Politician Data

Before each election, Swedish political parties are required by law to submit their ballot papers to the government with the personal identification (ID) code of every candidate. We digitize this information to generate a list of all nominated politicians in all municipal and national elections between 1973 and 2018, including their political party and list rank (Statistics Sweden 2022a). We link this list via each politician’s ID code to Swedish Electoral Agency data on (i) whether a candidate was elected and (ii) how many preference votes they received. These political variables have little or no missing data. We exclude local politicians from parties not represented in parliament (~4% of the data).

Administrative Data

We further link the list of politicians to administrative data on individual traits (Statistics Sweden 2022b). These data combine individual-year panels for more recent decades with census data at 5-year intervals for 1970–1990. It includes all permanent residents in the country over 18 years of age.

Most administrative variables come from census data (prior to 1991) and from the longitudinal integrated database for health insurance and labor market studies (after 1991). These variables come from different government records, such as tax records, education records, birth records, and employer surveys. They include each permanent resident’s demographic traits, education level, citizenship, and various traits of the job that constitutes their largest source of labor income in each year.

Occupation data for the person’s main job come from the censuses (1970–1990) and the Swedish occupational register (2001–2018). Note here the 10-year gap in data availability for the 90s. Census data have complete coverage of the labor force, while the register data do not. The variable in the register data assembles information from a mandatory government survey administered each year to firms and workplaces, industry-specific surveys managed by employer organizations, and other smaller data sources.Footnote ³ Compiling these sources results in coverage of the entire public sector and all large firms, most medium-sized firms, and a subset of small firms, self-employed people, and individuals on short temporary contracts.

Two datasets provide additional information for our measurements of cognitive skills. One provides high school grade point averages (GPAs) for all graduates after 1973. The other provides scores from cognitive tests administered as part of Sweden’s mandatory military draft for men born between 1951 and 1979. A third measurement of ability relies on the earnings data from tax records mentioned above (Besley, Persson, and Rickne Reference Besley, Persson and Rickne2017). Our measurement of organizational ties to labor unions uses an itemized taxation-based dataset to identify union representatives and union employees (further explained below).

Survey Data

We use data on the share of workers who voted for each party in the 1970–2018 national elections from the Swedish Election Survey. We use the national and Western regional versions of the Society, Opinion and Media Survey (SOM Institute 2022a; 2022b) to measure ambitions and political preferences in the population. We employ data on ambitions among local politicians from the 2012 Local and Regional Councils Survey (KOLFU, Karlsson and Gilljam Reference Karlsson and Gilljam2014) and added new questions to the 2017 wave of this survey (Karlsson Reference Karlsson2017) to measure certain aspects of competence. Response rates among local councilors were 80% in 2012 and 67% in 2017.

We use two nationally representative surveys collected by the Swedish government. The Swedish Work Environment survey is a biannual survey that measures work conditions (about 165,000 respondents in total between 1991 and 2019) and the Swedish Labor Force Survey tracks the business cycle in the labor market (about 8.4 million individual-quarter observations between 1987 and 2019). Section W3 of the Supplementary Material provides more details about all the surveys used in this article.

Defining the Political Career Ladder

We propose a simplified political career ladder with five rungs that range from the local to the national level (Figure 1). Sweden has 290 municipalities, each with an elected assembly of 31–101 councilors. Municipal party systems mirror the national level and strictly local parties hold less than 5% of all council seats. The municipal branches of the national parties make highly autonomous personnel decisions.Footnote ⁷

In our first career step, ordinary citizens (L5) start by being nominated as candidates in local politics (approximately fifty thousand people in each election). In the next step, they go from being an unsuccessful nominated candidate (L4) to winning an unpaid local elected seat (L3) (approximately thirteen thousand people). The third step consists of advancing to the top political position in their local political party, the local party leader who is ranked first on the ballot paper (L2).Footnote ⁸ This person nearly always holds the top appointment afforded the party by its vote share, such as the chair of the council board if the party is the largest in the governing majority. The fourth step is advancing to one of the 349 seats in parliament (L1). We validate the model by calculating that about two-thirds of politicians at each career rung arrived on there from the one below. The robustness test replicates the results for alternative career ladders (details below).

