Corporate Demography and Wage Inequality: Revisited

Abstract

To further research how organizations influence workforce wage inequality, the authors replicate and extend Sorensen and Sorenson’s study on organizational demography and wage inequality in region-industries by (1) replicating original results from Danish regions from 1992 to 1998 using a close-to-identical dataset in Sweden during the same time period, (2) using multiverse analysis to gauge the effect of analytical choices on research results, and (3) expanding the scope of Sorensen and Sorenson’s model by two new measures of organizational diversity. The findings suggest strong to fair test-retest validity of the original model, but model extensions with nuanced measures of organization form diversity do not enhance the model’s explanatory power. The authors analyze and discuss replication anomalies and show how multiverse analysis can be gainfully used more generally in comparative organizational sociology.

Wage inequality has long been a central topic in the social sciences, with increasing attention in recent decades to rising inequality in countries across the world. Economists and sociologists alike have probed questions related to markets and institutions that affect how wage inequalities among individual workers are produced, sustained, or reduced (DiPrete and Fox-Williams 2021; Gamoran 2021; Lamont 2018; Moya, Adriaans, and Sauer 2023; van de Werfhorst 2022). Less research has focused on the direct role of employing organizations in shaping such inequalities (Dencker and Fang 2016). Organizations are important sources for how wages are distributed among individuals (Hedström 1991), and organizations stratify employees and their career prospects (see, e.g., Baron and Bielby 1980; Castilla 2011). However, we know significantly less about how the distribution of organizations in a society—the demography of organizations—shapes aggregate patterns of wage inequality in a society (Hannan 1988a, 1988b).

To this date, we know of but two large-scale empirical studies directly testing predictions derived from theories of corporate demography on workforce wage inequality. In a seminal study, Sorensen and Sorenson (2007) (S&S) investigated the relationship between organizational diversity and wage inequality among Danish townships and industries. In a more recent study, Cobb and Stevens (2017) examined the role of firm-size differentials between organizations as an explanation of workforce inequality in U.S. states. These two studies represent initial attempts to understand the relationship between the diversity of organizations and wage inequality, but empirical findings to date are confined to the specific tests of data in these studies. As these two studies used radically different designs and scales of analysis in distinct settings, their results are difficult to compare. ¹ Without systematic work that scrutinizes earlier studies and systematizes empirical patterns, we will not understand well how organizational diversity shapes workforce wage inequality, distinguish generic mechanisms from contextual effects, or determine how to best specify theoretical models for empirical research (Walder and Nguyen 2008). Answering questions about organizational diversity and wage inequality is imperative for the development of organizational theories that seek to answer questions about overall societal outcomes (Davis 2013; George et al. 2016). Furthermore, without a solid understanding of how organizations may contribute to population-level outcomes such as wage inequality, sociological research will be less applicable for policymakers who still rely primarily on research in the domains of economics when grappling with topics such as workforce inequality.

In this study, we replicate and extend S&S’s study of organizational diversity and wage inequality. We examine workforce wage inequality among industries in Swedish townships using an identical design and close to an identical dataset as used by S&S. We replicate S&S’s analysis for the same time period (1992–1998) concerning two types of organizational diversity: vertical differentiation (hypothesized to increased wage inequality) and horizontal differentiation (hypothesized to reduce wage inequality).

Our results are consistent with the findings of S&S in terms of vertical differentiation, as we find support that vertical differentiation increases income inequality. Our results are weaker and only partly consistent with the findings of S&S when it comes to horizontal differentiation, as we find support that horizontal differentiation reduces income inequality for some measures but not for others. Overall, our results largely support the validity of S&S’s models as a macro-organizational framework for comparing wage inequality across industries and townships.

We perform post hoc tests to examine the validity and sensitivity of our results using a computational multiverse approach. This allows a systematic examination of whether analytical decisions played a role in our replication’s findings. These analyses further validate the differences observed between our and S&S’s results when it comes to horizontal differentiation and income inequality. The multiverse analysis also provides some suggestions regarding factors that can explain the differences between our and S&S’s results, namely, factors of what defines the workforce population in townships.

We extend research on organizational demography and income inequality and S&S’s study by introducing more nuanced proxies for organizational form diversity suggested to affect wage inequality: the role of the globalization driving foreign ownership of firms and the differential hiring and enumeration practices in for-profit versus nonprofit firms (Casson 1987; Hwang and Powell 2009). We find differing effects depending on the exact measure of wage inequality and specification used, leading us to conclude that more nuanced measures of the diversity of organizational forms do not necessarily enhance the explanatory power of S&S’s basic model of corporate demography and workforce wage inequality.

Our study contributes to the expanding trend of replicability of theoretical models in sociology (Herring 2017; Liu and Salganik 2019; Walder and Nguyen 2008; Young 2018). We demonstrate how a multiverse framework can assess the robustness of results and systematically analyze the influence of specific analytical decisions. The results lend support to the generality of S&S’s model of corporate demography as an explanatory framework regarding the role of organizations’ vertical differentiation driving workforce wage inequalities. Replication results are, however, partly inconsistent when it comes to organizations’ horizontal differentiation dampening workforce wage inequality, an aspect of organizational diversity that should be a candidate for further scrutiny in future research. The multiverse analyses shed light on these inconsistencies by revealing that both the direction and strength of the effect of organizational differentiation vary depending on analytical decisions related to how the workforce sample population is defined. This highlights the value of using computational multiverse analyses to assess the sensitivity of specific analytical choices, especially when performing replication studies.

Theory and Hypotheses

Advancing arguments about corporate demography and workforce inequality, Hannan (1988a, 1988b) focused on the diversity in characteristics of organizations to sketch a theory of how such diversity shapes workers’ career trajectories and related distributions of rewards. The theory is built upon the assumptions that although employers are dependent on the availability of a competent workforce, employees are dependent on employers that can nourish and provide a rewarding career trajectory for the employees. As a corollary, organizations specialized in specific industries might be highly dependent on the set of skills and talents held by individuals residing in their geographic proximity. Furthermore and equally important, individuals’ career prospects depend on the type of organizations located in their geographic proximity. A central, and for the theory unique, prediction was that the level of diversity among organizational forms in a social system could serve to mute the impact of differences in workers’ initial advantage, whereas low diversity serves to magnify such differences.

