The Role of Firms in Wage Inequality

Policy Lessons from a Large Scale Cross-Country Study

Report

9 December 2021

2. Worker skills or firm wage-setting practices? Decomposing wage inequality across 20 OECD countries

Abstract

In many OECD countries, low productivity growth has coincided with rising wage inequality. Widening wage and productivity gaps between firms may have contributed to both developments. This chapter uses harmonised linked employer-employee data for 20 OECD countries to analyse the role of firms in wage inequality. The main finding is that, on average across countries, differences in average wages between firms explain about one-half of overall wage inequality. Two-thirds of between-firm wage inequality (i.e. about a third of overall wage inequality) reflect firms’ wage-setting practices or wage premia, i.e. the part of wages that is determined by the firm rather than the characteristics of its workers. The remaining third (i.e. a sixth of overall wage inequality) can be attributed to differences in workforce composition across firms. The contribution of differences in wage premia to wage inequality tends to be larger in countries with decentralised collective bargaining systems and lower levels of job mobility. Overall, these results suggest that firms play an important role in explaining wage inequality, as wages are to a notable extent determined by firm wage-setting practices rather than being exclusively by workers’ skills.1

In Brief

In many OECD countries, low productivity growth has coincided with rising income inequality. Widening wage and productivity gaps between firms may have contributed to both developments, as a significant share of firms has increasingly fallen behind the best performers. This paper presents comprehensive new evidence on the role of firms in the evolution of wage inequality from the mid-1990s to the mid- 2010s based on harmonised linked employer-employee data for 20 OECD countries.

On average across countries, changes in between-firm wage inequality (differences in average pay between firms) explain about one-half of the changes in overall wage inequality.
Changes in between-firm wage inequality reflect changes in the dispersion of firm wage-setting practices (“firm wage premia”) and skills-based sorting of workers across firms.
Two-thirds (65%) of changes in between-firm wage inequality are accounted for by changes in firm wage premia, i.e. differences in average pay between firms that are unrelated to skills and other worker characteristics.
The remaining one-third (35%) of changes in between-firm wage inequality can be attributed to changes in the sorting of workers across firms based on their skills, possibly related to increased specialisation along the value chain (e.g. outsourcing).
- About 15% of changes in between-firm wage inequality reflect the sorting of workers across firms based on firm wage premia (the sorting of high-skilled workers into firms that pay high wages to all workers).
- About 20% reflect changes in the sorting of workers into firms with similar co-workers (the clustering of similarly-skilled workers at given firm wage premia), which does not affect overall inequality as larger wage differences between firms are offset by narrower differences within firms.
Differences in wage premia between firms tend to be more pronounced in countries with decentralised collective bargaining systems and lower levels of voluntary job mobility.

These results suggest that firms play a crucial role in explaining aggregate wage inequality. Rather than being fully determined by workers’ skills, wages appear to partly reflect firms’ wage-setting practices, which depend on their productivity as well as their wage-setting power. In addition to worker-centred policies, such as education and training, that may narrow the skill premium (which is estimated to have risen over the sample period), firm-centred policies that promote productivity in low-wage firms, increase competition for workers (e.g. by lowering barriers to voluntary job mobility) and limit the wage-setting power of firms (e.g. by collectively-agreed wage floors) are key to address concerns around inequality.

2.1. Introduction

At a time when many OECD countries are grappling with low productivity growth and rising inequality, gaps in business performance have also widened. While a small fraction of high-performing businesses continue to achieve high productivity and wage growth, the remaining ones are increasingly falling behind (Andrews, Criscuolo and Gal, 2016[1]; Berlingieri, Blanchenay and Criscuolo, 2017[2]). This raises the question whether growing performance gaps across businesses can at least partly account for aggregate productivity and inequality developments.

Designing better public policies for broadly shared productivity growth requires an understanding of the mechanisms through which firms affect both aggregate productivity and inequality. Firms may not only determine the distribution of market income between capital and labour, but also drive the distribution of labour income between workers, i.e. wage inequality.2 In particular, addressing concerns about rising inequality may not only require policies to support workers, such as in the areas of skills and wage-setting, but also business-focused initiatives that allow lagging firms to catch up or exit the market.

Uncovering the mechanisms linking growing performance gaps between businesses and wage inequality requires granular information on the characteristics of both workers and their employers. Previous cross-country studies relying on firm-level information have provided evidence of a close link between trends in productivity dispersion and trends in wage inequality (Berlingieri et al., 2019[3]). But quantifying the extent to which this correlation is due to worker composition as opposed to firm wage setting practices requires information on workers and the firms for which they work (i.e. linked employer-employee data). Such information allows quantifying the contributions to wage inequality of wage dispersion between (i) different workers within firms and (ii) similar workers across different firms. It also helps understanding the extent to which such differences are explained by workforce composition, differential technology adoption, or differences in market power between firms, which may in turn be driven by technology, domestic and international value chains, as well as policy.

In an effort to enhance the understanding of the role of firms in wage inequality across a large set of countries, this chapter makes use of a novel harmonised linked employer-employee dataset covering 20 OECD countries based on a strict data protocol that ensures cross-country comparability to decompose overall wage inequality within and between firms. The analysis covers a broad range of countries exhibiting widely different inequality dynamics and institutional settings. The chapter assesses for the first time the extent to which differences in the between-firm component of wage inequality reflect differences in firm wage-setting practices rather than differences in worker skills in a cross-country context.

