What makes workers happy: Empowerment, unions or both?

Does the negative effect of union membership on job satisfaction, as shown in Anglophone countries, also hold for Continental Western Europe? Given the differences in industrial relations, I hypothesize that the effect will be different. I also test hypotheses about the effect of empowerment on job satisfaction, which might explain the negative union effect, and broaden the analysis to include pay satisfaction. Analyses of European Social Survey data show that the negative union effect does not exist for Continental Western Europe and that this can be explained by empowerment of employees.


Introduction
Over a long period, research has shown that unions improve wages and working conditions, but somewhat paradoxically that unionized workers are less satisfied than average with their job. This result has been reproduced in many studies (such as Bender and Sloane, 1998;Freeman and Medoff, 1984;Gordon and Denisi, 1995;Guest and Conway, 2004;Miller, 1990;Renaud, 2002), mainly conducted in the USA and other Anglophone countries. Laroche (2016) shows in a meta-analysis that these results are robust. However, their explanations are not.
In the literature, two explanations prevail. The first is based on the 'voice' thesis, which states that union members voice their discontent, whereas non-members leave the firm. Furthermore, unions raise discontent to increase their bargaining power (Freeman and Medoff, 1984;Gordon and Denisi, 1995). The second derives from the thesis that workers in bad jobs and, relatedly, dissatisfied workers are more likely to unionize. This is also known as reverse causality (Bender and Sloane, 1998;Bryson et al., 2010;Hammer and Avgar, 2005;Schwochau, 1987). Because empirical evidence of these explanations is mixed, I put forward an additional explanation based on empowerment. Freeman and Rogers (2006) show for the USA that workers like to be empowered, but mostly are not.
Empowered employees have a say, sometimes small sometimes large, in how an organization is managed. By empowerment I mean that employees gain decision rights, or influence, over how the organization is managed, which goes beyond autonomy in their own job. Empowered workers not only have a say in how their own job is performed best (autonomy) but also in how the organization is or could be run, how their colleagues could do their jobs and how teams are organized and tasks are allocated among colleagues and team members. It goes from irregular informal work meetings, via official works councils to employee representatives on the board of the organization. Empowerment covers both unionized and non-unionized workers and firms. However, empowerment is connected to unionization: Freeman and Lazear (2006) and Lazear and Gibbs (2009) present an economic model of empowerment and connect this directly to unionization of firms. But employees in non-unionized firms can be empowered too. The elaboration towards empowerment is based on the fact that workers with high levels of autonomy, one form of empowerment, are happier ( Van der Meer and Wielers, 2013). This robust result links empowerment to job satisfaction and would suggest that union members are more satisfied than non-members, because they have higher levels of empowerment (Guest and Conway, 2004).
I test this idea not only on Anglophone data but on other Continental Western European (CWE) countries too, to add different industrial relations systems into the research about the effect of union membership on job satisfaction. Such research outside Anglophone countries is rare, with a few notable exceptions such as Donegani and McKay (2012), García-Serrano (2009), Hipp and Givan (2015) and Laroche (2017). The effect of union membership on job satisfaction could be different in CWE countries, given the differences in industrial relations institutions and practices. This idea is linked to the effect of the industrial relations climate on job satisfaction (Hipp and Givan, 2015).
Research into the combined effect of unions and empowerment on job satisfaction is rare. I could only find two studies (Holland et al., 2011;Wood, 2008) that report on these effects, but these two studies treat unions and empowerment as separate features. They do not hypothesize any linkages between these concepts.
To test the effects of empowerment and union membership, I also look into pay satisfaction. This concerns only the pecuniary reward of a job, not the non-pecuniary rewards, so is therefore a somewhat more limited concept than job satisfaction. I use the fifth round of the European Social Survey (ESS) held in 2010 to test my hypotheses. I use data from 13 different European countries, covering CWE and Ireland and the UK. The data set is a representative random sample from these populations and contains information about union membership, work, different forms of empowerment and job and pay satisfaction. The ESS is the only survey that I could find, which contains information on both union membership and various forms of empowerment for different countries.
