Workplace Gender Pay Gaps: Does Gender Matter Less the Longer Employees Stay?

Abstract

Research indicates men often receive greater merit rewards than women for the same performance. It is unclear, however, whether gender differences in merit rewards narrow with increasing firm tenure or whether gender differences in merit-rewards stay constant across employees' firm-internal career. Using longitudinal personnel records of a private U.S. employer (2005–2014), the author finds no evidence for declining gender effects on pay when employees stay longer, not even among nonprofessionals where performance is easier to assess. Results contradict information-based theories and speak to status characteristics theory. Moreover, gender disparities are significant only when supervisors have discretion over merit increases.

Keywords

income inequality gender organizations pay-setting practices

Gender pay differences in the United States have stagnated since the 1990s (Blau & Kahn, 2017), and researchers are increasingly turning toward pay-setting practices in workplaces to understand proximate causes of disparities (Blau & Kahn, 2017; Goldin, 2014; Reskin, 2000; Tomaskovic-Devey & Avent-Holt, 2019). Within firms, pay disparities first arise at the point of hire (e.g., Brett & Stroh, 1997; Dreher, Lee, & Clerkin, 2011; Fuller, 2008; Kronberg, 2013; Petersen & Saporta, 2004; Quintana-García & Elvira, 2016). Holding constant occupation and human capital, gender disparities at hire may result from gender-specific negotiation behavior (Babcock & Laschever, 2003), social networks (Ibarra, 1992; McDonald, Lin, & Ao, 2009), reasons for switching employers (Cooke, 2003; Dwyer, 2004), lack of information (Sterling & Fernandez, 2018), or differential treatment by hiring managers (Petersen & Saporta, 2004). Beyond the point of hire, gender pay gaps may change further as employees stay with their firm. Evidence suggests men receive greater rewards for annual performance scores than women, indicating a gender bias in performance-rewards (e.g., Belliveau, 2012; Castilla, 2008; Castilla & Benard, 2010). However, we know little about the temporal patterns of posthire biases. The present study therefore asks: “Does the effect of gender on pay lessen as employees stay longer with their firm or do biases affect employees across their entire firm-internal career? Further, how does organizational context shape gender pay gaps over time?”

To understand how gender pay gaps change with employees’ firm tenure, I build on Correll and Benard (2006) and distinguish between information- and status-based theories of pay disparities. Information-based approaches, such as statistical discrimination, emphasize that managers are uncertain of applicants’ future productivity (e.g., Akerlof, 1970; Bidwell, 2011; Halaby, 1988; Jovanovic, 1979). Lacking better information, supervisors rely on easily observable characteristics, such as gender, to predict applicants’ future performance (Aigner & Cain, 1977; Arrow, 1973; Phelps, 1972). Once hired, firms learn of employees’ actual performance and adjust pay accordingly to reward the most productive employees. Thus, among equally performing employees, the effect of gender should lessen with increasing tenure (Altonji & Pierret, 2001).

In contrast, status characteristics theory highlights nonconscious social processes that are independent of firm tenure (Correll & Benard, 2006). Status characteristics theory posits gender is one of the primary frames of reference in social interactions, meaning we automatically sex-categorize others in social encounters (Brewer, 1988; Brewer & Lui, 1989; Ridgeway, 2011). As the importance of gender is related to salience and not lack of information per se, status characteristics theory predicts that gender affects pay of newly hired and long-time employees alike.

To examine posthire pay trajectories, I use longitudinal personnel records (2005–2014) of a large, private U.S. employer (“B2G”¹) located in the service industry. Personnel records include detailed job and earnings data which allow me to follow employees for up to 9 years after they were hired. Given the detailed nature of personnel data, I can account for performance differences between employees and organizational contexts, both of which influence pay decisions. Moreover, B2G heavily regulates hiring salary, resulting in pay equity among equally qualified men and women at hire. Equal pay at hire allows me to isolate posthire processes without the interference of initial pay differences that often arise at hire in less regulated organizations.

Business units within B2G also differ on how much discretion supervisors have over their employees’ merit increases, and I distinguish between units in which supervisors have high versus low discretion. If supervisory biases give rise to gender differences, disparities should grow faster in high-discretion units (e.g., Baron, 1984; Baron, Hannan, Hsu, & Koçak, 2007; Castilla, 2015; Dobbin, Schrage, & Kalev, 2015; Kalev, Dobbin, & Kelly, 2006). Moreover, I differentiate between professional and nonprofessional employees to better distinguish between information- and status-based processes. Because professional work is often more abstract, productivity is arguably more difficult to observe in professional positions compared to nonprofessional positions. Statistical discrimination theory implies information-based biases are more persistent when productivity is harder to assess, causing gender pay gaps to narrow at a slower pace among professionals.

Counter to the information-based approach, my findings indicate performance-reward biases affect new and old employees alike. I find no evidence suggesting the effect of gender on pay weakens over time. Instead, gender pay gaps widen at a remarkably constant rate among professionals and widen at an accelerating rate among nonprofessionals. Moreover, total pay differences reach significance only when supervisors have individual discretion over merit raises.

This article contributes to our understanding of gender dynamics within workplaces by providing an empirical test of status- and information-based processes affecting pay decisions within an organizational environment. I also extend research on performance-reward biases (e.g., Belliveau, 2012; Castilla, 2008; Castilla & Benard, 2010) and show that biases affect employees over their entire firm-internal career instead of weakening over time. I discuss theoretical and policy-related implications in the conclusion.

Gender Pay Differences in the Workplace

After considerable advances in the 1980s, progress on gender pay equality has stagnated in the U.S. in recent decades. Among full-time employees aged 25 to 69 years, the raw female-to-male earnings ratio moved from 59% in 1980 to 77% in 2010 (Goldin, 2014). The female-to-male ratio narrows to 92% when adjusting for gender differences in human capital and sex segregation into specific occupations and industries. Adjusted earnings ratios, however, have remained remarkably constant at around 92% since the 1990s (Blau & Kahn, 2017). In this context, there is a renewed focus on firms’ employment practices and their role in generating and alleviating pay inequality (Blau & Kahn, 2017; Goldin, 2014; Reskin, 2000; Tomaskovic-Devey & Avent-Holt, 2019).

Within firms, gender pay gaps often start at the point of hire. Noticeable pay differences between men and women at hire persist even when controlling for education, work history, occupation, and industry (Brett & Stroh, 1997; Dreher & Cox, 2000; Fuller, 2008; Kronberg, 2013; Lam & Dreher, 2004; Pearlman, 2018; Petersen & Saporta, 2004; Quintana-García & Elvira, 2016). Pay gaps among men and women may widen even further after the point of hire as supervisors decide on employees’ annual merit increases. In this regard, experimental and organizational research has demonstrated that men receive greater rewards for the same performance than their women counterparts (Belliveau, 2012; Castilla, 2008; Castilla & Benard, 2010). It is unclear, however, whether posthire performance-reward biases affect pay particularly among newly hired employees or whether biases affect new and old employees alike.

Temporal Patterns of Posthire Pay Differences

Information-Based Mechanisms: Declining Effect of Gender

In most cases, information about external applicants is particularly limited during the hiring process, leaving managers uncertain of external applicants’ future job performance (e.g., Akerlof, 1970; Bidwell, 2011; Halaby, 1988; Jovanovic, 1979). In absence of better information, managers rely on easily observable characteristics, such as gender, to predict productivity (Arrow, 1973; Phelps, 1972). For instance, because women take on caregiving responsibilities more often than men (Bianchi, Robinson, & Milkie, 2006), supervisors may expect women’s productivity to be lower or to be less predictable than men’s (Arrow, 1973; Oettinger, 1996; Phelps, 1972). Acting on “known or assumed differences between groups” (Stainback, Tomaskovic-Devey, & Skaggs, 2010, p. 233), hiring managers may offer women less pay.

Once hired, managers are better able to observe employees’ actual productivity and adjust pay accordingly to retain the most productive employees. With more performance information available, supervisors may no longer rely on gender as performance proxy (Altonji & Pierret, 2001). Therefore, the gender effect on pay should lessen as employees accumulate tenure, meaning biases disproportionally affect newly hired employees.

In support of information-based explanations, Sterling and Fernandez (2018) found women received higher starting pay in firms in which they also completed an internship, whereas men’s pay was unaffected by previous internships. Gender gaps in starting salaries therefore disappeared when firms were able to observe applicants’ performance during the internship. Sterling and Fernandez’s (2018) results mirror studies by Petersen and Saporta (2004) and Woodhams, Lupton, Perkins, and Cowling (2015), which found gender gaps at hire gradually vanished as employees accrued firm tenure.²

Given theoretical and empirical evidence, the information-based approach predicts that gender biases decrease over time. I depict this prediction in Figure 1. The black bars indicate statistical assumptions are most pronounced in the first year, but then decrease in subsequent years as supervisors observe employees' performance. Put differently, statistical biases should decrease the longer employees stay. As annual pay differences still accumulate with tenure, the total gender gap continues to grow, but at a decreasing rate. Figure 2(a) illustrates how the accumulation of pay differences slows down several years after hire. Thus, information-based theory predicts:

Figure 1.

Change in annual biases, information- versus status-based theories. Values are hypothetical and illustrate rates of change.

Figure 2.