The linked employer-employee data used in this chapter are based on administrative records designed for tax or social security purposes or, in a few cases, mandatory employer surveys. These data have the major advantage of being very comprehensive (covering the entire population of workers and firms in most countries) and of very high quality, notably with respect to information on wages, given the potentially important financial or legal implications of reporting errors and extensive administrative procedures for quality control. While such data are increasingly used for research on single countries, their use in a cross-country context remains rare.3

The analysis covers a broad range of countries that differ significantly in terms of their exposure to global trends related to globalisation and technology and the nature of policies and institutions, resulting in widely diverging inequality dynamics (Austria, Canada, Costa Rica, Denmark, Estonia, Finland, France, Germany, Hungary, Italy, Japan, Netherlands, Norway, New Zealand, Portugal, the Slovak Republic, Spain, Sweden, United Kingdom and the United States). The sample encompasses low-inequality countries (e.g. Sweden) as well as high-inequality ones (e.g. United States), and countries with large increases in wage inequality (e.g. Germany) as well as countries with pronounced declines (e.g. Estonia).

The decomposition of wage inequality between and within firms proceeds in three steps. First, to provide an indication of the role of firms in inequality, it starts with a raw decomposition of wage inequality into a part related to inequality in average wages between firms and a part related to inequality between workers’ individual wages within firms, similar to Tomaskovic-Devey et al. (2020[4]). Second, the decomposition is augmented with controls for the observable characteristics of workers following Barth et al. (2016[5]; 2018[6]). This allows decomposing the between-firm component into a part related to the wage-setting practices of firms (e.g. wage premia), and a part related to the skill and demographic composition of the workforce. Third, for countries for which this is possible, the decomposition is repeated controlling for both the observable and unobservable characteristics of workers following Abowd et al. (1999[7]) and Song et al. (2019[8]).

The main finding of this chapter is that firm wage-setting practices play an important role in explaining aggregate wage inequality. This suggests that concerns about high or rising inequality may not only require policies to support low-wage workers, such as in the areas of skills and wage-setting, but also business-focused initiatives that allow lagging firms to catch up or leading firms to create new jobs. As shown in Chapter 3 such policies would not only help to strengthen aggregate productivity growth, but also contribute to smaller wage inequality between firms as reduced productivity dispersion results in reduced wage-premia dispersion between firms.

The challenge for policy makers is to simultaneously promote productivity gains from the adoption of new and possibly skill-biased technologies and the corresponding efficiency-enhancing sorting of workers across firms, while ensuring a broader sharing of these gains. Policies that promote the adoption of productivity-enhancing technologies in low-wage firms are likely to be key, as they promote increased access to adequate skill upgrading for all workers, providing them with pathways to climb the job ladder. More generally, worker-centred policies, such as education and training, may need to be complemented with firm-centred policies that promote productivity in low-wage firms to effectively address concerns around high inequality and low productivity growth.

The remainder of the chapter is organised as follows. Section 2.2 describes the analytical framework that links technological change, globalisation and public policies to within and between-firm wage inequality. Section 2.3 outlines the construction of a harmonised cross-country linked employer-employee dataset and compares the resulting measures of wage inequality with other available data sources. Section 2.4 uses this dataset to provide a statistical decomposition of wage inequality into within- and between-firm parts for a range of OECD countries. Section 2.5 provides evidence on the role of worker sorting across firms and differences in firm wage premia in between-firm wage inequality. Section 2.6 concludes.

2.2. A framework for dissecting the role of firms in wage inequality

2.2.1. Conceptual framework

Aggregate wage inequality can be decomposed into wage dispersion between firms and within firms (Figure 2.1). Wage dispersion between firms may reflect differences in workforce composition or differences in revenue-based productivity at given workforce composition due to technology or market power, and the extent to which market rents are shared with workers. Wage dispersion within firms reflects worker heterogeneity in terms of a range of earnings characteristics – including education, experience and gender – and returns to these characteristics. The digital transformation, trade integration and demographic change, as well as public policies affect aggregate wage inequality through these channels.

In a perfectly-competitive labour market without frictions, where firms pay workers according to their marginal productivity (e.g. skills, unobserved ability, motivation etc.), pay differences between firms entirely reflect differences in workforce composition. For instance, one firm may mainly employ high-skilled workers at high wage rates, whereas another one may mainly employ low-skilled workers at low wage rates, because they perform different economic activities or use technologies with different skill requirements. Put differently, in a perfectly-competitive labour market, such worker-to-worker sorting fully explains wage differences between firms. However, since workers’ wages are fully determined by their own skills worker-to-worker sorting has no impact on aggregate wage inequality: higher between-firm wage inequality due to higher skill dispersion between firms is fully offset by lower within-firm wage inequality due to more homogeneous workforces within firms.

In an imperfectly-competitive labour market with frictions, firms and workers bargain over market rents (Pissarides, 2000[9]; Mortensen, 2003[10]). In this case, average pay between firms may differ even when they employ identically-skilled workers because of differences in firm wage premia due to differences in firms’ revenue productivity and/or in the sharing of market rents with workers. For instance, one firm may adopt more advanced technologies than another one employing identically-skilled workers, because it benefits from better access to finance or has reached the minimum scale to cover the fixed cost of adopting advanced technologies. Revenue productivity may also differ between firms with identically-skilled workers because of differences in product market power, which allows some firms to charge higher prices at given technology and may partly reflect product innovation but also barriers to competition due to sunk costs or the policy environment. The scope for firms to align wages with productivity in an imperfectly competitive labour markets may depend on the presence of wage-setting institutions which impose minimum wage floors but in some cases also limit wage growth through coordinated wage bargaining (OECD, 2019[11]).

In reality, pay differences between firms are likely to be explained by both differences in workforce composition and differences in firm wage premia, with worker sorting across firms not only reflecting the clustering of similarly-skilled workers in the same firms (worker-to-worker sorting), but also the concentration of high-skilled workers in the best-performing firms (and of low-skilled workers in low-productivity firms, i.e. worker-to-firm sorting). Worker-to-worker sorting represents specialisation based on the preferences and skills of workers or the technology-based skill requirements of firms. This type of sorting is not driven by differences in pay between firms and does not generate changes in the distribution of productivity-related rents across workers. By contrast, worker-to-firm sorting may result from the presence of firm wage premia, based on complementarities between workers’ skills and firms’ production technology or labour market frictions. Firms may also aim at limiting the sharing of productivity-related rents with low-skilled workers, for instance by outsourcing the least skill-intensive production stages.

