Unions and the Educational Wage Premium: Collective Bargaining and Wage Inequality in the Knowledge Economy
Declining unionization has led to rising wage inequality. While most analyses of union effects have focused on low-wage workers, increasing wage inequality has also included growing differences between college and non-college educated workers. The role of unions in shaping these educational wage premium has however so far remained unexplored.
Using unique comparative data on union membership compositions, union organizations, bargaining structures and education wage premia for the period 2002 to 2020, we ask:
1. Is the educational composition of union membership related to the education premium? Unions are expected to increase inequality between members and non-members and reduce differences among members, yet these theories ignore union membership composition and union effects may depend on the relative unionization of unskilled and highly educated labor.
2. Is the union education effect related to unions’ organizational structure? Unions may increase membership by organizing both low- and high-wage earners, or by organizing specific occupational groups. These choices could have implications for union bargaining goals and strategies, and for the education wage premium.
3. Is the union education effect related to bargaining structure? Unionization and the education premium may also be linked through the structure of wage bargaining, e.g. the coordination of wage bargaining. This theoretically reduces wage dispersion, a mechanism that here will be examined empirically.
Using unique comparative data on union membership compositions, union organizations, bargaining structures and education wage premia for the period 2002 to 2020, we ask:
1. Is the educational composition of union membership related to the education premium? Unions are expected to increase inequality between members and non-members and reduce differences among members, yet these theories ignore union membership composition and union effects may depend on the relative unionization of unskilled and highly educated labor.
2. Is the union education effect related to unions’ organizational structure? Unions may increase membership by organizing both low- and high-wage earners, or by organizing specific occupational groups. These choices could have implications for union bargaining goals and strategies, and for the education wage premium.
3. Is the union education effect related to bargaining structure? Unionization and the education premium may also be linked through the structure of wage bargaining, e.g. the coordination of wage bargaining. This theoretically reduces wage dispersion, a mechanism that here will be examined empirically.
Final report
Final report for the project Unions and the Educational Wage Premium: Collective Bargaining and Wage Inequality in the Knowledge Economy (RJ 22-0697)
1. Aim and development
The background to the project was the decline in unionization observed in many countries, a decline that had been shown to have led to rising wage inequality. In addition, there had been indications of dramatic changes in union membership composition with regard to educational background and gender, changes that however had not been systematically documented. The importance of these changes in union membership in shaping the evolution of wage inequality had thus remained unexplored. The overarching purpose of this project was to document changes in the membership structure of trade unions in industrialized countries and to examine the relationship between union membership change, unions’ organizing structure, bargaining arrangements, and education wage premiums.
The main research questions outlined in the application were:
1. Is the educational composition of union membership related to the education premium? Unions are expected to increase inequality between members and non-members and reduce differences among members, yet these theories ignore union membership composition and union effects may depend on the relative unionization of unskilled and highly educated labor.
2. Is the union effect on the education premium related to unions’ organizational structure? Unions may increase membership by organizing both low- and high-wage earners, or by organizing specific occupational groups. These choices could have implications for union bargaining goals and strategies, and for the education wage premium.
3. Is the union effect on the education premium related to bargaining structure? Unionization and the education premium may also be linked through the structure of wage bargaining, e.g. the coordination of wage bargaining. This theoretically reduces wage dispersion, a mechanism that here will be examined empirically.
Furthermore, secondary research questions were related to wage gaps according to gender and age. Educational attainment among women was thus quickly overtaking that among men, and attainment among younger cohorts was outpacing that of older cohorts. These changes would most likely be related to changes in the educational composition of union members and impact on the gender and age gaps.
The purpose and research questions have remained largely unchanged.
2. Implementation
The empirical basis for the project was to be new and unique panel data on union membership composition covering 29 industrialized democracies from 2002 to 2020. No comparative information on the relative unionization rates of college- and non-college educated labor had until recently been available. However, using an approach pioneered by OECD (2017) we proposed to estimate union membership data from either European Social Survey (ESS), from national labor force or population surveys, or from national election surveys.
This we also have done, compiling membership data for 33 countries. We have thus been able to extended the data collection beyond what we originally thought would be feasible, and now have data from Australia, Austria, Belgium, Canada, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, the Republic of Korea, Lithuania, Mexico, the Netherlands, New Zealand, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, Taiwan, the UK, and the USA.
