What Works? How Combining Equal Opportunity and Work–Life Measures Relates to the Within-Firm Gender Wage Gap

Abstract

In this paper, we investigate how organizational equal opportunity and work–life measures when combined are associated with the gender wage gap within large German establishments. By looking at both kinds of measures as parts of a comprehensive personnel strategy, we provide a novel way to distinguish between a career-supportive implementation of work–life measures versus implementation as a compensating differential. Using a linked employer–employee dataset, we performed a cross-sectional multilevel regression analysis with fixed effects for 6439 respondents within 122 establishments. The results indicate that work–life measures that support employees in their parenthood responsibilities are linked to a significantly higher GWG, particularly for parents. Our results indicate that the implementation of work–life measures is used as compensating differentials, primarily for mothers. Particularly, this can be found when looking at the combination of multiple measures. In combination with equal opportunity measures for women, work–life measures that highlight the long-term absence from the workplace are associated with a higher GWG for parents.

1. Introduction

Gender inequality in the labor market is a long-standing topic in social inequality research. To understand how gendered wage differences emerge, it is essential to consider the role of establishments. These organizations are where wages are negotiated—in other words, they are “local inequality regimes” (Tomaskovic-Devey and Avent-Holt 2019, p. 82) embedded in the larger context of institutional policies of welfare state regimes. Therefore, focusing on the role of establishments does not deny the role of the institutional environment and political actors. However, it is at the workplace where important decisions about hiring, promotion, and wages are made (Acker 2006; Tomaskovic-Devey and Avent-Holt 2019).

These considerations also apply to the widely discussed issue of the gender wage gap (GWG) (Abendroth and Diewald 2019; Abendroth et al. 2017; Huffman et al. 2017; Tomaskovic-Devey and Avent-Holt 2019). Germany represents a case in which gender differences in earnings are relatively large and persistent. Between 2012 and 2015, when the study was carried out, the unadjusted GWG in Germany was around 22%, and still is 18%—which is significantly above the European average (Eurostat 2021). Moreover, the glass ceiling effect is strongly pronounced in Germany, with a small share of women in higher status positions and distinct gender inequality in the higher income quantiles (Holst and Wrohlich 2019; Huffman et al. 2017). Likewise, in other countries, these differences can be explained by gendered segregation into different occupations and branches (Kunze 2005), more overworking of men (Lott and Chung 2016), and, specifically in Germany, a high share of women (especially mothers) who work part-time (Matteazzi et al. 2018). Nevertheless, when these characteristics are controlled for, the adjusted GWG, it is about 6% (Eurostat 2021). In this paper, we focus on this adjusted GWG within large German establishments.

Germany is a particularly interesting choice for studying gender inequalities because of the pronounced co-existence of institutional and organizational inducements for more traditional as well as for more egalitarian life scripts. The German welfare state is characterized by a relatively poorly developed early childcare system in West Germany and an educational system in which the school day often ends at lunchtime. Moreover, there are policies that encourage one parent to stay at home (e.g., offering advantageous tax incentives for partnerships with an unequal income distribution) or that permit part-time employment after childbirth as well as a long parental leave period (Boll and Lagemann 2019; Dieckhoff et al. 2016; Steiner and Wrohlich 2004; Esping-Andersen 1990). These conditions contribute to the persistence of a traditional gendered division of labor. Women work part-time more often than they do in most other countries and take on a disproportionately larger share of housework and care work (Trappe et al. 2015). Nevertheless, more egalitarian gender arrangements are becoming more widespread. Focusing on the relatively conservative aspects of the German welfare state neglects the manifold policies that support gender equality (e.g., anti-discrimination laws and initiatives to better reconcile competing demands at work and in home life) (Federal Anti-Discrimination Agency 2019; Henninger et al. 2008) or workplaces offering both gender equality and work–life measures that support more egalitarian life scripts as a means to recruit and bind sought-after employees. Thus, Germany combines a plurality of gendered life scripts, including more traditional and egalitarian arrangements. Given these contradictions in institutional imperatives and often half-hearted policies, it is crucial to examine how employers react to the plurality of gendered life scripts and living arrangements and their perceived costs and benefits. In this regard, Germany may be considered an interesting case for the international reader.

When looking at the role of establishments in shaping the adjusted GWG, our interest is focused on organizational personnel policies and, more specifically, on the association between the GWG and two kinds of potentially relevant components of such policies: equal opportunity measures (EQM) and work–life measures (WLM). EQM are designed to meet gender inequalities in job chances, regardless of their parental status (Wirth 2001). They include preferential staffing to meet a women’s quota1, career planning and mentoring programs that focus on women, and mixed teams that aim to better integrate women in work teams. WLM that are targeted to help employees reconcile job and private demands include flexible working hours and home-based telework (primarily for everyone), as well as support for parental leavers or corporate childcare (specifically for parents) (Kossek and Ollier-Malaterre 2013). All these work–life measures may reduce gender inequality in earnings if they support women—since women face greater competing demands between work and family life (e.g., Bracke et al. 2008).

However, research about the association of EQM or WLM with gender-specific wage inequalities shows no clear association with higher wages for women, as well as some other somewhat ambivalent patterns. Recent results point to differences in the effect of measures depending on their type and organizational norms, and between women with higher or lower occupational statuses (e.g., Glass 2004; Huffman et al. 2017; Peters et al. 2020; Van der Lippe et al. 2019). An explanation for these results is that their implementation in establishments is subject to different rationales in organizational strategies (Abendroth and Diewald 2019; Filer 1985; Tomaskovic-Devey and Avent-Holt 2019).

In previous research, the two types of measures are usually explored as single measures that only stand for their specific concern—and are interpreted as such—rather than set in context with other measures. In contrast, we assume that it is essential to understand the measures in their embedding in a comprehensive personnel strategy. In other words, it is not only that both EQM and WLM contribute separately to the GWG, it is the cumulation of single measures for both WLM and EQM, as well as the combination of both kinds of measures, that can help to reveal the heterogeneity in their effects on the GWG across establishments. This applies to the inherent ambiguity of WLM, i.e., the question of whether they are meant as a concession at the expense of pay or to support the employability of those with competing family demands. A follow-up question is whether the cumulation and combination of measures work in the same direction or work differently for different parts of the workforce. According to existing research in this field, relevant boundaries could apply to parenthood and its intersection with gender and the divide between lower and higher qualifications (e.g., Glass 2004; Huffman et al. 2017; Peters et al. 2020; Van der Lippe et al. 2019). We ask whether looking at the comprehensive personnel strategy—the cumulation of EQM or WLM and the moderation of EQM—is important to disentangle the association between WLM and gender inequalities in wages.

This article contributes to existing research by emphasizing the role of work establishments and their personnel policies in the genesis of gender inequalities in wages. Moreover, we look beyond single measures and toward a more practical approach for establishments’ policy strategy. Thereby, we get better insights into the rationales behind the implementation of gender equality measures, differences in the addressed groups of employees, and addressing existing or even solidifying gender inequalities within establishments.

2. Human Resource Management: The Role of Equal Opportunity and Work–Life Measures

During the last few decades, workforce heterogeneity increased considerably with the rise of women at work and an increasing variety of living arrangements and life scripts. Consequently, employees’ needs and preferences for flexibility and agency, but also for reliability and certainty, grew (Beauregard et al. 2009; Coyle-Shapiro and Shore 2007). Thus, the importance of gender equality and the possibility of reconciling work and private life gained unprecedented relevance for hiring and binding required employees. The role of establishments in providing EQM and/or WLM has been emphasized in public and academic discussions (Moen 2015; Tomaskovic-Devey and Avent-Holt 2019). Establishments are places where employment chances and working conditions are negotiated and gratifications and resources are distributed (Tomaskovic-Devey and Avent-Holt 2019, p. 82). Offering measures like EQM or WLM as part of employment relationships possibly addresses specific groups of the workforce differently and emphasizes certain gratifications and demands more than others.