A renewed focus on how corporate demography affects important societal outcomes such as the level of economic inequality is warranted for several reasons. First, predominant micro-oriented explanations in organization theory seeking to explain workforce inequality within firms are problematic as explanatory frameworks for aggregate workforce inequality because the association between employer characteristics and inequality within firms does not necessarily explain how variance in employer characteristics affects overall workforce inequality. Scholars have found weak and sometimes even negative correlation between how employer characteristics affect inequality within firms and how the same characteristics affect total workforce inequality, constituting what Davis and Cobb (2010) called “the paradox of hierarchy” (p. 36). Second, there is a paucity of theoretical and empirical work attending to the societal outcomes of variability in corporate demography (Davis 2013; George et al. 2016). Specifically, the few extant studies on the role of regional variability in organizational form and workforce wage inequality exhibit somewhat inconclusive evidence and employee divergent model specifications at different levels of analysis (e.g., Cobb and Stevens 2017; Sorensen and Sorenson 2007).

In what follows we first describe the micro-mechanism of employee sorting, from which we develop our hypotheses on the vertical and horizontal differentiation of organizational diversity.

The Micro-mechanisms of Employee-Employer Sorting for Regional Wage Inequality ²

Although few studies have directly sought to test and develop the propositions about corporate demography and workforce inequality offered by Hannan (1988a, 1988b), a large literature in sociology and economics studies the matching processes between workers and employers that leads to the sorting of individuals into particular jobs at particular firms (e.g., Jovanovic 1979; Mouw and Kalleberg 2010). Scholars focusing on individual characteristics such as human capital find that such sorting may lead to strong cumulative advantages or disadvantages over individuals’ life courses (Frank and Cook 2010; Merton and Barber 2011). ³ Matching theory is based on the tenet that a “match” between a worker and an employing organization shapes the wage profile of the worker: good matches yield a high wage profile that increases over time, whereas “mismatches” lead to lower wage profiles unless the worker can change jobs (Jovanovic 1979; Petrongolo and Pissarides 2001). The mechanisms posited by matching theory are twofold. On one hand, employers seek to recruit candidates who can carry out the required work for payment. On the other hand, workers seek employment positions to participate in the economy and develop their careers. Yet a particular worker’s productivity is unknown ex ante to employers. Employers and workers may thus need to go through phases of trial and error to achieve good matches. The likelihood for good matches increases with the regional supply (and variety) of employers in labor market areas, as “workers could easily move to a competitor and improve the quality of their match” (Glaeser 1994:19). ⁴

Matching theory thus proposes that the enhanced sorting of workers to employers in more organizationally dense labor markets would lead to high-ability workers’ amassing at more specialized and highly productive firms. From a microeconomic matching perspective, the connection between workforce inequality and the number and types of organizations participating in a local labor market would thus depend solely on the sorting of more productive employees to more productive firms, with workforce inequality being a function of between-firm differences in productivity. Although recent studies have provided some support for such a relationship by showing that inequality in firms’ productivity has increased in recent decades with effects on overall workforce inequality (e.g., Card et al. 2018; Faggio, Salvanes, and Van Reenen 2010), the productivity-inequality relationship in economics is incomplete as a theory for how organizations shape workforce wage inequality. First, highly productive firms tend to pay all of their employees higher wages, not just employees with notable high productivity (e.g., Böhm, Metzger, and Strömberg 2015; Frank 1985). Second, the proliferation of international, family-owned, and other forms of organizations has documented importance for the distribution of wages in regions and industries in which such organizations are prevalent (e.g., Carrasco-Hernandez and Sánchez-Marín 2007; Ferner, Almond, and Colling 2005; Leete 2000). Although a host of studies in organization theory have challenged microeconomic explanations of inequality by developing theories for wage inequalities within firms, to our knowledge the only macro-oriented theory of organizations that has sought to explain how the structure of organizations in society can lead to systematic differences in the allocation of rewards is the theory of corporate demography.

Corporate Demography and Wage Inequality: Vertical and Horizontal Differentiation

Corporate demography, originating largely from organizational ecology, focuses on how industries evolve and transform over time by analyzing the diversity of organizational forms and how these organizations grow, decline, transform, and become extinct (Carroll and Hannan 2000b). An important implication derived from the theory of corporate demography is that organizational diversity might act as a central force shaping workforce inequality, as such organizational diversity also shapes matching processes between organizations and workers.

An important and arguably the first study to examine the long-standing predictions from corporate demography was S&S’s study of how horizontal and vertical differentiation in firm-size distribution across industries in Danish townships explains a sizable portion of workforce wage inequality. S&S used two central dimensions to theorize organizational diversity: vertical and horizontal diversity.

Vertical differentiation refers to the varying capabilities of organizations to leverage and use employee skills in industries and townships, contributing to increased wage inequality when some organizations invest more in their employees than others. The more firms there are in an industry-township, the more likely high-quality workers can match with firms that can maximize the utility from their labor. Following matching theory, this should lead these employees to receive higher wages (e.g., Jovanovic 1979; Mouw and Kalleberg 2010). In industry-townships with fewer firms, opportunities for workers to find an employers at which their specialization can be fully used are lower, decreasing perceived return of their labor and, in turn, their wages. Hence, an increased number of firms amplifies the effect of human capital–specific characteristics on wage as it enhances quality sorting.

Horizonal differentiation allows for increased quality sorting through the presence of better career alternatives in a particular township. With more types of employers, workers can match with employers close to their skillsets. In townships with less industry diversity, a worker might have to choose to work in an industry in which their full potential will not be realized. Thus, a highly horizontally differentiated region includes organizations of many types or forms, whereas a region that is less horizontally differentiated includes fewer types of forms of organizations (in the extreme, a region would include only one type or form of organization). In regions with high horizontal differentiation among employers, workers’ unique skillsets should be easier to match with employers that value their unique sets of skills, which should decrease wage inequality.

In models that did not account for workers’ human capital characteristics, S&S found that horizontal differentiation in terms of the industry firm-size distribution and the number of industries in a township was not directly related to wage dispersion among workers but instead moderated the inequality-increasing effect of the aggregate number of firms in the industry. When controlling for workers’ human capital characteristics, however, S&S found that horizontal differentiation among firms exhibited both direct negative effects on wage dispersion among workers in the industry and negatively moderated the effects of the aggregate number of firms in the industry. This suggests that failure to control for the sorting of employees with different human capital to particular firms, a key tenet of matching theory, could render studies of how variation in corporate demography predicts various workforce outcomes inaccurate.