Evidence for Germany and the United States suggests that domestic outsourcing of supporting service activities, such as cleaning, security and catering, has contributed to increased worker-to-worker and worker-to-firm sorting (Dorn, Schmieder and Spletzer, 2018[12]; Goldschmidt and Schmieder, 2017[13]). Moreover, improved access to imported inputs and services offshoring have allowed firms to replace tasks previously conducted in-house by imports, making worker skills within firms more homogeneous (Autor, Dorn and Hanson, 2015[14]; Bloom, Draca and Van Reenen, 2016[15]; Carluccio, Fougère and Gautier, 2015[16]; Weil, 2014[17]).

This framework allows for the possibility that rather than being fully determined by workers’ marginal productivity, wages may at least partly be driven by firms’ productivity-related rents. Such rents may affect wage inequality both directly by affecting the dispersion of average wages between firms and indirectly by affecting workers’ incentives to sort across firms with different wage premia. Therefore, worker-centred policies that have traditionally focused on addressing the gap between skill demand and supply may fall short of fully addressing the drivers of wage inequality. Instead, worker-centred policies may need to be complemented with firm-centred policies that address differences in productivity-related rents between firms while supporting overall productivity growth.

2.2.2. Empirical implementation

The analysis of the separate channels underlying aggregate wage inequality is implemented empirically as follows. Wage inequality is measured as the total variance of logarithmic wages, which is additively decomposable, scale independent and provides a more comprehensive measure of inequality compared to partial measures, such as the 90^th/10^th percentile ratio. In a first step, the total variance of wages is decomposed into the variance of average wages between firms and the variance of individual wages within firms. The results from this analysis are presented in Section 2.4 below.

In a second step, the estimation of a traditional human-capital earnings equation augmented with firm-fixed effects allows further decomposing between- and within-firm wage inequality into the four parts highlighted by the analytical framework in Figure 2.1 (Box 2.1):4

(i) the variance of wages at given observable workforce composition (dispersion of firm wage premia);
(ii) the covariance between the predicted wages of workers based on their observable earnings characteristics and firm-specific wage premia (worker-to-firm sorting);
(iii) the covariance between the predicted wages of workers based on their observable earnings characteristics and the firm-level average of predicted wages (worker-to-worker sorting);
(iv) the variance of wages related to workers’ observed and unobserved earnings characteristics and the returns to these characteristics.

The results from this analysis are presented in Section 2.5 below.

Box 2.1. Using a traditional human capital earnings equation to decompose wage inequality

Isolating the contribution of sorting of workers across firms to between- and within-firm wage inequality involves estimating a traditional human capital earnings equation augmented with firm fixed effects (Barth et al., 2016[5]):

ln⁡wij=xiβ+γj+εij

Equation 2.1

where $w_{i j}$ denotes the wage of worker i in firm j; $x_{i}$ denotes a vector of observable worker characteristics; $β$ denotes the estimated return to these characteristics; $γ_{j}$ denotes estimated firm fixed effects; and $ε_{i j}$ denotes the error term. The observable earnings characteristics included in the empirical model generally include education and/or occupation, age, gender, indicators for part-time work and interaction terms between these variables.

Based on Equation 2.1, denoting estimated coefficients and variables with superscript ^ and defining $\hat{s} \equiv x_{i} \hat{β}$ (workers’ predicted wages based on observable earnings characteristics) the total variance of $\ln w_{i j}$ can be written as follows:

Vtotal=V(s^)+V(γ^)+2cov(s^,γ^)+V(ε^)

Equation 2.2

where $V (\hat{s})$ is the variance of predicted wages based on observable earnings characteristics; $V (\hat{γ})$ is the variance of firm-specific wage premia; $c o v (\hat{s}, \hat{γ})$ is the covariance of predicted wages with firm-specific wage premia and $V (\hat{ε})$ is the variance of residual wages.

Defining $ρ_{γ} \equiv \frac{c o v (\hat{s}, \hat{γ})}{V (\hat{s})}$ and $ρ \equiv \frac{c o v (\hat{s}, \hat{S})}{V (\hat{s})}$ , where $\hat{S}$ is the average of all individual workers’ $\hat{s}$ in the firm, the total variance of $\ln w_{i j}$ can be re-written as:

Vtotal=[V(s^)ρ+2V(s^)ργ+V(γ^)]+[V(s^)+V(ε^)-V(s^)ρ]

= Vbetween + Vwithin

Equation 2.3

where $ρ_{γ}$ is the correlation of workers’ predicted wages based on observable earnings characteristics with the estimated firm-fixed effects (a measure of worker-to-firm sorting) and $ρ$ is the correlation of workers’ predicted wages with the average predicted wage in their firm (a measure of worker-to-worker sorting).

The between-firm variance can thus be decomposed into contributions from worker-to-worker sorting $V (\hat{s}) ρ$ , worker-to-firm sorting $2 V (\hat{s}) ρ_{γ}$ and the variance of firm-specific wage premia $V (\hat{γ})$ . The within-firm variance can be decomposed into contributions from the returns to observed and unobserved earnings characteristics $V (\hat{s}) + V (\hat{ε})$ and worker-to-worker sorting $- V (\hat{s}) ρ$ .

The positive contribution of worker-to-worker sorting to overall wage inequality through between-firm wage inequality $V (\hat{s}) ρ$ is exactly offset by the negative contribution through within-firm wage inequality $- V (\hat{s}) ρ$ . This reflects the fact that increased worker-to-worker sorting raises the dispersion of workforce composition between firms but makes workforce composition within firms more homogeneous, with no net effect on overall wage inequality.