EQM and WLM are signals that the concerns and the groups primarily addressed are indeed essential employer concerns when it comes to developing personnel strategies (Grover and Crooker 1995). The mechanisms behind EQM are straightforward. Equal opportunity policies are designed to overcome inequality-generating mechanisms that disadvantage women, such as stronger claims-making by men, unfair evaluations due to negative stereotyping of women (Klammer et al. 2018), or opportunity-hoarding in within-firm networks that exclude women (Tomaskovic-Devey and Avent-Holt 2019). These policies aim to promote women, irrespective of their marital status or motherhood. In contrast, WLM do not explicitly address gender but instead focuses on duties in employees’ personal lives that compete with work responsibilities, which is nominally irrespective of gender (Kossek and Ollier-Malaterre 2013). However, since such duties are still more of a burden for women, women should be the primary addressees of work–life policies. Women still face more obligations in their personal life, primarily as the main caregivers for children (Dechant and Blossfeld 2015). They also act as housekeepers (Grunow et al. 2012) and gatekeepers of kin networks (Bracke et al. 2008). Consequently, they should profit more from work–life policies than men—both fathers and non-fathers. These measures can act against the GWG in two ways: First, they can serve as important signals to relevant actors in the wage negotiation process, primarily human resource managers and direct supervisors (Grover and Crooker 1995). They may signal that women are seen as productive workers and should earn the equivalent of their male counterparts (Acker 1990). Previously held notions about women’s lack of competence and similar negative stereotypes, whether conscious or unconscious, should not be underestimated. Such attitudes are an essential part of the generation of gender inequalities within organizations (Tomaskovic-Devey and Avent-Holt 2019). Second, offering such measures can influence the claims-making behavior of the groups being addressed by the respective measures. Within organizations that implement these policies, the targeted groups will presumably be sheltered from negative self-perceptions and uncertainties (Balafoutas and Sutter 2012) and are more self-confident in their wage expectations when bargaining about wages (Kiessling et al. 2019). With respect to reducing the GWG, this is relevant insofar as women are known to be more hesitant and reluctant than men in this type of bargaining (Mazei et al. 2015).

It is often discussed that the actual implementation of those measures, particularly WLM, does not necessarily follow these theoretically-derived expectations. Aside from an actual investment in women’s productivity and employability, personnel policies can be counteracted in two ways: Employers have to respond to public opinion and demands imposed by the political system to some degree (Ball and Craig 2010); thus, demands to invest in equal opportunity or work–life reconciliation might be perceived as an obligation that employers can hardly ignore. However, it might be seen to conflict with the primary interest of the establishment, which is optimal productivity. Since it is assumed that the best way to achieve this goal would be to have ideal workers who dedicate their time and engagement entirely to work (Acker 1990), supporting women, especially mothers, would not be worthwhile (Den Dulk et al. 2012; Steele 2019). In this case, the solution can be a strategy of “organized hypocrisy”, with a marked difference between “talk” and “action” (Brunsson 1989): EQM and/or WLM would be formally offered but not actively communicated to employees and certainly not actively promoted.

Alternatively, employers can grant these measures as a concession to employees they consider less productive than those who do not need them (Heilman 2012; Filer 1985). Offering WLM is seen as alternative gratification combined with comparably lower wages—either in general or only for those who use such measures or benefit from them. In other words, these measures constitute compensating differentials (Filer 1985) at the expense of higher pay.

For an observer who lacks insights about specific establishments, such ambiguities make it difficult to distinguish among these three alternative options when it comes to how these measures are actually implemented. Consequently, without additional information, it would be hard to predict what influence these measures might have on the GWG in general and on more specific groups within the workforce.

Unlike previous studies, we conceive EQM and WLM as more comprehensive personnel strategies components. Here, two aspects must be considered. First is the mutually exclusive or parallel implementation of EQM and WLM. Second is whether the recipients of these measures overlap or whether these measures address different segments of the female workforce. In our view, this conceptual approach allows for coping with the often-ambiguous character of WLM, which, in other studies, was approached by looking at the organizational culture as a moderator (Abendroth and Diewald 2019; Van der Lippe et al. 2019). This suffers from being unable to distinguish between organizational culture as being induced by the employer to support its human resource management, or as something independent from the employer that moderates the impact of these measures. Our approach is dedicated to identifying the heterogeneity of the rationales behind the offer of these measures by the employer.

2.1. Comprehensive Personnel Policies as Support against Gender Inequality in Wages

We assume that the offer of single measures does not necessarily send a signal that would be strong enough to positively influence the income prospects of women or mothers. We selected two indicators to investigate whether the measures we hypothesize for reducing the GWG are indeed of serious concern to the employer and are part of a comprehensive personnel strategy toward this end. First, we assume that it is more convincing if more than just a single measure is offered, whether it be an EQM or WLM. Implementing more than one measure would send a stronger signal that gender equality, or reconciliation, respectively, is a priority (Butts et al. 2013; Van der Lippe et al. 2019). Moreover, Butts et al. (2013) noted that a greater number of measures may lead to reinforcement and synergies between them. This means fewer instances of organizational hypocrisy (i.e., when work organizations introduce a measure solely to act as a “fig leaf” to gain legitimacy) (Brunsson 1989). Thus, we hypothesize the following:

Hypothesis 1 (H1).

For EQM and WLM, having a higher number of measures than a single measure is associated with a smaller GWG.

When looking at EQM and WLM in combination, we gain additional insights into their contribution to comprehensive HRM strategies, which is foremost to cope with the often-ambiguous character of WLM as a support for employability or compensating differential. In the first case, WLM represent investments in the presumably lower productivity of employees with reconciliation conflicts, especially those that bear the risk of compromising an employee’s energy and time, such as when caring for children (Kelly et al. 2011). When taken at face value, offering these measures would be an option for employers to address serious concerns about the employability and productivity of their female workforce. By reducing an employee’s work–family conflicts, employers can expect a higher level of productivity (De Sivatte et al. 2015). It is when WLM facilitates reconciliation between competing work and family demands, which on average primarily relieves women without damaging labor force integration, that the GWG might be reduced (Huffman et al. 2017; Van der Lippe et al. 2019).

EQM provide a means for avoiding discrimination and possible adverse effects on the productivity of all women (Balafoutas and Sutter 2012). In other words, they signal the employability of the female workforce in general and provide a thoroughly supportive organizational environment for women, which might back-up WLM.

Hence,

Hypothesis 2 (H2).

The combination of EQM and WLM should contribute to less discrimination due to reconciliation demands and, therefore, a smaller GWG within organizations that offer WLM.

2.2. Personnel Policies as a Separation Strategy and Compensating Differentials

Hypotheses 1 and 2 do not differentiate between male and female workforce subgroups but assume that the addressees of both kinds of measures are the same. In the following, we abandon this assumption. WLM could be offered as an alternative to higher pay for those members of the female workforce who are strongly committed to duties at home, whereas EQM are intended for those women who are seen as highly committed to work. In this case, establishments follow a kind of “separation” strategy. They still adhere to their preference for the traditional full-time worker (if not even full-time plus overwork), based on the presence-oriented ideal worker norm with higher pay. This model also applies to parts of the female workforce. Nevertheless, these establishments also accept other life scripts by offering workplaces with lower pay for employees who do not conform to this standard as a way to exploit the existing labor supply fully. Employers, therefore, draw a clear distinction between ambitious workers and those who are more family-oriented. Since women (mothers in particular) tend to be more involved in family duties, differences between the two sexes and between mothers and non-mothers should be further accentuated when both types of measures are implemented in parallel. It can be assumed that either type of measure addresses different groups of employees, so that,

Hypothesis 3a (H3a).

