Off to a slow start: which workplace policies can limit gender pay gaps across firm tenure?

Abstract

Much of the gender pay gap is generated within workplaces, making it paramount to understand which workplace policies effectively address gaps. Our article asks when policies limit gender pay gaps across employee tenure to identify potential temporal weak points. We analyze a representative panel of 10,000 establishments with over 850,000 employees using the 2005–19 waves of German-linked employer–employee data (LIAB). Two key findings emerge. First, a temporal perspective on workplace policies reveals that no policy under study—formalization, identity-based career programs, and child care assistance—reduces gender pay gaps at hire. Instead, policies only address additional disparities that accumulate after hire. Second, only identity-based career programs narrow gender disparities for all women. In contrast, seemingly gender-neutral formalization is insufficient, while providing employer-sponsored child care has mixed effects depending on employees’ education. We conclude by discussing the implications of these findings for organizational policy and future research.

1. Introduction

Despite many advances, gender pay gaps remain persistent in modern economies (Gartner and Hinz 2009; Blau and Kahn 2017). Within workplaces, pay gaps develop gradually across employees’ firm tenure (Kronberg 2020). Men and women hired into the same job often start with different pay (e.g. Petersen and Saporta 2004), and then gaps widen further post-hire as gender disparities in annual merit raises and promotion opportunities accumulate with tenure (e.g. Yap and Konrad 2009; Castilla 2011; Roth, Purvis and Bobko 2012; Cassidy, DeVaro and Kauhanen 2016). Research has examined several policy approaches aiming to reduce gender pay gaps within workplaces, including formalization (e.g. Anderson and Tomaskovic-Devey 1995; Castilla 2015; Huffman, King and Reichelt 2017; Busch-Heizmann and Rinke 2018), career programs for women (e.g. Besley et al. 2017; Bertrand et al. 2018; Peters, Drobe and Abendroth 2020; Dobbin and Kalev 2022), and work–life policies, such as child care assistance (e.g. Glass 2004; Huffman, King and Reichelt 2017; Fuller and Hirsh 2019). However, how these policies affect employees across their firm tenure remains unclear.

These three policy approaches (formalization, career programs for women, and child care) differ along two dimensions: First, while formalization is seemingly identity-neutral, child care assistance and women’s career programs take an identity-conscious approach. Formalization of personnel practices may include training plans, planning future staffing needs, recruitment procedures, job descriptions, employee goal setting, and performance evaluations. Capturing these procedures in writing aims to make information accessible and decisions transparent (e.g. Tetlock 1985; Castilla 2015). Moreover, formalization tends to elevate achieved status characteristics, such as education and seniority, and deemphasize ascribed characteristics, such as gender (Sauer et al., 2021). In contrast, programs for women (including mentoring, training, gender quotas, and programs for parents on leave) and child care assistance (including referral programs, financial assistance, and in-house care) are explicitly identity-based and draw attention to categorical distinctions, such as gender or parental status. In addition to being identity-neutral versus identity-based, these three policy approaches differ regarding what they regulate. Formalization and career programs for women alter the opportunity structure of work itself, whereas child care assistance focuses on creating structures that accommodate work without changing its nature.

We build on recent research evaluating policies (e.g. Abendroth et al., 2017; Huffman, King and Reichelt 2017; Dobbin and Kalev, 2022) and introduce a temporal perspective by considering how policy effects unfold across employees’ firm-internal careers. As most identity-based programs, such as mentoring or child care, take effect after employees enter the organization, we expect their effects to develop gradually. In contrast, formalization includes recruitment procedures and job descriptions, which are central to hiring. Therefore, we expect formalization to reduce pay gaps starting at hire. Additionally, comparing different policies, we expect career programs for women to reduce gaps most effectively because they combine work- and identity-focused approaches. A better understanding of how policy effects unfold with employees’ firm tenure provides us with a better toolkit to design policies that address all temporal sources of workplace pay gaps.

We use German linked employer–employee LIAB data (Ruf et al., 2021) between 2005 and 2019 to examine how policies affect gender pay gaps with increasing firm tenure. The LIAB is a nationally representative panel of establishments. We divide our analyses by employees’ education, as policy effects typically depend on employees’ education. Models examining college graduates include 279,873 employees working in 6,963 establishments, while analyses focusing on employees with a vocational degree or less include 573,812 employees working in 8,208 establishments.

Our findings extend the literature in two ways. We show that policy effects unfold unevenly across employees’ tenure. None of the examined policies limit pay gaps at hire—even though women earn 9–12 per cent less than comparable men at entry. Instead, policy effects develop gradually with tenure, if at all. Such delays are problematic as many employees may leave before policies take effect.

We also contribute to research comparing workplace policies. Formalization as a seemingly identity-neutral approach fails to narrow gender gaps. In contrast, identity-based approaches, including career programs and child care, strongly affect gender disparities. However, identity-based policies work best when directly addressing work (e.g. mentoring). Identity-based programs that “only” accommodate work (i.e. child care) mostly reproduce education-specific advantages, meaning they benefit college graduates while penalizing mothers not holding a college degree. We discuss policy implications in the discussion section.

2. Theory and hypotheses

2.1 How gender pay gaps unfold with employees’ tenure

As in other OECD countries, German gender pay gaps have narrowed slowly over the past three decades (Gartner and Hinz 2009; Minkus and Busch-Heizmann 2020). The median gender gap in gross hourly pay among German full-time employees declined from 27 to 14 per cent between 1992 and 2020, with most changes occurring in the early 1990s (OECD 2023). It remains unclear precisely which policies have generated these changes.