The variance of firm-wage premia to overall wage inequality in the above framework represents an upper-bound estimate of its true contribution due to the role of unobservable worker characteristics (as shown in Box 2.4 following Abowd et al. (1999[7]), while it represents a lower bound estimate of the contribution of worker-to-firm sorting due to the presence of sorting on unobservable ability. This issue is particularly pronounced in countries where information on neither occupation nor education are available (Austria, Canada, Estonia and New Zealand).

2.3. Constructing a cross-country dataset based on employer-employee data

In order to empirically quantify the contributions of each of the elements of the above framework to levels and changes in wage inequality and the scope for firm-centred policies, data are needed that map workers to the firms that employ them. The linked employer-employee data used in this project are drawn from administrative records designed for tax or social security purposes or, in a few cases, mandatory employer surveys.5 In most countries, the project takes a distributed micro-data approach that relies on partners based in participating countries to provide relevant aggregations of individual-level data using a harmonised statistical code. In order to develop and test the statistical code, as well as to develop an in-house data infrastructure, the project has also gained direct access to a number of anonymised individual-level data sets.6

Linked employer-employee data have the major advantage of being very comprehensive and, in some cases, covering the entire population of workers and firms in a country. The information is generally also of very high quality, given the potentially important financial or legal implications of reporting errors and extensive administrative procedures for quality control. Since tax and social security systems differ in their administrative requirements across countries, with potentially important implications for their comparability across countries, considerable effort has been made to harmonise the data (see Annex on Data and Disclaimers for on overview of the data used for each country). The analysis is restricted to the private sector and excludes the self-employed, where possible, and own-account workers everywhere by focusing on firms with two employees or more. Including the self-employed and public sector firms would increase the importance of between-firm wage inequality at the expense of the within component, since the self-employed constitute overwhelmingly single-worker firms and the distribution of public sector wages is typically highly compressed. When information on public status is unavailable the “public government and defence” and “education” sectors are excluded. Information on self-employment is not always available, but a large fraction of self-employed workers is excluded by restricting the analysis to firms with at least 2 employees.

The main analysis focuses on total monthly earnings since information on working time is not available in several countries. In an attempt to exclude part-timers, all workers with earnings below 90% of monthly earnings of a full-time worker at minimum wage are dropped and in the absence of a minimum wage, those below 45% of the monthly median wage for a full-time worker. Using hourly wages for the subset of countries where this is possible does not change the main results of this chapter. Earnings information is reported in gross terms, i.e. total labour cost minus employer social security contributions and based on all taxable earnings, including overtime and other bonuses. To deal with the issue of top coding at the contribution threshold in social security data, censored wages are imputed based on regression analysis using the predicted wage and the distribution of estimated error terms based on methods developed by Dustmann et al. (2009[18]) and Card et al. (2013[19]).

The definition of an employer differs across countries. While some datasets link workers to their establishments, others link them to their firms (which may encompass several establishments) or to an administrative reporting unit somewhere between the firm and the establishment (Vilhuber, 2009[20]). Although this could matter for decomposing wage dispersion into between and within-employer components, empirical work suggests that in practice the unit of observation may only have a limited impact on such decompositions. This may partly reflect the fact that most firms have only a single establishment.7 Where both definitions are available, the analysis focuses on firms rather than establishments, which is typically the level at which wages are set.

While the administrative data typically cover the universe of workers and their employers, the data made available for analytical purposes are in some countries based on a representative sample of workers or firms. Worker-based samples only cover a fraction of workers in a firm, introducing measurement error in average firm wages. This tends to bias within-firm wage dispersion down relative to between-firm wage dispersion. The analysis corrects for sampling error in worker-based samples which tends to bias down within-firm wage dispersion relative to between-firm wage dispersion using the correction proposed by (Håkanson, Lindqvist and Vlachos, 2015[21]).

The resulting dataset generally covers the past two decades and is broadly consistent with other national and cross-country data sources in terms of levels and changes in overall wage inequality (Box 2.2).8 Deviations in terms of levels of the 90^th/10^th percentile ratio are generally very small, but there are significant deviations in terms of changes for a number of countries, which may reflect differences in samples or definitions of wages across the two data sources.

Box 2.2. Comparison of wage inequality measures based on LinkEED and official sources

This box assesses the extent to which the patterns in overall wage inequality based on the new linked employer-employee dataset (LinkEED) correspond to those reported by official sources from national agencies or international organisations. Since the variance of wages – the preferred measure of wage inequality used in this chapter – is generally not available from official sources, this is done by comparing the 90th/10th percentile ratio for the latest available year and the change in this ratio between the first and the last available year in both sources (Figure 2.2). Deviations in terms of levels of wage inequality are generally very small, with the correlation between the two data sources being around 0.9. The correlation is somewhat lower in terms of changes (around 0.6), which mainly reflects significant deviations for New Zealand and Sweden. Such deviations could signal differences in samples or wage definitions between the two data sources rather than fundamental disagreement on wage inequality developments. For instance, the European Union Structure of Earnings Survey that underlies the official statistics for European countries in Figure 2.2 only covers a relatively small sample of workers (generally around 5-10%) as opposed to the universe of workers for most countries covered by LinkEED. Moreover, the European Union Structure of Earnings Survey excludes firms with less than 10 employees as opposed to firms with less than 2 employees in LinkEED.

Figure 2.2. Comparison of LinkEED with official statistics

2.4. Key stylised facts on wage inequality between and within firms

A number of stylised facts emerge by decomposing aggregate wage inequality developments according to the analytical framework in Figure 2.1 using the harmonised linked employer-employee data.