(a). Hypothesis 1—posthire gender pay gap increases at a decreasing rate and (b) Hypothesis 2—posthire gender pay gap increases at a constant rate. Values are hypothetical and illustrate rates of change.

H1: With increasing firm tenure, gender pay differences grow at a decreasing rate.

Status-Based Mechanisms: Constant Effect of Gender

Counter to information-based approaches, status characteristics theory emphasizes social-cognitive, nonconscious processes that operate independent of employees’ tenure (Correll & Benard, 2006; Ridgeway, 2011; Ridgeway & Correll, 2004). Gender is not merely a proxy for performance in the absence of better information, but an important cognitive category. According to status characteristics theory, to navigate social interactions, individuals automatically categorize and act upon visible personal characteristics. When categorizing others, individuals process a person’s gender (along with race and age) before they assign people to more specific categories, which makes gender a primary cognitive category (Brewer, 1988; Brewer & Lui, 1989; Ridgeway, 1997, 2011). Once sorted, we apply gender status beliefs, which often assign men higher social status. With higher social status, men are typically believed to be more competent and deserving of rewards (e.g., Berger, Fisek, Norman, & Wagner, 1985; Pugh & Wahrman, 1983; Rashotte & Webster Jr., 2005; Ridgeway, 1997, 2011). In workplaces, gender status beliefs may therefore lead to nonconscious double standards of performance, with higher standards applying to women (e.g., Foschi, 1989, 1996, 2000).

Several experimental studies demonstrate that gender beliefs affect reward decisions even when decision makers have concrete performance metrics at hand (Auspurg, Hinz, & Sauer, 2017; Belliveau, 2012; Castilla & Benard, 2010; Steinpreis, Anders, & Ritzke, 1999). The aforementioned experiments asked participants to review fictitious employee profiles that included occupation, experience, tenure, parental status, health status, previous pay, and merit increases, along with detailed performance records. Fictitious employees’ gender was manipulated by randomly assigning male and female first names to each profile. Participants were asked whether they considered a specific salary to be fair (Auspurg et al., 2017), to determine applicants’ hiring pay (Steinpreis et al., 1999), or to divide a pool of performance rewards among a group of employees (Belliveau, 2012; Castilla, 2008). In all studies, participants assigned higher pay to male candidates than to equally performing women.

Thus, when compared with information-based approaches, status characteristics theory argues individuals do not rely on gender beliefs because they lack better information, but because gender is a socially relevant characteristic in the situation. Research suggests gender becomes particularly relevant or salient when social groups include men and women or when the task at hand is typically associated with one gender (Ridgeway, 1997, 2011), neither of which are immediately related to employees’ firm tenure. Given gender is a primary cognitive category, gender status beliefs serve as a relatively stable lens through which we filter new information (e.g., Berger, Cohen, & Zelditch, 1972; Correll & Benard, 2006; Ridgeway, 2011). Status characteristics theory, therefore, predicts gender status beliefs apply equally to new and old employees.³

I depict the prediction of status-based theory with the gray bars in Figure 1. Because performance-reward biases are time-constant the gray bars remain constant with increasing firm tenure. Gender pay differences accumulate over time and result in a constant increase of gender pay gaps, which I illustrate in Figure 2(b). In sum, status characteristics theory predicts:

H2: With increasing tenure, gender pay differences grow at a constant rate.

The Effect of Organizational Context

Gender disparities are not ubiquitous. Instead, organizational practices strongly affect the extent of workplace inequality (e.g., Abendroth, Melzer, Kalev, & Tomaskovic-Devey, 2016; Acker, 2006; Bidwell, Briscoe, Fernandez-Mateo, & Sterling, 2013; Dobbin et al., 2015; Kalev et al., 2006). In this article, I examine two contextual variations: supervisors’ discretion over annual merit raises and observability of performance (via professional status).

Low Versus High Supervisory Discretion

Information- and status-based approaches both assume gender differences result from supervisory decisions. To test this assumption, I distinguish between subunits in which supervisors have high versus low discretion over merit pay. While some research cautions that bureaucratic rules are inherently gendered (Acker, 2006; Dobbin et al., 2015; Ferguson, 1984; Williams, Kilanski, & Muller, 2014), other studies found that firms were able to limit the effect of biases by formalizing decision processes (Baron, 1984; Baron et al., 2007; Bielby, 2000; Castilla, 2015; Elvira & Graham, 2002; Reskin & McBrier, 2000).

In several experiments, participants showed weaker biases in hiring and pay decisions when they were held accountable for their decisions (e.g., Salancik & Pfeffer, 1978; Tetlock & Mitchell, 2009). Similarly, using personnel data from large employers, Elvira and Graham (2002) and Abraham (2017) found that gender pay gaps were greater when pay was less formalized. Castilla (2015) showed gender and race-based performance-reward biases vanished after supervisors had to report their decisions to upper-level management and a review committee. Finally, using a representative sample of U.S. firms, Kalev et al. (2006) and Dobbin et al. (2015) demonstrated accountability-based policies increased managerial diversity more than other policies. Taken together, these studies suggest posthire gender gaps should grow faster in units where supervisors have more discretion over employees’ annual merit increases.

H3: Gender pay differences increase faster in high-discretion units than in low-discretion units.

Professional Versus Nonprofessional Occupations

To assess the relative role of information- and status-based processes, I distinguish between employees in professional and nonprofessional jobs.⁴ The type of job may affect pay differences because work outcomes are arguably more observable and easier to attribute to employees in nonprofessional jobs.⁵ Professionals often rely on abstract and specialized knowledge they acquired through education and credentialing (Freidson, 1988). Reliance on abstract knowledge gives professionals more discretion and autonomy (Hodson & Sullivan, 2012). Because professional work is so specialized and autonomous, it is likely more difficult to observe and assess professionals’ productivity (e.g., Alvesson, 2001).

Several studies linked job type to the size of gender pay gaps. For instance, Petersen and Morgan (1995) found the within-establishment-occupation gender pay gap was 1.7% among blue-collar and clerical workers, whereas professional women earned 3.1% less than men in the same establishment and occupation. Greater gender pay gaps among professionals have since been replicated in several subsequent studies (Alkadry & Tower, 2006; Milgrom, Petersen, & Snartland, 2001; Petersen, Snartland, Becken, & Olsen, 1997).

Information-based theory therefore implies that statistical biases vanish faster when employees’ productivity is easier to evaluate. The gender effect on pay should therefore narrow faster in nonprofessional jobs than in professional jobs because performance is harder to assess among professionals. Put differently, information-based theory predicts the curve in Figure 2(a) flattens faster in nonprofessional jobs than in professional jobs. Conversely, a lack of slow down (i.e., learning) may indicate biases are not information-based.

H4: Statistical biases decrease faster among nonprofessional than professional employees.

Studying Gender Pay Gaps Using Personnel Data

Data

To examine pay growth of men and women once they are hired, I follow the tradition of previous studies on organizational pay disparities (e.g., Castilla, 2008; Petersen & Saporta, 2004) and use longitudinal personnel records of a large, private-sector U.S. employer, B2G.⁶ B2G primarily provides professional services and is located in a single metropolitan area. Between 2005 and 2014, B2G employed approximately 12,800 regular, full-time employees of which 44% were professionals. B2G is very bureaucratic, meaning policies aimed at preventing disparities govern most of B2G’s employment practices. Most prominently, B2G heavily regulates pay at hire by allowing hiring managers little room for negotiations. At-hire pay regulation makes B2G an ideal site to isolate and study posthire processes that are not confounded by previously accumulated pay inequality.

To reconstruct trajectories, I examine (anonymous) personnel records. Each employee receives an initial entry in their record at hire. B2G adds additional entries every time something changes (e.g., change of supervisor, pay raise, or promotion). Each record indicates the type of change, the effective date, job information, and employee demographics.

Although B2G’s personnel records go back to 1997, I focus on the time between 2005 and 2014 because B2G did not electronically record annual performance scores, a central variable in my analyses, until after 2005. I exclude temporary and part-time employees, as they do not experience the same earnings growth as regular employees. The analyses exclude employees who have missing information on education or pay grade.⁷ Similar to other organizational studies of pay (e.g., Bidwell, 2011), I exclusively focus on employees hired in or after 2005 to capture employees’ entire earnings trajectory after hire. After excluding employees with missing data (8%) and employees hired before 2005 (48%), my final sample includes 5,631 regular, full-time employees hired in or after 2005.

Some of my coding and analysis decisions were informed by semistructured background interviews I conducted with 19 B2G supervisors in 2014. Interviews focused on individuals working as frontline, mid-level, and executive-level managers in HR, IT, or finance. Interview questions asked how supervisors made hiring, promotion, and merit increase decisions and how decision processes may have changed over the years.

One disadvantage of using personnel data from a single organization is I have no information about employees’ experiences before they entered B2G. This means I cannot account for gender differences on prehire characteristics, even though factors such as previous experience or reasons for leaving affect subsequent pay (e.g., Dwyer, 2004; Fuller, 2008; Hachen, 1990; Rider & Negro, 2015; Topel & Ward, 1992).⁸ When better prehire experiences improve employees’ performance, however, performance scores should capture these differences (e.g., Dokko, Wilk, & Rothbard, 2009). B2G included annual performance scores in the personnel records, allowing me to control for annual performance. Moreover, the goal of the article is not to explain pay differences at hire, but to understand how gender gaps change over time after employees are hired.