If EQM are implemented in addition to WLM, both types are mutually exclusive. In this case, there are no differences in the GWG between establishments with WLM and establishments without WLM.

On the other hand, it has to be considered that, within establishments that offer both measures, WLM are even used as compensating differentials, especially for employee groups that are thought to be more dedicated to private obligations. In contrast, other employee groups that are assumed to be more productive—i.e., highly qualified workers—are less affected by this offer. Thus,

Hypothesis 3b (H3b).

If EQM are implemented in addition to WLM, WLM are used to compensate differentials. Thus, the GWG of parents is higher within establishments that offer WLM compared to those not offering WLM. For highly qualified employees, there is no such association.

3. Data and Methods

3.1. Data

We used the LEEP-B3 data, a linked employer–employee panel dataset, as the database for our analyses (for all further information, see Diewald et al. 2014). The sample was restricted to large German establishments with at least 500 employees who pay social security taxes. Data collection was motivated by investigating organizational inequality regimes in Germany across the whole industrial structure in both East and West Germany, including the public sector. The sample was drawn from administrative operational data (employment history data [BeH]) provided by the Institute for Employment Research (IAB), which includes 3934 establishments based on a disproportionate random sampling strategy (stratified by industry sector2 and region). The focus on large establishments provides the advantage of a larger number of employees per firm who could be sampled within the establishments, which allows for investigating more within-firm variation and minimizes possible survey effects or reactivity. However, this means that the explanatory power of our results is restricted to large establishments and their employees in Germany3. This has to be kept in mind when drawing our conclusions.

The study has been conducted within the DFG-funded project “Interactions Between Capabilities in Work and Private Life”. Wave 1 took place in 2012/2013 and wave 2 in 2014/2015. The final employer sample consisted of 100 large establishments. For the employee survey, a random sample of 65 employees—on average—who pay social security taxes from each employer was drawn, resulting in a total of 6454 employees who took part in the first of two waves of the survey. Employers were represented by human resource managers who were interviewed with standardized questionnaires in face-to-face interviews. Employees were surveyed by computer-assisted telephone interviews. The second wave included panel respondents who had participated in the first wave, along with new respondents who were sampled by the same sampling method used in the first wave, thus ensuring the same quality of data for the analyses. The dataset is representative of employees in large establishments, except for employees without German nationality and those without vocational training or whose education level was unknown (Diewald et al. 2014). We performed a cross-sectional analysis with a pooled sample of the two waves to answer our research questions. Following research from Rainey and Melzer (2021), we chose the approach of a pooled sample of both waves since it enabled us to include more cases in our analyses. Therefore, we added the newly sampled respondents of the second wave (1959 employees within 32 establishments) to the respondents of the first wave (6454 employees within 100 organizations).4 Thus, we included only one observation for each respondent (employers as well as employees) to avoid violating the independence of the cases. Including more than one observation would bias estimates because having same cases in two waves would be more similar than with others interviewed in only one wave. We chose to stick with a cross-sectional design since our focus was not on income development but the current state of income inequality. Moreover, there is a very low variation in the implementation of the measures between the two waves.

The initial sample consisted of 8413 employees within 132 establishments. Two establishments provided only male or female respondents and were excluded, as were eight establishments that lacked information concerning reconciliation and/or equal opportunity policies. Approximately 23% of the samples had to be dropped because permission to link the data with the administrative data from the IAB was refused5, or because of item nonresponse6. Moreover, this also applies to two cases that indicated zero contractually agreed working hours. As a result, our final sample included 6439 respondents within 122 establishments.

To avoid a possible time bias in our analyses because we used a pooled dataset including two time points or policy changes7, we did additional checks for robustness. We calculated models for each wave and controlled for the year within the overall models. We did not find any remarkable differences to the models we publish here.

3.2. Measures

An overview of all variables and their distribution differentiated by gender (significance tested by a two-sample t-test) can be found in Section 4.1 (see Table 1 and Table 2). For a general overview, see

Table A1. Characteristics and distributions of the observed variables (N = 6439).

	Mean	SD	Min	Max
Log hourly wages	3.09	(0.45)	0.54	6.54
Hourly wages in Euro	24.69	(19.705)	1.711	693.14
Gender	0.45	0.50	0	1
Human capital characteristics
Age	40.65	(8.86)	17	54
Age2	1730.79	(687.21)	289	2916
Qualification
Without vocational training	0.03	(0.17)	0	1
With vocational training	0.63	(0.48)	0	1
Tertiary degree	0.34	(0.47)	0	1
Labor market experience	16.70	(8.39)	0.50	38.36
Labor market experience2	349.28	(321.56)	0.25	1471.20
Employment characteristics
Occupation
Managers	0.03	(0.16)	0	1
Professionals	0.22	(0.42)	0	1
Technicians, associate professional	0.30	(0.46)	0	1
Clerical support workers	0.16	(0.36)	0	1
Services and sales workers	0.07	(0.26)	0	1
Skilled Agricultural	0.002	(0.05)	0	1
Craft and Related Trades Workers	0.10	(0.30)	0	1
Plant & Machine Operators & Assemblers	0.06	(0.25)	0	1
Elementary occupations	0.06	(0.23)	0	1
Firm tenure	8.24	(7.59)	0.50	36.30
Supervisory responsibility	0.36	(0.48)	0	1
Contractual work hours	35.29	(6.99)	3	50
Overwork hours	5.24	(6.97)	−33.5	30
Family characteristics
Age of the youngest child	12.447	(6.33)	0	44
Number of children
No child	0.34	(0.47)	0	1
1–2 children	0.54	(0.50)	0	1
3 and more children	0.12	(0.32)	0	1
Partner	0.83	(0.37)	0	1
N				6439

in Appendix A.

We used the logarithmized hourly wage as the dependent variable, including bonus payments and overtime compensation. To calculate the GWG, we integrated the respondent’s gender (0 = men; 1 = women) as the main independent variable, with men representing the reference category. Therefore, the coefficient of gender shows the income differences between men and women in the logarithmized hourly wages, thereby representing the GWG.

As other independent variables, we analyzed the possible relevance of EQM and WLM to the GWG using the employers’ answers to our question about whether their establishments offered these measures.

For equal opportunity measures, three EQM were distinguished: mentoring programs/networking groups for women, a voluntary women’s quota for staffing, and mixed teams focusing on the integration of women. Here, the representative of the establishments has been asked “On the following list you will find several measures that support equal opportunities for women and men. Please indicate which ones are available in your establishment/operating unit.”, followed by a list of the measures used in this article. The respective answer categories were dummy-coded (0 = no; 1 = yes).

For work–life measures, to measure WLM, we integrated the employers’ information about the availability of two dimensions of the work–life interface: dependent care on the one hand and flexible workplace arrangements on the other, with two measures for each dimension. Here, the establishment’s representative has been questioned, “On the following list you will find several family-friendly measures. Please indicate which ones are available in your establishment/operating unit?”, followed by the various measures, including the ones used within this article. Dependent care is represented by the offer of company daycare, financial support or other assistance for childcare, and opportunities for employees who are out on parental leave8 (0 = no; 1 = yes). Flexible workplace arrangements are represented by home-based telework and flexible working times. Home-based telework and flexible working times that we used for the summary index of WLM were dummy-coded (0 = no; 1 = yes). However, for flexible working times, the implementation of flexible working times was almost ubiquitous in our sample of large establishments, which led to a low variance for this variable. Therefore, instead of the dummy variable asked on the employer level, we used a variable that asked the employees for the availability of flexible working times on the individual level. This means that employees, not employers answered whether their establishment offers flexible working times. To adapt this information to the establishment level, we aggregated the answers of all employees of each establishment. This led to a share of employees (10 percent steps) within each establishment who stated that flexible working times were offered. This can be understood as the aggregated knowledge about flexible work hours in the workforce and as an indicator of the signaling effect of this measure in the workforce.