Procedures for Career Advancement

The processes for selecting and promoting politicians are similar across parties and at the local and national levels (e.g., Buisseret et al. Reference Buisseret, Folke, Prato and Rickne2022; for detailed descriptions of the institutions summarized in this subsection, see Johansson Reference Johansson1999; Widenstjerna Reference Widenstjerna2020). Election committees and party members are the main selectors. Such committees usually comprise prominent party members and play a key role in collecting nominations and organizing the ranking procedure at either the municipality or parliamentary district level. They often use advisory internal primaries among party members to gauge candidates’ internal party support. Election committees usually suggest a rank order of candidates on the list, which is usually formally approved without changes in an at-large meeting.Footnote ⁹

An individual’s political career usually progresses within a single party and municipality. Parties have strong seniority principles that operate mostly as norms rather than formal regulations (details in Brothén Reference Brothén2010). Longer service as an elected municipal councilor is strongly associated with becoming a local party leader or parliamentarian in all parties (Supplementary Figure W3) and among both workers and nonworkers (Supplementary Figure W4). Labor unions occupy a specific organizational position within the Social Democratic Party and its candidate nomination procedure, which we discuss in more detail below.

Voters

Voters may have negative views about workers’ competence or perceive them as too ideologically leftist or unlikely to understand the problems they face in their own lives. Such negative views may depress workers’ vote counts and explain their political exclusion. Evidence from previous research has largely rejected this conjecture. Empirical studies using methods ranging from observational studies to conjoint survey experiments have found little or no evidence of negative voter bias (recently reviewed by Albaugh Reference Albaugh2020; Carnes and Lupu Reference Carnes and Lupu2023a; for a specific example from this literature, see Carnes and Lupu Reference Carnes and Lupu2016).

Individual Resources

Working-class politicians may have lower levels of resources that generally increase the likelihood of political career advancement. As in most organizations, people who are more ambitious and possess competencies for a specific job are more likely to be promoted. We consider three such resources—political ambition, attitudes and personality traits (including cognitive skills), and time.

Workers might have less political ambition if the white-collar dominance in politics leads to negative beliefs about their career success (e.g., Lawless and Fox Reference Lawless and Fox2010). The same white-collar overrepresentation might also result in fewer social contacts with politicians or party organizers, leading to less exposure to role models and a lower likelihood of receiving personal encouragement. Prior research offers mixed findings about workers’ political ambitions (reviewed by Carnes and Lupu Reference Carnes and Lupu2023b; see also Allen and Cuts Reference Allen and Cutts2018).

Besley (Reference Besley2005) outlines types of competencies relevant to political office. One type describes attitudes and personality traits, such as public service motivation, honesty, and work effort. Another type describes a person’s cognitive skills, or innate “ability” as the term is used in the economics literature. These traits broadly overlap with those that citizens and party leaders often say they want in a politician (Berg et al. Reference Berg, Oscarsson, Folke, Persson, Rickne and Oleskog2015; Carnes Reference Carnes2020; Carnes and Lupu Reference Carnes and Lupu2023b).

There is little reason to expect working-class politicians to have lower levels of desirable attitudes or personality traits. Cognitive ability is a trickier question. A higher payoff from tertiary education among high-ability people will produce sorting by ability into tertiary education and, in turn, an overrepresentation of high-ability people in white-collar occupations. However, ability-based selection into tertiary education has been exposed as imperfect at best, since structural life circumstances fundamentally determine these choices (e.g., Chetty, Hendren, and Katz Reference Chetty, Hendren and Katz2016; Lynch and O’riordan Reference Lynch and O’riordan1998).

Previous research cautions against equating advanced degrees or white-collar jobs with political competence. Parliamentarians from different social classes display similar levels of policy efficacy and influence (Eriksson and Josefsson Reference Erikson and Josefsson2019), and politicians’ education levels have been shown not to meaningfully affect the results they produce while in office (see, e.g., Carnes and Lupu Reference Carnes and Lupu2015b; recently reviewed by Curto and Gallego Reference Curto, Gallego, Lupu and Pontusson2023). We return to the issue of tertiary education and specific occupation-related skills in the sensitivity analysis.

The time demands of political participation put time-constrained individuals at a participation disadvantage (Brady, Sidney, and Schlozman Reference Brady, Verba and Schlozman1995). Politics requires the flexibility to invest time in meetings and activities on evenings and weekends, as well as intense work around elections. Because political careers usually start alongside a person’s regular job, the time–space flexibility of those jobs might matter for political career investments.

