that any new colleagues are already known to the respondent. Still, the social connections that contribute recommendations about work placements are primarily weak ties (Engzell and Wilmers Reference Engzell and Wilmers2025; Gee et al. Reference Gee, Jones and Burke2017; Granovetter Reference Granovetter1973), thus aligning with the focus of this study. To ensure that the results are not driven by the hiring of a single individual – who thus may be part of an existing social network of the respondent – I estimate additional models focusing on more substantial changes in workplace composition (refer to Tables D3 and D4). The results remain robust under these specifications.
The first-difference regression estimates the relationship between a change in the independent variable – the share of coworkers with university education – and a change in the dependent variable for the same individual. This approach accounts for all unobserved, time-invariant characteristics that do not change between the two time points. Given that changes in coworker composition tend to be modest over a four-year period and attitudes are relatively stable after the formative years (Stubager Reference Stubager2008), the first-difference estimator – which already provides conservative estimates – is particularly constrained in this context (de Chaisemartin and D’Haultfoeuille Reference de Chaisemartin and D.’Haultfœuille2020). I thus consider even small effects as substantively meaningful, as they capture attitudinal change within individuals in response to shifts in their immediate social environment at work, which constitutes only one of several social contexts in which the respondent is embedded.
The control variables include changes in the individual’s salary to account for a potential positional upgrading, and changes in the size of the workplace as an estimate of the general success of the employer. A change in coworker composition can further occur because of the individual changing her workplace, or due to the collegial composition changing within the same workplace.Footnote 15 I thus include a control for whether the respondent stayed within the same workplace or not, and run the analysis separately for each group (Table D1). Similarly, if the respondent changes workplace involuntarily, this experience could have an attitudinal effect unrelated to the collegial composition in the new workplace. While it is not possible to empirically separate the motivations underlying respondents’ job mobility, Table D2 shows that the differences in socio-economic characteristics and baseline levels of cultural liberalism between the two groups are small.Footnote 16
Table 3 shows that an increase in the share of university-educated colleagues is associated with a statistically significant increase in cultural liberalism within individuals. A full-scale increase (from 0 to 1) in the share of university-educated employees predicts a 0.28-point increase on the index, suggesting a small but notable shift in political attitudes in response to changes in workplace composition. In other words, a standard-deviation increase in the share of university-educated coworkers (0.30) corresponds to just above 2 per cent of the full scale. While the magnitude is modest, it represents a non-negligible shift in attitudes given the stability of political preferences and the conservative nature of the first-difference estimator.Footnote 17
Within-individual (first-difference models) estimates show an effect of changes in coworker education on cultural liberalism
Table 3. Long description
A table with three rows and four columns. The columns are labeled as follows: Δ Share university educated, Δ Income, Δ Workplace size, Same workplace, and Constant. The row labels are Δ Share university educated, Δ Income, Δ Workplace size, Same workplace, and Constant. The values in the table are as follows: Row 1: Δ Share university educated, 0.28*, 0.28*, 0.28*; Row 2: Δ Income, 0.04, 0.04, 0.04; Row 3: Δ Workplace size, -0.01, -0.01, -0.01; Row 4: Same workplace, 0.01, 0.01, 0.01; Row 5: Constant, -0.11***, -0.12***, -0.12***. The table also includes standard errors in parentheses and the sample size (N) at the bottom.
Note: ***, **, * note statistical significance at the 1, 5, and 10 per cent level. First-difference regression of change in coworker position on cultural liberalism, SNES 2018–2022. Cluster-robust standard errors at the individual level in parentheses. The control variables include controls for change in the respondent’s logged salary, change in logged size at the workplace, and whether the respondent works at the same or a new workplace.