Like S&S, our first set of hypotheses relates to two specific types of organizational diversity in labor markets. S&S’s analyses of residual wage dispersion, having taken employees’ individual characteristics into account, provided ample support that (in particular vertical) differentiation increased workforce inequality, whereas horizontal differentiation decreased inequality (Sorensen and Sorenson 2007, Table 3). Yet S&S’s analyses of gross wage dispersion, perhaps the most parsimonious and easily observable difference in wage inequality, provided no evidence of an inequality-decreasing effect of horizontal differentiation (Sorensen and Sorenson 2007, Table 2). In their discussion of potential explanations, S&S suggested that the greater relative importance of vertical differentiation might reflect that purely size-based measures of organizational diversity do not appropriately capture horizontal differentiation among organizations. To probe the generality of their results and thus the validity of corporate demography as an explanation of workforce wage inequality, we first test two hypothesis related to vertical differentiation, identical to S&S’s, directly derived from the arguments of vertical differentiation as outlined earlier:

• Hypothesis 1a: Wage dispersion increases with the number of firms in the industry-region.

• Hypothesis 1b: Wage dispersion is higher in industry-regions with more than a single employer.

Second, we derive two hypotheses related to horizontal differentiation from the theoretical arguments outlined earlier, identical to S&S:

• Hypothesis 2a: Wage dispersion decreases with the variety of employers available outside the industry in the region.

• Hypothesis 2b: Wage dispersion decreases with the size heterogeneity of firms within the industry-region.

S&S also introduced a contingency measure between the main measure of vertical differentiation (number of firms in an industry-region) and two ⁵ of their three measures of horizontal differentiation (the variety of employers available outside the industry in a region and the concentration of employment shares across industries in the region). Their analysis of these contingencies showed that as the number of employers in an industry rises, organizational diversity increasingly reduces wage inequality (pp. 776–777). We test the same predictions:

• Hypothesis 3a and 3b: The association between the number of firms in the industry-region and wage dispersion (hypothesis 1a) decreases with (a) the number of industries in the region and (b) concentration of employment shares across industries in the region.

Methods and Data

Designing a study of corporate demography and workforce wage inequality demands detailed information about a large number of organizations as well as detailed information about individual workers’ characteristics and incomes. Furthermore, we need information that links employees with employers and where they are located. We use a rich register-based dataset covering all employees in Sweden from 1992 to 1998. ⁶ Besides allowing us to examine the effect of corporate demography on wage inequality, this allows us to closely replicate S&S’s study, including a set of theoretically salient extensions to their study.

We took several steps to ensure that we were unaware of any patterns in the data prior to formulating our hypotheses. The specific variables we use in this study were computed specifically for this project and have thus not been used before. Before constructing variables, we filed a preregistration plan with Open Science Framework (see https://doi.org/10.17605/OSF.IO/XFS6M) with research questions, hypotheses, methods used (including model specifications), variable measurement, and transformations. Explorative analysis was limited to data extraction and variable creation. These steps helped us avoid subconsciously tailoring our findings to the theoretical arguments or extend analyses beyond what initially planned (Gelman and Loken 2013). Codes for analysis are publicly available in Zenodo at https://doi.org/10.5281/zenodo.13807554 (Bomark 2024). For legal reasons, the data cannot be shared, but it can be requested from Statistics Sweden. ⁷

Wage Setting and Institutional Differences in Sweden versus Denmark

Our data are on organizations and individuals in Sweden, a country closely resembling Denmark (the setting of S&S’s original study) in terms of economic development, taxation levels, workforce skill composition, and gender equality. Yet Sweden and Denmark differ on two important points. First, employer flexibility in hiring and firing of employees is higher in Denmark than in Sweden, although both countries are above the European average in labor force flexibility (Auer and Cazes 2003). This is noticeable in that Denmark has somewhat high rates of mobility between firms and shorter durations of firm tenure than in Sweden (Madsen 2014). Second, the geography and thus local labor market boundaries as well as the firm-size distribution in those local labor markets differ between the two countries. The Danish economy is characterized by a large share of the private sector workforce employed at small or medium-sized firms, whereas a larger share of the Swedish workforce is employed at large firms (Henrekson and Johansson 1999). These differences pose an attractive feature for replication purposes with unobservable characteristics such as economic, political, and labor market institutions being similar but firm-size distribution within and across industries and regions differing. If S&S’s model of organizational diversity-wage inequality exhibits high external validity, the results should by and large replicate in settings with differing key explanatory and outcome variables. ⁸ Related to differences in industrial structure are also geographic differences, with Sweden being 10 times as large as Denmark but having almost as many township regions. ⁹

Data

The data we use were commissioned by Statistics Sweden for a larger research program on segregation. We access the raw data using an online protocol from which we can download aggregated data and results. Our sample consists of all employees in Sweden from 1992 to 1998 (we also perform robustness checks for the period from 1992 to 2012; see note 6). Specifically, the data comprise time-varying information on all residents of Sweden 16 years and older, employment, demographic, and educational characteristics, organizational (employer) affiliations, places of residence, and wages. The data come from the Longitudinal Integration Database for Health Insurance and Labor Market Studies.

We aggregate organizational-level data (size in terms of employees), industry affiliation, and organizational form (for profit or nonprofit and domestically or foreign owned) to the regional (township and labor market region) levels and aggregate individual data (wage dispersion and residual wage dispersion by industry) to the same regional levels for 1992 to 1998. As we use register data, there are none or few missing data points. We sample only people who are employed each year.

Unit of Analysis and Variables

The unit of analysis in our study is similar to that of S&S, an industry-region. Industry-regions are created from the underlying data by aggregating individual level data to all Standard Industrial Classification–equivalent industries (two-digit level) in Swedish municipalities (akin to Danish “townships”). In robustness tests, we replace the township definition of regions with labor market catchment areas. As this is a population study, we include all industry-regions in Sweden during the period available to us (1992–1997). In the following sections, we describe our dependent, independent, and control variables.

Dependent Variables

Two different variables measure workforce wage inequality in industry-regions. First, we compute gross wage dispersion by taking the logged standard deviation of wages in the workforce for each observed combination of industry-region, time variant (Sorensen and Sorenson 2007:773). Second, we compute residual wage dispersion by taking the natural log of the standard deviation residual income for workers in an industry-region. Similarly to S&S, we account for sampling error in estimating the standard deviation by adding 1/(2 × n) to each industry-region cell, where n is the number of employees in each industry-region cell. Residual income is computed from an ordinary least squares (OLS) model in which we regress the log of wages from an individual’s primary job (dependent variable) on the following same predictor variables: labor market experience, labor market experience squared, years of education, gender, marital status (dummy), interaction between gender and marital status, and number of children younger than two years, with fixed effects for industry and region (Sorensen and Sorenson 2007:773). This allows us to estimate the degree of wage dispersion net of individual observable characteristics to indirectly control for employees sorting to specific employers.