Our study contributes to growing research evaluating which workplace policies best address disparities (e.g. Williams, Kilanski and Muller 2014; Castilla 2015; Huffman, King and Reichelt 2017; Dobbin and Kalev 2022). We extend research by examining when policies affect employees’ firm-internal pay progression. With few exceptions (e.g. Glass 2004; Zimmermann and Collischon 2023), it remains unclear how policy effects unfold across employee tenure. We address this gap by using longitudinal employer–employee data.

Temporal trends matter because they improve our understanding of underlying mechanisms. Even when supervisors have known their employees for years, nonconscious gender-based performance expectations can shape merit and promotion decisions. Gender is a primary social category that functions as a lens through which people evaluate other information about that person (Berger, Cohen and Zelditch 1972; Ridgeway 2011). Status distinctions continuously affect how supervisors see employees, how supervisors evaluate employees’ claims for pay and promotions, and what positions employees will pursue (Correll and Benard 2006). Consequently, gaps widen with tenure as women receive lower merit-based pay increases (Castilla 2015), are promoted less often, or receive smaller pay increases after promotions than men (e.g. Barnett, Baron and Stuart 2000; Yap and Konrad 2009; Roth, Purvis and Bobko 2012; Blau and Kahn 2017; Cassidy, DeVaro and Kauhanen 2016). Disadvantages accumulate and flatten women’s pay growth compared to men (Kronberg 2020).

Pay gaps are not set in stone. Instead, each organization is a unique inequality regime, where local policies and practices affect how categorical distinctions translate into pay gaps (e.g. Acker 2006; Baron and Bielby 1980; Tomaskovic-Devey and Avent-Holt 2019). To understand workplaces’ ability to shape gaps, we examine three policy approaches: Formalization, career programs for women, and employer-sponsored child care assistance. Policies differ in two aspects: first, whether policies are identity-based or seemingly for everyone. Formalization takes a seemingly identity-neutral approach by deemphasizing the role of gender, while child care assistance and identity-based career programs are intended for employees of specific identities. We acknowledge some conceptual blurriness as many identity-based policies, such as mentoring programs, also benefit from formalization (Dobbin and Kalev 2022). Thus, our terminology emphasizes whether policies are identity-based or seemingly apply to everyone. Second, policies differ in whether they change the structure of work (i.e. formalization and career programs) or merely accommodate work (i.e. child care). Below, we hypothesize how these three policy approaches unfold across employees’ establishment tenure.

2.2 Formalizing employment practices to reduce pay gaps

In theory, formalization limits pay disparities by preventing nonconscious status expectations from translating into biased decision making (Baron and Bielby 1980; Reskin 2000). When organizations define processes in writing, they provide more explicit guidelines and reduce ambiguity in supervisory decision making (Tetlock 1985; Castilla 2015; Tetlock and Mitchell 2009; Tomaskovic-Devey et al., 2009; Hermans, Cools and van den Abbeele 2021). Formalization also elevates achieved status characteristics, such as education and training, over ascribed distinctions, such as gender (e.g. Tomaskovic-Devey et al., 2009; Sauer et al., 2021).

We think about formalization in two ways. First, establishments may develop an overarching practice of formalization. Formalizing most human resource (HR) procedures might indicate a deeper equity-focused organizational logic (Baron et al., 2007). Several studies use an additive index capturing formalization across several HR practices. Formalization correlates with smaller gender pay gaps, especially among low-earning employees (Anderson and Tomaskovic-Devey 1995; Huffman, King and Reichelt 2017). However, these studies do not address when formalization affects gender pay gaps across employees’ tenure.

Second, we consider which specific HR practices benefit from formalization. Studies examining recruitment procedures (including firm-internal recruitment via promotions) found that formalizing recruitment narrowed gaps, especially among highly qualified women (Abendroth et al., 2017). Written recruitment procedures especially empower mothers to request promotions (Lükemann and Abendroth 2018), limiting gender differences in negotiations.

Similarly, formal career planning via written training plans and employee goals makes expectations visible. Explicitly stating expectations might help employees understand what it takes to receive pay raises or promotions. In a cross-sectional study, Abendroth et al. (2017) found that formalizing career planning reduced gender gaps but only among highly educated workers. Likewise, written performance evaluations provide clear feedback to employees. While perceptions of performance are highly gendered (e.g. Foschi 1996; Ridgeway 2011), written records might hold supervisors accountable and help employees legitimate demands for raises or promotions. Abendroth et al. (2017) observed nonsignificant effects, while Lükemann and Abendroth (2018) found that formalizing evaluations encouraged mothers to ask for promotions more often.

Prior studies highlight that written job descriptions can be gendered, leading to greater between-job disparities (Acker 1990). However, including job descriptions narrows within-job gender pay gaps, especially among college graduates (Abendroth et al., 2017). Less research exists on written plans for future staffing needs, but formalizing the planning process might make the underlying opportunity structure more visible and help individuals anticipate future changes. In sum, formalizing recruitment, training, setting employee goals, evaluating employee performance, describing jobs, and staff planning should provide more transparency in merit and promotion decisions, thereby affecting overall pay progression.

H1: Gender pay gaps narrow with increasing tenure when establishments formalize more HR practices.