These results are consistent with an interpretation in which changes in the social environment at the workplace are associated with changes in attitudes. Importantly, the estimates in Table 3 are robust to controlling for workplace mobility, suggesting that the results are unlikely to be driven by selection into specific workplaces or by differences between individuals who change workplace and those who do not.Footnote 18 Table D1 shows analyses on the two groups separately, revealing positive coefficients of similar magnitude for both groups. However, the smaller sample size leads to larger standard errors, which reduces the statistical significance of the results. Taken together, these results point towards an interpretation in which parts of the compositional effect of colleagues result from the time spent at the workplace.
Linearity of the Effect
Dividing the share of highly educated employees into three broad categories may conceal non-linear patterns in the effect. Figure 6 displays results from a model in which the workplace share of highly educated colleagues is divided into quintiles. The plot shows that the effect on cultural liberalism remains modest up to the midpoint but increases more sharply once a majority of coworkers are university educated, thus indicating that there is a quasi-dichotomous relationship with the workplace environment. This pattern aligns with the logic of the liberalizing hypothesis: when university educated form a numerical majority, signals about the workplace norms are clearer and exert a stronger attitudinal influence.Footnote 19 Appendix Figure B3 presents the same relationship using deciles. Similarly, Figure B2 shows that the marginal effect of workplace composition varies with the length of tenure. Although the marginal effect remains positive across all groups, the effect is slightly stronger among employees with shorter tenure. This result is consistent with models of a heightened receptivity to social norms and a greater incentive to conform when entering a new workplace rather than a linear effect of the duration of social exposure.
The effect of workplace education level increases with the share of university educated (workplaces with up to 20 per cent university educated is the reference category).
Note: the figure is based on Model 5 from Table 2 but share of highly educated is divided into quintiles with 0–20 per cent as the reference category. 95% confidence intervals. The final quintile includes workplaces with 100 per cent university-educated employees. The figure is based on data from SNES 2022 and administrative records from Statistics Sweden. All control variables are held at their means.
Varying the Model Specifications
There is also a risk that the specific model specifications of individuals’ and workplace-level education drive the results. Figure 7 tests the individual-level education effect across multiple specifications: a linear measure (years of education), a three-category model (low, medium, high), and a seven-category model distinguishing between specific educational levels. Each specification is estimated both with and without controls for the educational composition at the workplace (black squares vs. white circles). The results confirm that individual education consistently predicts culturally liberal attitudes, regardless of specification, and independently of the social context at work.
Coefficients in Table 2 are robust across alternative specifications for individual-level education.
Note: N: 3,336. The coefficient plots are based on Table 2, Model 5. The figure is based on data from SNES 2022 and administrative records from Statistics Sweden. All control variables are held at their means. 95% confidence intervals. The circle corresponds to the baseline model, and the square to the adjusted model controlling for the educational composition at the workplace.
Figure 8 examines alternative specifications to measure workplace-level education, including the share of employees with specific degree types (bachelor’s, master’s, PhD), the overall share of university-educated employees, and broader categories such as post-secondary or PhD attainment. The results consistently show a positive association across the various specifications, although the strength and precision of the effect vary depending on the measure. Similarly, although this paper focuses on education level, recent research highlights the attitudinal heterogeneity within the group of university educated (Hooghe et al. Reference Hooghe, Marks and Kamphorst2025). It is therefore reasonable to expect that not only the overall share of university educated matters for the prevailing norms at the workplace, but also the fields in which they obtained their university degree. Indeed, the share of employees with a university degree in the humanities, art, social sciences or education – fields high in CECTFootnote 20 – has a liberalizing effect about half as large as the overall share of university-educated employees, when both are included in the same model (see Figure 8).Footnote 21
Coefficients in Table 2 are robust across alternative specifications for workplace-level education, and there is an effect of the educational field composition.
Note: N: 3336. The figure is based on data from SNES 2022 and administrative records from Statistics Sweden. The coefficient plots include the full set of controls as specified in Model 5, Table 2. 95% confidence intervals. All control variables are held at their means. Bachelor’s degree corresponds to three years of university education and master’s degree to 4–5 years of university education. Post-secondary education includes studies up to three years.