For all measures, the data contain no information from the employer about which specific groups are addressed by these measures or how often these measures are used.

Group differences: To compare the possible effects of these measures on the GWG for different groups, we also looked at gender in terms of the interaction of gender with parenthood and different levels of qualification. Parenthood is a dummy variable (0 = no child; 1 = at least one child). Qualification was operationalized by the CASMIN (Comparative Analysis of Social Mobility in Industrial Nations) classification (Brauns and Steinmann 1999). The categories were dichotomized into highly qualified workers (i.e., those with a tertiary degree) (=1) versus low- and medium-qualified workers (i.e., those with vocational training or lower) (=0) to get more robust results since low-qualified workers were underrepresented in our sample. Therefore, we had too few cases of low-qualified people to allow for separate analyses. Second, our theoretical perspective rested on highly qualified workers compared with others, which obviates the differentiation of categories below this distinction.

Controls: Since we wanted to look at the role of establishments rather than differences in individual or labor market structural characteristics, we calculated the adjusted GWG, which reflects equal pay for equal work. Therefore, we controlled for several individual and occupational indicators that are known to be related to wage differences between men and women to adjust the GWG. The individual characteristics are the respondent’s age in years and the squared age in years (since we assume a nonlinear correlation between age and income) (BMFSFJ 2009). Qualification levels reflective of an employee’s productivity were included since they account for a significant variance in wage differentials (Matteazzi et al. 2018). As an indicator, we used the employee’s level of education and vocational training, again based on the CASMIN classification. We distinguished between (a) not completed school or completed school without vocational training, (b) completed school with vocational training, or (c) obtained a tertiary degree (reference category). In addition, we included labor market experience (in years and squared, nonlinear) and firm tenure (in years) (Beblo and Wolf 2000; Kunze 2005). To control for the selection into different jobs and career statuses, we included the occupational status, which was measured by the one-digit ISCO-08 code (International Standard Classification of Occupations; for more information, see (ILO 2016))9, and whether or not respondents had supervisory responsibility irrespective of the number of subordinates (0 = no; 1 = yes). Since work hours, and especially overtime work, are highly relevant for earnings and the GWG (Cha and Weeden 2014), we considered the contractual work hours (limited to a maximum of 50 h to restrict the impact of outliers) and overtime hours. Overtime hours were defined as the difference between actual and contractual work hours (limited to 30 overtime hours). Moreover, we controlled for the family situation (Budig and England 2001). We considered the presence of a partner (0 = no partner; 1 = partner), the number of children (none, 1 or 2, or 3 or more), and the age of the youngest child. Structural effects such as the branch or region were considered automatically in a multilevel fixed-effects analysis (Rabe-Hesketh and Skrondal 2012).

3.3. Methods and Analytical Strategy

The linked employer–employee study design requires that the employees are nested within establishments. Thus, it must be considered that individuals within the same establishment are more likely to be exposed to specific working conditions than workers in other establishments. This violates the assumption of independence required for conventional ordinary least-squares regressions (Hox 2010; Rabe-Hesketh and Skrondal 2012) To account for this two-level structure of the data, multilevel linear regression models with organizational fixed effects and clustered robust standard errors (clustered by establishments) were calculated10 (Hox 2010; Rabe-Hesketh and Skrondal 2012). Calculating fixed-effects regression models accounts for the specific data structure. Also, it has the advantage of calculating the GWG within an establishment, thus controlling for unobserved differences between different establishments (Rabe-Hesketh and Skrondal 2012).11 Regarding the GWG, which is also affected by structural characteristics such as the selection processes between men and women, this is a strong benefit.

This method focuses on the differences in the logarithmized hourly wages by gender within establishments and not between establishments. Since the availability of EQM and WLM does not vary within organizations when measured on the organizational level, the main effects of the measures were omitted and could not be calculated within the models (Wooldridge 2010)12. To calculate the differences in the gender wage gap depending on the availability of EQM and WLM, the measures had to interact with gender within the models13. Thus, the coefficients must be interpreted like this: A negative interaction coefficient of gender and the measure means that the gap between women’s and men’s wages within establishments with the specific measure(s) is significantly larger than within establishments without the measure(s). Thus, women earn relatively less within establishments with the measure(s) compared to establishments without them.

To provide all information necessary for the reader, our procedure is as follows: In the first step, we report the adjusted GWG without the interaction with measures. Second, we show how single equal opportunity and work–life measures are associated with the GWG and what differences can be found between employee groups regarding these associations (see Table 3)14. To test hypothesis H1, we calculated models that included the interaction between gender and the numbers of equal opportunity measures and work–life measures (Table 4; for a similar strategy, see Van der Lippe et al. 2019). Last, we tested the rather explorative hypotheses H2 and H3a & b regarding the co-implementation of equal opportunity measures by calculating the association of work–life measures and the log hourly wages in organizations that had no EQM to the establishments with at least one EQM separately (Table 5).

4. Results

4.1. Descriptives

Table 1 shows the descriptive outcomes of the individual characteristics included in our analyses. Two-sample t-tests were calculated to test for the statistical significance of the mean values. On average, women earned 25.7% less per hour than men did (€20.77 vs. €27.96)—a difference that was statistically significant. These quite high differences represent the unadjusted wage gap and correspond with results for 2012–2015 from Eurostat, which displayed the unadjusted gender pay gap between European countries for each year (Eurostat 2022)15. Germany was and still is among the countries with the highest absolute wage differences between men and women, which is often explained by a high divergence in working hours due to a high share of part-time work among women but not men (Trappe et al. 2015). In line with previous research, women had less labor market experience and less tenure. Moreover, women were less represented than men in higher-level occupations and had less supervisory responsibility. Furthermore, our results showed well-known patterns regarding work hours. Men worked significantly more contractual hours as well as overtime hours. With regard to family characteristics, when compared with men, the average age of a woman’s youngest child was slightly higher. On average, women often had one or two children, whereas men were more often a parent of more than two children. In addition, men were slightly more often in a partnership than women were.

Table 1. Individual characteristics (N = 6439) split by gender.

	Men (N = 3513)		Women (N = 2926)		t-Test
	Mean	SD	Mean	SD
Log hourly wages	3.21	(0.46)	2.95	(0.39)	***
Hourly wages in Euro	27.96	(24.41)	20.77	(10.55)	***
Human capital characteristics
Age	40.73	(8.86)	40.55	(8.86)
Age2	1737.59	(686.57)	1722.64	(688.00)
Qualification
Without vocational training	0.03	(0.16)	0.03	(0.18)	*
With vocational training	0.63	(0.48)	0.64	(0.48)
Tertiary degree	0.35	(0.48)	0.33	(0.47)
Labor market experience	17.66	(8.71)	15.55	(7.85)	***
Labor market experience2	387.65	(321.54)	303.22	(266.09)	***
Employment characteristics
Occupation
Managers	0.04	(0.20)	0.01	(0.10)	***
Professionals	0.24	(0.43)	0.20	(0.40)	***
Technicians, associate professional	0.23	(0.42)	0.39	(0.49)	***
Clerical support workers	0.11	(0.32)	0.21	(0.41)	***
Services and sales workers	0.05	(0.23)	0.10	(0.29)	***
Skilled Agricultural	0.003	(0.051)	0.002	(0.045)
Craft and Related Trades Workers	0.17	(0.37)	0.02	(0.13)	***
Plant, Machine Operators & Assemblers	0.10	(0.30)	0.02	(0.14)	***
Elementary occupations	0.06	(0.23)	0.06	(0.23)
Firm tenure	8.58	(7.83)	7.83	(7.27)	***
Supervisory responsibility	0.43	(0.50)	0.27	(0.44)	***
Contractual work hours	37.89	(3.78)	32.18	(8.53)	***
Overwork hours	5.65	(7.14)	4.74	(6.73)	***
Family characteristics
Age of the youngest child	11.64	(5.99)	13.42	(6.59)	***
Number of children
No child	0.35	(0.48)	0.34	(0.47)
1–2 children	0.52	(0.50)	0.57	(0.50)	***
3 and more children	0.14	(0.34)	0.10	(0.29)	***
Partner	0.85	(0.36)	0.83	(0.37)	*

Note: * p < 0.05, *** p < 0.001.