Political Parties

Sociology research defines a glass ceiling as a combination of two main empirical patterns that together imply discrimination against women in job promotions. The share of women should decline in higher positions in organizations and observable job qualifications should fail to explain women’s promotion disadvantage (Cotter et al. Reference Cotter, Hermsen, Ovadia and Vanneman2001; applied to politics by Folke and Rickne Reference Folke and Rickne2016). We apply this concept to the working class and change the label to a class ceiling. Footnote ¹¹

Empirical research on party selectors’ bias against workers is rare (for exceptions, see Carnes Reference Carnes2016; Norris and Lovenduski Reference Norris and Lovenduski1995), but theory lets us outline several types of relevant negative beliefs that selectors might hold. Selectors might perceive workers to have lower popularity with voters for example if they use information shortcuts to equate a lower share of workers in politics with a lower level of electability (Carnes Reference Carnes2020; Norris and Lovenduski Reference Norris and Lovenduski1995). They might also perceive workers’ ideologies as straying too far from the electorate, potentially because they overestimate the electorate’s alignment with white-collar interests (e.g., Broockman and Skovron Reference Broockman and Skovron2018).

Beliefs about appropriate qualifications for political officeholding constitute another relevant type of belief. These beliefs might be “classed” so that notions of “merit” and “fit” limit workers’ career advancement (Friedman and Laurison Reference Friedman and Laurison2020). Party elites may overestimate the value of tertiary education or degrees from specific elite universities. They may also underestimate the importance of life experiences resulting in overlapping issue priorities, values, and policy objectives by social class—perhaps leading them to downplay candidate ideology as a qualification for political office (Norris and Lovenduski Reference Norris and Lovenduski1995; Widenstjerna Reference Widenstjerna2020).

A more subtle kind of negative belief might stem from psychological processes and human interactions based on social identity. Class is a social identity that delineates social circles in which people form positive opinions about their “in group” and negative opinions about their “out group” (e.g., Kraus and Keltner Reference Kraus and Keltner2013; Tajfel and Turner Reference Tajfel, Turner, Jost and Sidanius2004). This might result in negative presumptions about workers’ competence among selectors, who are less likely to be working-class. Consistent with this theory, Rehmert (Reference Rehmert2022) finds that highly educated party selectors prefer highly educated candidates, while less educated selectors do not.

Beyond direct decisions about nominations and promotions, selectors’ negative beliefs might shape the division of other career-relevant party decisions during the election period. This might include decisions about the assignment of tasks and roles or who to feature in the media or in political campaigns. Negative beliefs might extend to groups outside of politics in ways that affect working-class politicians’ work environments and chances to build qualifications. They might exist, for example, among bureaucrats, the media, or unelected party officials.

Labor Unions

Organized labor may affect workers’ political inclusion via broad-based unionization in the labor market, running training schools for political candidates, or economic and in-kind support to political campaigns. We focus on paid positions in blue-collar labor unions, for example, as a union representative. These positions build ideological and political knowledge as well as practical skills in leadership and negotiation, which are helpful in a political career.

Some labor parties, such as Sweden’s Social Democrats, have strong historical roots in the labor movement and formalized organizational linkages that allow unions to directly influence political nominations (Aylott Reference Aylott2003; Fouirnaies Reference Fouirnaies2021; Høyer Reference Høyer2015). Aylott (Reference Aylott2003) describes how local union sections and workplace-based union clubs were traditionally affiliated with local Social Democrats’ party organizations as “units,” putting them on the same footing as units for women, youth, or neighborhoods in local and national candidate nominations. This structure shaped strong norms in support of having people with different labor union backgrounds on electoral ballots and including union representatives on nomination committees. Unions’ influence remained significant after the practice of affiliating whole union sections ended in the 1990s and includes, in addition to local influence, union representation in the party’s executive committee and collaboration to organize courses in political work methods and ideology.

WORKERS’ NUMERICAL REPRESENTATION THROUGHOUT THE CAREER TRAJECTORY

The left-hand side of Figure 2 depicts the share of workers at each career level in Swedish politics. The share clearly decreases on higher career steps. From 1973 to 2018, workers averaged 50% of the voting-age population but 34% of nominated local politicians, 28% of municipal councilors, 17% of municipal list-leaders, and 13% of parliamentarians. The right-hand side of the figure displays underrepresentation in the percent of the population share, that is, dividing the difference between the share of workers at a specific career level in the population by their share in the population. According to this metric, workers are underrepresented by 34% among nominated local politicians, 48% among elected councilors, 70% among local party leaders, and 75% among parliamentarians. This pattern differs little between women and men politicians (see Figure 3).