Discussion
Labor market transformations contribute to a skill-based segregation of workplaces. This article studies how such segregating dynamics relate to the development of political attitudes using a data combination that identifies both an individual’s attitudes and her colleagues. I examine the formative role of colleagues’ educational background, given that education today is the most influential characteristic shaping political attitudes and voting behavior in post-industrial societies (Ford and Jennings Reference Ford and Jennings2020; Garritzmann Reference Garritzmann2025; Marks et al. Reference Marks, Attewell, Hooghe, Rovny and Steenbergen2023). As a core contribution, I disentangle the political implications of the workplace as a site of political mobilization and political identity formation, or simply as a context for local social belonging. By relying on administrative data that identify the respondents’ complete set of colleagues, the study partially avoids the selection effects associated with studies of self-assessed ties, thereby reducing measurement errors that challenge studies of social networks.
Having a job in a workplace with a considerable proportion of university-educated coworkers is associated with an increase in culturally liberal attitudes. The effect is independent of the individual’s own educational level and further panel analysis suggests that not only selection effects drive the results. The findings add to recent studies regarding educational networks that focus on voters’ close social ties, highlighting the necessity to look beyond an individual’s personal experiences and accounting for the social environment when studying political conflicts. However, a prior study of close educational networks convincingly demonstrates a dependence of the effect by the respondents’ own educational level, which further polarizes the attitudinal ‘diploma divide’ (Attewell and Zollinger Reference Attewell and Zollinger2025). These differences may arise from the inherently different social context of workplaces compared to close social networks, which likely involve higher emotional intensity, greater political transparency, and stronger pressure for norm adherence. The difficulties of effective mobilization processes at the workplace could be even more prevalent in the case of the education divide compared to other social divides, such as the class cleavage. Despite the education divide being the most central contemporary social divide in structuring party support (Garritzmann Reference Garritzmann2025; Hooghe and Marks Reference Hooghe and Marks2025; Marks et al. Reference Marks, Attewell, Hooghe, Rovny and Steenbergen2023), the micro-foundations for this relationship differ from other social divides in terms of the mobilizing appeals from parties and the group identifications at the individual level (Zollinger and Attewell Reference Zollinger and Attewell2026).
On the more speculative side, the results may have far-reaching implications for ongoing debates about the aggregate effects of rising educational levels and the unequal distribution of university educated in different parts of the labor market. In light of these findings, the growing share of highly educated individuals in the population suggests a broader shift towards cultural liberalism, which could be viewed as a countervailing process to the polarization of the education divide. Still, the clustering of individuals by educational background at different workplaces implies that such a liberalizing move will primarily occur among highly educated themselves, thereby aligning with the findings by Attewell and Zollinger (Reference Attewell and Zollinger2025) and de Jong and Kamphorst (Reference de Jong and Kamphorst2025). As heterogeneous networks at the workplace become increasingly rare, the distinctness of the workplace as a context that clusters individuals with different backgrounds diminishes, making the workplace increasingly similar to other social contexts with relatively weak ties.
One central implication of this study is thus that socializing (counter-)dynamics within workplaces are not the primary mechanism for the growing educational polarization in advanced democracies (Garritzmann Reference Garritzmann2025; Marks et al. Reference Marks, Attewell, Hooghe, Rovny and Steenbergen2023). Instead, a more fruitful approach to understand the relationship between such polarization and the social role of work is to focus on the structural segregation of the labor market. The extent to which the workplace composition correlates with an individual’s preferences is likely shaped by both personal and contextual factors. Individuals in more sought-after occupations, who reside in cities and enjoy financial stability, have greater opportunities to change workplaces if their current environment does not align with their values. Relatedly, macro-level economic developments influence overall labor market fluctuations and hiring opportunities across industries (Berglund and Ulfsdotter Eriksson Reference Berglund and Ulfsdotter Eriksson2024). As a final point, trends towards an oversupply of university-educated workers in certain sectors and countries sometimes compel degree holders to apply for jobs with lower skill requirements (Ansell and Gingrich Reference Ansell and Gingrich2017; Häusermann et al. Reference Häusermann, Kurer and Schwander2015), thereby contributing to a greater educational mix in workplaces. This development implies that the share of unemployed highly educated individuals increases in the population as a whole (Berglund and Ulfsdotter Eriksson Reference Berglund and Ulfsdotter Eriksson2024), although the vast majority of individuals who are not exposed to any workplace socialization still are rooted in the low-educated population.