Operationalization of the dependent variables remains close to that of S&S, with two deviations. First, S&S had access to data on annual hourly working time. We have access to individuals’ incomes from their largest employment each year. Second, S&S also included employer firm tenure in their OLS model. We do not have access to firm tenure but instead include individuals work experience.

Independent Variables

We compute a number of independent variables, measured identically as in S&S for the replication part of our study. Log mean industry wage is computed as the logged mean of wages for each observed combination of industry-region (time variant). The point estimate or directional effect of this variable does not constitute a test of any hypothesis, but the variable is essential in the model of residual wage dispersion, as the magnitude of the standard deviation depends in part of the mean (Sorensen and Sorenson 2007). Number of firms in the industry-region (hypothesis 1a) is measured by the log number of firms in the industry-region. Industry-regions with a single employer (hypothesis 1b) is measured by a dummy variable taking the value 1 for industry-region cells with only one employer. Variety of employers available outside the industry in the region (hypothesis 2a) is measured by the number of industries in the region. Size heterogeneity of firms within the industry-region (hypothesis 2b) is measured using three separate proxies. First, we calculate the standard deviation of firm sizes in the industry-region. Second, we count the number of separate industries employing labor in the region. Following S&S, we also calculate a more nuanced entropy measure that measures the number of industries in the region and the concentration of employment shares across these industries. The entropy measure is calculated as: −Σp_iln(p_i), where p_i is industry i’s share of private employment in a region (often called the Shannon index).

We also construct two independent variables for the extension part of our study: ratio of foreign-owned firms in the industry-region (hypothesis 4) is measured as the proportion of firms in the industry-region that are controlled by firms registered outside of Sweden according to firm ownership statistics. Ratio of nonprofit firms in the industry-region (hypothesis 5), measured as the proportion of firms in the industry-region that are controlled by a noncorporate legal entity (i.e., a religious organization, a charitable foundation, or similar, defined according to the Swedish sector code classification (codes greater than 50, excluding code 96, which is “foreign-owned entities”).

Control Variables

We also include the same control variables used by S&S: industry employment in a region is measured as number of employees in the industry-region. Total number of employers in a region across all industries is measured as sum of firms in the region with employees, scaled by 100. Year dummies for each of the years 1992 to 1998 control for economic cycle conditions.

Analytical Approach

To analyze the effect of corporate demography on wage inequality in industry-regions over time, we create a cross-sectional panel data of industry-region pairs, similar to S&S. Standard errors are clustered on the region level, with year and industry indicators (dummy variables) included in all models except models 4 and 5. Similar to S&S, we also mean center the interaction effects between (1) number of firms in the industry-region and overall number of industries in the region and (2) number of firms in the industry-region concentration of employment shares across industries in the region (hypothesis 3a and 3b). As described earlier, we log-transform both dependent variables and several of the independent variables.

Exploratory Analysis

In the exploratory analyses of variable distribution and correlations, we compared our yearly cross-sections of Swedish industry-regions to ascertain that there are no apparent coding errors or implausible kinks in the data. We ran scatterplot matrices for all combinations of variables to search for potential nonlinearities and outliers in our data. As inequality models are attentive to variance, especially in small clusters, we do not exclude outliers in the main analyses. We compare the exploratory analyses of our variable distribution and correlations with those of S&S. Figure A1 shows that both dependent variables (gross wage dispersion and residual wage dispersion) have markedly different mean values in S&S’s dataset of 97,489 Danish industry-regions than in our comparable dataset of 55,295 Swedish industry-regions. Figure A1 also shows that most of the independent and control variables have similar mean values in the two datasets, except for n industries in township and entropy of industry shares, which were higher in S&S’s dataset, likely because many Swedish townships in the north are sparsely populated.

Descriptive Statistics

Correlations among variables used are shown in Tables A2, A8, A14, and A20. Focusing on Table A2, which concerns the period from 1992 to 1998, some variables are moderately to highly correlated, as is common in regional-level analysis: total n firms in township (scaled down by a factor of 10^-3) with industry employment in township (0.50) and n industries in township (0.50), log n industry firms in township with single industry employer (−0.58), and importantly entropy of industry shares with n industries in township (0.64). A strong correlation between total number of firms and total employment in a township was expected, as regions with many employees also most likely have many firms. S&S reported similar correlations except for our strong correlation between entropy of industry shares with n industries in township. However, when we compare the two datasets, we see that the correlations between all independent and control variables with the two dependent variables are similar: for gross wage dispersion, differences in correlation coefficients across our dataset and that of S&S range from −0.31 to 0.06, with a mean difference from S&S of 0.04. For residual wage dispersion, differences in correlation coefficients range from −0.48 to 0.07, with a mean difference from S&S of 0.06. This means that even if intercepts and point predictions are likely to differ, we see no reason that beta coefficients—the test of each hypothesized variable on the two dependent variables—would necessarily differ from those of S&S because of the differences in mean gross wage dispersion and mean residual wage dispersion in the two datasets.

Hierarchical OLS Analyses

We perform hierarchical OLS analyses identical to those of S&S (their Tables 2 and 3) for the period from 1992 to 1998. Also, we add two new variables to examine hypotheses 4 and 5, the ratio of foreign-owned firms and the ratio of nonprofit firms, both measured at the industry-region level.

Similar to S&S, we conducted robustness tests with all independent variables measured on the labor market areas to explain wage inequality within industry-region clusters (see Tables A13–A24). We assessed p values for all coefficients using two-tailed tests of significance. When comparing multiple conditions or testing multiple hypotheses, like S&S, we both examine a separate OLS specification as well as a fully saturated model that includes all predictor variables, except for the interaction effects.

Results

We first report results for the replication analysis of S&S. We estimate the effects using gross wage dispersion (logged standard deviation of wages), before rerunning the models using residual wage dispersion as the dependent variable. All models include dummy variables for year and industry and standard errors clustered on the regional level to account for serial correlation. Results are presented visually in Figures 1 and 2 (complete regression tables can be found in Tables A3 and A4), and for ease of comparison, we report both our and S&S’s estimates in the figures and in the text. Estimates of control variables in the figures can be found in Tables A3 and A4. We start by presenting the hypotheses with results similar to those of S&S (hypotheses 1a, 1b, 3a, and 3b) before we present the hypotheses for which results deviate somewhat from those of S&S (hypotheses 2a and 2b).