Table 1. Individual characteristics (N = 6439) split by gender.

	Men (N = 3513)		Women (N = 2926)		t-Test
	Mean	SD	Mean	SD
Log hourly wages	3.21	(0.46)	2.95	(0.39)	***
Hourly wages in Euro	27.96	(24.41)	20.77	(10.55)	***
Human capital characteristics
Age	40.73	(8.86)	40.55	(8.86)
Age2	1737.59	(686.57)	1722.64	(688.00)
Qualification
Without vocational training	0.03	(0.16)	0.03	(0.18)	*
With vocational training	0.63	(0.48)	0.64	(0.48)
Tertiary degree	0.35	(0.48)	0.33	(0.47)
Labor market experience	17.66	(8.71)	15.55	(7.85)	***
Labor market experience2	387.65	(321.54)	303.22	(266.09)	***
Employment characteristics
Occupation
Managers	0.04	(0.20)	0.01	(0.10)	***
Professionals	0.24	(0.43)	0.20	(0.40)	***
Technicians, associate professional	0.23	(0.42)	0.39	(0.49)	***
Clerical support workers	0.11	(0.32)	0.21	(0.41)	***
Services and sales workers	0.05	(0.23)	0.10	(0.29)	***
Skilled Agricultural	0.003	(0.051)	0.002	(0.045)
Craft and Related Trades Workers	0.17	(0.37)	0.02	(0.13)	***
Plant, Machine Operators & Assemblers	0.10	(0.30)	0.02	(0.14)	***
Elementary occupations	0.06	(0.23)	0.06	(0.23)
Firm tenure	8.58	(7.83)	7.83	(7.27)	***
Supervisory responsibility	0.43	(0.50)	0.27	(0.44)	***
Contractual work hours	37.89	(3.78)	32.18	(8.53)	***
Overwork hours	5.65	(7.14)	4.74	(6.73)	***
Family characteristics
Age of the youngest child	11.64	(5.99)	13.42	(6.59)	***
Number of children
No child	0.35	(0.48)	0.34	(0.47)
1–2 children	0.52	(0.50)	0.57	(0.50)	***
3 and more children	0.14	(0.34)	0.10	(0.29)	***
Partner	0.85	(0.36)	0.83	(0.37)	*

Note: * p < 0.05, *** p < 0.001.

Looking at the availability of EQM and WLM at the organizational level, Table 2 shows that EQM were generally less common than WLM. On average, nearly 3 out of 4 WLM were available, whereas only 0.89 of 3 EQM were implemented; mentoring programs were the most likely to be available (<40%). More than one-fourth of the organizations had a women’s quota, and approximately 22% of all the establishments had implemented mixed teams to integrate women. However, more than 50% of the firms had implemented daycare opportunities for children, and more than three-fourths provided support for employees during parental leave. Even more common were flexibility measures. In 62% of the firms, some form of home-based telework had been implemented. The establishments’ most common measure was flexible work times, which was available in 95% of the firms.

Table 2. Availability of work–life and equal opportunity measures at establishments (N_{establishments} = 122; N_employees = 6439).

Measures	Mean	SD	Min	Max
Equal opportunity
Mentoring	0.41	0.49	0	1
Women’s quota	0.27	0.44	0	1
Mixed teams	0.22	0.41	0	1
Work–Life
Childcare support	0.53	0.50	0	1
Support for parental leavers	0.77	0.42	0	1
Flexible working hours (org.)	0.95	0.21	0	1
Awareness flex. hours (agg.)	0.63	0.38	0	1
Home-based telework	0.62	0.49	0	1
Sum of equal opportunity measures	0.89	0.96	0	3
Sum of work-life measures	2.88	0.92	0	4
Sum of all measures	3.77	1.54	0	7

Note: Calculated on the employee level but similar to calculation by hand on the establishments’ level; Awareness of flexible working hours displayed here in 1%-steps for a better interpretation.

Table 2. Availability of work–life and equal opportunity measures at establishments (N_{establishments} = 122; N_employees = 6439).

Measures	Mean	SD	Min	Max
Equal opportunity
Mentoring	0.41	0.49	0	1
Women’s quota	0.27	0.44	0	1
Mixed teams	0.22	0.41	0	1
Work–Life
Childcare support	0.53	0.50	0	1
Support for parental leavers	0.77	0.42	0	1
Flexible working hours (org.)	0.95	0.21	0	1
Awareness flex. hours (agg.)	0.63	0.38	0	1
Home-based telework	0.62	0.49	0	1
Sum of equal opportunity measures	0.89	0.96	0	3
Sum of work-life measures	2.88	0.92	0	4
Sum of all measures	3.77	1.54	0	7

Note: Calculated on the employee level but similar to calculation by hand on the establishments’ level; Awareness of flexible working hours displayed here in 1%-steps for a better interpretation.

4.2. Impact of Equal Opportunity and Work–Life Measures on the GWG

Table 3 concisely shows the results of the multi-level linear regressions with organizational fixed effects. Models 1 to 4 show the effects for the different groups of employees (i.e., Model 1 includes all workers, Model 2 includes low- and medium-qualified employees, Model 3 includes high-qualified employees, and Model 4 includes parents only). In the first step of each model, we calculated the adjusted GWG.16

Table 3. The association of equal opportunity/work–life measures and log hourly wages (multi-level linear regression models with organizational fixed effects).

	(1)	(2)	(3)	(4)
	Overall	Low/Medium Qualification	High Qualification	Parents
Step 1:GWG only
Women	−0.1147 ***	−0.1252 ***	−0.0837 ***	−0.1536 ***
	(0.0116)	(0.0128)	(0.0203)	(0.0153)
Constant	2.8910 ***	2.8133 ***	2.4260 ***	3.0729 ***
	(0.1217)	(0.1426)	(0.2596)	(0.2181)
Step 2: Equal opportunity measures
Women	−0.1260 ***	−0.1237 ***	−0.1044 **	−0.1568 ***
	(0.0150)	(0.0156)	(0.0311)	(0.0210)
Mentoringxwomen	−0.0022	−0.0169	−0.0119	−0.0233
	(0.0257)	(0.0320)	(0.0401)	(0.0340)
Women’s quotaxwomen	0.0492	0.0257	0.0773	0.0676 *
	(0.0253)	(0.0315)	(0.0394)	(0.0324)
Mixed teamsxwomen	−0.0128	−0.0094	−0.0079	−0.0349
	(0.0276)	(0.0291)	(0.0411)	(0.0374)
Constant	2.8884 ***	2.8113 ***	2.4189 ***	3.0775 ***
	(0.1222)	(0.1427)	(0.2598)	(0.2192)
Step 3: Work–Life measures
Women	−0.0996 **	−0.0874 **	−0.1325 *	−0.1460 **
	(0.0341)	(0.0329)	(0.0635)	(0.0420)
Childcare supportxwomen	−0.0491 *	−0.0134	−0.0570	−0.0778 **
	(0.0214)	(0.0233)	(0.0378)	(0.0286)
Support for parental leaversxwomen	0.0052	−0.0032	0.0053	0.0049
	(0.0290)	(0.0310)	(0.0385)	(0.0355)
Home-based teleworkxwomen	0.0065	0.0067	−0.0218	0.0020
	(0.0239)	(0.0250)	(0.0439)	(0.0333)
Flexible work hoursxwomen	0.0006	−0.0053	0.0127	0.0045
	(0.0029)	(0.0028)	(0.0068)	(0.0038)
Constant	2.8649 ***	2.7679 ***	2.4615 ***	3.0495 ***
	(0.1238)	(0.1455)	(0.2690)	(0.2187)
N	6439	4266	2173	4238

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual work hours, overwork, partner, and number of children and the age of the youngest child; all steps within this table are calculated independently from each other.