2.4.1. Inequality between firms accounts for a sizeable share of the levels and changes in overall wage inequality

On average across countries, the dispersion of average wages between firms accounts for about half of the overall dispersion of wages (Figure 2.3). While the share of between-firm inequality in overall wage inequality in no country falls below 30%, it approaches 70% in some, suggesting that there may be large cross-country differences in terms of worker sorting and the dispersion of firm wage premia.9 These may partly reflect cross-country differences in productivity dispersion between firms, but also the extent to which labour market institutions such as collective bargaining influence the sharing of productivity-related rents with workers.

The orders of magnitude are broadly in line with those of previous studies, which found that wage dispersion between firms accounts for up to 60% of overall wage inequality. Recent research using cross-country data for European countries estimates that wage dispersion between establishments explains around 60% of aggregate wage inequality (International Labour Organization, 2016[27]). A previous cross-country study covering European countries and the United States found that wage dispersion between firms accounts for around 20-40% of aggregate wage inequality (Lazear and Shaw, 2009[28]).10

Changes in the dispersion of average wages between firms also account for around half of changes in overall wage inequality (Figure 2.3, Panel B).11 Except for the United Kingdom, where between-firm inequality has increased despite declining overall wage inequality, in most countries changes in between-firm wage inequality have contributed significantly to overall wage inequality developments, highlighting the crucial of role of firms in aggregate wage inequality developments. Large cross-country differences in absolute changes in wage inequality partly reflect large differences in initial levels, with overall wage inequality typically changing by 10-20% over the sample period (Annex Figure 2.A.1). However, the fact that the direction of changes differs across countries suggests that changes in between-firm wage inequality most likely also reflect differences in the extent to which policies and institutions shape the impact of global trends, such as globalisation and technological change, on worker sorting and inequality in firm-level productivity and wages.

Figure 2.3. A significant share of overall wage inequality is between firms

2.4.2. Between-firm inequality partly reflects differences in workforce composition

Dispersion in average wages between firms partly reflects differences in workforce composition. For instance, high-skilled workers earning high wages may predominantly work in firms that employ other high-skilled workers or pay high wage premia. Defining high-skilled workers based on education or occupation, the evidence suggests that the share of high-skilled workers in high-wage firms is higher than in firms at the bottom of the firm wage distribution (Figure 2.4). On average across countries, in the last year of the sample, the share of high-skilled workers in firms at the top decile of the firm wage distribution was about 32 percentage points higher than in firms at the bottom decile. Moreover, the difference between the top and the bottom decile was about 8 percentage points higher than in the first year of the sample, suggesting that high-skilled workers increasingly cluster in the same firms as firms get more specialised or better-performing firms pay higher wages to attract better workers. Dispersion in average wages between firms partly also reflects the fact that women tend to work in low-wage firms, although this is less the case than about two decades ago (Box 2.3).

Figure 2.4. Skill dispersion across firms is large and increasing

Deviations of shares of high skilled employees from average across firm-wage deciles, in %-points

Box 2.3. Women are increasingly working in high-wage firms

Traditionally, women are much more likely to work in low-wage firms than men (Figure 2.5). About two decades ago, the share of women in the highest-paying firms (top decile of average wages) was about 15 percentage points lower than in the lowest-paying firms (bottom decile), but the difference has shrunk to about 11 percentage points. This likely reflects rising labour market skills among women, the changing nature of high-pay occupations (e.g. manufacturing versus services), a more supportive institutional environment (e.g. working time flexibility, childcare) and reduced gender discrimination as a result of changing social norms, which has increasingly allowed women to find jobs in higher-paying firms. This issue is discussed in more detail in Chapter 5 of this Volume.

Figure 2.5. Women increasingly work in high-wage firms

Deviations of shares of women from average across firm-wage deciles, in %-points

2.5. Decomposition results

2.5.1. Distinguishing between firm-wage premia and worker sorting

The between-firm component of wage inequality can be further decomposed into differences in firm-specific wage premia (due to productivity-related rents) and the sorting of workers into firms paying different average wages.

On average across countries, the dispersion of firm wage premia accounts for around two thirds of the level of between-firm wage inequality while worker sorting across firms accounts for around one third (Figure 2.6). The contribution to changes in between-firm wage inequality over the past 20 years has been similar, suggesting that there has been no major break in the role of firm wage premia over the period.The contribution of firm wage premia to between-firm wage inequality varies substantially across those countries, ranging from about 10% in Sweden to more than 50% in Germany. In Austria, Canada, Estonia and New Zealand, where only information on age and gender is available the estimated contribution of firm wage premia tends to be larger, as differences in occupational or educational composition of workers are incorporated into the estimated firm wage premia.

Accounting for differences in workforce composition between firms related to unobservable earnings characteristics slightly reduces the contribution of firm-wage premia to the overall level of wage dispersion, but has no systematic impact on their contribution to changes in overall wage dispersion (Box 2.4). These results strongly suggest that inequality in average wages between firms does not just reflect differences in workforce composition, but mainly differences in productivity-related rents or the extent to which such rents are shared with workers.

Figure 2.6. Distinguishing between firm-wage premia and worker sorting

Box 2.4. Accounting for unobservable earnings characteristics

Compositional differences between firms may not only relate to workers’ observable earnings characteristics (e.g. age, gender, education and/or occupation) but also unobservable ones (e.g. innate ability or motivation). As a result, the component of wage dispersion associated with firm fixed effects may not just reflect differences in firm wage premia, but also unobservable differences in workforce composition. This is likely to be particularly important for countries with limited information on the skills of workers such as Austria, Canada, Estonia and New Zealand. This box analyses the extent to which accounting for unobserved earnings characteristics affects the estimated contribution of firm-wage premia to the level and change in wage inequality in selected countries.