Figure 1.

Gross income dispersion: replication 1992 to 1998.

Figure 2.

Residual income dispersion: replication 1992 to 1998.

Hypotheses 1a and 1b Vertical Differentiation

We find strong support for vertical differentiation among organizations (hypotheses 1a and H1b) when explaining workers’ wage inequality, using both gross wage dispersion and residual wage dispersion.

Gross Wage Dispersion

Model 2 in Figure 1 shows that the number of firms in the industry-region, log n industry firms in region, is positively associated with workforce wage dispersion (β = 60.84, p < .001). Furthermore, workforce wage dispersion is higher in industry-regions with more than a single employer (hypothesis 1b), as shown by the negative coefficient for single industry employer (β = −197.00, p < .001). The coefficients used to test hypotheses 1a and 1b using gross wage dispersion are large with small standard errors, and the coefficients remain robust across all our model specifications. The β coefficients for our models using gross wage dispersion correspond to those of S&S (model 2: log n industry firms in region [β = 141.13, p < .01] and single industry employer [β = −130.69, p < .01) (Sorensen and Sorenson 2007, Table 2).

Residual Wage Dispersion

We find strong support when we regress residual wage dispersion on our variables measuring vertical differentiation (Figure 2, model 2): log n industry firms in region (β = 26.33, p < .001) and single industry employer (β = −72.51, p < .001). The coefficients used to test hypotheses 1a and 1b using residual wage dispersion are large with small standard errors, and the coefficients remain robust across all our model specifications. The β coefficients using residual wage dispersion correspond to those of S&S (model 2: log n industry firms in region [β = 61.92, p < .01] and single industry employer [β = −101.89, p < .01) (Sorensen and Sorenson 2007, Table 3). Thus, just like S&S’s results, we find that wage dispersion, measured using gross wage dispersion and residual wage dispersion, increases with the number of firms in the industry-region (hypothesis 1a) and is higher in industry-regions with more than a single employer (hypothesis 1b).

Hypotheses 3a and 3b Interactions

We find mixed results when testing if the association between number of firms in the industry-region and wage dispersion (hypothesis 1) is conditional on (a) the number of industries in the region and (b) the concentration of employment shares across industries in the region (hypotheses 3a and 3b). For gross wage dispersion, model 4 in Figure 1 shows that the interaction between log n industry firms in township and n industries in township is negatively related to gross wage dispersion (β = −1.86, p < .001), as is the interaction between log n industry firms in township and entropy of industry shares (β = −25.61, p < .001) in model 5 in Figure 1. The β coefficients for these two interaction terms correspond well to those of S&S (Sorensen and Sorenson 2007, Table 2) (β = 1.58 [p < 0.01] and β = −25.26 [p < .05], respectively). For residual wage dispersion (Figure 2, model 4), we do not find statistically significant interaction terms. The interaction between log n industry firms in township and n industries in township is β = −0.51 (p > .05). The interaction between log n industry firms in township and entropy of industry shares is β = −2.09 (p > .05) (Figure 2, model 5). S&S present β coefficients for these two interaction terms of −1.25 and −22.57 (p < .01 for both) (Sorensen and Sorenson 2007, Table 3).

Hypotheses 2a and 2b Horizontal Differentiation

Like S&S, we do not find strong and consistent support for the hypotheses that organizations’ horizontal differentiation (hypotheses 2a and 2b) reduces gross wage inequality in our replication. The coefficient for standard deviation of employer sizes is small and not statistically significant (β = 0.00, p > 0.05; Figure 1, model 2), and the coefficient for entropy of industry shares (Figure 1, model 3) is β = −18.18 (p > .05). ¹⁰ In contrast to S&S, who found a negative but not statistically significant effect for n industries in township (β = −0.55, p > .05), we do find a negative and statistically significant coefficient for n industries in township (β = −1.50, p < .05) (Figure 1, model 2).

We next regressed residual wage dispersion, which accounts for workforce human capital characteristics, on the three variables measuring horizontal differentiation. Like S&S, we find support for the hypotheses that organizations’ horizontal differentiation (hypotheses 2a and 2b) reduces residual wage dispersion in our replication, albeit with less conclusive results than those of S&S. Although we do find a negative and statistically significant β coefficient for standard deviation of employer sizes (Figure 1, models 2 and 3: β = −0.23, p < .001), and n industries in township (β = −2.73, p < .001), different from S&S’s study, we do not find a statically significant β coefficient for entropy of industry shares (β = −2.50, p > .05). S&S reported β coefficients with negative direction and relatively small standard errors (Sorensen and Sorenson 2007, Table 3, model 3). In the following section we present a set of multiverse analyses that seek to unearth the reason for these discrepancies and their underlying methodological or contextual rationales.

Robustness Tests

Like S&S, we reestimate all our models using local labor markets instead of townships (see Tables A13–A24); we follow Cobb and Stevens (2017) and rerun our analyses controlling for the proportion employed by large firms (see Tables A25 and A26). Last, we reestimate our models for the period from 1992 to 2012 to examine the theory’s validity in recent times (see Tables A9 and A10). In none of these robustness tests did the effects of horizontal and vertical differentiation change substantially.

Sensitivity Analyses Using a Multiverse Approach

Why do our results on horizontal differentiation differ from those of S&S, who found stronger support for organizations’ horizontal differentiation to decrease residual wage dispersion? To investigate this discrepancy, we probe whether differences in the distribution of key variables in S&S and the present study may stem from specific analytical decisions taken throughout the data analysis process. It is possible that we have misinterpreted or missed some analytical decisions of the original study, which could explain differences in the result of the replication (Freese and Peterson 2017; Young 2018). To assess this, we use a multiverse approach, which is a computational multimodel framework (Muñoz and Young 2018).

A multiverse analysis entails systematically performing analyses of the data from different reasonable choices related to the empirical analyses (e.g., data processing, sample selection criteria, and variable construction). Doing so allows researchers to assess the results’ robustness and sensitivity to multiple different reasonable research choices that may influence the findings (Gelman and Loken 2014; Young 2018). Therefore, multiverse analyses can be used to enhance inference and can serve as a tool to analyze the robustness of findings.