Compared with other U.S. firms, B2G is likely more bureaucratic, equity conscious, and female-dominated. To some extent, B2G self-selected into the study because they agreed to provide research access when other organizations did not. B2G’s leadership is possibly more open to scholarly research or more confident in its employment practices than nonparticipating organizations. Research on discretion, organizational culture, and gender composition of management suggests all of these factors should minimize gender pay differences, making B2G a particularly conservative site to study gender inequality. Pay gaps are likely even greater in less bureaucratic, less equitable, and less female-dominated workplaces. Similarly, existing research shows gender gaps vary widely between industrial sectors. Arguably, this variation is due to differences in workforce composition, regulation, or power of institutional actors such as unions or work councils. Net of differences in context, basic psychological processes, such as information- or status-based biases, should still function in similar ways.

Using personnel records also has several advantages. For instance, the personal data from B2G include employees’ exact pay grade and performance scores. To my knowledge, current employee or establishment panels (such as the Longitudinal Employer-Houshold Dynamics Data) do not include employees’ job title, rank, or performance scores, making it very difficult to compare pay of men and women in a meaningful way. Most important, personnel data provide important contextual controls (e.g., gender of supervisor and % female in job cells) and allow examination of de facto variation in pay-setting practices (e.g., supervisory discretion over pay). Such data are invaluable in developing a better understanding of organizational dynamics and policies surrounding pay.

Pay-Setting Practices

B2G standardizes initial pay for newly hired employees across all subunits. Before supervisors make job offers, they provide HR with applicants' résumé, the job description, and the job’s pay grade. Based on these records, compensation specialists determine a binding salary range. Multiple supervisors at B2G emphasized that narrow pay ranges effectively limit hiring managers’ ability to negotiate salary with applicants. Pay at promotion is equally standardized, such that pay increases by 3% to 6% for each ascending pay grade.

Once hired at B2G, each employee receives annual written performance evaluations. Supervisors score employees on a scale from 1 to 5 where 1 indicates unacceptable, 2 indicates below expectations, 3 indicates meets expectations, 4 indicates exceeds expectations, and 5 indicates far exceeds expectations. To arrive at a total score, supervisors aggregate scores from eight subdimensions. The total score is reported in the personnel records and used to determine annual merit increases. To enforce consistent evaluations, the HR department provides example rubrics to illustrate differences between scores. HR also refuses to process merit raises for employees without a performance score on record. While performance evaluations are standardized and enforced across all units, units differ in how they reward performance scores. In some units, the unit’s leadership decides centrally how to reward performance without supervisors’ input beyond the official performance evaluation. For instance, one supervisor in a low-discretion unit describes the process as automatic:

Once we have given the person the performance rating, then every year within – within the entity where I work, we come up with an entity-wide pay plan that assesses a merit increase for whichever level of performance you achieve. So everybody that achieves a level five is going to get the same merit increase so that we have salary equity within the organization.

In other subunits, supervisors have more individual discretion over merit increases, meaning managers can divide the merit pool among their employees as they see fit. A manager in a high-discretion unit describes:

You get the money from the Chief Financial Officer; this is what you have, and our VP doesn’t have prior – doesn’t do prior approval of them [merit raises]. He lets you decide what you’re going to do [ . . . ] and I do the same. I don’t ask [the supervisors in my department] who they’re giving what to. I trust that they’re going to do the right thing.

These accounts demonstrate how supervisors’ role in the merit process varies between units. This article uses between-unit differences to test how discretion affects posthire pay trends.

Overall, B2G’s policies appear to be similar to the organization studied by Castilla (2008, 2015), where entry salary is strongly regulated,⁹ but then supervisors are given discretion over employees’ merit increases posthire. This stands in contrast with the organization studied by Petersen and Saporta (2004) and Woodhams et al. (2015), where discretion was greatest at hire and then the organization made efforts to close gaps posthire. Being able to observe gender pay differences in an environment without initial pay differences adds significant insights because we can better isolate posthire mechanism. I will discuss policy implications in the Conclusion section.

Measures

Dependent variable

I use the natural logarithm of annual full-time equivalent earnings from wages and salaries in 2013 dollars. Full-time equivalent earnings are the annual pay rates associated with a job, independent of how many weeks and hours/week employees actually worked that year. Occasionally, B2G also pays bonuses, but these are rare events and apply to less than 10% of the total workforce. Additional analyses (available on request) show results were virtually identical when including bonus pay in the analyses.

Central explanatory variables

Gender

Employees self-identify as male (0) or female (1) during the initial hiring process.

Firm tenure

Firm tenure measures the number of years an employee has worked for B2G and is 0 at the point of hire.

Professional status

As part of their annual Equal Employment Opportunity reporting, B2G categorizes all employees into one of nine occupational groups. Employees are professionals when B2G categorized them as upper-level management or professional employees. Employees hold nonprofessional jobs when they work as technical worker, sales worker, administrative support worker, craft worker, operative, laborer, or service worker.

Low versus high-discretion units

Even though background interviews indicated substantial differences in how B2G’s 32 subunits reward performance, interview evidence is too anecdotal to gauge systematic differences in supervisory discretion. For a more systematic approximation of unit-level differences in supervisory discretion, I measure how much annual merit rewards vary among equally performing employees in the same unit. I therefore use unit-level variation in pay raises as a proxy for supervisory discretion over merit increases. High variation among similar employees indicates supervisors have more discretion and act independently. In contrast, low variation among similar employees suggests supervisors are more constrained in their decisions. Constraint may stem not only from policies such as centralized merit decisions but also from unwritten equity norms, which encourage supervisors to apply criteria more consistently.

Supervisors generally used two criteria to determine relative merit increases: annual performance scores and employees’ relative position within their pay grade (i.e., the compa-ratio). The compa-ratio describes employees’ pay relative to the pay grade’s midpoint. For example, if a grade’s midpoint is $80,000 and the employee earns $75,000, her compa-ratio is 93.75%, meaning her salary represents 93.75% of the grade’s midpoint (i.e., she is below the midpoint). Supervisors frequently used the following cutoffs: below midpoint (compa-ratio of below 90%), at midpoint (90%–110%), and above midpoint (110% or higher). The compa-ratio matters because below midpoint employees receive higher raises than employees who have the same performance score but are already at midpoint.

Table 1 illustrates how I calculated unit-level variation in merit increases within each unit-year. In each year, I divided each unit’s employees into five performance groups (performance rating: 2 or less, 3, 4, 5, or missing) and three compa-ratio groups (below, at, and above midpoint). Within each performance-compa-ratio cell, I calculated the average annual merit increase and the standard deviation around that mean. Cells have a standard deviation of 0 when equally performing employees with similar compa-ratios receive the same merit increase (e.g., all below midpoint employees with a performance score of 4 receive a 3% merit increase). The standard deviation is higher when similar employees receive different relative increases.

Next, I average the variation across the 5 × 3 cells within each unit. To gauge broader trends, I average units’ variation across time, meaning my discretion measure varies by unit but not time. Consequently, my discretion measure assesses discretion on the unit level instead of the supervisor level. I focus on units, because the interviews indicated merit policies were decided and enforced on the unit level. Put differently, the extent to which supervisors stick to or can subvert unit-level policies is a characteristic of the unit itself.

Table 1.

Exemplary Calculation of Discretion Measure for Business Unit A in Year X.

	Below midpoint (< 90%)		At midpoint (90%–110%)		Above midpoint (>110%)
	% Raise	Standard Deviation	% Raise	Standard Deviation	% Raise	Standard Deviation
Performance Score: 1 or 2	0.0%	0.00	0.0%	0.00	0.0%	0.00
Performance Score: 3	1.7%	0.63	1.5%	0.79	1.7%	0.21
Performance Score: 4	2.2%	1.36	1.9%	1.32	2.3%	1.60
Performance Score: 5	2.5%	3.53	1.2%	1.86	1.6%	2.28
Performance Score: Missing	0.0%	0.00	0.0%	0.00	0.0%	0.00

Note. Average standard deviation in merit raises for Unit A in year X is 1.06.

To calculate the final discretion measure for Unit A, I repeat the calculation above for every year (2005–2014) and then take the average across years. I repeat this process for Units B through W separately.

Although continuous in nature, the original discretion measure was too clustered to justify a continuous measure. Instead, I created a “high-discretion unit” dummy that is “1” when units fall above the median of the original discretion measure and “0” when units score below the median. Supervisors in low-discretion units still have some discretion, but overall discretion is noticeably lower than in high-discretion units.

Controls

Demographics and human capital

Race is self-recorded during the hiring process using standard Equal Employment Opportunity categories. My analyses control for being White (reference group), Black, Asian, and Other Race. The Other Race category includes less than 4% of the workforce and captures Native Americans, employees of Hispanic origin, and employees who identify with multiple racial categories or as Other/Unspecified.

I measure education in terms of highest degree, distinguishing between less than college (reference group), bachelor’s degree, master’s or professional degree, and PhD.¹⁰ I assess potential labor market experience at the point of hire by subtracting years of education from age (minus 6 years). Although not ideal, this is the best approximation of experience, as B2G does not record employees’ actual years of experience in the personnel records. Similarly, I control for age using eight age-group dummies (5-year categories).