What makes the findings particularly interesting from a political science perspective is that they lay a new foundation for exploring the conditions under which education influences cleavage formation (Ford and Jennings Reference Ford and Jennings2020; Lipset and Rokkan Reference Rokkan and Lipset1967; Scott Reference Scott2022; Stubager Reference Stubager2008), as well as how and when homogeneous social networks contribute to polarizing group dynamics (see also Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024). While other compositional characteristics beyond colleagues’ educational background may also be politically relevant, the consistent effects of education underscore that the social dimension of the workplace has considerable traction for attitude formation. Future studies should proceed by testing the direct mechanisms that drive such effects, which may include separating issue salience and attitudinal positions in different types of workplaces (Dennison and Kustov Reference Dennison and Kustov2023; Kustov Reference Kustov2023). Another fruitful avenue for future research is to consider sorting based on educational field and occupational task structure to a higher extent. As the share of highly educated individuals increases in a population, differences in experiences and opportunities within the group of highly educated are likely to play an increasingly important role for both group belonging and work placements (Hooghe et al. Reference Hooghe, Marks and Kamphorst2025; Scott Reference Scott2025).
Both classic conjectures and more recent studies highlight the political role of the people we work with. The recent availability of large-scale administrative micro data linked to election surveys makes it possible to test this notion directly. The findings demonstrate the potential of including colleagues in processes of political socialization and should encourage future research in this direction. Such studies could strive to more precisely identify how the workplace composition influences attitude formation, potentially by introducing quasi-experimental methods to the field. Likewise, this empirical analysis could be further developed by accounting for the hierarchical structure of the workplace (Frymer and Grumbach Reference Frymer and Grumbach2021). The socializing effect of cross-educational group interactions appears particularly relevant in workplaces with egalitarian organizational structures rather than in workplaces with strict hierarchies between occupational groups. Such an approach would more explicitly connect to the work of Kitschelt and Rehm (Reference Kitschelt and Rehm2014) by considering the different collegial dynamics associated with different labor market positions not only horizontally, but also vertically. How individuals perceive political conflicts in society depends on their social networks, and their position within them.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123426101781.
Data availability statement
Do-files and log-files are available at the BJPS dataverse and can be found at https://doi.org/10.7910/DVN/TABFXZ. The data used in this paper are provided by Statistics Sweden in association with the Swedish National Election Study (2022) and are subject to access restrictions. I have obtained permission from the Ethical Review Board to analyze the data at a secure server, and there are accordingly no local copies of the data. The data can therefore not be shared publicly for replication purposes. Researchers who want access to the data may apply to Statistics Sweden, after gaining permission from the Ethical Review Board.
Acknowledgements
I am grateful to the four anonymous reviewers, the editors of the BJPS, Mikael Persson, Johannes Lindvall, and Thomas Kurer for helpful and detailed comments on this manuscript. I am also thankful for the valuable feedback provided at the EUI workshop ‘Comparative Politics of Europe’, organized by Liesbet Hooghe and Gary Marks in Florence 2025, and at the EPSA panel ‘Perceptions and Preferences in the Knowledge Economy’ in Madrid in 2025. I also want to thank Moa Frödin Gruneau and Maria Solevid for their generous assistance in gaining access to the data, and Felix Bäckstedt for excellent research assistance in the early stages of this project.
Financial support
Support for this research was provided by the Swedish Research Council (grant agreement no. 2022-02399, project ‘No wo/man is an island – A social theory of economic voting’).
Competing interests
The author declares none.
Open access