In the next section, I summarize the main literature about union membership and job satisfaction and elaborate on the effects of empowerment. Next I explain the differences between the countries. This results in several hypotheses which will be tested. I then address the data and methods, before presenting the results of the analyses. In the final section, I summarize the research and draw some conclusions, including policy recommendations. Freeman and Medoff (1984) wrote that in the USA, unions improve workers' working conditions and wages, but that unionized workers were less satisfied than non-unionized with their job. At least since these results have been published, this has become a theoretical puzzle. A prominent explanation in the literature is the 'exit-voice' thesis, which reads that dissatisfied non-members leave the organization, whereas dissatisfied members raise their voice (Bryson et al., 2004(Bryson et al., , 2010. Implicit in this thesis is that raising voice does not diminish complaints; these are at best partially resolved, otherwise the dissatisfaction would disappear. It further raises the question why these dissatisfied union members do not look for another job. This is addressed by a second explanation, which focuses on specific individual characteristics and states that the dissatisfied union members are sorted into jobs with bad working conditions. It is primarily these workers who are dissatisfied who become unionized, and additionally complainers who join unions. Thus, some kind of sorting mechanism explains the differences in job satisfaction between union members and nonmembers (Bender and Sloane, 1998;Hammer and Avgar, 2005;Laroche, 2017). Holland et al. (2011) show that the possibility of voice explains the negative effect of union membership in Australia. They studied the effect of direct voice (one form of empowerment) and union voice on job satisfaction, because an important role of unions is to represent workers and express their voice in their relations with employers. Now that union membership is declining and more and more employers give direct voice to workers instead of unions, they wanted to know how this relates to job satisfaction. In their research, they find a positive effect of direct voice and autonomy on job satisfaction, but they do not find any effect of union voice and union membership.

Unions, job satisfaction and empowerment
In these studies, voice is measured or defined as the extent to which managers at the workplace consult employees or their representatives and whether a union is present at the workplace. I follow this definition. The first part refers to what I call empowerment, in which employees have direct contact with their colleagues, supervisors and higher managers to improve work and/or working conditions. The second part refers to indirect or collective voice in which union representatives either through presence at the workplace or through central collective negotiations try to improve work and working conditions. Empowerment or direct voice is studied more often in Human Resource Management, whereas indirect or collective voice is studied more often in Employment or Industrial Relations (Mowbray et al., 2015). I use both concepts, because empowerment and collective voice are complements, as Holland et al. (2011) show. I further use measures of autonomy, because these are linked to empowerment (Carr and Mellizo, 2013;Mowbray et al., 2015). Holland et al. (2012) examined the relationship between employee voice arrangements and employees' trust in management. They found a positive relation between trust in management and empowerment but a negative relation with union voice. This might explain why a negative relation is found between job satisfaction and union membership and a positive relation with empowerment. In unionized firms, there is a lack of trust between employees and management, which has a large negative effect on job satisfaction (Helliwell and Huang, 2011). They further show that union members find trust less important than non-unionists but also that trust is lower, on average, in unionized firms than in non-unionized. Managerial opposition to unions lowers trust in management (Bryson, 2001), and thereby job satisfaction, explaining the effect of union membership on job satisfaction in the Anglophone context. So the level of empowerment and the way employees are empowered might explain the negative effect of union membership on job satisfaction in these countries. Empowerment without union involvement increases job satisfaction, whereas empowerment through unions decreases job satisfaction, because of poor relations between employees and management.
Employers do not trust unions, and an increasing number in Anglophone countries oppose unions. They make it more difficult for workers to unionize and many try to drive out unions. The main reason is to increase shareholder value and reduce the labour share in added value (Freeman and Lazear, 2006;Lazear and Gibbs, 2009). Given such distrust between management and employees, union members in unionized firms are less satisfied. If employers find a way to empower employees without involving unions, these employees will be more satisfied.
So my first hypothesis is that unions have a negative effect on job satisfaction. My second is that empowerment has a positive effect. My third is that the association between empowerment and non-union firms can explain the negative union effect. I also expect, fourth, that the effect of direct voice, one form of empowerment, on job satisfaction will be higher than that of collective voice, because of the role trust plays in the employment relation.