For the context of this study, multiverse analyses offer an additional layer to assist replication studies. This potential comes from their capacity to serve two purposes. First, they enable us to gauge the sensitivity of the original findings, thereby enhancing credibility and robustness of the results. Second, multiverse analyses enable us to scrutinize the impact of research choices that may not have been explicitly articulated but still have the potential to significantly influence research findings. By encompassing a wide range of methodological and analytical choices within the multiverse, it helps uncover hidden nuances and potential sources of bias, contributing to a more comprehensive understanding of the validity and reliability issues.

A multiverse analysis consists of several steps. As described by Simonsohn, Simmons, and Nelson (2020), the researcher starts off by defining a set of reasonable specifications and research choices. Next, the analyses are performed on each of the unique combinations of these specifications to assess the variability of statistical estimates across these choices. One specific tool for evaluating variability is the specification curve. This curve represents the estimated effect size for all specifications, sorted by magnitude, and includes a chart revealing the specific choices made for each estimate. This allows readers to visually discern both the differences in effect size among various specifications and how these differences correlate with operationalization decisions and provides an intuitive presentation of the multiverse analyses enchilada allows a visual identification of variability across research choices, as well as the identification of choices that may drive variability in results obtained.

In the process of replicating S&S’s analysis, we identified three such analytical decisions related to sampling and data processing decisions underlying the construction of the dependent variable, residual wage dispersion, for which differences in variable distributions between S&S and the present study are most noticeable (see Figures A2–A8).

The concerns identified relate to analytical decisions regarding (1) the removal of organizations belonging to certain industries. ¹¹ We also evaluate specifications that can be considered arbitrary, such as (2) the threshold for excluding public sector workforce ¹² and (3) different definitions of and how much an individual needs to earn to be considered part of the workforce. ¹³ Specifically, S&S included the complete working population and performed analyses on workers’ hourly wages. As we do not have access to data on hourly wages and instead use annual wages, differences in results obtained could be driven by the share of the population who worked part-time or only worked parts of the year. To address these differences in our data compared with S&S, we exclude individuals whose annual wage is below a set threshold to see if that is causing the differences.

In addition to these three analytical decisions, we also identified apparent discrepancies in the descriptive statistics (see Figure A1) of S&S and our data, indicating that the workforces in Sweden and Denmark may differ. More precisely, the descriptive statistics indicate that the demographic composition of Swedish companies differ from their Danish counterpart: Sweden has more large companies compared with Denmark (Henrekson and Johansson 1999). Moreover, although Sweden is almost 10 times as large and has a workforce about the double that of Denmark, the two countries have a similar number of township regions. Danish townships are thus denser in terms of both employers and employees. One way to transform the industrial structure of Sweden to more closely resemble that of Denmark is to reestimate our models excluding Swedish regions with small workforce sizes. To investigate these factors and if there are any analytic decisions that could be taken to reduce these differences (which also could explain the discrepancies between our results and S&S when it comes to horizontal differentiation), we identify two more analytic decisions: (4) large organizations and (5) workforce size in townships.

For each of the five analytical decisions, we define a set of parameters that reasonably could be chosen. We identify 22 possible choices related to these six factors outlined above and rerun the analyses with all possible combinations of these (in total 1,200). Table 1 outlines the specific parameter specifications. We replicate S&S’s analysis (i.e., for 1992–1998) and focus on the model on residual wage, as that is where we observe a difference. We present the result from these following Simonsohn et al.’s (2020) specification curve (see Figures A2–A8).

Table 1.

Decisions and Parameters for Multiverse Analysis.

Decision	Description	S&S’s Decision	Parameters
(1) Industries	Inclusion criteria for industries	S&S excluded industries in agricultural and extractive.	Omit industries in the agricultural and extractive industries or keep all sectors.
(2) Sectors with public employees	Threshold for dropping industries with large public employee share (sensitivity analysis of S&S’s 15% threshold)	S&S removed industries in which the public sector employed >15 percent of the workforce.	We omit industries in which the public sector employed more than a percentage of the workforce; where possible, percentages were 0, 0.1, 0.15, 0.2, and 0.3.
(3) Wages	Omitting individuals with no income (but employed) and assess removal of individuals in lower income brackets (to exclude part-time workers)	S&S used hourly wage, so their analyses should be less sensitive to the influence from part-time workers.	Omit employees whose annual wage was below a threshold. We used the following thresholds: 0, 25,000, 50,000, 75,000, and 100,000 Swedish kronor.
(4) Large organizations	We omit large companies in several steps.	S&S did not exclude large firms	Omit firms that had more employees than a threshold. We used the following thresholds: 0, 500, 5,000, and 10,000 employees.
(5) Workforce size	Number of smaller township deciles removed	S&S did not exclude townships on the basis of their workforce sizes.	Omit deciles of smallest townships in terms of workforce using the following thresholds: 0, 0.1, 0.2, 0.3, 0.4, and 0.5.

Note: In deciding (1) cutoffs for industrial codes, we use the parameters that exclude industries related to agricultural and extractive as well as parameters that do not exclude them. For (2), we alternate the threshold for dropping industries with large public employee share in 10 percent intervals between 0 percent and 30 percent, as well as including S&S’s decision on 15 percent. For (3), we omit employed individuals with no income and gradually remove of individuals in lower income brackets to exclude part-time workers (the median wage in Sweden was 178,800 kronor in 1992 and 231,600 kronor in 1998). The main reason for this is that S&S used hourly wage, whereas we use annual wage. For contextual factor (4), we omit large organizations with more than 500, 5,000, or 10,000 workers. We also include S&S’s decision (i.e., not excluding firms on the basis of their size). For contextual factor (5), we omit declines from 0 to 0.5 of the smallest townships in terms of workforce.

The specification curves of our multiverse analyses (see Figures A2–A8) indicate that most of the findings from the replication are relatively robust to different analytic decisions. In the majority of the “universes” generated, we find similar results to our main replication of S&S and the original study, both in terms of statistical significance and direction of association. More important, the analytic decisions are evenly distributed along different coefficients, suggesting that they do not substantially influence the analyses. However, we can observe that one factor has an especially large impact: wages. For all independent variables, omitting different parts of the workforce with the lowest wages influences the direction and significance of our results. This could be an artifact of our not being able to differentiate between part-time work and full-time work, which might influence the observed wage inequality.

In analyzing the effects of vertical differentiation, the higher wage threshold used to omit a part of the workforce, the further away our results diverge from those of S&S. Our main results show the effects to be not statistically significant or indicate that the association switches direction. As seen in the figures (Figures A2–A8), across all variables measuring vertical differentiation, the analytic decision space for wage restrictions of the sample is unevenly distributed. Thus, restricting the sample on the basis of wage seems to influence the analyses to a large degree.

We observe the opposite for horizontal differentiation: the higher the wage threshold we use to omit the lower parts of the wage distribution, the closer the associations become to those of S&S. Regarding the interaction terms capturing vertical and horizontal differentiation, patterns such as those of vertical differentiation emerge; no threshold or a low wage threshold yields results similar to those of S&S and the main results in our replication. For most other analytic decisions, the distribution in the decision space seems to be relatively evenly distributed, indicating that they are not decisions influencing the analyses to a larger degree. A factor that plays a role for the strength of these correlations is the size of the workforce: when omitting townships with a smaller workforce—to resemble the Danish labor market—the analyses yield results similar to those of S&S. Alternating other central analytic choices (Table 1) does not yield any larger systematic effects on our results.

Model Extensions

To extend the original model of corporate demography and wage inequality, we incorporate two key notions of differentiation in organizational form within industries that are linked to divergent employment and wage setting practices: foreign-owned and nonprofit organizations.

Foreign-owned organizations are either subsidies of or majority owned by foreign entities. Many of those entities are multinational enterprises known to predominate in high value-added industries, on average paying higher wages, but are also frequently found in industries with high ratios of employees on nonpermanent contracts (Casson 1987). As a reflection of globalization, multinational enterprises commonly face trade-offs between adopting local modes of organizing such as levels of management or acceptance of unionization or global modes of organizing on the basis of economies of scale (Kerrissey 2015; Western and Rosenfeld 2011).

For a generic model of how corporate demography affects the wage structure of employees, the relative level of distinctiveness of foreign owned compared with other organizations in an industry is, however, not of key interest. Our interest is in how variation in the kinds, or forms, of organizations in a particular labor market affects the sorting of employees with various skills sets differentially, whether a sizable fraction of those organizations are foreign-owned companies or not, and consequently, whether this affects the overall wage dispersion in the industry-region in which they operate. Hence, foreign-owned firms could enhance high-quality sorting on the labor market compared with their native counterparts (Malchow-Møller, Markusen, and Schjerning 2013). If matching is enhanced in regions with higher shares of foreign-owned firms, this should lead to increased wage dispersion. The ratio foreign-owned firms thus add another dimension of organization diversity that is likely to influence the sorting on the labor market and, in turn, wage inequality. Given the evidence that large foreign-owned firms concentrate in high value-added industries or industries with a more employees on nonpermanent contracts (Casson 1987), we expect that the overall differentiation mechanisms will increase wage dispersion in the industry-region:

• Hypothesis 4: Wage dispersion increases with the ratio of foreign-owned firms in the industry-region.

A second but important source of horizontal differentiation among organizations highlighted in the literature is the role of nonprofit organizations as supplementing goods and services for sale in competitive markets. A distinctive feature of nonprofit organization is that they are prohibited from distributing profits, if any, to individuals who exercise control over the organization (Hansmann 2021). In our setting of Sweden, this entails voluntary associations, religious organizations, or foundations active primarily in markets such as health and elderly care, schooling and training, and artistic goods and services (Wijkström 2004). Corporate demography argues that nonprofit status is a crucial sort of distinctiveness in organizational form (Hsu and Hannan 2005) but does not contain clear predictions whether and how such distinctiveness may shape employee sorting. Yet the literature on nonprofits in economics and organizational sociology portrays nonprofits as entities more likely to emerge in markets characterized by asymmetric information and the need for consumer trust, recruit managers with a mindset aligned with that of the nonprofit entity, seldom use incentives or varying wage levels, and tend to have more compressed wage structure than their for-profit counterparts (Ben-Ner, Ren, and Paulson 2011; Leete 2000). This suggests asymmetric matching of employee driven in part by intrinsic reasons to work for a nonprofit firm, leading us to propose the following hypothesis:

• Hypothesis 5: Wage dispersion decreases with the ratio of nonprofit firms in the industry-region.

Hypothesis 4: Foreign-Owned Firms

Figures 3 and 4 examine whether the ratio of foreign-owned and nonprofit firms in the industry-region adds explanatory power to the model of corporate demography in explaining wage inequality. ¹⁴ Looking first at model 1 in Figure 3, the coefficient ratio of foreign-owned firms in industry-region is not related to gross wage dispersion. Yet in the analogous test for residual wage dispersion in model 1 in Figure 4, the β coefficient is negative and statistically significant (β = −151.93, p < .05), leading us to conclude that the level of foreign ownership in an industry-region is not systematically related to wage inequality when individual sorting is accounted for, as in the residual wage dispersion model.

Figure 3.

Gross income dispersion: extension 1992 to 1998.

Figure 4.

Residual income dispersion: extension 1992 to 1998.

Hypothesis 5: Ratio of Nonprofit Firms

In model 2 in Figure 3, when we use gross wage as our dependent variable, the coefficient ratio of nonprofit firms in the industry-region (hypothesis 5) is not statistically significant. However, we do find a positive and statistically significant coefficient in model 2 in Figure 4 (β = 297.00, p < .001), when we use residual wage as the dependent variable, which is the opposite than the hypothesized direction.

Discussion

In this study we have empirically investigated the role of organizational diversity in the production of workforce wage inequality. Despite its theoretical rich history, limited empirical studies have tested the validity of these theoretical propositions. Replicating Sorensen and Sorenson’s (2007) study on the topic on a close-to-identical dataset using the same time period but in Sweden, a country similar to Denmark in background conditions but different in the type and number of organizations, similarly to S&S, we find strong support for vertical differentiation increasing workforce wage inequality. However, we find less support compared with S&S for less wage dispersion in industry-regions with a high size heterogeneity of firms when accounting for workforce human capital characteristics in the analysis of residual wage dispersion. To understand this discrepancy, we perform a computational multiverse analysis to identify what potentially could explain the discrepancy. We find that the definition of what constitutes the working population strongly influence our results and could be the factor explaining why we obtain different results than S&S when it comes to horizontal differentiation. Another explanation for the replication discrepancy suggested by our multiverse analyses concerns the geographic differences in population density across regions in Denmark and Sweden. This could indicate that the generalizability of S&S’s findings is limited to certain contextual factors (i.e., population density and others). Thus, future research examining the effects of corporate demography across regions should benefit from contrasting regions that are more or less dense (Dahl and Sorenson 2010; Moretti 2011). Another potential explanation could be that Swedish workers—similar to workers in the United States, where inequality has always been among the highest of the Organisation for Economic Co-operation and Development countries, are differentially exposed to the “firm-size wage premium” than in Denmark (Cobb and Lin 2017). As the upper left box plots in Figure A1 indicate, wage dispersion was significantly higher in Sweden than in Denmark from 1992 to 1998, a period that overlaps with the largest economic setback in Sweden since the depression era. ¹⁵

In extensions of S&S’s original model examining how variations in organizational forms in the form of foreign ownership and nonprofit organizations, we found no systematic support for the ratio of foreign ownership or nonprofit organizations being systematically related to workers’ wage inequality in the industry. A plausible conclusion is that these more fine-grained measures of organizational form variation in the industry-region are by and large captured by the systematic differences in firm-size distribution that form the core components of the theory of corporate demography.