Following Castilla (2008), I also control for individual turnover risk. This measure addresses the possibility that growing pay gaps result from changes in the workforce composition. I first use cox event history models to predict turnover hazards, controlling for individual, job, and labor market characteristics. I show and discuss the results from these analyses in detail in Table 4 in the Findings section. After estimating the models, I predict employees’ individual turnover hazards in a given year and use these predicted hazards as control variables in the pay analyses.

Job and unit characteristics

B2G employees work in 30 broad occupations across 32 business units. To compare men and women in the same jobs, all analyses include dummy variables for each job cell, with each job cell representing a unique Business Unit × Occupation combination.

Before managers can fill jobs, they work with HR to assign each job to a pay grade based on the job description. Pay grades range from 1 to 28 and define the possible range of pay in this job. Annual pay may range from $20,000 to $28,000 in the lowest grade and from $125,000 to $250,000 in the highest grade. To compare men and women in the same position, I control for employees’ current pay grade and initial pay grade at hire by adding dummy variables for each current and initial pay grade. This means, all gender differences in the multivariate analyses indicate differences between men and women in the same occupation, unit, and pay grade.

To account for other job characteristics, I control for % women and average % raise in each job–year. In addition, I include supervisors’ gender, which is “1” for male supervisors. I include a control for when supervisors’ gender is missing (6% of final sample). Finally, a measure for unit size gauges the number of employees (in 1,000 s) in each unit-year.

Performance measures

Supervisors score employees’ annual performance on a scale from 1 (unacceptable) to 5 (far exceeds expectations). The analyses distinguish between a performance rating of 3 or below (reference category), 4, and 5.¹¹ For employees without a score, I include a dummy indicating missing performance scores. Additional analyses (available on request) showed employees are most likely to miss performance scores in the first two years. Men and women are equally likely to miss a performance score, but scores are missing more often in high-discretion units. To account for other performance-related factors, I also control for employees’ time on paid or unpaid leave¹² and for employees' number of previous promotions within B2G.

Labor market characteristics

Additional analyses indicated pay increases are smaller when unemployment increases. To account for cyclical patterns of pay growth, I control for the annual, state-level unemployment rate (Bureau of Labor Statistics). I use state instead of county-level unemployment rates because B2G often recruits outside the boundaries of the county, but rarely outside of the state. I also control for the current year using year fixed-effects, for whether the current year is a recession year (i.e., 2007–2009, based on National Bureau of Economic Research), and whether employees were originally hired before, during, or after the recession.

Analysis Method

This article aims to determine how gender pay gaps develop posthire. Hence, I follow employees after they have been hired and model gender differences in pay trajectories. For that purpose, I use multilevel growth curve models, which assess whether slopes vary systematically with employees’ gender, while also taking into account that repeated observations are nested within employees (Raudenbush & Bryk, 2002; Singer & Willet, 2003).

The first level of these models estimates the effect of within-individual growth over time. Here, pay of person i at time t is a function of pay at hire ( $π_{0}$ ), plus earnings growth over time ( $π_{1}$ and $π_{2}$ ), plus other time-varying factors $(π_{k})$ , plus the time-varying person-error $(ε_{it})$ . I include a squared tenure term ( $π_{2}$ ) to test whether earnings and disparities in earnings grow at a nonlinear (Hypothesis 1) or linear rate (Hypothesis 2).

pa y i t = π_{0} + π_{1} {(tenure)}_{it} + π_{2} {(tenur e^{2})}_{it} + π_{k} X_{kit} + ε_{it}

On the second level, the model considers time-constant variables, such as gender. Level one intercepts and slopes now become a function of time-constant person characteristics. For example, the average effect of firm tenure ( $π_{1}$ ) becomes a function of men’s average pay growth $(β_{10})$ , plus the gender difference in pay growth $(β_{11})$ . By modeling growth as a function of gender, I essentially add cross-level interactions between gender and firm tenure. These allow me to examine whether individual pay trajectories differ by gender.

\begin{array}{l} π_{0} = β_{00} + β_{01} {(Woman)}_{i} + β_{0 k} X_{ki} + μ_{0 i} \\ π_{1} = β_{10} + β_{11} {(Woman)}_{i} \\ \dots \end{array}

I also employ event history models, which test for gender-specific timing of turnover and promotions. For that purpose, I use cox event history models. All tables report hazard rates, which express how the predictors modify the (unobserved) instantaneous baseline hazard rate.

Findings

Raw Gender Pay Differences

Figure 3 shows how raw gender wage gaps develop as employees accumulate tenure. Average gender pay differences vary dramatically by professional status, meaning professional men earn on average 14% more than professional women. This equates to a raw pay gap of approximately $10,000 at hire. After hire, pay disparities initially narrow slightly among professionals before they widen again after Year 6. Gender pay differences point in the opposite direction among nonprofessionals, meaning newly hired men in nonprofessional jobs earn on average 8% less than nonprofessional women. These differences stay relatively constant over time with exception of Year 6. Together, descriptive analyses suggest raw gender differences exist among all occupational groups and that gender pay differences are more likely to be in favor of men in professional jobs, whereas differences are in favor of women in nonprofessional jobs.

Figure 3.

Raw gender pay gap by firm tenure and professional status. Gender pay gap (men-women) relative to men's pay.

Workforce Composition at Hire

Table 2 shows employees’ individual and job characteristics at hire. The distribution of characteristics varies by unit and job type, but overall women are overrepresented among Black employees, whereas men are overrepresented among Asian employees. Women are more likely to hold a master’s degree, while men are more likely to hold either a PhD or less than a bachelor’s degree. When hired, women have less potential labor market experience than men. Men are hired into jobs that are associated with a higher pay grade. Moreover, men enter into gender-balanced jobs, whereas women’s jobs are heavily female-dominated. Thus, men and women differ considerably at the point of hire, and these differences do account for pay gaps at hire, as shown in the multivariate analyses later.

Table 2.

Workforce Characteristics at Hire, by Gender, Supervisory Discretion, and Professional Status.

	Nonprofessionals						Professionals
	Low discretion			High discretion			Low discretion			High discretion
	Women	Men	Δ	Women	Men	Δ	Women	Men	Δ	Women	Men	Δ
Race
White	45%	38%	*	43%	48%		64%	64%		56%	64%	*
Black	42%	50%	**	36%	19%	*	26%	20%	*	30%	19%	***
Asian	8%	5%	*	16%	25%	*	6%	9%	*	8%	13%	**
Other race	5%	6%		5%	7%	**	4%	7%	**	6%	4%
Education and experience
Less than bachelor	34%	63%	***	24%	11%	***	6%	12%	***	7%	11%	*
Bachelor	42%	26%	***	48%	54%		44%	45%		41%	38%
Master	21%	8%	***	25%	27%	**	46%	37%	**	47%	45%
PhD	3%	3%		4%	8%		4%	6%		5%	7%
Work experience	13.3	18.0	***	11.5	8.9	*	13.5	14.6	*	13.6	14.5
Unit and job characteristics
Pay grade	12.3	10.6	***	13.3	13.7	***	18.2	20.1	***	18.6	19.7	***
Average raise in job	1.52	1.56		1.41	1.40		1.47	1.47		1.7	1.8
% Women in job	72%	40%	***	78%	69%	***	72%	51%	***	73%	59%	***
N (employees)	686	590		1,288	416		823	521		965	342

Note. *p < .05. **p < .01. ***p < .001 (two-tailed t test between men and women in same job and unit type).

Changes in Performance and Workforce Composition After Hire

To examine posthire trends, I first focus on performance evaluations because annual merit decisions are heavily based on employees’ performance that year. Figure 4 shows average performance scores by firm tenure, supervisory discretion, and professional status. In their first year, employees receive a score of approximately 3.6. After hire, average scores increase in all units, although scores are noticeably lower among nonprofessionals in low-discretion units.

Figure 4.

Average performance scores by tenure, professional status, and discretion. Diamond-shaped line markers indicate significant gender performance differences within the respective unit-profession group (p < .05, two-tailed test).

Raw gender differences in performance are nonsignificant in most groups. The exception is nonprofessionals in low-discretion units, where women receive significantly higher performance evaluations than men throughout the first 5 years (indicated by the diamond-shaped line markers). These gaps vanish, however, when controlling for individual and job characteristics (results available upon request).

In the next step, I examine whether gender affects how quickly employees receive a promotion. Promotions are associated with significant pay increases (Bidwell & Mollick, 2015; Waldman, 1984) and therefore affect average pay trajectories. Table 3 shows the results from cox event history analyses, which I run separately for each Discretion × Professional Status cell.

Table 3.

Cox Event History Analyses: Timing of Promotion.