To test these hypotheses fully, I would need data about job satisfaction, union membership, empowerment and trust. Unfortunately, I was unable to find a data set that contains all this information. So, I have to find a way around. One way of doing this is by looking at differences between countries, especially between those with different systems of industrial relations. A main characteristic of the Anglophone system is the activity of unions within firms (Bamber et al., 2016). Firms are unionized or not: in some, unions and union members play an active role in negotiations between employees and management about wages and working conditions, and these negotiations are the cause of mistrust between management and unions. The unions do not play such a role in nonunionized firms. In these firms, individual workers or non-union representatives negotiate, through direct voice. Given the absence of the unions at the firm level in these firms, the level of trust between management and employees is higher.
In CWE, industrial relations are organized differently. Unions play less of a role at the firm level, and it is therefore not possible to distinguish between unionized and nonunionized firms, only between those covered or not covered by collective agreements. There are often big differences between union density and bargaining coverage (Visser, 2010). So improvements achieved by unions are not restricted to union members, whereas empowerment is restricted to the employees directly involved. When unions negotiate about wages and working conditions, they do so mostly at the industry, regional or even national level (Bamber et al., 2016). Although there might be distrust between unions and employers, this distrust is not felt directly in the workplace, because negotiations take place at a higher level. The amount of trust between employees and management is less affected by what the unions do, because they are less present on a day-to-day basis in the firms themselves. I therefore hypothesize, fifth, that the effect of union membership on job satisfaction will be less negative in CWE than in Anglophone countries. The effect could even be positive, because of the improvements in working conditions and wages.
Empowerment will have the same positive effect on the relations between employees and management in CWE as in the Anglophone countries. The reason is the same. Empowerment improves trust between employees and management, and thereby increases job satisfaction. For CWE, I expect to find the same positive effect of empowerment on job satisfaction as for the Anglophone countries.
Because negotiated wage increases are much more restricted to union members in the Anglophone countries than in CWE, I hypothesize, sixth, that the effect of union membership on pay satisfaction will be stronger in the Anglophone countries than in CWE.

Data and method
Data I test my hypotheses on the basis of ESS round 5, held in 2010. I use data from 13 different European countries, comprising two subpopulations: CWE (Belgium, Switzerland, Denmark, Germany, Spain, Finland, France, the Netherlands, Norway, Portugal and Sweden) as coordinated market economies, and Anglophone countries (UK and Ireland) as liberal market economies. The data set is a representative sample from the population and contains information about union membership, work, empowerment, direct voice, collective voice and job and pay satisfaction. The ESS has already been used by Donegani and McKay (2012) to test the effect of union membership on job satisfaction. They used waves 3 and 5 and more countries. However, they did not test for the effects of empowerment. The ESS, round 5, contains a model on work-life balance and has therefore more information about job characteristics than the other rounds of this survey.
My subsample consists of persons in paid employment and between 25 and 60 years of age. Those younger could still be in full-time education and those older could already be retired. Those in the sample have a job and can become a union member. I further restrict the sample to those who answered the questions of interest. For the control variables, I applied a strategy of mean substitution. Although the number of missing answers per variable is not high, I would lose too many respondents, given the number of control variables.
Job satisfaction is measured by the question 'How satisfied are you in your main job?' with a range from 0 to 10. As a second dependent variable I use pay satisfaction, measured on a five-point scale by a single question.
Trade union membership is captured by a question about current and past union membership. I created one dummy variable indicating current union membership. I measure empowerment in three different ways: autonomy, direct voice and influence. To measure autonomy, I combine three questions into one measure: being allowed to decide how to organize one's daily work, the influence to choose or change the pace of work and the ability to decide the time to start and finish work. I rescaled all answers on a scale from 0 to 10 to make them comparable. The reliability of this scale of autonomy is 0.74. I measure direct voice by a question about how much influence discussions have on decisions that affect working conditions, on a scale from 0 to 4. I measure influence by a question on the possibility to influence policy decisions about activities of organization, on a scale from 0 to 10. I construct a measure of collective voice on the basis of a question about the influence of trade unions on working conditions and practices on a scale from 1 to 4. This could be through trade union representatives in works councils or through central collective agreements. I cannot distinguish between these two. Because a union is not present in all the firms, I created a dummy variable indicating the absence of a union and replaced the missing values on collective voice by the country group mean and centred this variable around its group mean. In this way, I capture the true effect of collective voice.