Research on gendered organizations questions whether formalization is sufficient to address inequality. Employees may use formalized procedures inconsistently to legitimate their own, sometimes biased and shifting, standards (e.g. Foschi 1996; Bridges and Nelson 1989; Roscigno 2007; Castilla 2012; Smith-Doerr et al., 2019; Dobbin and Kalev 2022). Moreover, bureaucratic rules form around gendered logics and thereby legitimate existing gender disparities. For instance, job descriptions of women-dominated jobs are vaguer than men’s, resulting in fewer promotions for women (e.g. Ferguson 1984; Acker 1990, 2006; Williams, Kilanski and Muller 2014). Formalization can also create a false sense of achieved procedural justice. For example, participants favored men more when experimenters framed processes as meritocratic (Castilla and Benard 2010; Dobbin and Kalev 2022; Portocarrero and Carter 2022). This literature motivates the null hypothesis that formalization does not affect gender pay gaps.

2.3 Reducing gaps via career programs for women and parents

Scholars of gendered organizations, such as Acker (1990, 2006) and Ferguson (1984), argue that bureaucracies incorporate gendered logics and thereby workplaces maintain existing inequality. Instead of trying to reduce the salience of gender, firms might take the opposite route by elevating gender as something in need of group-specific solutions. Identity-based policies make social distinctions, such as gender or parental status more salient and legitimize women’s claim to existing positions and pay (Abendroth et al., 2017; Tomaskovic-Devey and Avent-Holt 2019).

Career programs may include mentoring, training, or voluntary quotas specifically for women. Because our data bundle all career programs into a single survey question, we cannot distinguish between programs, such as mentoring and quotas in our analysis. However, overall, we expect career programs for women will gradually narrow gender wage gaps after hiring, as many programs only go into effect once employees enter the establishment.

Specifically, mentoring and training programs support women’s career development and may lead to more promotions (Zimmermann and Collischon 2023) or performance bonuses. In support, Dobbin and Kalev (2022) demonstrated that mentorship programs in the USA increased women’s representation in management, especially when all employees receive mentors. In contrast, when examining the pay-related effects of mentoring programs in Germany, Peters, Drobe and Abendroth (2020) and Marx and Diewald (2022) found no relationship between mentoring and gender pay gaps. However, their sample was limited to establishments with more than 500 employees. Large businesses likely already have more formalized career development.

Firms can also support women’s career growth by setting voluntary gender targets. Quotas may empower qualified women to reach for higher-paying positions (Besley et al. 2017; Bertrand et al. 2018). Likewise, quotas may reduce pay disparities as supervisors need to offer higher pay to attract female talent. While quotas receive backlash in workplaces (Williams, Kilanski and Muller 2014; Dobbin, Schrage and Kalev 2015), diversity goals across Europe and the USA have reduced gender pay gaps, especially among highly qualified employees (Leslie, Manchester and Dahm 2017; Maume, Heymann and Ruppanner 2019; Peters, Drobe and Abendroth 2020; Marx and Diewald 2022).

All policies mentioned above focus on active employees. However, women’s pay particularly stagnates after parental leave (Schmelzer, Kurz and Schulze 2015; Musick, Doherty Bea and Gonalons-Pons 2020). While all parents in Germany are eligible for paid parental leave, mothers take leave more often and for more extended periods than fathers (e.g. Gerst and Grund, 2023). Therefore, training programs connecting parents on leave with their employers and occupations might particularly help mothers retain pre-birth skills and social networks. Additionally, mothers may use these programs to signal work commitment. When programs ease the reentry of new mothers, all women benefit as the mental association between (potential) motherhood and performance weakens.

Since career programs for women and parents address the structure of work and typically go into effect after hire, we expect them to narrow gender pay gaps gradually. Likewise, we expect additive effects, meaning organizations that implement career programs for women and parents reduce gaps more effectively than when addressing only one group:

H2: Gender pay gaps narrow with increasing tenure when establishments implement more career programs for women and parents.

2.4 Reducing gaps via employer-sponsored child care assistance

As women do more unpaid care work than men, combining work and family can be particularly challenging for mothers (Yavorsky, Kamp Dush and Schoppe-Sullivan 2015; Grunow 2019; Musick, Doherty Bea and Gonalons-Pons 2020). Many establishments design work around the ideal worker norm where the ideal employee is uninterrupted by other responsibilities. Firms may ease this tension by providing child care assistance, including in-house care or financial assistance for child care elsewhere.

The literature is divided regarding how child care affects pay. The work–life facilitation model postulates that providing child care lowers barriers for mothers (e.g. Johnson and Provan 1995; Fuller and Hirsh 2019). Two studies in the UK and France find that employer-sponsored child care is associated with smaller gender pay gaps (e.g. Heywood, Siebert and Wei 2006; Pailhé and Solaz 2019). At the same time, studies of employer-sponsored child care in the USA and nine European countries found no significant effects (e.g. Baughman et al., 2003; Huffman, King and Reichelt 2017; van der Lippe, van Breeschoten and van Hek 2019).

In contrast, the flexibility stigma literature predicts that child care assistance widens gender pay gaps. Adverse effects arise because employer-sponsored child care hightens the salience of employees’ parental status and (perceived) violations of the ideal worker norm (Cech and Blair-Loy 2014; Munsch, Ridgeway and Williams 2014; Fuller and Hirsh 2019). Likewise, employers may perceive child care as an expensive burden and try to offset costs. Glass (2004) makes a strong case for the adverse effects of child care assistance on pay. In her longitudinal panel, US mothers experienced substantial pay penalties after using their employer’s child care assistance.

H3: Gender pay gaps widen more with increasing tenure in establishments providing child care assistance than in establishments without child care.