	Nonprofessionals				Professionals
	Low discretion		High discretion		Low discretion		High discretion
Gender and performance scores
Woman	1.398		1.425		0.690		0.754
Perform: 4	3.089	***	1.865		2.047	***	1.831	*
Perform: 5	5.360	***	1.256		2.490	***	1.513
Perform: missing	2.590	*	0.966		1.421		0.753
Woman × Perform: 4	0.519		0.844		0.986		0.978
Woman × Perform: 5	0.554		1.970		1.490		1.653
Woman × Perform: Missing	0.679		1.366		1.572		1.482
Controls
Demographics and human capital
Black	0.395	***	0.712	*	0.928		0.694	*
Asian	0.668		0.498	***	0.679		1.099
Other race	0.768		0.637		0.846		0.862
Bachelor	1.313		1.389		0.896		1.150
Master	2.128		1.229		1.040		1.303
PhD	1.364		1.327		1.070		1.403
Work experience	1.067		0.985		0.970		0.991
Work experience2	0.998		1.001		0.999		0.999
Categorical age	Yes		Yes		Yes		Yes
Job and unit characteristics
Unit size	29.06	**	1.475		2.540		1.095
Job-cell: average raise	0.777		0.832		0.992		1.078
Job-cell: % female	0.067		9.391		2.904		1.220
Job-cell	Yes		Yes		Yes		Yes
Current rank	Yes		Yes		Yes		Yes
Rank at hire	Yes		Yes		Yes		Yes
Performance
Days on leave	0.981		0.969		1.031		1.026
Previous promotions	Yes		Yes		Yes		Yes
Labor market characteristics	Yes		Yes		Yes		Yes
N (employees)	1253		1,617		1332		1280
N (failures)	250		355		402		309
Log likelihood	–1,321		–2,098		–2257		–1755

Note. Errors clustered within employees. Event = employee receives a promotion.

p < .05. **p < .01. ***p < .001.

The nonsignificant effect of gender across all four models indicates women and men are promoted at similar speed, regardless of their professional status or unit type. Consistent with organizational policy, promotions depend heavily on performance scores, which is especially true in low-discretion units. The nonsignificant interactions between gender and performance scores show high performance scores are equally beneficial for men and women. These findings run counter to previous studies in which women had to outperform men to receive a promotion (e.g., Joshi, Son, & Roh, 2015; Lyness & Heilman, 2006).

In Table 4, I examine whether there is selective attrition among men and women at B2G. On average, employees stay with B2G for 3 years (firm tenure is relatively short, because the analyses include only individuals hired after 2005). Again, nonsignificant gender coefficients indicate men and women leave at the same rate. Employees with higher performance evaluations are less likely to leave B2G, but differences are significant only among professional employees. Employees with missing evaluations are significantly more likely to leave across all groups. This may be because employees with missing scores are often new employees, who are also more likely to leave. Most important, performance evaluations have the same effect for men and women, as indicated by the nonsignificant interaction between gender and performance scores. The exceptions are nonprofessionals in high-discretion units, where B2G retains more high-performing women than high-performing men. To control for turnover patterns, I save the predicted hazards from Table 4 and use them as control variable in the earnings analyses later.

Multivariate Results: Changes in Gender Gaps After Hire

In the next step, I examine how gender pay gaps develop posthire, controlling for individual, job, and labor market characteristics. Table 5 shows the multivariate growth curve models in which I estimate intercepts (i.e., pay at hire) and average within-employee earnings trajectories. Earnings are inflation-adjusted, meaning positive growth indicates an increase in real earnings. To assess whether average wage trajectories vary by gender, I model cross-level interactions between gender and firm tenure. To distinguish between information- and status-based explanations of inequality, the interaction “Tenure2 × Woman” indicates whether gender pay differences slow down (Hypothesis 1) or grow at a steady rate over time (Hypothesis 2).

Table 4.

Cox Event History Analyses: Timing of Exit From B2G.

	Nonprofessionals				Professionals
	Low discretion		High discretion		Low discretion		High discretion
Gender and performance scores
Woman	0.924		1.055		0.718		0.967
Perform: 4	0.804		0.885		0.636	*	0.460	**
Perform: 5	0.833		0.822		0.487	*	0.639
Perform: missing	3.523	***	2.459	***	1.751	*	1.408
Woman × Perform: 4	1.066		0.637	*	1.361		1.300
Woman × Perform: 5	1.176		0.820		1.470		0.925
Woman × Perform: Missing	0.683		0.643	*	1.227		1.337
Controls
Demographics and human capital
Black	0.960		0.869		1.198		0.966
Asian	1.017		1.046		1.221		0.913
Other race	1.113		1.224		1.514		0.982
Bachelor	0.720		0.867		0.969		1.315
Master	0.669		0.736		1.162		1.286
PhD	0.782		0.688		1.049		1.361
Work experience	0.861	***	0.897	***	0.910	*	0.956
Work experience2	1.002	*	1.001		1.002	*	1.001
Categorical age	Yes		Yes		Yes		Yes
Job and unit characteristics
Unit size	1.023		1.611		1.437		1.058
Job-cell: average raise	1.049		0.873		0.899		0.826
Job-cell: % female	1.980		9.160		1.961		0.937
Job-cell	Yes		Yes		Yes		Yes
Current rank	Yes		Yes		Yes		Yes
Rank at hire	Yes		Yes		Yes		Yes
Performance
Days on leave	1.135	***	1.067	**	1.081	**	1.075	**
No. of previous promotions	Yes		Yes		Yes		Yes
Labor market characteristics	Yes		Yes		Yes		Yes
N (employees)	1,244		1,613		1,342		1,273
N (failures)	542		858		552		528
Log likelihood	–3,142		–5,324		–3,262		–3,120

Note. Errors clustered within employees. Event = employee exits the organization via involuntary termination or voluntary quit.

p < .05. **p < .01. ***p < .001.

Table 5.

Growth Curve Models: Nonlinear Growth of Gender Pay Gaps.

	Nonprofessionals				Professionals
	Low discretion		High discretion		Low discretion		High discretion
Woman	–0.0053		0.0070		0.0034		–0.0059
Tenure	0.0072	***	0.0019		0.0030		0.0077	**
Tenure × Woman	0.0002		0.0000		–0.0012		–0.0022
Tenure2	0.0005	***	0.0003		0.0011	***	0.0012	***
Tenure2 × Woman	0.0003	*	0.0004		0.0001		–0.0001
Demographics and human capital
Black	–0.020	***	–0.009		0.000		–0.013
Asian	–0.038	***	–0.036	***	–0.013		–0.014
Other race	0.007		–0.028	**	–0.003		–0.003
Bachelor	0.009		0.011		0.032	**	0.006
Master	0.031	***	0.037	***	0.057	***	0.017
PhD	0.074	***	0.040	***	0.080	***	0.029
Work experience	0.007	***	0.008	***	0.008	***	0.010	***
Work experience2	0.000	***	0.000	***	0.000	*	0.000
Exit hazards	–0.007		0.005		0.001		–0.037	**
Categorical age	Yes		Yes		Yes		Yes
Unit and job characteristics
Unit size	0.000		–0.002		0.002		–0.002
Supervisor: male	0.003	*	0.001		0.001		0.000
Supervisor: unknown	0.000		–0.004	*	0.001		–0.008	*
Job-cell: average raise	0.003		0.007	*	0.004	*	–0.004	**
Job-cell: % female	–0.010		–0.025		–0.008		–0.018
Job-cell	Yes		Yes		Yes		Yes
Rank (current and at hire)	Yes		Yes		Yes		Yes
Performance
Perform: 4	0.005		0.011	*	0.008	*	–0.009
Perform: 5	0.017	***	0.017	***	0.016	*	0.006
Perform: missing	–0.001		–0.008		–0.004		0.017	*
Days on leave	0.000		–0.001		–0.001		0.000
No. of previous promotions	Yes		Yes		Yes		Yes
Labor market characteristics	Yes		Yes		Yes		Yes
N (employee-years)	5,189		6,161		5,789		5,245
N (employees)	1,341		1,729		1,411		1,350
Bayesian Information Criterion (BIC)	–17,892		–19,548		–16,265		–12,251

Note. Errors clustered within employees. Dependent variable: log of annual pay rate from salary and wages.

p < .05. **p < .01. ***p < .001.

The nonsignificant “woman” coefficients in Table 5 show men and women receive the same initial pay at hire, once I account for individual, job, and labor market characteristics. Pay equity at hire suggests B2G’s policies aimed at minimizing differences at entry are effective. Beyond the point of hire, a positive effect of tenure and tenure-squared implies that real earnings grow at an accelerating rate.

Dependent on organizational context, posthire earnings growth is, however, unevenly distributed among employees. To visualize relative differences in earnings growth, Figure 5 displays predicted gender earnings differences based on estimates in Table 5. When the gap is greater than zero, men earn more than women. In contrast, a gap below zero indicates men earn less than women, controlling for human capital, job, performance, and labor market characteristics.

Figure 5.

Predicted relative gender pay gap, all employees. Pay predicted based on estimates in Table 5. Relative gap above 0% means men earn more than women. Relative gap below 0% means women earn more than men. Diamond-shaped line markers indicate years in which gender pay gaps are significantly different from 0 after controlling for individual, job, and labor market characteristics.

Figure 5 reveals that gender pay differences are greater in high-discretion units (dashed lines) than in low-discretion units (solid lines), regardless of occupational status. Among professionals (gray lines), gender pay differences increase dramatically over time when supervisors have discretion. These disparities become statistically significant 5 years after hire, as indicated by the diamond-shaped markers. In contrast, gender pay differences never become significant among professionals in low-discretion units. While gender differences reverse among nonprofessionals (black lines) such that women earn more than men, gender gaps are still greater in high-discretion units than in low-discretion units. Greater gaps in high-discretion units are consistent with Hypothesis 3, and literature showing supervisory discretion allows for more opportunity for biased pay decisions (e.g., Baron & Pfeffer, 1994; Bielby, 2000; Castilla, 2015; Elvira & Graham, 2002; Petersen & Saporta, 2004; Reskin, 2000).