To take account of sorting effects, I control for both personal and job characteristics that could influence union membership. As personal characteristics, I include age (and its square), gender, having a partner, having children living at home, educational level, subjective health (reverse coded), subjective income (reverse coded), working experience, tenure in the occupation and tenure at the firm.
For job characteristics, I use information about being a supervisor, the employment contract (temporary or open-ended), contractual working hours and irregular working hours (a combination of working evenings or nights, overtime on short notice and at weekends), on a scale from 1 to 5 (reliability = 0.67). I measure the length of job-specific training and whether the respondent has attended a work-related course, lecture or conference. I further include an indicator variable of establishment size, consisting of five categories. Further measures of job characteristics include variety in work, learning new things and support from co-workers and job security, all measured on a scale from 1 to 4. I include two measures of work pressure, 'my job requires that I work very hard' and 'I do not have enough time to get everything done', on a scale from 1 to 5. I also have a measure of opportunities for advancement on a scale from 1 to 5. As additional controls in the treatment effects analyses, I use the first digit of the statistical classification of economic activities in the European Community (NACE), as union membership and job characteristics differ between industries.
Descriptive information on most variables and the correlations between my variables of interest for both subpopulations is presented in the Online Annex.

Method
I start with a regression model using job satisfaction and pay satisfaction as dependent variables. To control for country differences, I use robust regression where the standard errors are clustered at the country level and include country fixed effects. I estimate (versions of) the following equation

H = + U + A + DV + I + CV + C +
where H i is job satisfaction or pay satisfaction of a person i. U indicates union membership; A, autonomy; DV, direct voice; I, influence; CV, collective voice; and C, the control variables consisting of the personal and job characteristics and country fixed effects. I use a linear model instead of an ordered model, because these models hardly show any differences in results when the dependent variable has an 11-point scale like job satisfaction (Van der Meer and Wielers, 2013). The 'exit-voice' hypothesis suggests that union membership is an endogenous variable and is dependent on the type of job and type of worker. To take account of the endogeneity of union membership, I estimate an endogenous treatment effects model. The general model is In this system, the error terms ε i and w i are correlated. I estimate the model with the probit of joining a union (equation (3)) and use this result to calculate the Inverse Mills Ratio (IMR). The IMR is included in equation (2). Applying maximum likelihood methods produces consistent estimates of the parameters. To fit the model, it is necessary to include instrumental variables for joining a union. I use industry dummies (Ind) as instruments because other research (Bryson et al., 2004) shows these to be good instruments. Union membership varies over industries and previous research has shown that these can help to predict union membership. In Stata, a module is implemented to estimate such a model. I use the survey data module (svy) to weight properly the data as the ESS organization suggests. This module also calculates robust standard errors. Table 1 contains the results on job satisfaction for the UK and Ireland and for CWE. In the basic model, which contains only the effect of union membership on job satisfaction, I find no effect on job satisfaction in the UK and Ireland. In model 2, I find that the effect of union membership is positive but also that the effect of union presence is negative. Their combined effect is negative (-0.142 (0.030) t = -4.78) and significant, corroborating earlier research on the effect of unions on job satisfaction in Anglophone countries and supporting my first hypothesis. This result supports the idea regarding trust between workers and management, which is violated once the union is present in a firm or organization. The effect stays negative after controlling for personal characteristics. In the full model, controlling for both person and job characteristics, the effect of union membership is small and insignificant and the effect of union presence is positive but also small. This result suggests that some kind of sorting mechanism contributes to the explanation of the negative union effect.

Results
Furthermore, I find effects of empowerment on job satisfaction: both direct voice and influence have positive and sizeable effects. I also find a positive effect of collective voice on job satisfaction. This supports my second hypothesis. Given the insignificant union effect, I did not find support for hypothesis 3 about the mediating effect of empowerment. The effect of direct voice is stronger than that of collective voice, supporting hypothesis 4.