Figure 2. Workers’ Numerical Representation across the Career Ladder

Note: A working-class person is defined based on their primary occupation in a working-class job according to the EGP class scheme. The population includes all permanent residents over 18 years. A local leader is the top-ranked politician on a municipal electoral ballot. The data on municipal politicians are restricted to parties represented in parliament. In the right-hand side graph, the share of workers in the population is measured at the municipal level for municipal positions and at the national level for parliament. N (population) = 59,760,437; N(nomiAnated) = 397,511; N(municipal councilors) = 159,743; N(local party leader) = 7,125; N(parliamentarian) = 4,068.

Figure 3. Workers’ Numerical Representation by Sex at Birth

Note: The figure reproduces the left-hand side of Figure 2 according to sex at birth.

We replicate the impact of the gradual decline in the share of workers across political career levels for five alternative definitions of the working-class in Supplementary Figures W5 and W6. Alternatives are (1) the ISCO-based definition by Carnes and Lupu (Reference Carnes, Lupu, Lupu and Pontusson2023c), which includes clerical workers in the working-class, (2) having below tertiary education, (3) having below median labor earnings, (4) using the person’s most common worker or nonworker category across all observable years on our main variable, or (5) using the observations closest in time to the age of occupational maturity at age 37. Supplementary Figure W7 shows highly similar results when restricting the data to people with Swedish citizenship, which is a requirement for voting and office holding in national elections.

Figure 2 shows some improvement in workers’ numerical representation over time. The share of workers in parliament rose from one-fifth of their share in the population in the 1970s and 1980s to one-third since the 2000s. Figure 4 contrasts this development with three other political minorities: women (female sex at birth), young people (under 35 years old), and foreign born. It indicates that the gains in representation for workers pale in comparison with those made by these other groups over the last 50 years.

Figure 4. Political Minorities’ Numerical Political Representation in Parliament

Note: See the note of Figure 2.

We split the analysis by political party in Figure 5. This analysis pools all elections and plots the share of workers at each rung of the career ladder in each party relative to the share of workers in the population (top) and relative to the share of workers among that party’s voters (bottom, measurement details in the figure note).Footnote ¹² Extended descriptions in the Supplementary Material show party-by-year developments (Supplementary Figure W8) and replicate the results in Figure 5 for an alternative career ladder that includes appointments to local executive positions (Supplementary Figures W9 and W10).

Figure 5. Numerical Representation by Party and Career Step

Note: Pooled cross-sectional data for elections from 1973 to 2018. Data voter composition over time is data from Swedish National Election Survey reported in Hedberg (Reference Hedberg2020). We use these together with administrative data to impute the expected share of the party’s voters who are workers. The share of workers among the voters of party p in election t and municipality m is given by $ {W}_{pt m}=\left({Vs}_{pt}^W\ast Po{p}_{mt}^W\right)/\left({Vs}_{pt}^W\ast Po{p}_{mt}^w+{Vs}_{pt}^{NW}\ast Po{p}_{mt}^{NW}\right) $ , where $ {Vs}_{pt}^W $ and $ {Vs}_{pt}^{NW} $ are national-level vote shares among workers and non-workers, and $ Po{p}_{mt}^W $ and $ Po{p}_{mt}^{NW} $ are population shares. We divide the share of workers at each career level and in each party by these numbers for each municipality and election and report averages of these fractions in the figure. The calculation for MPs follows the same method at the national level.

The share of worker declines at all career steps in all parties, except for parliamentarians in the Left Party. In left-ideological parties and the radical right, these declines start from a higher level, which echoes findings from other countries where these party families have more politicians from the working class (e.g., Durose et al. Reference Durose, Gains, Richardson, Combs, Broome and Eason2011; Hemingway Reference Hemingway2020; Matthews and Kerevel Reference Matthews and Kerevel2022; Norris and Lovenduski Reference Norris and Lovenduski1995).

The results change if we benchmark parties’ share of working-class candidates against their share of working-class members (calculations in Supplementary Table W5). The Social Democrats now perform no better than center-left parties. The radical right (Sweden Democrats) performs the best; its rapid growth in recent elections has contributed to the improvement over time in workers’ numerical representation observed in Figure 2. Previous research links the greater presence of working-class politicians in this party to patterns in candidate self-selection rather than conscious recruitment strategies (e.g., Art Reference Art2011; Dal Bó et al. Reference Bó, Ernesto, Finan, Folke, Persson and Rickne2023). Government austerity policies and economic shocks may have acted as a greater pull factor for workers, while the party’s political and social stigma has reduced its relative attractiveness for high-socioeconomic status groups.