Accounting for the role of unobservable earnings characteristics for the variance of wages, involves augmenting the human capital earnings equation in Box 2.1 with a person fixed effect using the method developed by Abowd et al. (1999[7]) (henceforth AKM):

ln⁡wijt=xitβ+πi+φj+θt+εijt

Equation 2.4

where $w_{i j t}$ denotes the wage of worker i in firm j at time t; $x_{i t}$ is a vector of observable worker characteristics and $β$ the estimated return to these characteristics; $π_{i}$ , $φ_{j}$ and $θ_{t}$ are person-, firm- and year-fixed effects, respectively; and $ε_{i j}$ is the error term. Since the person fixed effects are identified from worker mobility across firms, Equation 2.4 is estimated over periods of at least five years. The decomposition of the between-firm variance into the components associated with firm-wage premia and sorting is analogous to that described in Box 2.1.

Accounting for unobservable workforce differences between firms typically reduces the contribution of firm-wage premia to the overall level of wage dispersion, but has no systematic impact on the contribution to changes in overall wage dispersion (Figure 2.7). On average, across the countries covered by this analysis, the contribution of firm-wage premia to the level of between-firm wage variance declines by about one-third relative to the baseline model. However, the contribution of changes in firm-wage premia dispersion to changes in overall wage dispersion is typically similar whether or not we account for unobservable worker differences between firms, even in countries with very limited information on observable worker characteristics such as Estonia. The results with respect to worker-to-firm sorting remain broadly unchanged when compared with those obtained by applying the method proposed by Borovičková and Shimer (2017[29]). In sum, these results suggest that sorting of workers across firms based on unobservable characteristics matters significantly for the level of between-firm wage inequality but only marginally for changes in between-firm wage inequality.

Figure 2.7. Contribution of firm-wage premia to variance of wages

A. Levels B. Changes

2.5.2. Dissecting the contribution of sorting to between-firm wage inequality

Turning to the role of worker sorting, the evidence suggests that in many countries sorting has also tended to exacerbate between-firm wage inequality and, to a lesser extent, overall wage inequality developments (recall that only worker-to-firm sorting contributes to overall wage inequality). Moreover, within countries, worker-to-worker sorting and worker-to-firm sorting have often moved in the same direction (Figure 2.8). Thus, from the perspective of firms, specialisation in tasks with different skill requirements – be it to take advantage of pure gains of specialisation or to limit rent-sharing with low-skilled workers – has increased over time. From the perspective of workers, increases in the dispersion of firm-wage premia may also have raised incentives for sorting into higher-paying firms. Consistent with this hypothesis, Spain and Portugal, which are the only countries that experienced declines in the dispersion of firm-wage premia (in the group of countries with measures of occupation and/or education), also experienced a decline in worker-to-firm sorting.

Figure 2.8. Worker-to-worker and worker-to-firm sorting have often moved together

With increased sorting of workers and more homogenous workforces (in terms of observable earnings characteristics), one would expect a declining contribution of within-firm wage differences to inequality (not reported, see Criscuolo et al. (2020[30]). However, many countries have also experienced widening wage gaps within firms. This is because, on average across the countries covered in this chapter, returns to worker skills, which represent the main part of within-firm difference in wages, have increased by around 6 percentage points.12 This points to skill shortages due the failure of education systems to keep pace with developments in demand for certain skills by firms (OECD, 2018[31]; OECD, 2019[32]). For instance, digitalisation may have raised the demand for highly skilled engineers by more than the education system can rapidly supply.

2.6. A tentative exploration of the determinants of firm wage premia dispersion

The variation in the contribution of firm wage premia dispersion to overall wage dispersion across countries raises important questions about the role of policies and institutions. At a given level of labour market frictions, policies and instutions may shape the dispersion of firm productivity and thereby the dispersion of firm wage premia (Andrews, Criscuolo and Gal, 2016[1]). But policies and institutions may also shape the transmission of productivity to firm wage premia at a given level of productivity dispersion, either by affecting the degree of frictions in the labour market or by institutional limits on the dispersion of wage premia.

To provide a first indication of the possible role of policies and institutions in firm wage premia dispersion, Figure 2.9 compares the contribution of wage premia dispersion to overall wage inequality across different groups of countries according to the degree of centralisation of their collective bargaining systems and the degree of voluntary job mobility between firms. These simple descriptive statistics provide a number of insights. First, the share of firm wage premia dispersion is higher in countries with more decentralised collective bargaining institutions (Panel A). These are countries where collective bargaining predominantly takes place at the firm-level or wages are set through individual-level bargaining. Second, conditional on the collective bargaining arrangements, the share of firm wage premia dispersion tends to be higher in countries with low job mobility (Panel B). This is consistent with the view that more productive firms offer higher wage premia to attract and retain the workers required to reach their desired employment levels in a frictional labour market. The results are qualitatively similar when using the level of wage premia dispersion instead of its share in overall wage dispersion.13

Figure 2.9. The role of collective bargaining and job mobility in firm wage premia dispersion

2.7. Conclusion

Wage premia dispersion is in part determined by differences in productivity between firms. This suggests that productivity developments matter for wage inequality, both directly, by affecting firms’ wage-setting practices, and indirectly, by affecting incentives for sorting of workers across firms. The implication is that a better understanding of the factors driving productivity dispersion between firms, the extent to which productivity-related rents are shared with different types of workers, and the effect of these developments on worker sorting across firms are crucial for developing public policies that address concerns around slowing productivity growth and increasing wage inequality.

Apart from directly influencing productivity dispersion between firms, public policies and institutions may also shape the link between productivity and wage premia dispersion. This link is determined by the degree of competition for workers among employers and the presence of institutional constraints on the wage-setting power of firms. The exploratory evidence in this chapter suggests that wage premia differences between firms tend to be more pronounced in countries with decentralised collective bargaining systems, i.e. fewer institutional restrictions on the wage-setting behaviour of firms, as well as countries with higher rates of job mobility and thus stronger competition for workers between firms.