For the UK and Ireland, I find a positive raw effect of union membership on pay satisfaction. This supports hypothesis 6 and corroborates the results of research that shows that union members earn higher wages than non-union members. This raw effect is explained away in the full model. The indicators for empowerment and collective voice reduce the effect by approximately 50 percent. The effect becomes even negative after controlling for personal characteristics but disappears in the full model in which I also control for job characteristics. In the full model, I find a strong positive effect of union presence. The combined effect of union membership and union presence is significant (0.221 (0.021) t = -10.14). I also find a positive effect of collective voice. All other indicators of empowerment and collective voice also have a positive effect on pay satisfaction. All results for pay satisfaction are presented in the Online Annex. Table 1 also presents the results for CWE. Union membership has no effect on job satisfaction in any of the models. Contrary to the UK, I do not find effects of union presence on job satisfaction. This supports hypothesis 5 about the differences between the UK and Ireland as against CWE. Furthermore, I find strong effects of autonomy, direct voice and influence, and in the case of CWE, empowerment does mediate to a small extent the union effect, supporting hypotheses 2 and 3. In the case of CWE, I do not find a difference in effect between direct and collective voice.
There is no raw effect of union membership on pay satisfaction. The effect becomes negative in the full model. This supports hypothesis 6, which states that this effect will be smaller in CWE than in the UK and Ireland. There is a positive effect of autonomy and a strong positive effect of collective voice on pay satisfaction.
Because the sorting explanation of the negative union effect on job satisfaction finds some support in these analyses, I also estimated a treatment effects model, to model explicitly the sorting mechanism. These results are shown in Table 2, which shows that union membership has a strong positive effect on job satisfaction. This effect is much stronger than the effect of union presence, so unions do increase job satisfaction for the UK and Ireland. I also find positive effects of autonomy, direct voice and influence. So empowerment increases job satisfaction; it also mediates the union membership effect, although not completely. For the UK and Ireland, I find an insignificant effect of union membership on pay satisfaction and a positive effect of union presence, making the combined effect positive as well. I also find positive effects of autonomy, direct voice and influence, but these effects are much smaller. This is the same as in the regression model. Again the effect of direct voice is larger than that of collective voice. Table 2 shows a strong positive effect of union membership on job satisfaction for CWE. This is different from the regression analyses, so unions do improve job satisfaction in CWE. I also find effects of autonomy, direct voice and influence on job satisfaction. These effects of empowerment hardly mediate the effect of union membership. I do not find an effect of union membership on pay satisfaction. The only significant effect is that of autonomy.

Summary and conclusion
In this research, I tested an additional explanation for the negative union effect on job satisfaction. In the literature, two explanations stand out. The first is based on the exitvoice hypotheses as put forward by Freeman and Medoff (1984) and the second on sorting effects (Bender and Sloane, 1998). Because support for these explanations is mixed, I put forward an additional explanation based on empowerment: this may have a positive effect on job satisfaction because it improves the employment relationship. Empowered workers have a trust relation with management, which improves job satisfaction. However, in unionized firms, this trust relation is absent. Management does not trust unions and many firms try to prevent from becoming unionized. So I test whether empowerment has a positive effect on job satisfaction and if this can explain the union effect. I do so first for the Anglophone countries, because most research into the union effect stems from these countries. Second, I also test this hypothesis for 11 CWE countries, for two reasons: first to estimate the union effect for these countries, because research into the union effect outside Anglophone countries is still rare; and second because it provides a further test of the relation between empowerment, the union effect and job satisfaction. This is important because there is an essential difference in industrial relations systems between the Anglophone countries and CWE, which is the presence of the union within firms, where unions are active in individual firms (the unionized ones) in Anglophone countries and unions are less active in the other CWE countries. Here, negotiations between unions and employers take place at the industrial, regional or national level and much less at the firm level. In these countries, the distrust between union and employers is much less felt on the work-floor.
My first hypothesis is a replication of the negative union effect for the Anglophone countries. Although I did not find a direct negative effect of union membership on job satisfaction, I did find a negative effect of union presence. So unions do have a negative effect on job satisfaction. My second hypothesis reads that empowerment has a positive effect on job satisfaction. I did find support for this hypothesis in the Anglophone countries: all my indicators of empowerment showed strong positive effects on job satisfaction. However, for these countries, I did not find a mediating effect of empowerment on the negative union effect, my third hypothesis. Thus, I cannot say that empowerment explains the union effect in the UK and Ireland.