Similar to S&S, we found that the dummy variable single industry employer was negatively related to wage inequality. A plausible interpretation is that this variable, rather than being a proxy for vertical differentiation in local labor markets, measures the well-known “fair wage hypothesis” in behavioral economics (Akerlof and Yellen 1990) on the basis of equity theory in social psychology and social exchange theory in sociology (see also Cobb and Stevens 2017 and the behavioral mechanisms of internal labor market of, e.g., Frank 1985 and Thurow 1975). The fair wage hypothesis holds that internal wage structure is significantly more compressed than what would be expected from the marginal productivity of each employee in a competitive labor market. Hence, although industry-regions dominated by a sole employer “offer no opportunity for quality sorting between firms” (Sorensen and Sorenson 2007:774), the inequality-reducing effects of large internal labor markets in dominating organizations seem to offset the inequality-increasing effects of workers’ not being able to move to competing employers in the same industry-region.

In conclusion, our replication of S&S’s study provides further support that organizations’ vertical differentiation drives wage inequalities. However, the findings are less consistent regarding the impact of organizational horizontal differentiation on wage inequality, an aspect that needs to be further explored in future research. The results of our multiverse analyses demonstrate that analytic decisions, particularly how the workforce sample population is defined, significantly influence the direction and strength of the effects of organizational differentiation. In doing so, we highlight how multiverse analyses can be used to assess the sensitivity of findings to analytic choices and ensure the reliability of the findings. Moreover, we also showcase how multiverse analyses can be a tool for replication studies, especially for investigating analytic decisions that were vaguely described or were not justified. This contribution moves beyond the topic-specific scope of our article and highlights the utility of multiverse analyses as a methodological framework in sociological research generally, especially in replication studies. Overall, our study offers insights for the cumulating research in organizational sociology regarding the growing wage inequality in societies, emphasizing the usefulness of replication studies and specifically multiverse analyses to discern testable causal mechanisms from contextual effects.

Limitations and Future Research

This study is a first attempt to demonstrate the generality of theories of wage inequality from corporate demography beyond the setting in which those theories were originally developed. Given the overall parsimonious results obtained with those of the replicated study, even when extending the period of study to that of a more recent period, the contributions offered are intentionally refinements to rather than upheaval of the theory. Our study also provides food for thought for research on corporate demography and for organizational scholars interested in wage inequality more broadly.

Several challenges remain to grapple with for organizational sociology to contribute to the broader discussion on how to address the rampant workforce wage inequality in most countries. First, there are still few agreed-upon standardized measures of wage inequality and a dearth of theoretical models on which to formulate and test theoretical predictions. We have sought to test and extend perhaps the most well-developed theoretical framework designed specifically for addressing macro-questions from an organizational perspective, that of corporate demography. Still, challenges remain in how to bring fragments from this theory together in well-designed studies that can be tested on empirical data. Our study identifies three major areas for further development of how corporate demography may explain workforce inequality in outcomes:

First, research on organizational forms could help further probe, extend, and problematize how we measure corporate demography in the rapidly changing economies of today. Although bountiful research on organizational forms exists and could be used to develop measures of, for example, nonprofits (Perrow 1961), network-based (Ahuja and Carley 1999), franchise (Sorenson and Sørensen 2001), and international (Bucovetsky and Haufler 2008) forms of organizations ¹⁶ compatible with the overarching theory of corporate demography, our study shows that introducing such measures imply a trade-off between taxonomic detail on one hand and model parsimony and generality on the other hand (Carroll and Hannan 2000a).

Second, comparative testing of models of corporate demography and workforce inequality against other counterfactual explanations such as globalization or skill-biased technological change ¹⁷ would help further validate and refine the extent to which models of corporate demography can explain the rising inequality in labor markets of today. For organizational models of wage inequality to influence the broader social science literature, comparative scrutiny of such competing theoretical explanations is vital.

Finally, future research could also expand upon the “middle range” analyses of industry-regions here presented to also examine (1) the micro-level dynamics by which vertical differentiation among organizations brings about better quality sorting of employees within industries and over time as well as (2) the macro-level implications for societies (national, international, or global) by which corporate demography affects career paths, inequality and economic attainment. The first line of inquiry would entail constructing more detailed models of how workers of different ability and characteristics sort within and across regional labor markets (Kalleberg and Sorensen 1979) and how this is shaped by the prevailing corporate demography. A fundamental challenge for such research is that workforce sorting in itself affects the size distribution of firms. The second line of inquiry would involve connecting models of the type articulated by S&S and here extended to macro-patterns of workforce inequality. Also, isolating the effects of corporate demography for workers’ career paths, inequality, and economic attainment represents a major methodological challenge. The broader research on organizations and wage inequality may seek to further tease out cause-and-effect relationships in observational data by, for example, exploiting exogenous shocks to the stock and structure of organizational populations caused by legal changes or economic shocks (Autor, Dorn, and Hanson 2015). Further examination of changes in corporate demography due to digitalization and decreasing transaction costs provides avenues for further probing the important role of organizations for workers’ careers and earnings. Meeting these and similar challenges may help scholars of organization advance the research on corporate demography and further probe questions related to inequality in workforce career paths and economic attainment articulated by Hannan (1988a, 1988b) more than three decades ago.

Supplemental Material

sj-docx-1-srd-10.1177_23780231241287858 – Supplemental material for Corporate Demography and Wage Inequality: Revisited

Supplemental material, sj-docx-1-srd-10.1177_23780231241287858 for Corporate Demography and Wage Inequality: Revisited by Niklas Bomark, Elis Carlberg Larsson and Karl Wennberg in Socius