2.5 Education and policy effects

Broader research shows that gender-related public policy effects often vary by education (e.g. Morrissey 2017; Hook and Paek 2020). However, prior studies contradict each other regarding which educational group benefits most from workplace policies. For instance, Huffman, King and Reichelt (2017) find that more formalized workplaces narrowed the gender wage gap among lower income employees. In contrast, Abendroth et al. (2017) demonstrate that formalizing hiring practices primarily benefitted college-educated mothers. Regarding career programs for women, voluntary quotas and diversity goals tend to narrow gender gaps among highly educated employees (e.g. Leslie, Manchester and Dahm 2017; Bertrand et al. 2018; Peters, Drobe and Abendroth 2020), while a more general measure of policies promoting women (including mentoring) closed gaps among lower income employees (Huffman, King and Reichelt 2017). Thus, it remains unclear how formalization and career programs vary by education.

In contrast, employees’ education is critical in moderating child care effects. We expect worse outcomes for less-educated mothers because child care primarily accommodates employees’ current jobs. These programs might amplify education-specific differences, given that child care signals parental status (and creates costs for employers) without changing the nature of work. Managers might perceive employees in lower skill jobs as more interchangeable and are less willing to invest in retaining these employees. Likewise, jobs requiring less education might have fewer opportunities for advancement. Thus, while all mothers receive penalties for not conforming to the ideal worker norm (e.g. Williams, Blair-Loy and Berdahl 2013; Cech and Blair-Loy 2014), mothers with less education arguably have fewer means to counter stigma.

2.6 Studying gender pay gaps in the German context

As in other OECD countries, German laws prohibit employment discrimination, especially after 2006 with the passage of the German anti-discrimination law (Allgemeines Gleichbehandlungsgesetz). Policy effects on gender gaps in Germany are similar to those in other countries. For instance, formalization and career programs have comparable effects when using US data (e.g. Castilla 2015; Leslie, Manchester and Dahm 2017) and German data (e.g. Huffman, King and Reichelt 2017; Busch-Heizmann and Rinke 2018; Peters, Drobe and Abendroth 2020).

Historically, unions and work councils have played a central role in determining wages in Germany. However, the percentage of jobs covered by collective bargaining agreements has declined from 85 per cent in 1960 to 54 per cent in 2018 (OECD 2022). With a growing share of establishments operating outside of collective bargaining agreements, wage inequality has increased in Germany, especially between establishments (Card, Heining and Kline 2013; Tomaskovic-Devey et al., 2020). A 2017 survey found that 59 per cent of respondents had the opportunity to individually negotiate their salary, which primarily benefitted men (Sauer et al., 2021). As we limit our analyses to establishments with at least 20 employees, our sample’s share of establishments covered by collective bargaining is higher than the German average. Therefore, our results represent a conservative estimate of policy effects.

3. Data, measures, and analytic strategy

3.1 Data

To examine how workplace policies shape firm-internal wage trajectories, we analyze linked employer–employee data (LIAB) from the German Institute for Employment Research (IAB) (Ruf et al., 2021). The LIAB surveys a representative panel of German establishments yearly (i.e. the Establishment Panel) and then links the Establishment Panel with employees’ register data to create employer–employee linked data. These data include all employees working at participating establishments for at least one day as of 30 June each year and who are subject to social security contributions or are marginally employed, which excludes civil servants and self-employed (Bechmann et al., 2017).

We start with the 2005 wave because the Establishment Panel first included survey questions regarding formalization in that year. We only analyze individuals for whom we can observe complete pay trajectories, meaning who entered participating establishments in or after 2005. We exclude part-time and marginal employees, as the LIAB does not say how many hours per day they worked, making it impossible to compare daily pay. Similar to prior research, we exclude small establishments with less than 20 employees (e.g. van der Lippe, van Breeschoten and van Hek 2019). Overall, our study examines firm-internal careers of 573,812 employees (1,735,490 person-years) with a vocational degree or less who work in 8,208 different establishments. We observe 279,873 college graduates (889,645 person-years) across 6,963 establishments.

3.2 Measures

3.2.1 Dependent variable

We use two different measures of earnings. For employees whose highest degree is vocational training, we examine the natural logarithm of gross daily pay in 2015 euros, including bonus payments. For employees with a college degree, we use a binary measure that assesses whether their wages are top-coded. We use different outcome measures for college graduates because the LIAB uses pension insurance records to calculate employee pay. Pension insurance contributions only apply to earnings up to a certain threshold. Consequently, the LIAB cannot calculate pay for employees earning above the contribution limits (Ruf et al., 2021). In 2019, employees earning at least €80,400/year in West Germany or €73,800/year in East Germany fell above the contribution limit (see Supplementary Table S1 for annual, region-specific limits).

Top coding affects less than 3 per cent of employees with vocational training or less. We drop these individuals from the analyses. However, among college graduates, 18 and 40 per cent of women’s and men’s person-years, respectively, include top-coded pay. Given how common and gender-specific top-coding is among college graduates, dropping top-coded observations or imputing pay may bias results, as gender is typically one of the variables used in imputation (e.g. Tomaskovic-Devey and Melzer 2020). We include college graduates in our analysis because workplace policies, like child care, often have education-specific outcomes (e.g. Abendroth et al., 2017; Huffman, King and Reichelt 2017; Morrissey 2017). Thus, while we run analyses separately for employees with and without a college degree, using different dependent variables prevents us from testing education-specific hypotheses.

In additional analyses (Supplementary Tables S6 and S7), we divide the noncollege sample into employees with a vocational degree and those without occupational qualifications. None of the policies reached significance among the small group of employees lacking occupational degrees, although effects trended in the same direction as those with vocational training. The results for vocational school graduates remained virtually unchanged. For parsimony, we combined both groups into a single sample.