My fourth hypothesis about the difference in effect on job satisfaction between direct and collective voice is also supported. As expected, I find a larger effect of direct than of collective voice.
Given the differences in industrial relations, I expected to find a less negative effect of unions on job satisfaction in CWE, hypothesis 5. The results of the analyses support this hypothesis, because I did not find significant effects of union membership or union presence on job satisfaction.
Hypothesis 6 concerning the difference in union effect on pay satisfaction between the UK and Ireland and CWE finds support in the analysis. As expected, the effect for the UK and Ireland is larger than for CWE, because of the more decentralized negotiations.
All in all, I can say that most hypotheses were supported by the empirical results and that the negative union effect found in Anglophone countries does not hold for CWE. This can be explained by the differences in industrial relations. In the Anglophone countries, unions are organized within firms and are much more present there. In these countries, firms are unionized or non-unionized, meaning that, where present, unions can negotiate with management about working conditions and pay, representing the union members. In CWE, negotiations between unions and employers take place mainly outside the firm; so unions are much less present and visible within firms than in the Anglophone countries. In CWE, employee representation takes place in regular work meetings, works councils and sometimes even in the board of the organization, as in Germany. Employees are not necessarily represented by union members but can be represented by non-unionists too. Therefore, workers are more satisfied by empowerment than by collective voice, as shown in this research.
That is also the reason why empowerment explains the smaller union effect in CWE but does so less in the UK and Ireland. Furthermore, I see that in the Anglophone countries, union members are more satisfied with their pay, whereas I do not find such an effect in CWE. Within Anglophone countries, the union members are directly represented in wage negotiations which take place within the firm, whereas in CWE these negotiations take place at a higher level where unions negotiate the wages of all the workers, not necessarily union members.
The differences in industrial relations also explain the relations between workers and managers within firms. In the UK and Ireland, unions and their members are mistrusted. Many managers oppose, sometimes actively, the unionization of their firm. If a firm is unionized, despite opposition, relations are often poor. This affects job satisfaction and subjective well-being. Trust at the workplace, between colleagues and managers, is a very strong contributor to job satisfaction. Although I do not have a direct measure of workplace trust, I see in my results that supervisors in the UK and Ireland have lower than average job satisfaction, whereas in CWE they have at least the same level of satisfaction. My interpretation is that in the UK and Ireland, the supervisors are caught between the workers whom they supervise and higher management who do not trust the ordinary worker. In CWE, with the tradition of direct representation, supervisors feel less trapped between workers and higher management. This supports the idea that trust at the workplace is important and suggests that this is different between the Anglophone countries and CWE.
My research does not reject the sorting and 'exit-voice' hypotheses that have been formulated to explain negative effects of unions in Anglophone countries. I find support for these hypotheses, but they do not tell the complete story, and do not explain the differences between the UK and Ireland and CWE. Next to the sorting that is going on, we need to take account of the importance of empowerment. Empowerment, combined with trust, explains the differences between CWE and the Anglophone countries.
A shortcoming of this research is that I do not have a direct measure of trust at the workplace. I was unable to find a data set which includes the variables used here, job satisfaction, union membership, empowerment, collective voice and further job and personal characteristics, and also has a measure of trust at the workplace for both Anglophone countries and CWE. Maybe data from individual countries might be available to test the importance of workplace trust in CWE countries, as done by Helliwell and Huang (2011). This might shed more light on my proposed explanation of the negative union effect.
So this research suggests that managers should empower workers by giving them direct voice and possibilities to influence company policies and maybe even encouraging them to join a union. This empowerment not only increases job satisfaction but also increases productivity. Specific information is better used and this will increase productivity. In doing so, management should also increase autonomy and use more modern methods of job design.
However, nowadays in many European countries, union density is declining and union power as well. It has become more difficult for unions to organize workers; they do not see the benefits of the union or they free ride. In many instances, the negotiated pay and working conditions between unions and employers apply for both members and non-members. Moreover, some workers are still afraid to unionize, fearing that they might lose their job.
Workers in newer industries are less likely to unionize. Most unionized workers can be found in the public sector and in 'old' declining industries, so at national level unions are confronted with declining membership rates. Given the results of this research, this trend is bad for society at large. Governments could maybe help unions to increase membership density.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental material
Supplemental material for this article is available online.