Results also indicate women experience greater pay disadvantages in professional jobs than in nonprofessional jobs. This finding is consistent with earlier research, which suggested women experience greater disadvantages in white-collar jobs where productivity is more difficult to assess (e.g., Alkadry & Tower, 2006; Milgrom et al., 2001; Petersen & Morgan 1995; Petersen et al., 1997). Thus, in professional jobs, gender pay differences favor men. In contrast, pay gaps among nonprofessionals develop in the opposite direction and favor women. This finding suggests decision makers are more likely to favor men when productivity is harder to assess.

It is notable, however, that gender gaps among nonprofessionals do not merely close, but actually reverse, such that gaps favor women, discriminating against equally performing nonprofessional men. In this regard, Abraham (2017) found women supervisors in lower hierarchical ranks were more likely to be “agents of change” by using their discretion to boost female employees’ earnings. If greater support for women becomes the default among of nonprofessional supervisors at B2G, then men would fall behind in pay as they accumulate tenure in these jobs. Future research should investigate under what conditions nonprofessional men experience pay disadvantages compared to women in the same position.

Finally, Figure 5 provides little support for the presence of statistical biases. The information-based approach predicts the rate at which gender gaps grow should flatten with increasing tenure because gender is no longer a necessary proxy of performance (Hypothesis 1). This flattening of gender pay gaps should be particularly pronounced among nonprofessionals, where performance is easier to evaluate (Hypothesis 4). Counter to this hypothesis, gender differences grow at an accelerating rate among nonprofessionals and at a constant rate among professionals, which is consistent with status-based theory (Hypothesis 2).

In further support of Hypothesis 2, Table 6 shows the same analysis, but using only a linear time–gender interaction. Leaving out the “Tenure2 × Woman” interaction essentially forces gender pay differences to change at a constant rate over time. A comparison of model fit statistics shows that the linear term in Table 6 provides noticeably better model fit than the nonlinear effects in Table 5 (except for nonprofessionals in high-discretion units). Moreover, linear trends among professionals in high-discretion units are highly significant, providing further support for Hypothesis 2.

Table 6.

Growth Curve Models: Linear Growth of Gender Pay Gaps.

	Nonprofessionals				Professionals
	Low discretion		High discretion		Low discretion		High discretion
Woman	–0.011		0.006		0.003		–0.006
Tenure	0.006	***	0.000		0.002		0.008	***
Tenure × Woman	0.003	***	0.002	***	0.000		–0.003	***
Tenure2	0.001	***	0.001	***	0.001	***	0.001	***
Tenure2 × Woman	–		–		–		–
Demographics and human capital	Yes		Yes		Yes		Yes
Unit and job characteristics	Yes		Yes		Yes		Yes
Performance	Yes		Yes		Yes		Yes
Labor market characteristics	Yes		Yes		Yes		Yes
N (employee-years)	5,189		6,161		5,789		5245
N (employees)	1,341		1,729		1,411		1,350
Bayesian Information Criterion (BIC)	–17,898		–19,533		–16,273		–12,259
Δ BIC_linear– BIC_nonlinear	–6		15		–8		–8

Note. Errors clustered within employees. Dependent variable: log of annual pay rate from salary and wages. Same control variables as Table 5.

**p < .001. All other coefficient are non-significant (p > 0.05).

Intersection Between Gender, Class, and Race

Extensive intersectional research demonstrated how the gender effect on employment outcomes depends on employees’ race (e.g., Acker, 2006; Browne & Misra, 2003; Collins, 2015; Greenman & Xie, 2008; Mandel & Semyonov, 2016). To further examine the intersection between gender and race, Figures 6(a) and (b) replicate the analyses for Figure 5 separately for Black and White employees.¹³

Figure 6

(a). Predicted relative gender pay gap, Black employees and (b) predicted relative gender pay gap, White employees. Pay predicted based on estimates that are available on request. Relative gap above 0% means men earn more than women. Relative gap below 0% means women earn more than men. Diamond-shaped line markers indicate years in which gender pay gaps are significantly different from 0 after controlling for individual, job, and labor market characteristics.

Among professional employees, Figures 6(a) and (b) suggest that steady increase of women’s disadvantage applies equally to Black and White employees. Regardless of race, professional women’s pay gradually falls behind men’s pay after the point of hire. In contrast, gendered pay trends among nonprofessionals are more sensitive to race. Specifically, men’s earnings disadvantage among nonprofessional employees only applies to Black men and women. Gender pay differences among White nonprofessionals are in slight favor of men, but nonsignificant. Thus, gender differences among nonprofessionals are driven by Black women outearning their male counterparts, particularly in high-discretion units. Despite some racial differences, overall substantive findings hold for Black and White employees: The gender effect (in favor of men or women) on pay does not decrease with tenure, women experience greater disadvantages in professional jobs, and gender pay differences emerge faster in high-discretion units.

Alternative Explanations

Gender-specific negotiation behavior

One possible explanation for the observed patterns detailed in Figure 5 may be gender-specific negotiation strategies. If professional men or nonprofessional Black women challenged their merit increases more often, they may also receive greater annual merit increases. Qualitative interviews at B2G, however, revealed employees virtually never challenged posthire pay changes, regardless of unit type. Absent negotiation over merit increases stood in stark contrast with supervisors’ frequent accounts of men’s assertive negotiation behavior at hire. When asked why employees did not challenge merit increases, supervisors indicated that the seemingly standardized nature of the merit process is taken for granted by employees, discouraging them from challenging pay changes posthire.

Occupational sex segregation

Another possible explanation for gender pay differences is occupational sex segregation. Women’s pay may grow slower or faster because women are overrepresented in occupations that experience different overall pay growth. My analyses guard against this possibility in multiple ways. First, I control for each job’s gender composition. Second, I control for each job’s average pay by including fixed-effects for occupation-unit cells as well as fixed-effects for each pay grade. Finally, to account for job-specific pay growth, I control for the average pay increases in each job-year, thereby comparing men and women in jobs with identical earnings profiles. Men and women within the same job may still perform different tasks. Task segregation might contribute to rising gender inequality (e.g., Chan & Anteby, 2015); however, lower valuation of “female” tasks should lead to lower performance ratings for women overall. Counter to this argument, multivariate regressions (available on request) found no significant gender differences in annual performance scores.

Unobserved gender differences in performance

Proponents of statistical discrimination may argue that persistent gender pay disparities indicate actual and persistent gender performance differences. For example, firms may discover posthire that professional men are more productive than professional women, for instance because professional men are able to work overtime more often than women (Goldin 2004). In this case, firms would regularly afford higher pay to men, producing the pay trajectories observed among professionals at B2G. To guard against this explanation, I control for employees’ annual performance scores in all analyses. Annual performance scores represent annual performance as perceived by supervisors and are the basis on which employees receive merit raises. Moreover, I control for employees’ personal leave to account for employee absences (e.g., to take care of sick family members).

Discussion

This article examines whether the gender effect on pay lessens when employees stay longer with their employer and whether pay trajectories depend on organizational context. To trace men’s and women’s earnings growth after the point of hire, I examine personnel records of 5,631 employees of a private U.S. employer between 2005 and 2014.

I find once employees are hired, gender pay gaps grow significantly with employees’ firm tenure. The direction and severity of pay differences, however, depend heavily on organizational context such as supervisory discretion and job type. Gender pay differences are significantly greater in units with more supervisory discretion over merit pay decisions, which confirms Hypothesis 3. Gender pay differences reach statistical significance only in high-discretion units, whereas adjusted gender pay gaps fail to reach significance in low-discretion units.¹⁴ Moreover, results show women are more disadvantaged in professional jobs than nonprofessional jobs where performance is easier to assess.

Temporal patterns of gender pay gaps provide important insights into underlying inequality-generating mechanisms. Counter to Hypothesis 1, Figure 5 does not provide evidence that gender biases decrease when more performance information becomes available over the years. Instead, differences among equally performing professional men and women grow at a constant rate once hired. In support of Hypothesis 2, models that allow only for constant growth of gender pay differences (Table 6) fit the data noticeably better than models allowing for nonlinear growth of pay differences (Table 5). Not even among nonprofessionals, where outcomes are easier to observe, do gender gaps flatten; instead, they accelerate with tenure, which contradicts Hypothesis 4.

Conclusion

The present article builds on previous organizational studies that observed performance-reward biases in annual merit increases (Belliveau, 2012; Castilla, 2008; Castilla & Benard, 2010). I extend and contribute to this literature by showing performance-reward biases do not diminish when employees stay longer with their company. This means staying longer with the same firm does not reduce employees’ vulnerability to biases. Instead, gender has a constant effect on pay during employees’ entire firm-internal career. In this regard, annual differences compound noticeably over time. Among professionals, men and women both start at an average pay of $69,000/year. When professional women fall behind in pay posthire, they lose a total of $8,709 in compound earnings after working for B2G for 8 years. In comparison, nonprofessional men and women start at $35,700 per year, and men lose a total of $2,969 in compound earnings over the span of 8 years.

This article has several implications for pay-setting policies. Growing posthire gender pay gaps demonstrate how achieving pay equity at one point does not guarantee employees will stay at equity for the remainder of their firm-internal career. Consistent with previous research on pay discretion (e.g., Castilla, 2015; Elvira & Graham, 2002; Kalev et al., 2006), results show limiting differences at hire only prevents disparities in units where supervisors also have limited discretion over subsequent merit increases. Pay-setting policies at various points of employees’ firm-internal career (e.g., at hire, at promotion, or merit raises) should therefore be viewed as a whole. Failure to do so may undo successful regulation (e.g., equity at hire) and cause seemingly equitable policies to produce and legitimate inequality at later stages (Acker, 2006; DiPrete & Eirich, 2006; Ferguson, 1984).¹⁵

Given the data support status-based theories, pay policies should focus on intervening in cognitive processes (e.g., Pugh & Wahrman, 1983) or preventing biases from affecting decisions related to pay, for example, via accountability measures (Castilla, 2015; Dobbin et al., 2015; Tetlock & Mitchell, 2009). Pay policies also need to focus on all employees, not only on newly hired employees.

As mentioned earlier, B2G strongly regulates pay at hire but provides more leeway for some supervisors to determine merit increases later on. B2G’s pay policies appear to be similar to the organization studied by Castilla (2008, 2015). In comparison, organizations studied by Petersen and Saporta (2004) and Woodhams et al. (2015) seem to pursue the opposite strategy: provide more pay discretion at hire but then regulate posthire pay setting to narrow gender pay gaps over time. The present article therefore adds to our understanding of inequality-generating mechanisms by examining wage trajectories in an organization where employees start at pay equity.

Future research should examine the relative influence of status and statistical biases in firms that give supervisors discretion over hiring and merit pay. It is possible that statistical biases still have a noticeable effect on gender pay differences in these workplaces. Observing statistical biases when supervisors have discretion over hiring pay would imply that statistical biases are mostly limited to the point of hire. Likewise, the effect of status-based pay biases may be weaker in firms that fail to regulate hiring pay. Dobbin et al. (2015) and Williams et al. (2014) demonstrate how limiting discretion in pay decisions leaves supervisors frustrated and more likely to reestablish gendered pay hierarchies. While not indicated in the background interviews with B2G supervisors, it is conceivable that heavy regulation made gender more salient, thereby heightening status-based biases, driving supervisors to restore a status quo they perceive to be more appropriate posthire (Auspurg et al., 2017). In summary, the present article highlights that status-based process are a serious issue affecting men’s and women’s pay after the point of hire. Future research should examine to what extent initial regulation of hiring pay moderates the effect of subsequent information- and status-based biases in merit pay decisions.

Footnotes

Acknowledgments

I thank Irene Browne, Christopher Rider, Richard Rubinson, Arne Kalleberg, Cathryn Johnson, Joseph Dippong, and Lisa Slattery Walker for their thoughtful comments and Amanda Sargent for her research assistance. I also thank the participants at the following meetings for their feedback: Southern Sociological Society, People and Organizations Conference, American Sociological Association, Lectures of Organizations and Human Resources, and the Frankfurt Center for Leadership and Behavior in Organizations. Finally, I thank B2G and the 19 interviewees for their participation in the study.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Emory Goizueta Business School Dean’s Research Grant (PI: Anand Swaminathan) and the Emory Laney Graduate School Professional Development Grant.

Notes

Author Biography

Anne-Kathrin Kronberg is an assistant professor of sociology and organizational science at the University of North Carolina at Charlotte. Her research focuses on organizations, work, and social inequality. Her work has appeared in Social Forces, Work and Occupations, and Mobilization and was funded by the German Research Foundation.

References

Abendroth

A.-K.

Melzer

Kalev

Tomaskovic-Devey

(2016). Women at work: Women’s access to power and the gender earnings gap. ILR Review, 70(1), 190–222.

Abraham

(2017). Pay formalization revisited: Considering the effects of manager gender and discretion on closing the wage gap. Academy of Management Journal, 60(1), 29–54.

Acker

(2006). Inequality regimes: Gender, race, and class in organizations. Gender & Society, 20, 441–464.

Aigner

D. J.

Cain

G. G.

(1977). Statistical theories of discrimination in labor markets. Industrial and Labor Relations Review, 30(2), 175–187.

Akerlof

G. A.

(1970). The market for ‘lemons’. Uncertainty and the market mechanism. The Quarterly Journal of Economics, 84, 488–500.

Alkadry

M. G.

Tower

L. E.

(2006). Unequal pay: The role of gender. Public Administration Review, 66(6), 888–898.

Altonji

J. G.

Pierret

C. R.

(2001). Employer learning and statistical discrimination. Quarterly Journal of Economics, 116(1), 313–350.

Alvesson

(2001). Knowledge work: Ambiguity, image and identity. Human Relations, 54(7), 863–886.

Arrow

K. J.

(1973). The theory of discrimination. In Ashenfelter

Rees

(Eds.), Discrimination in labor markets (pp. 1--31). Princeton, NJ: Princeton University Press.

10.

Auspurg

Hinz

Sauer

(2017). Why should women get less? Evidence on the gender pay gap from multifactorial survey experiments. American Sociological Review, 82(1), 179–210.

11.

Babcock

Laschever

(2003). Women don’t ask: Negotiation and the gender divide. Princeton, NJ: Princeton University Press.

12.

Baron

J. N.

(1984). Organizational perspectives on stratification. Annual Review of Sociology, 10(1), 37–69.

13.

Baron

J. N.

Hannan

M. T.

Hsu

Koçak

Ö.

(2007). In the company of women: Gender inequality and the logic of bureaucracy in start-up firms. Work and Occupations, 34, 35–66.

14.

Baron

J. N.

Pfeffer

(1994). The social-psychology of organizations and inequality. Social Psychology Quarterly, 57, 190–209.

15.

Belliveau

M. A.

(2012). Engendering inequity? How social accounts create vs. merely explain unfavorable pay outcomes for women. Organization Science, 23(4), 1154–1174.

16.

Berger

Cohen

B. P.

Zelditch

(1972). Status characteristics and social interaction. American Sociological Review, 37(3), 241–255.

17.

Berger

Fisek

M. H.

Norman

R. Z.

Wagner

D. G.

(1985). The formation of reward expectation in status situations. In Berger

Zelditch

(Eds.), Status, rewards, and influence (pp. 121--154). San Francisco, CA: Jossey-Bass.

18.

Berger

Norman

R. Z.

Balkwell

J. W.

Smith

R. F.

(1992). Status-inconsistency in task situations – A test of 4 status processing principles. American Sociological Review, 57, 843–855.

19.

Bianchi

S. M.

Robinson

J. P.

Milkie

M. A.

(2006). Changing rhythms of American family life. New York, NY: Russell Sage Foundation.

20.

Bidwell

(2011). Paying more to get less: The effects of external hiring versus internal mobility. Administrative Science Quarterly, 56, 369–407.

21.

Bidwell, M., Briscoe, F., Fernandez-Mateo, I., & Sterling, A. (2013). The Employment Relationship and Inequality: How and Why Changes in Employment Practices Are Reshaping Rewards in Organizations. The Academy of Management Annals, 7, 61--121.

22.

Bidwell

Mollick

(2015). Shifts and ladders: Comparing the role of internal and external mobility in executive careers. Organization Science, 26(6), 1629–1645.

23.

Bielby

W. T.

(2000). Minimizing workplace gender and racial bias. Contemporary Sociology – A Journal of Reviews, 29, 120–129.

24.

Blau

F. D.

Kahn

L. M.

(2017). The gender wage gap: Extent, trends, and explanations. Journal of Economic Literature, 55(3), 789–865.

25.

Brett

J. M.

Stroh

L. K.

(1997). Jumping ship. Who benefits from an external labor market strategy? Journal of Applied Psychology, 82, 331–341.

26.

Brewer

M. B.

(1988). A dual process model of impression formation. In Scrull

Wyer

R. S.

Haslam

S. A.

(Eds.), Advances in social cognition ( Vol. 1) (pp. 1--36). Hillsdale, NJ: Lawrence Erlbaum Associates.

27.

Brewer

M. B.

Lui

L. N.

(1989). The primacy of age and sex in the structure of person categories. Social Cognition, 7(3), 262–274.

28.

Browne

Misra

(2003). The intersection of gender and race in the labor market. Annual Review of Sociology, 29, 487–513.

29.

Castilla

E. J.

(2008). Gender, race, and meritocracy in organizational careers. American Journal of Sociology, 113, 1479–1526.

30.

Castilla

E. J.

(2015). Accounting for the gap: A firm study manipulating organizational accountability and transparency in pay decisions. Organization Science, 26(2), 311–333.

31.

Castilla

E. J.

Benard

(2010). The paradox of meritocracy in organizations. Administrative Science Quarterly, 55, 543–576.

32.

Chan

C. K.

Anteby

(2015). Task segregation as a mechanism for within-job inequality: Women and men of the transportation security administration. Administrative Science Quarterly, 61(2), 184–216.

33.

Collins

P. H.

(2015). Intersectionality’s definitional dilemmas. Annual Review of Sociology, 41, 1–20.

34.

Cooke

T. J.

(2003). Family migration and the relative earnings of husbands and wives. Annals of the Association of American Geographers, 93(3), 338–349.