Table 3. The association of equal opportunity/work–life measures and log hourly wages (multi-level linear regression models with organizational fixed effects).

	(1)	(2)	(3)	(4)
	Overall	Low/Medium Qualification	High Qualification	Parents
Step 1:GWG only
Women	−0.1147 ***	−0.1252 ***	−0.0837 ***	−0.1536 ***
	(0.0116)	(0.0128)	(0.0203)	(0.0153)
Constant	2.8910 ***	2.8133 ***	2.4260 ***	3.0729 ***
	(0.1217)	(0.1426)	(0.2596)	(0.2181)
Step 2: Equal opportunity measures
Women	−0.1260 ***	−0.1237 ***	−0.1044 **	−0.1568 ***
	(0.0150)	(0.0156)	(0.0311)	(0.0210)
Mentoringxwomen	−0.0022	−0.0169	−0.0119	−0.0233
	(0.0257)	(0.0320)	(0.0401)	(0.0340)
Women’s quotaxwomen	0.0492	0.0257	0.0773	0.0676 *
	(0.0253)	(0.0315)	(0.0394)	(0.0324)
Mixed teamsxwomen	−0.0128	−0.0094	−0.0079	−0.0349
	(0.0276)	(0.0291)	(0.0411)	(0.0374)
Constant	2.8884 ***	2.8113 ***	2.4189 ***	3.0775 ***
	(0.1222)	(0.1427)	(0.2598)	(0.2192)
Step 3: Work–Life measures
Women	−0.0996 **	−0.0874 **	−0.1325 *	−0.1460 **
	(0.0341)	(0.0329)	(0.0635)	(0.0420)
Childcare supportxwomen	−0.0491 *	−0.0134	−0.0570	−0.0778 **
	(0.0214)	(0.0233)	(0.0378)	(0.0286)
Support for parental leaversxwomen	0.0052	−0.0032	0.0053	0.0049
	(0.0290)	(0.0310)	(0.0385)	(0.0355)
Home-based teleworkxwomen	0.0065	0.0067	−0.0218	0.0020
	(0.0239)	(0.0250)	(0.0439)	(0.0333)
Flexible work hoursxwomen	0.0006	−0.0053	0.0127	0.0045
	(0.0029)	(0.0028)	(0.0068)	(0.0038)
Constant	2.8649 ***	2.7679 ***	2.4615 ***	3.0495 ***
	(0.1238)	(0.1455)	(0.2690)	(0.2187)
N	6439	4266	2173	4238

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual work hours, overwork, partner, and number of children and the age of the youngest child; all steps within this table are calculated independently from each other.

Step 1 in Table 3 shows that the adjusted GWG in our sample was approximately 10.84%.17 We found that the GWG for highly qualified workers (8.03%) was smaller than that for workers with medium or lower qualifications (11.77%). An even more considerable difference was found among parents, with mothers earning 14.24% less than fathers. In the next step, specific EQM were included in the regression. Here, we only found a positive correlation for the women’s quota, and just for the GWG between parents—which means a lower parental GWG within establishments offering a quota (for a graphical illustration of the interaction see Figure A1 in the Appendix A).

Regarding WLM, we found a significantly higher GWG for parents, even for women in general, in establishments that offered childcare support. We found no significant associations with the GWG for the other measures.

4.3. From Single Measures to Personnel Strategies

Our results did not display solid or unequivocal associations with EQM or WLM. Foremost among our findings was that WLM seemed to be implemented according to varying rationales so that, on average, the differences remain small. The finding that the overall impacts on the GWG of EQM and WLM were modest raises the question of whether the number of measures or the combination of both bundles of measures matters more.

We expected a higher sum of measures to spill over to the whole female workforce because such measures should back up HRM policies and prevent organizational hypocrisy (H1). However, our analyses showed that two or more EQM had only a small and nonsignificant association with the GWG compared with no or only one such measure (Table 4). Regarding WLM, contrary to H1, the regression showed that a higher number of WLM was negatively correlated with mothers’ wages, meaning that the GWG of parents was even higher within these establishments compared to establishments that offered no or only one WLM. Thus, more clearly than for the single measures, we saw significant evidence that WLM were being predominantly offered to mothers as compensating differentials. Since the expected generalizing effect of a higher number of measures on all women within the firm was not found to a significant degree, the discriminatory compensating differentials strategy would appear to be targeted at mothers specifically, rather than at women in general. Thus, H1 must be rejected.

Table 4. The association of multiple equal opportunity/work–life measures and log hourly wages (multi-level linear regression models with organizational fixed effects).

	(1) Overall	(2) Low/Medium Qualifiation	(3) High Qualification	(4) Parents
Step 1: Equal opportunity measures
Women	−0.1173 ***	−0.1190 ***	−0.0952 ***	−0.1516 ***
	(0.0134)	(0.0149)	(0.0240)	(0.0186)
2 to 3 measuresxwomen	0.0087	−0.0231	0.0331	−0.0063
	(0.0238)	(0.0234)	(0.0384)	(0.0301)
Constant	2.8910 ***	2.8123 ***	2.4258 ***	3.0729 ***
	(0.1217)	(0.1426)	(0.2589)	(0.2180)
Step 2: Work–life measures
Women	−0.0670 *	−0.0781 **	−0.0362	−0.0665 *
	(0.0314)	(0.0243)	(0.1065)	(0.0317)
2 to 4 measuresxwomen	−0.0508	−0.0514	−0.0490	−0.0925 **
	(0.0339)	(0.0277)	(0.1083)	(0.0343)
Constant	2.8945 ***	2.8175 ***	2.4257 ***	3.0763 ***
	(0.1220)	(0.1429)	(0.2601)	(0.2193)
N	6439	4266	2173	4238

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual work hours, overwork, partner, and number of children and the age of the youngest child; all steps within this table are calculated independently from each other.

Table 4. The association of multiple equal opportunity/work–life measures and log hourly wages (multi-level linear regression models with organizational fixed effects).

	(1) Overall	(2) Low/Medium Qualifiation	(3) High Qualification	(4) Parents
Step 1: Equal opportunity measures
Women	−0.1173 ***	−0.1190 ***	−0.0952 ***	−0.1516 ***
	(0.0134)	(0.0149)	(0.0240)	(0.0186)
2 to 3 measuresxwomen	0.0087	−0.0231	0.0331	−0.0063
	(0.0238)	(0.0234)	(0.0384)	(0.0301)
Constant	2.8910 ***	2.8123 ***	2.4258 ***	3.0729 ***
	(0.1217)	(0.1426)	(0.2589)	(0.2180)
Step 2: Work–life measures
Women	−0.0670 *	−0.0781 **	−0.0362	−0.0665 *
	(0.0314)	(0.0243)	(0.1065)	(0.0317)
2 to 4 measuresxwomen	−0.0508	−0.0514	−0.0490	−0.0925 **
	(0.0339)	(0.0277)	(0.1083)	(0.0343)
Constant	2.8945 ***	2.8175 ***	2.4257 ***	3.0763 ***
	(0.1220)	(0.1429)	(0.2601)	(0.2193)
N	6439	4266	2173	4238

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual work hours, overwork, partner, and number of children and the age of the youngest child; all steps within this table are calculated independently from each other.