3.2.2 Central explanatory variables

Administrative records capture employees’ gender as a binary variable. Our models also include employees’ years of establishment tenure and tenure squared to examine nonlinear pay growth. To capture formalization, we analyze establishments’ answers in response to the following survey items. “Please indicate whether the following instruments exist in your establishment: a) planning of personnel development and advanced vocational training set out in writing, b) staff requirements set out in writing, c) stipulated procedures for staff recruitment, d) job descriptions for most of the existing jobs, e) written agreements on objectives with employees, f) written personnel reviews” (IAB, 2019, p. 4). Like Huffman, King and Reichelt (2017), we use an index combining all policies to capture a broader logic of formalization. The Cronbach’s Alpha for the index is in the acceptable range, with 0.75 and 0.70 for employees without and with college degrees, respectively (Singleton and Straits 2018). We break the index into three categories (no written policies, one to three written policies, and four to six written policies) to capture nonlinear effects.

The Establishment Panel includes two questions regarding career programs for women and parents. First: “Please indicate which measures are provided in your establishment: a) Targeted promotion of women (e.g., mentoring schemes, plans for the promotion of women, special further training, target agreement to increase the proportion of women).” We acknowledge that this single yes/no question combines several conceptually distinct approaches, such as mentoring and voluntary quotas. However, it allows us to assess the presence of any program that is both gender-specific and work-focused.

The second question asks about “Services for employees taking periods of parental leave (e.g., further training).” While these programs do not explicitly name women, mothers take parental leave more often than fathers. Therefore, we combine the question about programs for women with programs for parents because they are both identity-based and work-focused. Together, our career program measure indicates whether establishments have no policy, one policy, or both policies in place.

Finally, the child care assistance measure is “1” when establishments provide “Support with child care (e.g., company child care facility, cooperation with child care facilities, child care during school holidays) or financial contributions towards child care.” Because the LIAB asks about workplace policies (including formalization, career programs, and child care assistance) at intervals of 2–4 years, we interpolate all policy measures between years. Changes in workplace policies occur rarely, for instance only 10 per cent of establishments introduced or discontinued child care assistance between observation points. Thus, we believe that being able to observe employees’ pay trajectories over a consecutive 14-year period outweights minimal measurement error introduced by interpolation.

3.2.3 Control variables

Our control variables include employee-, job-, and establishment-level characteristics. On the employee level, we control for demographics and human capital measures. We include nationality (German, South E.U. & Balkan, Central/Northern EU & North America, East E.U., and Non-EU & Non-North American), age, and age squared. Among employees without a college degree, we also control whether individuals have attained an upper-secondary degree (“Abitur”) and whether they completed vocational training (“Berufsausbildung”). We calculate employees’ potential labor market experience by subtracting years of education from their age (minus 6 years) at hire.

We include employees’ turnover probability to account for gender differences in turnover. If turnover differs for men and women, then selective attrition might artificially widen or narrow gender wage gaps over time. We estimate a separate discrete event history analysis to account for selective attrition (see Supplementary Table S2). These additional event history models use the same control variables as our wage analysis. Based on the estimates, we predict turnover probabilities for each person-year and then use that predicted probability as a control variable in the wage analyses below.

The models account for an indicator of employment gaps with the current employer before the current observation. We sometimes observe an employment spell for several years, but the employee is missing in one or more of these years. Because the LIAB follows establishments, not individuals, we do not know what employees did during these gaps. Employees may have worked in another branch, worked for a different firm, or were unemployed before returning to their original establishment. Likewise, parents (mostly women) have such gaps when they go on parental leave. Because factors leading to gaps might be gendered, we control for gaps in our analysis.

On the job level, we account for employees’ two-digit occupation and the average years of schooling of all individuals working in the same occupation and establishment. To assess the relative power of an occupation versus other occupations in an establishment, we add the percent of the workforce employed in the focal occupation.

Our establishment-level controls include three sets of variables. First, establishment demographics include 12 industry dummies, 16 state dummies, and 6 establishment-size dummies. We further control for establishment age, the relative change in establishment size, and the establishment’s legal form.

Finally, the institutional controls include dummy variables indicating whether the establishment has an employee council or another form of employee representation, whether it is bound by a collective bargaining contract, and whether the establishment is a member of the Chamber of Trade, Chamber of Commerce, or another chamber. To account for the establishment composition, we include the percentage of women in an establishment, the percentage of women in management, and a dissimilarity index, which measures gender job segregation in each establishment. All models include year dummies.

3.3 Method

We examine how gender pay gaps develop after establishments hire employees. Therefore, we observe employees from the point-of-hire (“intercepts”) and how their pay develops post-hire with increasing establishment tenure (“slopes”). To this end, we employ multilevel growth curve models that assess whether intercepts and slopes vary systematically by gender and establishment policies. The models consider that repeated observations cluster within individual employees nested within establishments. Moreover, growth curve models handle unbalanced panels well, which arise because we observe each employee for different numbers of years (Raudenbush and Bryk 2002; Singer and Willet 2003).

We include random slopes for establishment tenure so that pay growth can vary between individuals. We add random slopes for gender at the establishment level so that the effect of gender on wages can vary across establishments. Our models include cross-level interactions between tenure, gender, and policies to assess whether pay trajectories depend on gender and workplace policies.

Our models consider whether policies are in place in a particular year. Thus, findings represent the average difference between employees with a given tenure in establishments with versus without policies. This interpretation is consistent with the idea that policy effects may unfold with time and that policies rarely change.