35.

Correll

S. J.

Benard

(2006). Biased estimators? Comparing status and statistical theories of gender discrimination. In Thye

S. R.

(Ed.), Social psychology of the workplace ( Vol. 23, pp. 89–116). Stamford, CT: Jai Press Inc.

36.

DiPrete

T. A.

Eirich

G. M.

(2006). Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments. Annual Review of Sociology, 32(1), 271–297.

37.

Dobbin

Schrage

Kalev

(2015). Rage against the iron cage. American Sociological Review, 80(5), 1014–1044.

38.

Dokko

Wilk

S. L.

Rothbard

N. P.

(2009). Unpacking prior experience: How career history affects job performance. Organization Science, 20(1), 51–68.

39.

Dreher

G. F.

Cox

(2000). Labor market mobility and cash compensation: The moderating effects of race and gender. The Academy of Management Journal, 43, 890–900.

40.

Dreher

G. F.

Lee

J.-Y.

Clerkin

T. A.

(2011). Mobility and cash compensation: The moderating effects of gender, race, and executive search firms. Journal of Management, 37, 651–681.

41.

Dwyer

R. E.

(2004). Downward earnings mobility after voluntary employer exits. Work and Occupations, 31, 111–138.

42.

Elvira

M. M.

Graham

M. E.

(2002). Not just a formality: Pay system formalization and sex-related earnings effects. Organization Science, 13(6), 601–617.

43.

Ferguson

(1984). The feminist case against bureaucracy. Philadelphia, PA: Temple University Press.

44.

Foschi

(1989). Status characteristics, standards, and attributions. In Berger

Zelditch

Anderson

(Eds.), Sociological theories in progress: New formulations (pp. 58–72). Newbury Park, CA: Sage Publications.

45.

Foschi

(1996). Double standards in the evaluation of men and women. Social Psychology Quarterly, 59, 237–254.

46.

Foschi

(2000). Double standards for competence: Theory and research. Annual Review of Sociology, 26, 21–42.

47.

Freidson

(1988). Professional powers: A study of the institutionalization of formal knowledge. Chicago, IL: University of Chicago Press.

48.

Fuller

(2008). Job mobility and wage trajectories for men and women in the United States. American Sociological Review, 73, 158–183.

49.

Glass

(2004). Blessing or curse? Work-family policies and mother’s wage growth over time. Work and Occupations, 31, 367–394.

50.

Goldin

(2014). A grand gender convergence: Its last chapter. American Economic Review, 104(4), 109–119.

51.

Greenman

Xie

(2008). Double jeopardy? The interaction of gender and race on earnings in the United States. Social Forces, 86(3), 1217–1244.

52.

Hachen

D. S.

(1990). Three models of job mobility in labor markets. Work and Occupations, 17, 320–354.

53.

Halaby

C. N.

(1988). Action and information in the job mobility process – The search decision. American Sociological Review, 53, 9–25.

54.

Hodson

Sullivan

(2012). The social organization of work ( Vol. 2012, 5th ed.). Belmont, CA: Wadsworth Cengage Learning.

55.

Iacus

S. M.

King

Porro

(2011). Multivariate matching methods that are monotonic imbalance bounding. Journal of the American Statistical Association, 106(493), 345–361.

56.

Iacus

S. M.

King

Porro

(2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1–24.

57.

Ibarra

(1992). Homophily and differential returns – Sex-differences in network structure and access in an advertising firm. Administrative Science Quarterly, 37, 422–447.

58.

Joshi

Son

Roh

(2015). When can women close the gap? A meta-analytic test of sex differences in performance and rewards. Academy of Management Journal, 58(5), 1516–1545.

59.

Jovanovic

(1979). Job matching and the theory of turnover. Journal of Political Economy, 87, 972–990.

60.

Kalev

Dobbin

Kelly

(2006). Practices or best guesses? Assessing the efficacy of corporate affirmative action and diversity policies. American Sociological Review, 71, 589–617.

61.

Knoke

Ishio

(1998). The gender gap in company job training. Work and Occupations, 25, 141–167.

62.

Kronberg

A.-K.

(2013). Stay or leave? Externalization of job mobility and the effect on the US gender earnings gap, 1979-2009. Social Forces, 91, 1117–1146.

63.

Lam

Dreher

(2004). Gender, extra-firm mobility, and compensation attainment in the United States and Hong Kong. Journal of Organizational Behavior, 25, 791–805.

64.

Lyness

K. S.

Heilman

M. E.

(2006). When fit is fundamental: Performance evaluations and promotions of upper-level female and male managers. Journal of Applied Psychology, 91(4), 777–785.

65.

Mandel

Semyonov

(2016). Going back in time? Gender differences in trends and sources of the racial pay gap, 1970 to 2010. American Sociological Review, 81(5), 1039–1068.

66.

McDonald

Lin

(2009). Networks of opportunity: Gender, race, and job leads. Social Problems, 56, 385–402.

67.

Milgrom

E. M. M.

Petersen

Snartland

(2001). Equal pay for equal work? Evidence from Sweden and a comparison with Norway and the U.S. The Scandinavian Journal of Economics, 103(4), 559–583.

68.

Oettinger

G. S.

(1996). Statistical discrimination and the early career evolution of the black-white wage gap. Journal of Labor Economics, 14, 52–78.

69.

Pearlman

(2018). Gender differences in the impact of job mobility on earnings: The role of occupational segregation. Social Science Research, 74, 30–44.

70.

Petersen

Morgan

L. A.

(1995). Separate and unequal: Occupation-establishment sex segregation and the gender wage gap. American Journal of Sociology, 101, 329–365.

71.

Petersen

Saporta

(2004). The opportunity structure for discrimination. American Journal of Sociology, 109, 852–901.

72.

Petersen

Snartland

Becken

L.-E.

Olsen

K. M.

(1997). Within-job wage discrimination and the gender wage gap: The case of Norway. European Sociological Review, 13(2), 199–213.

73.

Phelps

E. S.

(1972). The statistical theory of racism and sexism. American Economic Review, 62, 659–661.

74.

Pugh

M. D.

Wahrman

(1983). Neutralizing sexism in mixed-sex groups – Do women have to be better than men? American Journal of Sociology, 88, 746–762.

75.

Quintana-García

Elvira

M. M.

(2016). The effect of the external labor market on the gender pay gap among executives. ILR Review, 70(1), 132–159.

76.

Rashotte

L. S.

Webster

Jr. (2005). Gender status beliefs. Social Science Research, 34(3), 618–633.

77.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models. Applications and data analysis models (2nd ed.). Thousand Oaks, CA: Sage Publications.

78.

Reskin

B. F.

(2000). The proximate causes of employment discrimination. Contemporary Sociology- A Journal of Reviews, 29, 319–328.

79.

Reskin

B. F.

McBrier

D. B.

(2000). Why not ascription? Organizations’ employment of male and female managers. American Sociological Review, 65, 210–233.

80.

Rider

C. I.

Negro

(2015). Organizational failure and intraprofessional status loss. Organization Science, 26(3), 633–649.

81.

Ridgeway

Correll

(2004). Unpacking the gender system – A theoretical perspective on gender beliefs and social relations. Gender & Society, 18, 510–531.

82.

Ridgeway

C. L.

(1997). Interaction and the conservation of gender inequality: Considering employment. American Sociological Review, 62, 218–235.

83.

Ridgeway

C. L.

(2011). Framed by gender: How gender inequality persists in the modern world. New York, NY: Oxford University Press.

84.

Salancik

G. R.

Pfeffer

(1978). Uncertainty, secrecy, and choice of similar others. Social Psychology, 41, 246–255.

85.

Singer

J. D.

Willet

J. B.

(2003). Applied longitudinal data. Modeling change and event occurrence. Toronto, Canada: Oxford University Press.

86.

Stainback

Tomaskovic-Devey

Skaggs

(2010). Organizational approaches to inequality: Inertia, relative power, and environments. Annual Review of Sociology, 36, 225–247.

87.

Steinpreis

R. E.

Anders

K. A.

Ritzke

(1999). The impact of gender on the review of the curricula vitae of job applicants and tenure candidates: A national empirical study. Sex Roles, 41, 509–528.

88.

Sterling

A. D.

Fernandez

R. M.

(2018). Once in the door: Gender, tryouts, and the initial salaries of managers. Management Science, 64(6), 5444--5460.

89.

Tetlock

P. E.

Mitchell

(2009). Implicit bias and accountability systems: What must organizations do to prevent discrimination? Research in Organizational Behavior, 29, 3--38.

90.

Tomaskovic-Devey

Avent-Holt

(2019). Relational inequalities: An organizational approach. New York, NY: Oxford University Press.

91.

Topel

R. H.

Ward

M. P.

(1992). Job mobility and careers of young men. Quarterly Journal of Economics, 107, 439–480.

92.

Waldman

(1984). Job assignments, signaling, and efficiency. The RAND Journal of Economics, 15(2), 255–267.

93.

Williams

C. L.

Kilanski

Muller

(2014). Corporate diversity programs and gender inequality in the oil and gas industry. Work and Occupations, 41(4), 440–476.

94.

Woodhams

Lupton

Perkins

Cowling

(2015). Multiple disadvantage and wage growth: The effect of merit pay on pay gaps. Human Resource Management, 54(2), 283–301.