Figure 1 illustrates the interaction effect of a higher sum of WLM and parental gender. It can be seen that there was no significant GWG for parents in establishments with no or only one WLM. Within establishments that offer two or more WLM, the GWG of parents was significant18.

To investigate the contingency of both types of measures, we split the sample into firms that offered none of the three EQM (2920 employees and 57 establishments) and firms that implemented at least one of the three EQM (3519 employees and 65 establishments). Again, we differentiated the employees according to the qualification level and parenthood (see Table 5). Based on existing theories, we assumed that the combined measures could affect wages in opposite ways: either the implementation of EQM speaks to a more egalitarian organizational culture, making work–family measures supportive of instead of complementary to the higher wages of women, or these two bundles of measures are implemented exclusively or as compensating differentials, with EQM being implemented to promote committed and career-oriented women and WLM designed to achieve the opposite goal (i.e., as compensation for not offering higher wages). In the latter case, the GWG within workplaces that combine these two bundles of measures would be higher.

Table 5. The association of work–life measures and the log hourly wages dependent on whether organizations offer also equal opportunity measures; multi-level linear regression models with organizational fixed effects.

	Without Equal Opportunity Measures				With at Least One Equal Opportunity Measure
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Overall	Low/Medium Qualification	High Qualification	Parents	Overall	Low/Medium Qualification	High Qualification	Parents
Step 1:Single WLM measures
Women	−0.1212 ***	−0.1186 **	−0.1174	−0.1778 ***	0.0221	0.0529	−0.0464	−0.0171
	(0.0292)	(0.0339)	(0.1284)	(0.0432)	(0.0538)	(0.0639)	(0.0609)	(0.0734)
Childcare supportxwomen	−0.0895 **	−0.0259	−0.1287	−0.1088 **	−0.0251	−0.0063	−0.0297	−0.0563
	(0.0267)	(0.0300)	(0.0714)	(0.0393)	(0.0287)	(0.0323)	(0.0448)	(0.0379)
Support for parental leaversxwomen	0.0457 (0.0239)	0.0543 (0.0286)	0.0126 (0.0659)	0.0520 (0.0376)	−0.1232 ** (0.0388)	−0.1701 ** (0.0591)	−0.0578 (0.0429)	−0.1309 * (0.0529)
Home-based teleworkxwomen	−0.0061	−0.0055	−0.0183	−0.0206	0.0037	0.0112	−0.0348	0.0092
	(0.0317)	(0.0338)	(0.0779)	(0.0464)	(0.0336)	(0.0353)	(0.0494)	(0.0479)
Flexible work hoursxwomen	0.0011	−0.0052	0.0110	0.0070	0.0004	−0.0036	0.0101	0.0026
	(0.0038)	(0.0039)	(0.0116)	(0.0048)	(0.0033)	(0.0035)	(0.0064)	(0.0043)
Constant	2.8618 ***	2.7204 ***	2.5849 ***	3.1389 ***	2.8692 ***	2.8273 ***	2.2923 ***	2.9484 ***
	(0.1987)	(0.2195)	(0.4950)	(0.3022)	(0.1486)	(0.1894)	(0.3173)	(0.3196)
Step 2:Sum of measures
Women	−0.0987 ***	−0.0968 ***	−0.0873	−0.1011 ***	0.0528	0.0028	0.0653	0.0997
	(0.0221)	(0.0218)	(0.1550)	(0.0200)	(0.0869)	(0.1038)	(0.0484)	(0.0934)
2 to 4 measuresxwomen	−0.0368	−0.0330	−0.0262	−0.0700 *	−0.1552	−0.1273	−0.1355 *	−0.2478 **
	(0.0287)	(0.0283)	(0.1611)	(0.0292)	(0.0875)	(0.1047)	(0.0516)	(0.0924)
Constant	2.9183 ***	2.7950 ***	2.6035 ***	3.2064 ***	2.8822 ***	2.8651 ***	2.2641 ***	2.9509 ***
	(0.1961)	(0.2157)	(0.4667)	(0.3083)	(0.1476)	(0.1895)	(0.3068)	(0.3150)
N	2920	2103	817	1870	3519	2163	1356	2368

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual working hours, overwork, partner, and number of children and the age of the youngest child; steps are calculated independently from each other.

Table 5. The association of work–life measures and the log hourly wages dependent on whether organizations offer also equal opportunity measures; multi-level linear regression models with organizational fixed effects.

	Without Equal Opportunity Measures				With at Least One Equal Opportunity Measure
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Overall	Low/Medium Qualification	High Qualification	Parents	Overall	Low/Medium Qualification	High Qualification	Parents
Step 1:Single WLM measures
Women	−0.1212 ***	−0.1186 **	−0.1174	−0.1778 ***	0.0221	0.0529	−0.0464	−0.0171
	(0.0292)	(0.0339)	(0.1284)	(0.0432)	(0.0538)	(0.0639)	(0.0609)	(0.0734)
Childcare supportxwomen	−0.0895 **	−0.0259	−0.1287	−0.1088 **	−0.0251	−0.0063	−0.0297	−0.0563
	(0.0267)	(0.0300)	(0.0714)	(0.0393)	(0.0287)	(0.0323)	(0.0448)	(0.0379)
Support for parental leaversxwomen	0.0457 (0.0239)	0.0543 (0.0286)	0.0126 (0.0659)	0.0520 (0.0376)	−0.1232 ** (0.0388)	−0.1701 ** (0.0591)	−0.0578 (0.0429)	−0.1309 * (0.0529)
Home-based teleworkxwomen	−0.0061	−0.0055	−0.0183	−0.0206	0.0037	0.0112	−0.0348	0.0092
	(0.0317)	(0.0338)	(0.0779)	(0.0464)	(0.0336)	(0.0353)	(0.0494)	(0.0479)
Flexible work hoursxwomen	0.0011	−0.0052	0.0110	0.0070	0.0004	−0.0036	0.0101	0.0026
	(0.0038)	(0.0039)	(0.0116)	(0.0048)	(0.0033)	(0.0035)	(0.0064)	(0.0043)
Constant	2.8618 ***	2.7204 ***	2.5849 ***	3.1389 ***	2.8692 ***	2.8273 ***	2.2923 ***	2.9484 ***
	(0.1987)	(0.2195)	(0.4950)	(0.3022)	(0.1486)	(0.1894)	(0.3173)	(0.3196)
Step 2:Sum of measures
Women	−0.0987 ***	−0.0968 ***	−0.0873	−0.1011 ***	0.0528	0.0028	0.0653	0.0997
	(0.0221)	(0.0218)	(0.1550)	(0.0200)	(0.0869)	(0.1038)	(0.0484)	(0.0934)
2 to 4 measuresxwomen	−0.0368	−0.0330	−0.0262	−0.0700 *	−0.1552	−0.1273	−0.1355 *	−0.2478 **
	(0.0287)	(0.0283)	(0.1611)	(0.0292)	(0.0875)	(0.1047)	(0.0516)	(0.0924)
Constant	2.9183 ***	2.7950 ***	2.6035 ***	3.2064 ***	2.8822 ***	2.8651 ***	2.2641 ***	2.9509 ***
	(0.1961)	(0.2157)	(0.4667)	(0.3083)	(0.1476)	(0.1895)	(0.3068)	(0.3150)
N	2920	2103	817	1870	3519	2163	1356	2368

Note. Clustered robust standard errors in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001; Controlled for age, age (squared), qualification level, occupational status, labor market experience, labor market experience (squared), tenure, supervising responsibilities, contractual working hours, overwork, partner, and number of children and the age of the youngest child; steps are calculated independently from each other.

When we looked at the specific WLM offered by establishments that did not also implement EQM, we saw patterns that were similar to what we found when looking at the whole sample. Specifically, daycare measures and a higher number of WLM were associated with a higher GWG.

As shown in Figure 2, there is a GWG for parents within establishments that offer no EQM and no company childcare. This gap, however, is significantly larger when company childcare is implemented.

For establishments that simultaneously offered EQM and WLM, the patterns changed. Other than in organizations without EQM and the whole sample, childcare support had no significant negative association with the GWG. In contrast, the offer of re-integration possibilities for employees on parental leave showed a strong negative correlation with women’s wages in general and, consequently, a higher GWG within these establishments than those who did not offer such possibilities. This was driven mostly by mothers and female employees with lower- and medium-ranking qualifications. As Figure 3 shows, there was no significant GWG of parents within establishments that offer EQM only but within organizations with the offer of a re-integration of parental leavers.

For a higher number of WLM, the patterns were similar to those found for establishments that did not offer equal opportunity measures. However, these negative correlations were considerably stronger, mostly for women with higher qualifications and mothers (for a graphical illustration, see Figure A2 in the Appendix A). For flexibility measures, no significant results could be shown.

Taken together, the results point to possible separate strategies in dealing with parenthood and gender and differences between the types of measures. Thus, both our hypotheses, 3a and 3b, were partly supported. For flexibility measures H3a, which does not assume any harmful effect from the simultaneous implementation of both EQM and WLM, holds true. H3b applies to measures that highlight absence because of parenthood. These measures are obviously used as compensating differentials for this group of employees. Interestingly, for establishments with more WLM measures, H3b partly holds true. Here, the GWG of highly qualified employees was larger than within establishments with none or only one WLM.

5. Discussion and Conclusions

In many Western societies, gender differences in wages are a prominent and widely discussed topic in inequality research. However, the role of establishments in shaping the GWG is still neglected, with a few exceptions (e.g., Abendroth and Diewald 2019; Van der Lippe et al. 2019). In this article, we took a closer look at the within-firm GWG in three respects: (1) the general role of EQM and WLM for the adjusted GWG, (2) which groups of employees are primarily addressed by these measures, and (3) what role the specific mix of both types of measures plays. The relevance of WLM, in particular, is hard to assess since employers vary considerably in how these measures are actually intended by the employer and perceived by the employees.

We found less evidence that these measures had a general effect on the GWG. Regarding group differences, we found that mothers, in particular, as a group with a relatively high level of care responsibilities, were affected by these measures. Whereas a staffing quota for women can be seen as a counterweight against an employer’s reservations about paying higher salaries to fathers than mothers, the implementation of company childcare appeared to reinforce stereotypes regarding this group of workers. By combining EQM and WLM, we were able to determine that such reservations came mostly from establishments without EQM.

Our results demonstrate that aggregate perspectives on these bundled measures help to explain how they contribute to the GWG in their combination. In addition to single specific measures, we took the actual number of measures into account and looked not only at EQM and WLM separately, but also at how they interact to produce different GWG patterns. In investigating this combination in particular, we hoped to decode the heterogeneity in the strategic orientation of WLM across different establishments. With this strategy, we came closer to understanding how these measures are intended and how they address specific groups within the female workforce.

We could not confirm our assumption that a greater number of measures might signify a straightforward HRM strategy towards more gender equality in wages. On the contrary, a higher amount of WLM went along with a significantly higher GWG for parents in establishments that offered more than one WLM. Regarding the moderating effect of EQM as part of a comprehensive personnel strategy, we found interesting patterns. Within establishments that offer no EQM but do offer childcare support, the GWG was significantly higher than in establishments that offered none of them. This points to the fact that, within establishments that focus particularly on enabling support for childcare demands, the presence of these demands is even detrimental for gender equality in wages. Moreover, in establishments that offered both EQM and WLM, we found evidence that the implementation of support for parental leavers and a greater number of work–life measures goes along with a significant and higher GWG, especially for parents. Consequently, there was no indication that these establishments had a general concern for all women to be treated equally, or that they implemented WLM as a form of career support. Though both results point to the offer of WLM as compensating differentials primarily for mothers, there seem to underlie different boundary drawings within the female workforce. Whereas establishments without EQM seem to compensate for mothers in general, establishments with EQM motherhood per se is not penalized. However, it is instead the absence from the workplace that matters and provokes lower income. Ironically, this situation was evident in the case of support for parental leavers, which, at face value, is a measure precisely designed to avoid such losses for parents as well as for less-qualified employees. In other words, childcare responsibilities that are linked to parents’ decision to leave the workplace for a shorter or longer duration is still a trap, as is evident for the most part in establishments that also support career prospects for women. This result becomes even more interesting regarding the fact that we could not find a significant GWG within establishments that offered EQM but no support for parental leavers.

In other words, in establishments offering EQM, measures that highlight the absence from the workplace are even more detrimental for those groups among the workforce who are vulnerable to high private demands. Moreover, the lack of an association with home-based telework points to the fact that it is less a matter of an occasional absence from the workplace than it is a matter of interpreting employment interruptions. In the case of motherhood as a stigma, it puts mothers—and to a lesser degree, women in general—in a situation where reconciliation is seen as an alternative to high-pay employment. This is in line with research by, for example, Beblo and Wolf (2000), who found women’s employment interruptions to be an essential predictor of the GWG. We interpret this outcome as a diversity strategy perpetrated by firms that offer both EQM and WLM, but for different purposes and for different groups. In our study, we also noted that measures designed to counteract this danger, in fact, do the opposite. Moreover, a higher number of WLM showed a much more pronounced association with lower wages for mothers and high-qualified women. Particularly, the second result was quite surprising, since we, theoretically, expected high-qualified women to be less affected by the role of WLM in compensating differentials and, empirically, the analyses of single WLM showed no association of WLM and the GWG for high-qualified employees. Thus, in the case of a high amount of WLM in combination with EQM, there are negative associations with women’s wages for both, such as mothers—as the group of employees that is mostly associated with WLM—as well as with high-qualified women, the group which is mostly addressed by EQM. This points to the fact that emphasizing the availability of WLM may even harm the supportive effect of EQM for high-qualified women. In consequence, promises of either EQM or WLM failing to reduce the GWG should be seen critically.

Lastly, our results emphasize the importance of looking at WLM and EQM in a differentiated way—i.e., concerning the measures as parts of a bundle of human resource management measures and the respective addressees among the workforce.

Limitations and Further Research

Our study is not without limitations. First, because it is not longitudinal, we could not avoid the potential problems of reversed causality and selection effects—that is, the possibility that organizations with a higher GWG reacted by implementing these measures. Second, although we covered the entire industrial structure in our sample, this coverage was limited to large firms only. We do not pretend that the mechanisms we found work in the same way in small- and medium-sized companies. Third, though we were explicitly interested in the signaling effect of the mere offer of the specific measures we looked at, we cannot rule out the possibility that their individual use might have an additional and maybe more prominent effect on the GWG. Since the individual use of the measures was collected only for flexibility measures in our survey, we could not account for this possibility.

Thus, these limitations have implications for further research that would require a longitudinal perspective concerning either the influence of changes in the implementation of measures or the role of such measures in determining income trends. Also, the influence of the use of these measures on income inequality should be investigated to determine whether it emphasizes the bad signal of WLM or may eventually reduce the stereotypes linked to them.