Our outcome variable for university graduates is binary, and we use a growth curve model that estimates linear probabilities. We chose a linear probability model (LPM) over a logistic model because logit models are less suited for estimating interaction effects. Moreover, we are primarily interested in average effect estimates, and LPM coefficients are almost identical to average marginal effects derived from logistic regressions. Therefore, this method allows us to compare coefficients across groups and models (Mood 2010; Gomila 2021).

4. Results

4.1 Descriptive analyses

Table 1 presents our individual- and job-level characteristics by gender and education. On average, full-time employed men with vocational training earn €96.80/day, while equivalent women earn €11.60/day less—leaving a 12 per cent pay gap. Among college graduates, 40 per cent of men earn above the pension contribution limit at least once, compared to only 18 per cent of women. Because we only include employees who started employment during our 14-year observation period, our sample is limited to employees with 14 years of tenure or less. Still, over 50,000 employees with a vocational degree or less and 21,000 college graduates reach at least 8 years of tenure in our sample.

Table 2 summarizes our establishment-level characteristics. About 8–9 per cent of establishments in our sample lack written rules for personnel practices, whereas 32–34 per cent have one to three rules. About 60 per cent of establishments formalize at least four of six procedures. Formalization of specific HR policies varies widely, ranging from programs adopted by 77–88 per cent of establishments (job description) to programs adopted by about half of establishments (employee objectives). Career programs for women and parents are even less common, with 22–27 per cent of establishments having one policy and only 11–12 per cent having career programs for both women and parents. Similarly, only 19–20 per cent of establishments offer child care assistance.

4.2 Development of gender gaps over time

Next, we examine how gender pay gaps evolve over employees’ firm-internal careers. Figure 1 shows predicted gender pay gaps using a multivariate regression model (see Supplementary Table S3). Figure 1a depicts gender gaps among employees whose highest degree is vocational training. Despite accounting for detailed employee, job, and establishment characteristics, women earn about 9 per cent less at hire than men with the same education, occupation, and establishment type. Gaps continue to widen over time. After 14 years of firm tenure, full-time women earn about 13 per cent less than men. Thus, substantial pay gaps arise at hire and widen further with firm tenure.

Figure 1b shows results for college graduates and their gender-specific likelihood of receiving pay above pension insurance contribution limits. After including controls, women are ten percentage points less likely to earn above the contribution limit than men at hire. During the first years of firm tenure, gaps widen gradually until about 7 years after entry, when gender disparities narrow again. Please note that although the gender-specific chance of reaching top-coded pay narrows after 6 years, gender pay gaps above the contribution limit might still increase. Still, approximately less than 5 per cent of the total workforce reached contribution limits. Therefore, simply reaching this echelon is an achievement associated with substantial income.

In sum, women never earn the same wages as men, even after controlling for various individual, job, and establishment characteristics. Women’s relative earnings prospects worsen with longer establishment tenure, at least for employees without a university degree, while women with a university degree can catch up a little. Because we use different dependent variables for employees with and without a college degree, it remains unclear whether the differences are substantive or due to our outcome measures. Our subsequent interpretation will focus on how policies change these trajectories rather than interpreting trajectories themselves.

4.3 Policy effects on the development of gender pay gaps

4.3.1 Formalizing HR Practices

To test our hypotheses, we examine how gender differences in pay growth depend on workplace policies. We add interactions between our formalization measures, gender, and establishment tenure. Results for the formalization index suggest that gender pay gaps are independent of formalization for both educational groups. Put differently, formalizing more HR practices does not affect gender pay gaps (see Supplementary Table S4).

We also examine the separate effects of each HR policy. Analyses broadly support the conclusion that formalization leaves gender pay gaps unaffected. Most individual policies slightly reduce gaps, but the effects are weak and never reach significance. We find two smaller exceptions: Formalizing recruitment procedures mitigates widening gender gaps with tenure (see Supplementary Figure S1). However, effects only start 9 years after hire and only apply to employees with vocational degrees or less.

In contrast, formalizing employee objectives widens gender gaps in reaching top-coded pay (see Supplementary Figure S2). This effect is specific to college graduates and is only evident in the first few years after hire. When examining all results, we fail to find systematic evidence that formalization reduces gender pay gaps because most individual policies do not reach significance, and any significant results are contradictory. Therefore, we reject H1.

In additional analyses, we examine how collective bargaining affects gender pay gaps because bargaining agreements formalize pay decisions via external pressure. Agreements slow the widening of gaps over time (after about 8 years of tenure), but only among employees with a vocational degree or less. Since collective agreements bind more than 80 per cent of the establishments in our sample, our formalization effects on gender gaps are possibly muted.

4.3.2 Career programs for women and parents

To test our hypotheses regarding women-focused career policies, we model interactions between employees’ gender, tenure, and career programs (see Supplementary Table S5) while controlling for the simultaneous effect of child care programs on gender pay gaps. Figure 2 shows the predicted effect of career programs based on our multivariate models, once for employees with vocational degrees or less (Fig. 2a) and once for college graduates (Fig. 2b). The black line displays gender gaps in establishments without career support. In contrast, the dashed gray line shows establishments with career programs for women or parents. The solid gray line represents establishments with programs targeting both groups (parents and women).

Effects of career programs unfold within 2–3 years after employees join the establishment. Thus, as expected, career programs do not limit pay gaps at hire. However, identity-based career policies help all women keep up with their male counterparts over time. Without career programs, gender gaps widen much faster post-hire. After 14 years of tenure, the gender pay gap for employees with vocational training is three percentage points smaller in establishments with strong career support than in workplaces lacking support (∼20 per cent reduction in gender gaps). Thus, measures to promote women’s and parents’ careers slow the widening gaps over time. Hence, findings support H2.

In additional analyses (available on request), we examine career programs for women and parents separately. While both policies reduce pay gaps independently, combining them reduces inequality beyond what each policy can accomplish individually.

4.3.3 Child care assistance

Next, we examine interactions between employees’ gender, tenure, and employer-sponsored child care (see Supplementary Table S5), holding constant the effect of career support on gender pay gaps. Figure 3 depicts policy effects for employees whose highest degree is vocational training (Fig. 3a) or college (Fig. 3b). Among employees with a vocational degree, child care is associated with widening gender pay gaps. That is, gender pay gaps are greater when child care is present than when it is absent. The results support Hypothesis H3 for employees with less education. In essence, child care assistance is associated with disadvantages during early and medium tenure among women holding vocational degrees or no occupational degrees.

Figure 3b focuses on college graduates. While delayed, child care is beneficial and associated with narrowing gender gaps in reaching top-coded pay. After 14 years of tenure, the gap is 50 per cent smaller in the presence of child care compared to workplaces without assistance. The beneficial effect of child care among college graduates contradicts Hypothesis H3. Our different outcome variables make it difficult to compare Fig 3a and b directly.

Moreover, our results contradict Huffman, King and Reichelt (2017) and Glass (2004). The former found beneficial effects among lower income workers, while the latter demonstrated adverse child care effects among professional and managerial employees. However, Huffman, King and Reichelt (2017) excluded top-coded earners from the analysis, and Glass (2004) observed employees for 7 years. In contrast, our beneficial effects only become significant after 8 years, meaning prior studies likely missed late positive effects. While policy effects for college graduates develop late, more than half of German employees stay with their employer for 10 years or longer (Destatis 2021; IWD 2022).

We want to note that the effect of employer-sponsored child care is weaker and significant for a shorter period if we do not simultaneously control for career support for women and parents. In contrast, the effect of formalization and career programs was the same regardless of other policies in the model. We discuss the implications in the “Discussion” section.

In summary, formalization does not affect pay gaps, while career support prevents gaps from widening among all employees. Child care assistance widens gaps among employees with less than a college degree while it helps college-educated women reach top-coded pay in their mid to late careers.

4.4 Robustness checks

4.4.1 Motherhood penalty

Women in Germany typically take 1–3 years of parental leave per child, which slows women’s wage growth (Musick, Doherty Bea and Gonalons-Pons 2020). Unfortunately, the registry data does not include parental status by default. Still, the German Institute for Employment Research (IAB) developed an indirect identification of childbirth using health insurance records (Müller and Strauch 2017). The Research Data Center (FDZ) generated this measure for our project as part of the German Research Foundation Priority Programme 1764. The measure only identifies 40–70 per cent of all births for women aged 20–38 years and does not identify new fathers. The IAB provides information on motherhood between 2005 and 2017, limiting robustness checks to this period. Since relatively few university graduates in our sample are also mothers, we only estimate models for employees with vocational training.

To unravel the effects of gender and motherhood, we include an additional interaction with motherhood, establishment tenure, and a random slope for motherhood on the establishment level (results available on request). Once we separate the effect of gender and motherhood, childless women catch up to men in pay. Consistent with the literature (Musick, Doherty Bea and Gonalons-Pons 2020), having a child severely depresses full-time employed women’s wages, and mothers fall further behind compared to men and childless women. Next, we separate policy effects for women with and without children. All effects hold after controlling for motherhood, suggesting that the abovementioned effects apply to all women, not just mothers.

5. Discussion and conclusion

Our article examines how gender pay gaps evolve over employees’ firm-internal careers. We use German linked employer–employee data between 2005 and 2019 to examine how three policy approaches—formalization, career programs for women and parents, and child care assistance—impact gender pay gaps across employees’ tenure. Despite accounting for detailed employee, job, and establishment characteristics, earnings gaps for employees without a college degree are just shy of 10 per cent and then widen with each year of firm tenure. For college graduates, we examine the likelihood of being among top-earners (i.e. above the pensions insurance contribution limit) and find substantial gaps at hire that worsen for 7 years after hire and then reverse.

This article advances our understanding of when policies work. None of the policies examined reduce hiring gaps, indicating considerable room for improvement. The results are consistent with Zimmermann and Collischon (2023), who found that a combined measure of child care and career programs for women failed to close pay gaps among recently hired employees. Our article extends this research by showing that formalization, as measured in this article, cannot limit hiring gaps either. Even beyond the point-of-hire, most policies never actually narrow disparities that came about during the hiring process (except child care among college graduates) and, at best, prevent disparities from widening further. Thus, policies only prevent or minimize gradually compounding disparities over time. This study highlights the need to investigate which policies can address gaps at hire.

Our study also speaks to the broader literature comparing policy approaches (e.g. Abendroth et al., 2017; Huffman, King and Reichelt 2017). The nonsignificant findings suggest that formalizing personal practices fails to reduce gender pay differences (e.g. Huffman and Velasco 1997; Dobbin, Schrage and Kalev 2015). Nonresults may be partially attributable to the broad coverage of collective bargaining agreements in Germany, as bargaining coverage generally compresses gender wage gaps (Tomaskovic-Devey et al., 2020). Indeed, in additional analyses, we found that agreements primarily compress pay differences during the mid- and late-career among employees with a vocational degree or less. Thus, our findings add to the literature linking collective agreements and gender pay gaps by highlighting that agreements primarily compress gaps among higher tenure employees. We expect that existing bargaining agreements mute the effect of formalization. However, the muting should apply to career programs and child care as well. Thus, formalization may still be weaker in a less regulated national context than identity-conscious approaches.

Formalization significantly reduced hiring gaps when we estimated an additional simple ordinary least squares (OLS) regression with clustered errors. However, formalization no longer affects gaps once we introduce random slopes and intercepts in the multilevel models. Thus, between-establishment differences in pay and other unobserved establishment characteristics likely drove the results of the OLS regression. Most prior studies, which concluded that formalization reduced inequality, used a combination of cross-sectional data and OLS regressions (e.g. Anderson and Tomaskovic-Devey 1995; Busch-Heizmann and Rinke 2018; Lükemann and Abendroth 2018). The results may be partially attributable to how analysis methods treat the nested nature of workplace data. In support, a recent study using German employer–employee linked data with multilevel models also found mostly nonsignificant direct effects of formalization on gender pay gaps (Abendroth et al., 2017).

Critical scholars (Ferguson 1984; Acker 1990; Acker 2006) identified myriad ways gendered logics inform policies and how gender is embedded in bureaucratic structures. Our study shows that policies challenging structural disadvantages by creating career opportunities for women and parents benefit women of all educational backgrounds. The results contradict Huffman, King and Reichelt (2017), who only found beneficial effects for lower income employees. Inconsistencies might be due to differences in outcome variables or sampling, as Huffman, King and Reichelt (2017) excluded all top-coded earners from their analyses. However, our findings echo previous studies on women-focused programs with beneficial effects for women of all educational backgrounds (e.g. Maume, Heymann and Ruppanner 2019; Leslie, Manchester and Dahm 2017; Peters, Drobe and Abendroth 2020). Career programs for women and parents create supportive structures and empower women to claim important organizational resources, such as promotions and pay (Tomaskovic-Devey and Avent-Holt 2019). In short, despite women’s dramatic advances in workplaces over the past 50 years, gender is still a meaningful status distinction that needs tailored solutions. As about 60 per cent of establishments in our sample lack career programs for women or parents, there is room to grow.

Regarding child care assistance, employer-sponsored child care assistance penalizes women with less education but benefits women with a college degree. Results might diverge by education because we used different outcome measures for each group: For employees with a vocational degree, we examine daily pay, while we use a dichotomous indicator of whether pay falls above the pension insurance limits for college graduates, which collapses gender pay gaps that likely exist above the earnings cut-off.

However, the dichotomous measure is substantively meaningful because reaching top-coded pay is rare for women and indicates access to elite jobs as only about 5 per cent of the workforce reaches the contribution limit. More importantly, our divergent results are consistent with the idea that gender and class inequality are closely linked (Mandel and Rotman 2022).

Thus, the mixed results for child care assistance suggest that identity-based policies have heterogeneous effects that depend on what they regulate. Child care assistance enables parents to come to work, but this policy ignores what happens at work. “Just” showing up full-time and adhering to the ideal worker norm is insufficient to prevent motherhood penalties and instead amplifies whatever education-specific opportunity structures employees find at work. With limited career opportunities in non-professional jobs (often held by parents with less than a college degree), child care only makes parental status more salient (Budig and England 2001). Employers might be less willing to invest in less-educated workers; therefore, the flexibility stigma outweighs any work–life facilitating effects of child care. In contrast, facilitation might outweigh stigma among managerial and professional women because child care allows women to take advantage of existing career ladders, and companies might invest more to encourage highly trained mothers to return from parental leave. Thus, better opportunity structures among college graduates might offset the heightened visibility of motherhood status. In contrast, less-educated women have fewer opportunities to compensate for the greater salience of their parental status.

Several pieces of evidence support the intersectional interpretation. Additional analyses (available on request) show promotions to managerial positions typically occur after about 7–8 years of firm tenure. Without child care assistance, professional women might miss their promotion window and stay below top-coded pay. Moreover, child care effects strengthen after we account for career programs in our models, indicating that child care strongly covaries with opportunities at work. Hence, it is essential to distinguish between identity-based policies determining what happens at work (e.g. career programs) and policies enabling employees to come to work (e.g. child care).

Our study suggests that career programs for women and parents limit widening gender pay gaps most effectively. In contrast, formalization is ineffective, and employer-sponsored child care assistance even widens gaps unless accompanied by opportunities at work. However, we hesitate to give identity-based (career) policies an all-out endorsement because research needs to examine how career programs for women affect pay disparities among other demographic groups. While career programs narrow gender gaps, they may widen earnings disparities along other dimensions, such as race and ethnicity, religion, citizenship, age, or disability. For instance, past gender-based policies in the USA primarily benefited White women and ignored race-related barriers, thereby increasing race-based disparities among women (Stainback and Tomaskovic-Devey 2012).

Overall, this study advances our theoretical understanding of workplace policies in two ways: First, it supports theories of gendered organizations, which argue that formalization alone does not address gender gaps. Instead, policies need to be explicitly identity-based and work-focused to overcome gendered barriers. Second, policy effects are uneven throughout employees’ careers. At best, policies guard against the further widening of gaps after the point of hire. However, none of the policies addressed gender gaps at entry, and most failed to reverse existing gaps over time.

Acknowledgements

We would like to thank Steven Vallas, Jill Yavorsky, and the anonymous reviewers for their thoughtful comments. We also thank the participants at the following conferences for their feedback: the 2021 meeting of the American Sociological Association and the 2021 European Consortium for Sociological Research.

Supplementary data

Supplementary data is available at Socio-Economic Review Journal online.

Conflict of interest statement. None declared.

Funding

The German Science Foundation (DFG) funded this research via the Priority Programme 1764 (Funding #: GA 758/5-1, Co-PIs: Markus Gangl and Anne-Kathrin Kronberg).

References