Figure 6 displays the extent to which different parties contribute to existing workers (left) and “missing” workers (right) in parliament. Missing workers are those who would be in office if each party’s share of working-class politicians would correspond to its share of working-class voters. The graph highlights that the Social Democratic Party has by far the largest share of working-class parliamentarians but also the largest share of missing workers. Figure 6 also documents the decline of social democracy and the rise of the radical right Sweden Democrats in the most recent elections. This party now accounts for a growing share of existing and missing workers, while the Social Democrats’ share of both has sharply declined.

Figure 6. Party Composition of Existing and “Missing” Working-Class Parliamentarians

Note: Shares of missing workers are calculated relative to a situation where each party’s share of working-class parliamentarians equates its share of working-class voters. See the note of Figure 5 for a description of how the share among voters is calculated.

Preference Votes

An electoral reform in 2002 allowed Swedish voters to cast one optional preference vote by checking a box next to the candidate’s name on the party’s ballot. While few politicians amass enough votes to win a seat in this way, parties use these vote counts to guide future promotions (Folke, Persson, and Rickne Reference Folke, Persson and Rickne2016). Parties are free to print information on their ballots, and most ballots include candidates’ job titles (data collected by the authors).

Our analysis proceeds in two steps. The first step obtains a measurement of preference votes that isolates voter preferences from parties’ list-rank decisions and the mechanical tendency to vote for highly ranked candidates. We regress votes on dummy variables for each list rank, predict the residuals at the individual level, and transform these residuals into a z-score. The second step regresses this measurement on a dummy variable for being a worker and fixed effects for all combinations of municipality, party, and election. The analysis returns a coefficient on the worker dummy close to zero: smaller than 0.05 standard deviations when running the regression for either all local politicians or elected councilors only (details in Supplementary Table W6). Workers do not underperform (or overperform) compared to nonworkers in terms of how their votes deviate from the typical preference vote result for a politician on their list-rank. Including sociodemographic controls does not change this finding. In sum, we find no evidence that voters prefer nonworkers to workers when casting their preference votes.

Individual-Level Resources

We use survey data to test whether workers have lower political ambitions than nonworkers. The Western Sweden SOM survey (SOM Institute 2022b) asks about nascent ambition with the question “Would you consider taking on a political appointment in the municipality for the party you sympathize with?”Footnote ¹³ We create a dummy variable that takes the value 1 for affirmative responses and 0 for negative ones. The 2012 KOLFU survey asks local councilors how long they want to remain in office. Strong seniority principles make this a good proxy for politicians’ progressive political ambition, that is, their aspirations to seek a higher political position. The note of Figure 7 provides details on variable creation and sample sizes. We lack data on ambitions for the final career step to national office.

Figure 7. Gaps in Political Ambition between Workers and Nonworkers

Note: The figure shows estimated coefficients from a regression of self-rated political ambition on a binary indicator for having a working-class occupation. Black and gray markers show estimates from two regression specifications (see legend) and vertical lines denote 95% confidence intervals. Full regression results are available in dataverse log-files (Folke and Rickne Reference Folke and Rickne2024). Demographic controls include a dummy variable for female sex at birth, dummies for five age categories, two dummies for being born in Europe (excluding Sweden) and outside of Europe, and a dummy for being a full-time student. Pooled annual survey cross-sections come from the Regional SOM survey in Western Sweden in 1995–2003. Data on politicians’ ambition levels come from the 2012 KOLFU survey (Karlsson and Gilljam Reference Karlsson and Gilljam2014). Working class is self-reported on a list of class categories (details in Section W3 of the Supplementary Material). N(population) = 18,325. N(councilors) = 6,268; N(local party leaders) = 976.

We estimate the gap between workers and nonworkers by regressing each variable for ambition on a dummy variable for being a worker. Figure 7 plots these gaps with 95% confidence intervals. One regression specification includes fixed effects to narrow the comparison by geographic region, time point, and party, and the other adds controls for age, global region of birth, sex at birth, and a dummy for being a full-time student (details in figure note).

Workers in the population have a lower average level of nascent political ambition than nonworkers. They are 5–7 percentage points less likely to state a willingness to become a local politician, which corresponds to 24–33% of the population average on this measure.Footnote ¹⁴ This result contrasts with those for progressive political ambition, where workers have, if anything, a higher level among both local councilors and local party leaders.

Figure 8 shows estimated gaps for four measures of attitudes and personality traits (summarized below with details in Section W3 of the Supplementary Material). The first is self-reported public service motivation measured using an index based on a battery of questions about private versus altruistic motives (the Perry Score). The second measure is an index of morality contained within the HEXACO personality inventory developed by social psychologists (the Honesty–Humility Score).

Figure 8. Gaps in Political Competence between Workers and Non-Workers

Note: The figure shows estimated coefficients from a regression of competence measurements (listed on the X-axis) on a dummy variable for being a worker. See the note of Figure 7 for the list of demographic controls. Full regression results available in dataverse log-files (Folke and Rickne Reference Folke and Rickne2024). Population: municipality-year; nominated and councilors: municipality-year-party; local party leaders in administrative data: municipality, party, and year; local party leaders in survey data: party. High school grades are standardized within-cohort percentiles for graduates in 1973–2018; leadership scores from military draft tests taken by male cohorts in 1953–1979; the Earnings Score is an earnings-based competence measure calculated using data for the entire Swedish workforce. Data on attitudes and personality traits are from the 2017 KOLFU survey (Karlsson Reference Karlsson2017) and details on all measurements are presented in Section W3 of the Supplementary Material. N(KOLFU local councilors) = 4,638; N(KOLFU local party leader) = 1,009.

The third and fourth measurements are based on self-reported levels of effort in their current political position. We use survey data to calculate politicians’ hours of political work per week conditional on the number of formally remunerated hours by virtue of their political position(s) in the local government structure. We measure campaign work as their self-reported share of the local party’s political campaigns in which they participated during the last year. Both measurements are standardized to z-scores.

For each competence measurement, we compare workers and non-workers at different career levels. We find no substantively meaningful gaps in attitudes and personality traits. The sample of municipal councilors contains some negative coefficients that indicate a gap to workers’ disadvantage, but their small sizes (less than 0.1 standard deviations) classify them as substantively unimportant according to the standard Cohen’s d categorization (Cohen Reference Cohen1988).

The results for cognitive skills exhibit medium-sized gaps to workers’ disadvantage. Looking at the population of men born in 1959–1979 for whom we have cognitive test scores, workers have a 0.3–0.2 lower average score at all three lower career levels. We find a similar-sized disadvantage of 0.3 standard deviations for high school grades, which we measure as standardized within-cohort percentiles of the GPA distribution.

The third competence measure is derived from the assumption that ability is priced in the labor market. We use an earnings regression for all employed permanent residents to benchmark yearly deviations in each person’s annual earnings relative to their peers of the same sex, age, municipality of residence, education level, and industrial sector (Besley, Persson, and Rickne Reference Besley, Persson and Rickne2017, further details in Section W3 of the Supplementary Material). We call the standardized residual from this equation the Earnings Score. Workers’ disadvantage in this variable is relatively small, ranging from 0.25 standard deviations in the population to 0.1 for local party leaders.

For all three measurements of cognitive skills, the gaps between workers and nonworkers are half as large among politicians as they are for the entire Swedish voting-age population. This suggests that the selection of workers into politics is more positive than for nonworkers, which replicates Dal Bó et al.’s (Reference Bó, Ernesto, Finan, Folke, Persson and Rickne2017) results for parental social class. The remaining gaps offer a potential explanation of why political parties are less likely to promote workers to higher positions, which we formally test in the next section.

We use nationally representative survey data to create four binary indicators for inflexible work conditions (details in Section W3.3 of the Supplementary Material and Supplementary Table W3). In the average worker’s occupation, 60% report having no schedule flexibility compared to 25% of nonworkers; 34% vs. 12% report having non-daytime work, 11% vs. 3% report split-shift work,Footnote ¹⁵ and 91% vs. 63% report having no opportunities to work from home. Workers have a 1.44-standard-deviation lower value on a combined inflexibility index based on these four variables. This large gap comes with the empirical challenge of multicollinearity: the correlation coefficient between the flexibility index and the worker dummy is 0.7.

Political Parties

We test whether workers have a lower average probability of promotion from one career step to the next by estimating:

(1) $$ {Y}_{it}={\alpha}_{pmt}+\beta {W}_i+{\boldsymbol{D}}_{it}+{\boldsymbol{E}}_{it}+{\boldsymbol{Q}}_{it}+{e}_{it}, $$

where $ {Y}_{it} $ is a dummy variable for promotion to a higher career level in election t. It takes a value of 1 if a politician advances in t, and 0 otherwise (including candidates who do not appear on the ballot in t, subject to sensitivity results below). We run regression separately for entry as a nominated politician, promotion to councilor, local party leader, and parliamentarian. In each case, the sample includes everyone who held the lower adjacent level in the previous election (for parliamentarians, this includes local elected councilors and local party leaders, Figure 1). We cluster the standard errors at the individual level.

The main variable of interest in Equation 1 is the dummy variable for being a worker, $ {W}_i $ . Its estimated coefficient $ \beta $ captures workers’ relative probability of being promoted in percentage points. We do not report this metric but instead divide the estimated coefficient by the average of the outcome variable. The resulting number represents the gap in promotion probability measured as a percent: −0.5 equates to a 50% smaller likelihood of promotion among workers than nonworkers.

We always include fixed effects for combinations of party-municipality-election when estimating Equation 1. These dummies restrict the statistical comparisons of workers and nonworkers, which are averaged in the coefficient $ \beta $ , to politicians running for the same party, in the same municipality, and in the same election. Adding a vector of dummies for sociodemographic traits $ {\boldsymbol{D}}_{it} $ further nets out promotion gaps between social classes stemming from the potentially correlated identities and inequalities associated with age, sex at birth, global birth region, or being a full-time student. To make the promotion ladder meaningful, we restrict the sample to municipal-level parties with at least five elected councilors.

We add dummy variables $ {\boldsymbol{E}}_{it} $ to denote the politician’s number of previous election periods as a municipal councilor to hold constant political seniority in the estimations of promotions to the two highest career rungs. The strong seniority principles in Swedish parties make seniority a basic qualification for promotion, comparable to job or sector tenure in analysis of private sector careers.

The variable vector $ {\boldsymbol{Q}}_{it} $ contains the three competence variables. A lack of data overlap prevents us from adding all three at once. Table 1 starts with a specification including the fixed effects for sociodemographics only in column 1, adds dummies for seniority in column 2, and then adds the Earnings Score variable in column 3. The next specification reestimates the basic specification in the data sample with non-missing high school grades in column 4 and adds the two variables for grades and Earnings score in column 5. Columns 6 and 7 do the same for the cognitive score. Estimating a version of Equation 1 without the worker dummy shows that all three measurements correlate positively and strongly with the likelihood for political promotion (Supplementary Table W7).

All estimated promotion gaps are large, negative, and statistically significant at conventional levels. The estimates in column 1 of Table 1 indicate that workers are 50% less likely to enter politics as a local candidate, 40% less likely to move to the level of councilor, 30% less likely to become a local party leader, and 60% less likely to go from elected municipal office to parliament. Holding politicians’ seniority at the local level constant in column 2 reduces the disadvantage in party leadership somewhat and shrinks the disadvantage in parliamentary promotions by 15% (from 60% to 50%).

Comparing the estimates in columns 2 and 3 of Table 1 illustrates that the Earnings Score measure of competence explains very little of the workers’ promotion disadvantages. The same is true for combining the Earnings Score and high school grades (comparing columns 4 and 5); together, these explain one-third of the disadvantage of becoming a party leader. It explains less of the other gaps, however, and even after including these controls, workers’ disadvantage in that promotion remains large at 30%.

We add the index of inflexible work conditions to the career regression to see if this variable explains workers lower chances of promotion (results in Supplementary Table W9). Perhaps surprisingly, the regression coefficient on the index is near zero for lower-level promotions and small for higher levels. Including the control nevertheless reduces the estimated promotion disadvantage by about a quarter for the three higher promotions. Time constraints imposed by the occupation appear to partly explain the promotion gap, although the strong correlation between the worker dummy and the inflexibility index inflates standard error and prevents us from drawing clear conclusions about the exact importance of this variable.

The results in Table 1 persist for alternative operationalizations of the career ladder. Workers have a large disadvantage in directly entering politics at each of the three highest career levels (Supplementary Table W10) as well as a large disadvantage in entering parliament from being nominated on a parliamentary ballot (Supplementary Table W11). Defining people who leave politics as not promoted does not affect our results, because workers do not drop out at a higher rate than nonworkers (Supplementary Table W12). The results persist when we use Carnes and Lupu’s (Reference Carnes, Lupu, Lupu and Pontusson2023c) alternative definition of the working class (Supplementary Table W13).

Table 1. Gaps in Promotion Probabilities between Workers and Non-Workers

Note: The table reports gaps in promotion probabilities between workers and nonworkers across adjacent rungs on the political career ladder. OLS squares estimates in percentage points are rescaled by the mean of the outcome variable so that −0.5 in the table equals a 50% lower promotion probability. FEs for sociodemographic traits include a dummy variable for female sex at birth, dummies for five age categories, two dummies for being born in Europe (excluding Sweden) and outside of Europe, and a dummy for being a full-time student. Supplementary Tables W8A–D reproduces the table with control variable estimates. Standard errors clustered at the individual level in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 2. Gaps in Promotion Probabilities by Politicians’ Organizational Ties to Labor Unions

Note: The table reports estimates for Equation 1 after splitting the working-class dummy variable into two dummies for workers with vs. without organizational ties to labor unions. Standard errors clustered at the individual level in parentheses. See the note of Table 1 for the recalculation of the estimates to values in percent and the specification of control variables. Supplementary Table W17 reproduces the table with control variable estimates. ***p < 0.01, **p < 0.05, *p < 0.1.

A potential concern about our analysis is that tertiary education or working in specific white-collar occupations delivers key competencies for political office. Supplementary Table W14 runs two variants of the basic promotion regression in column 2 of Table 1 to address this point. One adds dummy variables for seven levels of education, and the other excludes individuals with a college or university degree. Of the eight estimated promotion gaps across these two variants, the largest reduction in the size of workers’ promotion disadvantage compared to the main analysis is 30%. Supplementary Table W15 shows that large promotion gaps persist after holding constant the number of tasks that occupations share with being a legislator according to the O*NET database (measurement details in Section W3.6 of the Supplementary Material). These results contradict the idea that a lack of skills learned in tertiary education or white-collar jobs constitutes the main explanation for workers’ promotion disadvantage.

Labor Unions

We identify people with union ties in a taxation register with detailed information about all labor income payments that sum to more than 10 USD in each calendar year (1985–2018). We define a person as having a union tie if they received such a payment from a blue-collar labor union before the election in which we observe them (Section W3.7 of the Supplementary Material provides more measurement details). About 5% of the population aged 18 or over has a union tie according to this measure. Only 3% of people with such ties are fully employed by a union (i.e., receive a relatively large sum of annual labor income from this main employer).Footnote ¹⁶ The remaining 97% receive smaller payments, presumably for being a union representative in a workplace. Supplementary Table W16 demonstrates a large socioeconomic similarity between these (presumed) union representatives and members of blue-collar unions. Because union ties become observable in 1985 and we cannot observe social class in the 1990s, we analyze union links from 2002 onward.

The share of politicians with union ties differs strongly between political parties and by the politician’s social class (see Figure 9). Ties are most common in the Social Democrats, especially among workers in parliament. Ties are also relatively common in the Left Party, which lacks official collaborations with specific unions but actively encourages union-party collaborations and union influence over candidate nominations (Left Party 2022). The near-total absence of union ties among Sweden Democrat politicians is unsurprising given this party’s antagonistic relationship with organized labor.

Figure 9. Proportion of Politicians with Organizational Ties to Blue-Collar Labor Unions

Note: An organizational tie is defined as having held a paid position (>10 USD) in a blue-collar union at any time since 1985. Pooled data for elected politicians in 2002–2018. N(local councilors) = 60,217; N(Parliamentarians) = 1,557.

We want to compare the promotion probabilities for workers with vs. without union ties and to compare both of these groups with nonworkers. This amounts to splitting the worker dummy in regression in Equation 1 into two dummies, one for workers who have a tie to a blue-collar labor union and one for workers who do not. The large disparities in organizational ties across parties motivate us to run this analysis separately for left-leaning parties and other parties. Because we are mainly interested in testing whether organizational ties help workers overcome the class ceiling, we include controls for seniority and cognitive skills (the Earnings Score variable) in these specifications.

Organizational ties to blue-collar labor unions help workers overcome their promotion disadvantage in politics, but only to lower-level political positions in left-leaning parties. In these parties, workers with union ties are more likely than nonworkers to enter politics and equally likely as nonworkers to become an elected local councilor (column 1 of Table 2, full list of estimates in Supplementary Table W17). Workers without union ties have familiar disadvantages in both steps. The advantage of having a union tie disappears for the two higher career rungs in left-leaning parties: we instead see similar-sized disadvantages for workers with or without union ties. While the two estimates have similar magnitudes for promotion to parliament, low statistical precision warrants some caution in this interpretation. In the other parties, union ties do not improve workers’ promotion prospects at the two lower career levels; if anything, they are associated with a larger promotion disadvantage at the two higher rungs.