References

[7] Abowd, J., F. Kramarz and D. Margolis (1999), “High wage workers and high wage firms”, Econometrica, Vol. 67/2, pp. 251-333.

[1] Andrews, D., C. Criscuolo and P. Gal (2016), “The Best versus the Rest: The global productivity slowdown, divergence across firms and the role of public policy”, OECD Productivity Working Papers, Vol. 2.

[14] Autor, D., D. Dorn and G. Hanson (2015), “Untangling Trade and Technology: Evidence from Local Labor Markets”, The Economic Journal, Vol. 125, pp. 621-646.

[33] Autor, D. et al. (2020), “The Fall of the Labor Share and the Rise of Superstar Firms”, Quaterly Journal of Economics, Vol. 135/2, pp. 645-709.

[5] Barth, E. et al. (2016), “It’s Where You Work: Increases in the Dispersion of Earnings across Establishments and Individuals in the United States”, Journal of Labor Economics, Vol. 34/S2, pp. S67-S97, http://dx.doi.org/10.1086/684045.

[6] Barth, E., J. Davis and R. Freeman (2018), “Augmenting the Human Capital Earnings Equation with Measures of Where People Work”, Journal of Labor Economics, Vol. 36/S1, pp. S71-S97.

[2] Berlingieri, G., P. Blanchenay and C. Criscuolo (2017), The great divergence(s), OECD Science, Technology and Innovation Policy Papers.

[3] Berlingieri, G. et al. (2019), Last but not least: Laggard firms, technology diffusion, and its structural policy determinants.

[15] Bloom, N., M. Draca and J. Van Reenen (2016), “Trade Induced Technical Change”, The Review of Economic Studies, Vol. 83/1, pp. 87-117.

[29] Borovičková, K. and R. Shimer (2017), “High Wage Workers Work for High Wage Firms”, NBER Working Paper Series, No. 24074, National Bureau of Economic Research, Cambridge, MA, http://dx.doi.org/10.3386/w24074.

[19] Card, D., J. Heining and P. Kline (2013), “Workplace Heterogeneity and the rise of West German wage inequality”, The Quarterly Journal of Economics, Vol. 123/3, pp. 967-1015.

[16] Carluccio, J., D. Fougère and E. Gautier (2015), “Trade, Wages and Collective Bargaining: Evidence from France”, The Economic Journal, Vol. 125/584, pp. 803-837.

[30] Criscuolo, C. et al. (2020), “Workforce composition, productivity and pay: The role of firms in wage inequality”, OECD Social, Employment and Migration Working Papers, No. 241, OECD Publishing, Paris, https://dx.doi.org/10.1787/0830227e-en.

[12] Dorn, D., J. Schmieder and J. Spletzer (2018), Domestic Outsourcing of Labor Services in the United States: 1996-2015.

[18] Dustmann, C., J. Ludsteck and U. Schönberg (2009), “Revisiting the German Wage Structure”, The Quarterly Journal of Economics, Vol. 124/2, pp. 843-881.

[26] Eurostat (2017), Structure of Earnings Survey: monthly earnings (http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=earn_ses_monthly).

[23] Federal Reserve Bank of St. Louis (2019), Usual weekly nominal earnings (first decile): wage and salary workers: 16 years and over (https://fred.stlouisfed.org/series/LEU0252911200Q).

[22] Federal Reserve Bank of St. Louis (2019), Usual weekly nominal earnings (ninth decile): wage and salary workers: 16 years and over (https://fred.stlouisfed.org/series/LEU0252911500Q).

[37] Garnero, A., A. Hijzen and S. Martin (2019), “More Unequal, But More Mobile? Earnings Inequality and Mobility in OECD Countries”, Labour Economics, Vol. 56, pp. 26-35.

[13] Goldschmidt, D. and J. Schmieder (2017), “The Rise of Domestic Outsourcing and the Evolution of the German Wage Structure”, Quarterly Journal of Economics, Vol. 132/3, pp. 1165-1217.

[21] Håkanson, C., E. Lindqvist and J. Vlachos (2015), “Firms and skills: the evolution of worker sorting”, IFAU Working Paper 9, pp. 1-84.

[27] International Labour Organization (2016), Global Wage Report 2016/17: Wage Inequality in the workplace.

[34] Kehrig, M. and N. Vincent (2019), “Good Dispersion, Bad Dispersion”, NBER Working Paper, Vol. No. 25923.

[28] Lazear, E. and K. Shaw (eds.) (2009), Wage Structure, Raises and Mobility: An Introduction to International Comparisons of the Structure of Wages Within and Across Firms, University of Chicago Press.

[10] Mortensen, D. (2003), Wage dispersion: Why are similar workers paid differently?, MIT Press.

[25] OECD (2019), Decile ratios of gross earnings (https://stats.oecd.org/Index.aspx?DatasetCode=DEC_I).

[11] OECD (2019), Negotiating Our Way Up: Collective Bargaining in a Changing World of Work, OECD Publishing, Paris, https://dx.doi.org/10.1787/1fd2da34-en.

[32] OECD (2019), OECD Skills Outlook 2019 : Thriving in a Digital World, OECD Publishing, Paris, https://dx.doi.org/10.1787/df80bc12-en.

[31] OECD (2018), Skills for Jobs Database.

[38] OECD (2015), The Quality of Working Lives, OECD Publishing.

[9] Pissarides, C. (2000), Equilibrium unemployment theory, MIT Press.

[35] Schwellnus, C. et al. (2018), “Labour share developments over the past two decades: The role of technological progress, globalisation and “winner-takes-most” dynamics”, OECD Economics Department Working Papers 1503.

[36] Skans, O., P. Edin and B. Holmlund (2009), Wage Dispersion Between and Within Plants: Sweden 1985-2000, University of Chicago Press.

[8] Song, J. et al. (2019), “Firming Up Inequality”, The Quarterly Journal of Economics, Vol. 134/1, pp. 1-50.