All studies were to use state-of-the-art regression methods, and we started examining pooled OLS, fixed effects (FE), and random effects (RE) regression models. However, the statistical properties of the two former types of models have lately been criticized, and we in the first paper therefore turned to recently introduced modifications of the random-effects models: the so-called random effects within and between models (REWB).
In the second paper, the long-term stability of some of the indicators made the REWB model less useful. We instead opted to use Prais-Winsten (PW) regression, a version of pooled OLS frequently used in political economy in these situations.
To bolster the casual interpretation of the regression results we looked into some of the countries that appear particular interesting from an analytical standpoint, for instance due to reforms or due to their influence on the estimated regression parameters. While not equal to synthetic control or process tracing models, these supplementary analyses provided useful insights into the processes generating the results.
3. Discussion of the project’s three main results
As noted above, there was evidence that trade union membership composition had changed in
many countries in recent years, yet no systematic comparative evidence had been available to date. The first task of the project was therefore to compile such information for as many countries and for as long as possible. So far information has been compiled for 33 countries for the period 2002 to 2020. In this period, the share of highly educated labour among union members in these countries increased from 27% to 47%. However, the overall trends conceal substantial cross-national variation. At the end of the period, university graduates made up about half or more of union membership in roughly half the countries. However, this contrasts starkly with the situation in the remaining countries. Although the membership share of non-tertiary educated had declined to some extent, they maintained a stable majority position of at least 55 %. Similar dramatic changes in mean trade union membership composition over time are observed with regard to gender and age, and again there are dramatic differences among the countries in both levels and trends. These, purely descriptive, results will be of major relevance for both scientific and policy-related discussion around the future of employment relations in industrialized countries.
Furthermore, two substantive papers explored the implications of the change in the educational composition of trade unions for wage inequality. The first paper examined the general relationship between the proportion university-educated union members and the education wage premium. Starting from the well-known median-voter theory, which postulates that the member with the median opinion on a distributional question will decide the organizations’ position, we hypothesized that the shift to a membership made up of primarily university-educated members would lead to a shift in the association between trade union membership and wage inequality. This is also what we found. Specifically, our results show that an increase of highly-educated labor among union members initially lowers the education premium, but that this relationship is reversed once the university educated become the median member.
The second paper, still under review, examined this result in the context of various forms and levels of wage coordination, i.e. the inter-union and inter-organizational context that characterize wage bargaining in many countries. Specifically, individual unions often coordinate bargaining demands with those of other unions, and similar coordination may take place among employers as well, both of which could be expected to influence the outcome of bargaining. Such wage coordination has traditionally been related to decreased wage inequality, in particular in cross-country comparisons. We again, in line with the median-voter theory, hypothesized that this may be reversed when the educational level of union members increases. We again found evidence in support of our hypothesis, yet this was limited to uncoordinated wage bargaining. In contrast, systems combining broad bargaining coverage with strong inter-industry coordination appear resistant to this shift. A shift in membership composition did in other words not lead to the same inequality increasing effect in the context of coordinated bargaining.
4. New research questions
The handbook chapter, Trade Unions and Social Policy in Post-industrial Economies: Changing Union Membership Composition and Its Policy Implications, was published in an edited volume on social policy and employment. The other chapters in the volume examined different aspects of social policy, but none of them shared our focus on changes in trade unions’ membership composition. The format of the chapter, and the available funding, did unfortunately not allow us to empirically examine the implications of the changes we documented. We therefore plan to extend our analyses beyond wage effects and examine the effects compositional changes on education, labor market as well as family policy. A first preliminary outline of one such study has been presented at the Swedish Institute for Social Research, and a research proposal on the topic is in preparation.
5. Research dissemination
Drafts of the papers have been presented at:
• the European Network for Social Policy Analysis’s annual conferences in 2023,
• the Swedish Network for Social Policy Analysis’s annual conference in 2023,
• the Uppsala Center for Labor Studies’ annual conference in 2023,
• the European Network for Social Policy Analysis’s annual conference in 2024,
• the Labor Movement’s Research Network’s conference in 2024,
• the Uppsala Center for Labor Studies’ annual conference in 2025,
• the Labor Movement’s Research Network’s conference in 2026.
Collaboration has not yet taken place, but is currently being planned.
6. List of publications
Publications
• Byun, Young-hwan, and Korpi, Tomas (2025). Trade Unions and Social Policy in Post-industrial Economies: Changing Union Membership Composition and Its Policy Implications, in Ramia, G., Irving, Z., Heins, E., and Velázquez Leyer, R. (eds.), Research Handbook on Social Policy and Employment, pp. 256–272. https://doi.org/10.4337/9781035317943.00028 (open access)
• Byun, Young-hwan, and Korpi, Tomas (2025). From compression to dispersion: how union transformation reconfigures wage inequality in the knowledge economy, Socio-Economic Review. https://doi.org/10.1093/ser/mwaf086 (open access)
Manuscripts
• Byun, Young-hwan, and Korpi, Tomas (2025). When Centralized Bargaining Ceases to Equalize Wages: Its Actors and Institutions in the Knowledge Economy, under review at the British Journal of Political Science
• Byun, Young-hwan, and Korpi, Tomas (2025). Collective wage bargaining, union membership composition and the gender pay gap (preliminary title). Manuscript in preparation for submission.
1. Aim and development
The background to the project was the decline in unionization observed in many countries, a decline that had been shown to have led to rising wage inequality. In addition, there had been indications of dramatic changes in union membership composition with regard to educational background and gender, changes that however had not been systematically documented. The importance of these changes in union membership in shaping the evolution of wage inequality had thus remained unexplored. The overarching purpose of this project was to document changes in the membership structure of trade unions in industrialized countries and to examine the relationship between union membership change, unions’ organizing structure, bargaining arrangements, and education wage premiums.
The main research questions outlined in the application were:
1. Is the educational composition of union membership related to the education premium? Unions are expected to increase inequality between members and non-members and reduce differences among members, yet these theories ignore union membership composition and union effects may depend on the relative unionization of unskilled and highly educated labor.
2. Is the union effect on the education premium related to unions’ organizational structure? Unions may increase membership by organizing both low- and high-wage earners, or by organizing specific occupational groups. These choices could have implications for union bargaining goals and strategies, and for the education wage premium.
3. Is the union effect on the education premium related to bargaining structure? Unionization and the education premium may also be linked through the structure of wage bargaining, e.g. the coordination of wage bargaining. This theoretically reduces wage dispersion, a mechanism that here will be examined empirically.
Furthermore, secondary research questions were related to wage gaps according to gender and age. Educational attainment among women was thus quickly overtaking that among men, and attainment among younger cohorts was outpacing that of older cohorts. These changes would most likely be related to changes in the educational composition of union members and impact on the gender and age gaps.
The purpose and research questions have remained largely unchanged.
2. Implementation
The empirical basis for the project was to be new and unique panel data on union membership composition covering 29 industrialized democracies from 2002 to 2020. No comparative information on the relative unionization rates of college- and non-college educated labor had until recently been available. However, using an approach pioneered by OECD (2017) we proposed to estimate union membership data from either European Social Survey (ESS), from national labor force or population surveys, or from national election surveys.
This we also have done, compiling membership data for 33 countries. We have thus been able to extended the data collection beyond what we originally thought would be feasible, and now have data from Australia, Austria, Belgium, Canada, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, the Republic of Korea, Lithuania, Mexico, the Netherlands, New Zealand, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, Taiwan, the UK, and the USA.
All studies were to use state-of-the-art regression methods, and we started examining pooled OLS, fixed effects (FE), and random effects (RE) regression models. However, the statistical properties of the two former types of models have lately been criticized, and we in the first paper therefore turned to recently introduced modifications of the random-effects models: the so-called random effects within and between models (REWB).
In the second paper, the long-term stability of some of the indicators made the REWB model less useful. We instead opted to use Prais-Winsten (PW) regression, a version of pooled OLS frequently used in political economy in these situations.
To bolster the casual interpretation of the regression results we looked into some of the countries that appear particular interesting from an analytical standpoint, for instance due to reforms or due to their influence on the estimated regression parameters. While not equal to synthetic control or process tracing models, these supplementary analyses provided useful insights into the processes generating the results.
3. Discussion of the project’s three main results
As noted above, there was evidence that trade union membership composition had changed in
many countries in recent years, yet no systematic comparative evidence had been available to date. The first task of the project was therefore to compile such information for as many countries and for as long as possible. So far information has been compiled for 33 countries for the period 2002 to 2020. In this period, the share of highly educated labour among union members in these countries increased from 27% to 47%. However, the overall trends conceal substantial cross-national variation. At the end of the period, university graduates made up about half or more of union membership in roughly half the countries. However, this contrasts starkly with the situation in the remaining countries. Although the membership share of non-tertiary educated had declined to some extent, they maintained a stable majority position of at least 55 %. Similar dramatic changes in mean trade union membership composition over time are observed with regard to gender and age, and again there are dramatic differences among the countries in both levels and trends. These, purely descriptive, results will be of major relevance for both scientific and policy-related discussion around the future of employment relations in industrialized countries.
Furthermore, two substantive papers explored the implications of the change in the educational composition of trade unions for wage inequality. The first paper examined the general relationship between the proportion university-educated union members and the education wage premium. Starting from the well-known median-voter theory, which postulates that the member with the median opinion on a distributional question will decide the organizations’ position, we hypothesized that the shift to a membership made up of primarily university-educated members would lead to a shift in the association between trade union membership and wage inequality. This is also what we found. Specifically, our results show that an increase of highly-educated labor among union members initially lowers the education premium, but that this relationship is reversed once the university educated become the median member.
The second paper, still under review, examined this result in the context of various forms and levels of wage coordination, i.e. the inter-union and inter-organizational context that characterize wage bargaining in many countries. Specifically, individual unions often coordinate bargaining demands with those of other unions, and similar coordination may take place among employers as well, both of which could be expected to influence the outcome of bargaining. Such wage coordination has traditionally been related to decreased wage inequality, in particular in cross-country comparisons. We again, in line with the median-voter theory, hypothesized that this may be reversed when the educational level of union members increases. We again found evidence in support of our hypothesis, yet this was limited to uncoordinated wage bargaining. In contrast, systems combining broad bargaining coverage with strong inter-industry coordination appear resistant to this shift. A shift in membership composition did in other words not lead to the same inequality increasing effect in the context of coordinated bargaining.
4. New research questions
The handbook chapter, Trade Unions and Social Policy in Post-industrial Economies: Changing Union Membership Composition and Its Policy Implications, was published in an edited volume on social policy and employment. The other chapters in the volume examined different aspects of social policy, but none of them shared our focus on changes in trade unions’ membership composition. The format of the chapter, and the available funding, did unfortunately not allow us to empirically examine the implications of the changes we documented. We therefore plan to extend our analyses beyond wage effects and examine the effects compositional changes on education, labor market as well as family policy. A first preliminary outline of one such study has been presented at the Swedish Institute for Social Research, and a research proposal on the topic is in preparation.
5. Research dissemination
Drafts of the papers have been presented at:
• the European Network for Social Policy Analysis’s annual conferences in 2023,
• the Swedish Network for Social Policy Analysis’s annual conference in 2023,
• the Uppsala Center for Labor Studies’ annual conference in 2023,
• the European Network for Social Policy Analysis’s annual conference in 2024,
• the Labor Movement’s Research Network’s conference in 2024,
• the Uppsala Center for Labor Studies’ annual conference in 2025,
• the Labor Movement’s Research Network’s conference in 2026.
Collaboration has not yet taken place, but is currently being planned.
6. List of publications
Publications
• Byun, Young-hwan, and Korpi, Tomas (2025). Trade Unions and Social Policy in Post-industrial Economies: Changing Union Membership Composition and Its Policy Implications, in Ramia, G., Irving, Z., Heins, E., and Velázquez Leyer, R. (eds.), Research Handbook on Social Policy and Employment, pp. 256–272. https://doi.org/10.4337/9781035317943.00028 (open access)
• Byun, Young-hwan, and Korpi, Tomas (2025). From compression to dispersion: how union transformation reconfigures wage inequality in the knowledge economy, Socio-Economic Review. https://doi.org/10.1093/ser/mwaf086 (open access)
Manuscripts
• Byun, Young-hwan, and Korpi, Tomas (2025). When Centralized Bargaining Ceases to Equalize Wages: Its Actors and Institutions in the Knowledge Economy, under review at the British Journal of Political Science
• Byun, Young-hwan, and Korpi, Tomas (2025). Collective wage bargaining, union membership composition and the gender pay gap (preliminary title). Manuscript in preparation for submission.