[24] Statistic Bureau of Japan (2019), Basic Wage Structure (https://www.mhlw.go.jp/english/database/db-l/wage-structure.html).

[4] Tomaskovic-Devey, D. et al. (2020), “Rising between-workplace inequalities in high-income countries”, Proceedings of the National Academy of Sciences of the United States of America, Vol. 117/17, pp. 9277-9283, http://dx.doi.org/10.1073/pnas.1918249117.

[20] Vilhuber, L. (2009), Adjusting Imperfect Data: Overview and Case Studies, University of Chicago Press.

[17] Weil, D. (2014), The Fissured Workplace: Why Work Became So Bad for So Many and What Can Be Done to Improve It, Harvard University Press.

Annex 2.A. Additional material

Annex Figure 2.A.1. Total log wage variance, all years and countries

Notes

← 1. This chapter has been written by an OECD team consisting of Chiara Criscuolo, Alexander Hijzen, and Cyrille Schwellnus with contributions of: Erling Barth (Institute for Social Research Oslo, NORWAY), Antoine Bertheau (University of Copenhagen, DENMARK), Wen-Hao Chen (Statcan, CANADA), Richard Fabling (independent, NEW ZEALAND), Priscilla Fialho (OECD, PORTUGAL), Katarzyna Grabska-Romagosa (Maastricht University, NETHERLANDS), Ryo Kambayashi (Hitotsubashi University, JAPAN), Valerie Lankester and Catalina Sandoval (Central Bank of Costa Rica, COSTA RICA), Michael Koelle (OECD), Timo Leidecker (OECD), Balazs Murakőzy (University of Liverpool, HUNGARY), Oskar Nordström Skans (Uppsala University, SWEDEN), Satu Nurmi (Statistics Finland/VATT, FINLAND), Vladimir Peciar (Ministry of Finance, SLOVAK REPUBLIC), Capucine Riom (LSE, FRANCE), Duncan Roth (IAB, GERMANY), Balazs Stadler (OECD), Richard Upward (University of Nottingham, UNITED KINGDOM) and Wouter Zwysen (ETUI, formerly OECD). For details on the data used in this chapter please see the standalone Data Annex and Disclaimer Annex.

← 2. The role of firms in determining the labour share has, for instance, been the subject of Autor et al. (2020[33]), Kehrig and Vincent (2019[34]) and Schwellnus et al. (2018[35]). The role of firms in determining wage inequality has, for instance, been the subject of Barth et al. (2016[5]; 2018[6]) and Song et al. (2019[8]).

← 3. Two notable exceptions are provided by Lazear and Shaw (2009[28]) and Tomaskovic-Devey et al. (2020[4]). Lazear and Shaw (2009[28]) focus largely on an earlier period that complements the analysis in the present paper and only make limited efforts to ensure results are comparable across countries. Tomaskovic-Devey et al. (2020[4]) focus on a similar period for 14 countries, but importantly from the perspective of the present paper do not account for the role of worker composition for wage differences between firms.

← 4. As a robustness check, Box 2.4 further augments the human capital earnings equation by including worker in addition to firm fixed effects (Abowd, Kramarz and Margolis, 1999[7]).

← 5. This is, for instance, the case in the United Kingdom.

← 6. The project currently has direct access to linked employer-employee data for Estonia, France, Italy and Spain.

← 7. Barth et al. (2018[6]) based on US data and Skans, Edin, and Holmlund (2009[36]) based on Swedish data show that the within-firm between-establishment variance in earnings is very small. Similarly, Song et al. (2019[8]) show that almost all of the increase in earnings inequality occurred between firms rather than between establishments within firms.

← 8. In a number of countries, including Japan and Norway, the sample period is significantly shorter than two decades, implying that overall changes in wage inequality may not be directly comparable across all countries.

← 9. The measurement of wage inequality in Japan is particularly sensitive to the inclusion of part-time workers because their average hourly wages are lower than those of full-time workers. When including all workers, wage inequality is among the highest in the OECD (OECD, 2015[38]; Garnero, Hijzen and Martin, 2019[37]). However, when focusing on full-time workers only, wage inequality in Japan is around the OECD average (Figure 2.2).

← 10. International Labour Organization (2016[27]) covers a limited sample period (2002-2010) using the European Structure of Earnings Survey data that consists of repeated cross sections of random samples of workers and their establishments. Lazear and Shaw (2009[28]) use national administrative data but do not cover the past two decades (their sample period typically covers 1980-2000).

← 11. Consistent with these results, most of the available evidence suggests that changes in wage dispersion between firms account for at least 60-70% of changes in overall wage dispersion (Lazear and Shaw, 2009[28]). Annual changes in wage dispersion are reported in Annex Figure 2.A.1.

← 12. The skill premium is defined as the wage gap between high-skilled and low-skilled workers (based on occupation or educational attainment) controlling for other earnings characteristics in Equation 2.1. The estimated increase of 6 percentage points is based on regressing the skill premium on a linear time trend and country fixed effects and using the estimated coefficient on the linear trend to predict the average gaps in 1990 and 2016. The sample for these regressions includes France (2002 to 2015), Italy (1991 to 2015), Japan (2005 to 2013), Netherlands (2001 to 2016), Norway (2004 to 2014), Portugal (1995 to 2009), Spain (1996 to 2016), Sweden (1999 to 2015), and the United Kingdom (1998 to 2018).

← 13. The role of wage-setting institutions and job mobility is analysed in more detail in Chapter 3 combining the present data on firm wage premia dispersion with data on productivity dispersion at the industry level in a regression framework. The results of that analysis confirm the insights obtained from the descriptive statistics presented here.

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Countries

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Publications

Publications

Featured publications

Data

Data

Featured data

News & Events

News & Events

Featured Events

About OECD

About

Engage with us

Work with us

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment