FIRM WAGE PREMIA, INDUSTRIAL RELATIONS,

AND RENT SHARING IN GERMANY

BORIS HIRSCH AND STEFFEN MUELLER*

The authors use three distinct methods to investigate the influence

of industrial relations on firm wage premia in Germany. First, ordi-
nary least squares (OLS) regressions for the firm effects from a two-
way fixed-effects decomposition of workers’ wages reveal that aver-
age premia are larger in firms bound by collective agreements and
in firms with a works council, holding constant firm performance.
Next, recentered influence function (RIF) regressions show that
premia are less dispersed among covered firms but more dispersed
among firms with a works council. Finally, in an Oaxaca–Blinder
decomposition, the authors find that decreasing bargaining cover-
age is the only factor they consider that contributes to the marked
rise in premia dispersion over time.

A substantial body of research documents that firms pay dissimilar wages

to similar workers. Based on the two-way fixed-effects methodology by
Abowd, Kramarz, and Margolis (1999; AKM hereafter), several studies have
found that 15 to 20% of the total variance in wages can be attributed to firm
wage premia, that is, to wage differences that are left after differences in
workers’ human capital and unobservable skills have been rewarded.1 Such
1
Among these studies are Abowd, Lengermann, and McKinney (2003) for the United States; Card,
Heining, and Kline (2013) for West Germany; Card, Cardoso, and Kline (2016) for Portugal; and Macis
and Schivardi (2016) for Italy.

*BORIS HIRSCH ( https://orcid.org/0000-0002-6328-4931) is Professor of Economics at Leuphana

University of Lüneburg and Research Fellow to Halle Institute for Economic Research (IWH) and IZA
Institute of Labor Economics. STEFFEN MUELLER is Head of the Department of Structural Change and
Productivity at Halle Institute for Economic Research (IWH), Professor of Economics at University of
Magdeburg, and Research Fellow to CESifo.
We thank Paul Beaudry, Michael Böhm, Alex Bryson, Michael Burda, Stefano Collonello, Bernd
Fitzenberger, Hanna Frings, Giovanni Gallipoli, Nicole Gürtzgen, Florian Hoffmann, Pat Kline, Claus
Schnabel, Heiko Stüber, and two referees for very useful suggestions. We further appreciate comments
by participants of the SOLE 2017 and VfS 2017 conferences and the Pakt Project Workshop on ‘‘Worker
Flows, Match Quality, and Productivity’’ and by seminar participants in Berkeley, Berlin, Lüneburg, and
Vancouver. Last but not least, we thank Alexander Giebler and Georg Märker for excellent research
assistance. An Online Appendix is available at http://journals.sagepub.com/doi/suppl/10.1177/
0019793920917105. For information regarding the data and/or computer programs used for this study,
please address correspondence to [email protected] or [email protected].

KEYWORDs: firm wage premium, industrial relations, trade unions, works councils, bargaining power, rent
sharing, wage inequality, Germany

ILR Review, 73(5), October 2020, pp. 1119–1146

DOI: 10.1177/0019793920917105. Ó The Author(s) 2020
Journal website: journals.sagepub.com/home/ilr
Article reuse guidelines: sagepub.com/journals-permissions
1120 ILR REVIEW

persistent firm wage premia are at odds with a competitive labor market in
which firms consider wages as given. Instead, the premia point at substantial
employment rents that are split between workers and firms.
Differences in firm wage premia may stem, on the one hand, from
differences in the firms’ surpluses to be shared between workers and firms
and, on the other hand, from differences in firms’ bargaining power in the
wage-formation process. Such differences in bargaining power, in turn, may
be shaped by firms’ industrial relations regime. Specifically, we expect
workers’ bargaining power to be higher when unions are present or when
worker codetermination exists through works councils. Conditional on the
amount of rents to be shared among the firm and its workforce, the exis-
tence of these institutions should thus yield higher firm wage premia. We
would arrive at the same predictions using an efficiency wage model, such
as Shapiro and Stiglitz’s (1984) canonical shirking model. In such a setting,
unions and works councils are expected to protect workers from being
dismissed by employers (as is found empirically for works councils by
Hirsch, Schank, and Schnabel 2010), thereby forcing employers to pay
higher efficiency wages in order to incentivize workers to exert effort. We
consider this effect of union and works council existence to be an improve-
ment in workers’ implicit bargaining position. We therefore regard this set-
ting as similar in spirit to the rent-sharing setting with explicit bargaining
whereby union presence or works council existence are expected to
strengthen workers’ bargaining power directly via collective action.
Evidence on the sources of firm wage premia in general and, more spe-
cifically, on the role of bargaining power is still sparse, however. In particu-
lar, we lack evidence on how firm wage premia differ between firms bound
by collective agreements and uncovered firms and on how plant-level code-
termination affects wage premia. One notable exception is Card, Heining,
and Kline (2013; CHK hereafter) who applied the AKM approach to West
German data. They were not only the first to document that rising disper-
sion in firm wage premia or, as they put it, rising workplace wage inequality,
is a major contributor to increasing wage inequality in West Germany, but
they also provided some suggestive evidence that the dispersion of premia is
larger among those firms that are unbound by collective agreements. The
rising dispersion of firm wage premia over time may thus reflect the falling
prevalence of collective agreements or works councils provided that these
institutions compress the premia distribution, which seems plausible. That
said, we lack evidence whether bargaining coverage or worker codetermina-
tion exerts a differential impact on firm wage premia along the premia dis-
tribution. Another study, by Goldschmidt and Schmieder (2017), found
that domestic outsourcing of cleaning, security, and logistics services led to
more dispersed firm wage premia that, in turn, explained 9% of the rise in
overall wage inequality in Germany. Changes in jobs’ industrial relations
regime triggered by outsourcing may to some extent be the source of this
finding.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1121

There exist both a large body of evidence (surveyed by Card, Cardoso,

Heining, and Kline 2018) on the relationship between firm performance
(measured as either productivity, profits, or quasi rents) and wages as well
as considerable research into the wage effects of unions (Bryson 2014) and
works councils (e.g., Addison, Teixeira, and Zwick 2010). As these two
strands of the literature investigate wages rather than firm wage premia,
however, they cannot shed light on how unions and worker codetermina-
tion affect these premia. In particular, wage differences across firms may
stem from workers with distinct unobserved abilities sorting into firms that
differ in performance and their industrial relations regimes, thereby con-
taminating estimates.
Against this backdrop, this article investigates the influence of industrial
relations on firm wage premia and whether the falling prevalence of collec-
tive bargaining and worker codetermination (Oberfichtner and Schnabel
2019) are likely candidates for explaining the rising dispersion of firm wage
premia in West Germany over time. To measure firm wage premia, we rely
on the CHK plant wage effect (in our data we observe plants rather than
firms) obtained from an AKM-type two-way fixed-effects approach. The
CHK plant wage effect, which is essential for our application, measures the
wage premium enjoyed by all workers in a plant’s workforce adjusted for
observed and unobserved worker quality. This approach therefore does not
suffer from worker sorting that is likely to contaminate other wage
measures, such as the wage bill per worker or individual workers’ wages, as
used in previous literature on union wage differentials. It captures that the
difference in wage premia paid by a worker’s current employer and a partic-
ular future employer equals the expected wage change when moving
between these two employers.
Our analysis proceeds in three steps. First, we analyze how the mean
plant wage premium varies with the plant’s industrial relations regime, hold-
ing constant plant performance. Specifically, we regress the CHK plant wage
effect on dummies that indicate coverage by a collective agreement and
works council existence as well as the plant-level quasi rent, that is, the sur-
plus left after the factors of production have been paid their outside options
(e.g., Abowd and Lemieux 1993), as a performance measure. Second, we
investigate whether the impact of the industrial relations regime on wage
premia changes over time by redoing our analysis from the first step sepa-
rately for the beginning and the end of our observational window. Third,
we examine how the industrial relations regime affects the dispersion of
plant wage premia and whether it exerts a differential impact on the lower
and the upper tail of the premium distribution. We further investigate
whether part of the rise in the premia variance that took place between
1994 and 2009 can be attributed to the falling prevalence of collective
agreements and works councils during that period.
1122 ILR REVIEW

Institutional Setting and Some Hypotheses

In Germany, the principle of bargaining autonomy gives trade unions and
employers the right to regulate wages and working conditions absent state
interference. Collective agreements are legally binding, are predominantly
concluded as multi-employer agreements between a single union and an
employers association at the sectoral level, and almost always apply to all of
the covered employers’ workers irrespective of workers’ union status. (More
recently, some employers’ associations have started to offer memberships
that do not force members to adopt collective agreements [Behrens 2013].)
Although sectoral negotiations mostly take place in regional bargaining
units, officials of the two bargaining parties closely coordinate the regional
negotiations within one sector, and consequently variations between them
are small. Furthermore, some cross-sectoral coordination by both parties
occurs, giving rise to some uniformity in collective bargaining policy across
sectors (for more details, see Hirsch and Schnabel 2014).
Collective bargaining in Germany predominantly concerns wages but also
determines job classifications, working time, and working conditions.
Norms stipulated in the collective agreement are generally minimum terms,
and employers bound by the agreements thus cannot undercut but only
improve upon these terms and conditions. Exceptions to this general rule
are in some cases laid down in so-called opening or derogation clauses that
allow renegotiating collective bargaining issues, mostly wages and working
time, at the plant level, typically under conditions of economic hardship.
Whereas a minority of employers do in fact pay wages higher than stipu-
lated in the collective agreements (for details on this wage cushion, see
Jung and Schnabel 2011), and opening clauses have grown in importance
in the past three decades, for most workers the wages set in the collective
agreements are crucial for the level and development of their actual wages.
In 2015, 59% of workers and 31% of plants were covered by collective
agreements in West Germany, which is the focus of our analysis (Ellguth
and Kohaut 2016). Moreover, 51% of West German workers employed by
uncovered plants were covered indirectly by collective agreements because
43% of uncovered plants report to act upon a sector-level agreement
(Ellguth and Kohaut 2016).
On average, plants covered by a collective agreement have a higher value
added per worker (Mueller 2011) and also pay higher wages (Gürtzgen
2009). Furthermore, studies by Dustmann, Ludsteck, and Schönberg
(2009), Biewen and Seckler (2019), and Baumgarten, Felbermayr, and
Lehwald (2020) have found that declining collective bargaining coverage
can account for a substantial part of the rise in German wage inequality.
Yet, evidence on these links between collective bargaining coverage and
wages does not necessarily carry over to plant wage premia. Further, these
findings may suffer from unobserved differences in worker quality between
bound and unbound plants. As a case in point, workers of better quality,
who arguably possess a stronger bargaining position than do less qualified
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1123

workers, may select into uncovered plants that are more flexible when
deciding on plant wage premia. Or, workers may sort into covered plants if
these have larger rents to be distributed.
Moreover, since collective agreements are predominantly concluded at
the sectoral level, they not only allow plants to save on transaction costs
(and thus to save on total labor costs despite higher wage bills) but also
offer plants employing workers with high bargaining power the opportunity
of hiding behind the average employer in collective wage negotiations. On
account of this ‘‘hide effect,’’ plants that employ workers with high
bargaining power may even pay lower wage premia when covered by a col-
lective agreement, thereby hiding behind the collective wage.2 We thus
expect the impact of collective bargaining coverage to be less pronounced
(or even negative) in the upper part of the wage premium distribution than
in the lower part, thereby compressing the premium distribution.
Investigating plant wage premia instead of wages permits us to test for these
possibilities.
Beyond collective bargaining typically conducted at the sectoral level, the
second backbone of Germany’s dual system of industrial relations is defined
by worker codetermination at the plant level through works councils, the
German counterpart of the workplace union in other countries. Works
councils are mandatory but not automatic in all plants with at least five per-
manent workers, since setting up a works council requires three workers or
a union representative to initiate an election procedure in the plant. In
2015, 42% of workers in West Germany worked in the 9% of plants with a
works council (Ellguth and Kohaut 2016).
Although works councils are formally independent of unions, in practice
most works councilors are union members (Behrens 2009). The size of the
works council is an increasing function of the plant’s employment level, and
the entire cost of the works council apparatus is borne by the employer,
with works councilors being exempted from work once certain plant size
threshold levels are reached. Yet, works councils are absent in many eligible
plants. Because actions by employers to block the introduction of a works
council are legally prohibited, since works councilors enjoy additional
employment protection, and since the time spent on work as a works coun-
cilor counts as regular working time, this absence implies that introducing a
works council imposes some cost on workers. This cost comprises, for
instance, the costs of becoming exposed as a works councilor, as many
employers have reservations against works councils, as well as the costs of
actively representing one’s colleagues’ interests and being personally
responsible for the negotiation outcomes. For a recent discussion of the

2
This possibility of hiding behind a sector-level collective agreement oriented toward less productive
and usually smaller employers in an industry has been recognized before in the industrial relations litera-
ture (e.g., Kohaut and Schnabel 2003). In a case study of a large firm in the German metalworking
industry, Arrowsmith, Marginson, and Sisson (2003) reported that the firm is committed to sector-level
bargaining because it feels that the powerful metalworkers’ union would achieve more otherwise.
1124 ILR REVIEW

causes of employer resistance against the introduction of works council, see

Mueller and Stegmaier (2020).
Works councils have extensive information, consultation, and codetermi-
nation rights (for details, see Addison 2009). In particular, and in contrast
to continental European counterparts of workplace representation,
German works councils have codetermination rights on what are termed
‘‘social matters,’’ which comprise remuneration arrangements, the com-
mencement and termination of working hours, the regulation of overtime
and reduced working hours, as well as health and safety measures. Works
councils can also negotiate social plans, which establish compensation for
the dislocation caused by (partial) plant closings and by major changes in
plant organization. Unlike unions, though, works councils may not call a
strike and they are excluded from reaching agreement with the employer
on wages and working conditions that are settled or normally settled by col-
lective agreements between unions and employer associations at the sec-
toral level. One exception to this general rule is that collective agreements
may contain opening or derogation clauses (as previously mentioned) that
explicitly authorize works councils to do so.
Even if opening clauses are absent, however, works councils’ extensive
information, consultation, and codetermination rights on many other issues
mean that works council existence is likely to improve workers’ bargaining
power and thus to spur rent-seeking activities. In line with this conjecture,
Addison, Schnabel, and Wagner (2001) showed that plants with works
councils pay higher effective wages (see also Addison et al. 2010). They also
found that these plants are less satisfied with their profit situation, whereas
Mueller (2011) showed they are in fact more profitable, which is readily
explained by a positive productivity effect originating in works councils’ col-
lective voice (Addison et al. 2001; Mueller 2012; as well as Mueller and
Stegmaier 2017) that dominates their adverse rent-seeking activities.
As in the case of union wage effects, however, this evidence on the works
councils’ effect on wages does not necessarily carry over to plant wage
premia, and it may be contaminated by worker sorting on unobservables.
Furthermore, Germany saw an increased decentralization in the wage for-
mation process since the mid-1990s, which is considered an essential ingre-
dient in Germany’s recovery from the ‘‘sick man of Europe’’ to an
‘‘economic superstar’’ (Dustmann, Fitzenberger, Schönberg, and Spitz-
Oener 2014: 168) and which should have increased works councils’ say in
wage formation. We thus expect not only a positive association between
works council existence and wage premia for a given plant performance but
also a rise in this positive association over time.
That said, we have no clear indication of whether councils compress or
widen the premia distribution. On the one hand, if workforces who possess
marked bargaining power from the outset and who are thus working for
high-premia employers use plant-level codetermination as a means to foster
rent extraction, works council existence will yield more dispersed premia.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1125

On the other hand, if workers facing a powerful employer, who are arguably
to be found in the lower part of the premium distribution, use plant-level
codetermination to ensure a minimum premium level, works council exis-
tence will compress the premia distribution from below. We regard it an
empirical question which of these two effects dominates.

Data and Empirical Approach

The LIAB Cross-Sectional Model
Our data come from the LIAB cross-sectional model, a linked employer–
employee data set for Germany provided by the Institute for Employment
Research (IAB) and described in Alda, Bender, and Gartner (2005). The
LIAB merges worker-level social security information with an establishment
panel survey, the IAB Establishment Panel described in Ellguth, Kohaut,
and Möller (2014). Starting in 1993, the IAB Establishment Panel has sur-
veyed West German plants from all industries that employ at least one
worker covered by the social security system on June 30 of the survey year
and is representative of the population of these establishments.3 Crucial for
our purpose, the IAB Establishment Panel contains information on plants’
value added, employment, wages, capital stock, and industrial relations
regime (i.e., collective bargaining coverage and works council existence).
To make use of this information in our application, we exclude plants in
the agricultural and mining sectors, as well as the financial and public
sectors, publicly owned plants, and plants with fewer than five employees.
Hence, our findings apply to plants in nonfinancial for-profit sectors that
meet the legal minimum plant size requirement for a works council to be
elected by workers.
Although the IAB Establishment Panel has no direct information on
plants’ capital stock, it can readily be computed from the included invest-
ment data as put forward by Mueller (2008). Note that data on wages and
employment are also contained in the social security data. In our analysis,
we use the wage and employment information from the survey, which is
arguably more consistent to the survey information on plants’ value added
and capital stock that we use to construct quasi rents than is the worker-level
information aggregated at the plant level. What is more, wages in the social
security data are top-coded at the contribution ceiling to the unemployment
insurance, whereas the survey data contain plants’ uncensored monthly
wage bill.
Since the worker-level data of the LIAB stem from the same source as
was used by CHK in their AKM-type wage decomposition, the LIAB data

3
The IAB Establishment Panel is among the first large-scale establishment surveys in Germany. As the
first wave of a newly established survey may suffer from a variety of issues, such as inexperienced
interviewers and interviewees and uncertainty regarding the extent of non-response, we decided against
using the first wave in our analysis. Including it, however, leaves our regression results qualitatively
unchanged.
1126 ILR REVIEW

allow us to merge the plant fixed effects from CHK, which will be our mea-
sure of plant wage premia (as detailed below). We further make use of the
administrative data to construct a consistent sector classification as put for-
ward by Eberle, Jacobebbinghaus, Ludsteck, and Witter (2011) and use the
information on plant age and detailed plant location contained in these
data. To minimize the impact of possible outliers in the survey data, we
truncate the bottom and top 1% of the distributions of value added per
worker and capital costs per worker within any two-digit industry–year cell.
For the same reason, we truncate the top 1% of the distribution of the wage
bill per worker and, similar to CHK, dropped plants that pay less than e 10
per worker as daily wage in the administrative data.
The IAB Establishment Panel oversamples large plants in all waves of the
survey. Because the number of sampled plants rises over time, and since this
rise is attributable to increased numbers of medium-sized and small plants,
the extent of oversampling of large plants drops over time. This change not
only leads to a drop in average plant size but also affects other plant
characteristics as, for example, smaller plants are on average less productive
and pay lower wages than do larger plants. For this reason, we present
results from population-weighted samples only. Besides, information on
plants’ wage cushions and on the distinction between sector-level and firm-
level bargaining, which would be of potential interest for our investigation,
is not available for some early waves of the survey data.4 Lest we lose too
many observations, we do not use this information.5

Measuring Wage Premia

Most of the literature on how rents are shared between workers and
employers (recently summarized by Card et al. 2018) relates workers’ wages
measured at the industry, firm, or individual level to some performance
measure, such as productivity, profits, or quasi rents. Using wages, however,
ignores that workers with high unobserved abilities may sort into jobs at
high-performance employers (as, for example, shown by CHK), thereby
overestimating the extent of rent sharing. Put differently, workers with high
unobserved abilities are expected to earn higher wages because they increase

4
Information on collective bargaining coverage and works council existence is available for the first
wave of the IAB Establishment Panel but is missing in some of the following waves. Since both variables
are highly persistent over time, we imputed missing values with the value of adjacent observations for the
same plant provided that these are unchanged. A minor complication arises because in the 1993 wave,
which we use together with the 1995 wave to impute missing information on wage bargaining in the
1994 wave, the survey asked for the existence of collective agreements at the sector but not at the firm
level. Because of the very high prevalence of collective wage bargaining in sectors having a collective
agreement in 1993 and since we condition on an unchanged collective wage bargaining status in 1995,
we believe that our imputation strategy does not introduce serious measurement error, in spite of the
imprecise question in the 1993 wave.
5
As suggested by a referee, we checked whether our results change when we exclude from our sample
the small number of plants in which collective agreements were negotiated at the firm level. As shown in
the final robustness check in Online Appendix C, our results did not change.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1127

performance, which results in a spurious positive relationship between

wages and performance even in the absence of rent sharing. Indeed, Card
et al. (2018) demonstrated that the rent-sharing coefficient obtained from
regressing wages on some performance measure mixes up returns to
observed worker skills, returns to unobserved worker skills, and true rent
sharing. Using employer wage premia that come from an AKM-type decom-
position approach rather than wages avoids the bias from unobserved
worker heterogeneity because these premia are net of observed and unob-
served worker quality as detailed below.
Our measure of the plant wage premium therefore builds on the decom-
position approach of AKM, which splits up individual wages into worker-
specific and plant-specific components. In the AKM framework, the log
wage of worker i in period t is decomposed as

ð1Þ log wit = ai + uj ði, t Þ + xit0 b + uit

where ai is a fixed log wage component specific to worker i, uj ði, t Þ is a fixed

log wage component specific to plant j employing worker i at time t, xit0 b is
a time-varying log wage component stemming from time-varying worker
characteristics xit that are rewarded equally across plants, and uit is an idio-
syncratic log wage component. In the AKM framework, ai reflects the
worker’s time-invariant human capital, such as education and ability, and
xit0 b mirrors the worker’s time-varying human capital, such as experience,
that affects his or her productivity regardless of where the job is held, while
uj ði, t Þ gives the percentage wage premium enjoyed by every worker
employed at plant j.
The crucial assumption for this interpretation of the AKM decomposition
to hold is that the idiosyncratic log wage component uit is unrelated to the
sequence of worker i’s employers fj ði, t Þgt . In their estimation of the AKM
model for West Germany, CHK not only demonstrate that this decomposi-
tion accounts for approximately 90% of the variation in the wages of full-
time workers but also present evidence that the assumptions of AKM hold.
Specifically, they show that match-specific log wage components are negligi-
ble in size and unrelated to the direction of worker flows across employers.
Hence, the AKM decomposition presented in Equation (1) provides a suit-
able approximation of the wage structure in West Germany and allows iden-
tification of the worker and plant wage components. Since CHK also
demonstrate that workers with distinct skills obtain approximately the same
log wage premium at a given plant, the interpretation of uj ði, t Þ as plant j’s
wage premium net of differences in observed and unobserved worker qual-
ity is supported. The estimated worker and plant effects from CHK are avail-
able to the scientific community (Card, Heining, and Kline 2015), and we
use their plant effects as our measure of plant wage premia.
1128 ILR REVIEW

Measuring Quasi Rents

Rents to be shared between workers and employers consist of the surplus
that is generated by the plant, and our aim is to control for this surplus to
investigate how industrial relations shape rent splitting among employers
and their workers. A proper measure of this surplus is the quasi rent, that is,
value added net of the costs for the competitively priced capital and labor
inputs (e.g., Abowd and Lemieux 1993). To arrive at the quasi rent, it is
necessary to properly account for workers’ outside options that, in turn,
depend on their unobserved abilities. Typically, studies that use quasi rents as
a performance measure proxy outside options by average wages in the same
industry (e.g., Abowd and Lemieux 1993; Van Reenen 1996; and Gürtzgen
2009). As with analyzing wages rather than wage premia, however, measuring
firm performance using the one-size-fits-all quasi rent neglects differences
in workforce characteristics and, thus, in workers’ outside options.
Depending on the available information on workforce characteristics, this
can lead to very poor estimates of the quasi rent, most likely to bias
estimates toward zero.
To our knowledge, Abowd and Allain (1996) are the only researchers to
account for workers’ unobserved abilities by basing their measure of quasi
rents on the worker wage effects from an AKM-type decomposition
approach. Even the most recent study by Card et al. (2018), which used firm
wage premia from an AKM-type decomposition as a dependent variable,
sticks to value added as a profitability measure and thus does not correct for
worker sorting and the resulting heterogeneity in workers’ outside options
across plants. To arrive at a proper measure of rents that accounts for
worker sorting and the resulting heterogeneity in workers’ outside options
across plants, we follow Abowd and Allain’s approach when constructing
the quasi rent. The quasi rent at plant level QR is defined as

ð2Þ QR = VAD  xN  rK

where VAD denotes the plant’s value added, that is, sales net of the value of
intermediates; x denotes workers’ outside wage; N denotes the plant’s num-
ber of workers; r denotes the competitive rental price of capital, which we
compute from the plant’s capital stock and in so doing distinguish between
prices for debt and prices for equity at the sector level; and K denotes the
plant’s capital stock.6

6
The IAB data do not contain information on debt and equity financing at the plant level. We there-
fore use the sector-level information provided by Aswath Damodaran at http://pages.stern.nyu.edu/
~adamodar/. Specifically, we assess data on the ‘‘costs of capital by industry sector’’ for Europe issued on
January 5, 2016. Using additional data on the long-run treasury bond rate for Germany gives an average
rental rate of capital of 10.8% during 1994–1996, 9.2% during 1997–2002, and 8.1% during 2003–2009.
In contrast to the quasi rent definition in Equation (2), workers and employers may split rents before
deducting the capital costs. Following the approach outlined in Card, Devicienti, and Maida (2014), we
did not find evidence for bargaining over the competitive returns to capital, which is in line with their
findings for Italy.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1129

When constructing workers’ alternative wage x we again follow Abowd

and Allain (1996) and calculate workers’ outside options as
 
wage

ð3Þ log x = log  s  ðu
+ aj  a  s  us10 Þ,
N s

separately for each of the three time intervals 1994–1996, 1997–2002, and
2003–2009, for which CHK provide worker and plant effects that can be
wage
merged to the establishment survey data. In Equation (3), N s is the aver-
age plant-level wage per worker in the respective one-digit sector, aj is the
average CHK worker effect (log wage component) in plant j, a  s is the aver-
age CHK worker effect in the one-digit sector, us10 is the 10th percentile of
the distribution of the CHK plant effects in the one-digit sector, and u  s is
the average CHK plant effect in the one-digit sector.7
In Equation (3), the term aj  a  s captures the deviation (in logs) in
worker quality between plant j and the sector average and thus accounts for
unobserved ability differences of employers’ workforces. Yet, note that CHK
estimate worker and plant effects separately for male and female workers
and thus the level of both worker and plant effects cannot be meaning-
fully compared across the genders. As a consequence, computing average
plant-level worker (plant) effects by simply averaging over the worker
(plant) effects at the plant level would not account for differences in the
gender composition of the workforce, and thus such an aggregation
would be uninformative on the average worker (plant) effect that mirrors
the average workforce ability (the plant wage premium). We therefore
purge within each year–gender cell the year–gender mean of the worker
(plant) effect from the individual worker (plant) effect and use these
demeaned CHK effects when computing the average worker (plant)
effect at the plant level.
Besides, CHK calculate worker effects for full-time workers only. Hence,
aj  a s is unknown for part-time workers, who make up 14% of the workers
in our sample. Under the assumption that full-time workers and part-time
workers
holding jobs
at the same plant did not differ in aj  a  s , that is
aj  a s FT = aj  a s PT , our measure of workers’ outside options would
not depend on the share of part-time workers in the plant’s workforce. We
impose this assumption in the following, and we provide additional
explanations and checks of robustness in Online Appendix A.
The outside options of plants’ workers also depend on the wage premia
paid by possible future
  employers.
In Equation (3), subtracting the term
 s  us10 from log wage
u N + aj  a s means subtracting the spread between
s
the average CHK plant effect in the respective one-digit sector and the 10th

7
As we discussed, previous studies typically use average industry wages, that is, the first component in
Equation (3), as their measure of workers’ outside options and thus account for neither the differences
in the quality of workforces nor the spread of plant wage premia in the respective sector.
1130 ILR REVIEW

percentile of the plant effects there. In so doing, we assume that workers

are risk averse and, independently of the wage premium obtained from
their current employer, expect to receive just a modest pay premium at the
10th percentile of the premium distribution when changing employers.
That workers, in their expectations, consider random draws from the pre-
mium distribution is plausible within the AKM framework, in which the
same plant pay premium is enjoyed by all workers of a certain plant and is
thus, in particular, independent of individual workers’ characteristics. We
impose this assumption in the following, and we will provide additional
explanations and checks for robustness in Online Appendix B.8 Also, the
choice of the 10th percentile of the sector-level plant premium distribution
as risk-averse workers’ reference point is somewhat arbitrary. We therefore
experiment with different choices, such as the 25th percentile of this distri-
bution, and obtain very similar results (see Online Appendix B).
CHK estimate plant wage and worker wage effects for four time intervals,
whereas we use those for the latter three time intervals, 1990–1996, 1996–
2002, and 2002–2009, and merge these to the plant-level survey information
for the years 1994–2009. Note that within these time intervals, CHK normal-
ize their plant wage effects by omitting the last plant dummy in their wage
equation (CHK 2013: 988). In other words, plant wage effects have to be
interpreted ‘‘relative to the last plant in the sample,’’ and for this reason
the level of these effects jumps across time intervals and has no clear-cut
interpretation. Because of this feature of the CHK plant wage effects (and
because of transitory fluctuation in average plant-level wages) it is not possi-
ble to infer workers’ outside options by simply subtracting the CHK plant
effect from the average plant-level log wage. Notwithstanding, within the
three time intervals differences in the CHK wage effects across plants can
be readily interpreted, as can their variances and other distributional
parameters, such as the interdecile range, across time intervals. Against this
background, another useful feature of subtracting the spread u  s  us10 in
Equation (3) is the invariance of workers’ outside options to the jumps of
the level of the CHK plant wage effects across time intervals, so that we can
meaningfully interpret the evolution of quasi rents over time.

Wage Premium Regressions

To investigate how mean plant wage premia vary depending on plants’
industrial relations regime, we regress the CHK plant wage effect uj on
dummies for collective bargaining coverage CWBj and works council

8
One objection to our approach is that some bargaining models consider receiving unemployment
benefits rather than holding alternative jobs as the relevant outside option available to workers. Note,
however, that the wage cut induced by subtracting u  s  us10 in Equation (3) roughly mimics the gap
between continued wage payments and receipt of unemployment benefits. At the end of our observa-
tional window, the replacement rate of the German unemployment insurance was 60–67% of the net
wage in the previous employment, which corresponds well with the average u  s  us10 of –0.37 in our
sample.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1131

existence WOCOj , on plant j’s quasi rent QRj calculated as in Equation (2)
to control for plant performance, and on a rich set of additional plant-level
controls.9 Our baseline specification is thus given by

ð4Þ uj = b0 + b1 CWBj + b2 WOCOj + b3 QRj + Controlsj0 g + ej

where the additional controls comprise 32 district dummies (Regierungsbezirke),

62 two-digit sector dummies, four plant-size dummies, four plant-age
dummies, a dummy indicating plants that do not belong to a multi-branch
company, and the percentages of women as well as part-time workers in the
plant’s workforce.10 Hence, conditioning on the quasi rent allows us to assess
whether union presence or works council existence raise workers’ bargaining
power in that they are associated with higher mean wage premia for a given
surplus to be split between the employer and the workers. We stress again that
both our wage premium and our rent measure are corrected for plant-level
differences in observed and unobserved worker quality and thus do not suffer
from bias originating in worker sorting.
We estimate Equation (4) by ordinary least squares (OLS), pooling the
observations from all three time intervals as well as separately for each of
the three time intervals 1994–1996, 1997–2002, and 2003–2009, whereby the
latter allows us to analyze whether the influence of the industrial relations
regime on average plant wage premia changes over time. Before pooling
data over the time intervals we purge the sample means in each interval
from the CHK plant wage effects. Doing so takes account of the jumps in
the levels of the plant effects attributable to CHK’s normalization, which we
discussed in detail in the previous subsection. To examine how industrial
relations shape the dispersion of wage premia across plants, we run uncon-
ditional quantile regressions using the recentered influence function (RIF)
approach (Firpo, Fortin, and Lemieux 2009) that regresses the variance of

9
As an alternative measure of the plant wage premium, we also experimented with using the differ-
ence between the average plant-level wage bill per worker net of the plant-level outside options of
workers. Note that this alternative measure is highly correlated with the CHK plant wage effect and using
it as alternative regressand did not change any of our conclusions. For the sake of comparability with the
existing literature, such as CHK, though, we decided against using this alternative measure in our main
specifications.
10
Apart from these controls, other plant characteristics contained in the IAB Establishment Panel
might be of potential interest, such as information on plants’ export activity, foreign ownership, the wage
cushion, or whether there exist formal profit-sharing agreements. Yet, including these variables would
lead to a massive drop in observations and is thus not viable in our application as some of them have a
considerable extent of missing values and others were completely absent in some of the waves of the sur-
vey. As stressed by a referee, though, exporters have been shown to pay higher wages compared to
nonexporters (e.g., Schank, Schnabel, and Wagner 2007, as well as Macis and Schivardi 2016) so that not
controlling for exporter status in the plant wage premia regressions may arouse concerns. Reassuringly,
running OLS wage premium regressions that control for exporting activity on the reduced sample where
this piece of information is available leaves the coefficients of the industrial relations dummies and the
quasi rent almost unchanged compared to Table 4. We therefore think that neglecting this control is
unlikely to introduce bias in the estimates of interest.
1132 ILR REVIEW

Table 1. Descriptive Statistics at Plant Level

1994–1996 1997–2002 2003–2009

Mean SD Mean SD Mean SD

In e 1,000 per year and worker

Wage 23.4 10.8 25.4 12.2 25.5 12.8
Value added 51.4 44.9 48.7 48.1 48.5 39.9
Workers’ outside wage 18.7 5.1 20.4 5.6 19.6 6.2
Costs of capital 10.2 15.4 7.8 12.0 7.0 10.6
Quasi rent 22.6 44.3 20.4 46.3 21.9 38.7
Other plant characteristics
Percentage part-time workers 18.3 20.1 18.3 20.9 21.1 21.2
Percentage female workers 33.9 27.6 34.3 27.4 36.3 27.5
Collective wage agreement 0.76 0.43 0.61 0.49 0.48 0.50
Works council 0.13 0.34 0.11 0.31 0.09 0.29
Single plant firm 0.88 0.33 0.88 0.33 0.86 0.35
Plant size:
 10 workers 0.50 0.50 0.47 0.50 0.45 0.50
11–50 workers 0.40 0.49 0.44 0.50 0.45 0.50
51–250 workers 0.09 0.28 0.09 0.28 0.09 0.28
. 250 workers 0.01 0.12 0.01 0.11 0.01 0.12
Observations 495,880 1,508,950 1,824,611
Plants (unweighted) 1,213 4,977 6,079

Notes: IAB Establishment Panel, 1994–2009, West Germany. Final regression sample. Weighted using
sample weights. SD, standard deviation.

Table 2. Standard Deviations of the Plant Wage Effects in CHK and LIAB Data

(1) (2) (3)

1990–1996 1997–2002 2003–2009 (3) – (1)

CHK (worker level, male workers only) 0.172 0.194 0.230 0.058
LIAB (plant level, final sample, weighted) 0.202 0.221 0.262 0.060

Notes: The first row refers to table III in CHK. The second row refers to our final sample for West
Germany encompassing the years 1994–2009 that aggregates male and female plant wage effects from
CHK at plant level in the LIAB data and weights the data using the sample weights of the IAB
Establishment Panel. CHK, Card, Heining, and Kline (2013); LIAB, Linked employer–employee data
set of the Institute for Employment Research.

the premia as well as their first and ninth deciles and their median on the
same set of regressors as in Equation (4).

Sample Description
For selective descriptive statistics on our final regression samples in the
three time intervals 1994–1996, 1997–2002, and 2003–2009, for which CHK
provide separate plant wage effects, see Table 1. We see a strong decline in
union coverage and, less so, in works council prevalence. Table 2 compares
the dispersion of the CHK plant wage effects at the worker level, that is, as
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1133

Table 3. Dispersion of the Plant Wage Effects

(1) (2) (3)

1994–1996 1997–2002 2003–2009 (3) – (1)

90–10 gap 0.435 0.496 0.570 0.135

90–50 gap 0.180 0.198 0.234 0.054
50–10 gap 0.255 0.298 0.336 0.081
75–25 gap 0.219 0.241 0.287 0.068

Notes: IAB Establishment Panel, 1994–2009, West Germany, and CHK plant wage effects provided by
Card, Heining, and Kline (2015). Final regression sample; weighted using sample weights. Numbers
shown are dispersion measures of the plant wage effects. For example, the 90–10 gap refers to the
difference between the ninth and the first decile of the distribution of the CHK plant wage effects.

Figure 1. Kernel Density Plots of the Demeaned CHK Plant Wage Effects in West Germany
in the 1994–1996, 1997–2002, and 2003–2009 Periods

Notes: CHK, Card, Heining, and Kline (2013).

in the original CHK article, with the dispersion in our final regression sam-
ple (using the sample weights of the IAB Establishment Panel). Notably, the
dispersion of the CHK plant effects at the plant level is similar to their dis-
persion at the worker level. Further, at both levels of aggregation their stan-
dard deviation rises by approximately 6 log points, or roughly one-third,
during our observational window, which reflects the rise in workplace wage
heterogeneity at the heart of CHK’s contribution.
The rise in the dispersion of wage premia over time is also evident in
Figure 1, which shows kernel density plots of the demeaned wage premia
for the three time intervals. Notably, not only does the spread of premia
increase over time but also the lower tail of the distribution seems to widen
more than its upper tail. This latter finding is further substantiated by Table
3, which reports several dispersion measures of the wage premia and their
1134 ILR REVIEW

Table 4. Wage Premium OLS Regressions

(1) (2) (3) (4)

1994–2009 1994–1996 1997–2002 2003–2009

Collective wage agreement 0.019** –0.012 0.012 0.035***

(0.008) (0.025) (0.011) (0.012)
Works council 0.050*** 0.044*** 0.030* 0.068***
(0.010) (0.017) (0.015) (0.012)
Quasi rent per worker (in e 100,000) 0.063*** 0.058*** 0.052*** 0.086***
(0.007) (0.016) (0.009) (0.012)
Observations 3,829,441 495,880 1,508,950 1,824,611
Plants (unweighted) 9,054 1,273 4,977 6,079
R2 0.24 0.35 0.37 0.26

Notes: IAB Establishment Panel, 1994–2009, West Germany, and CHK plant wage effects provided by
Card, Heining, and Kline (2015). Weighted using sample weights. The regressand is the CHK plant
wage effect. The control variables consist of 32 district dummies (Regierungsbezirke), 62 two-digit sector
dummies, four plant-size dummies, four plant-age dummies, a dummy for a single plant (as opposed to
a plant belonging to a multi-branch company), and the percentages of women as well as part-time
workers in the plant’s workforce. Standard errors (in parentheses) are clustered at the plant level. OLS,
ordinary least squares.
***; **; * indicate statistical significance at the 1; 5; 10% level.

evolution over time. What emerges is that in all three time intervals the pre-
mium inequality is stronger in the lower part of the premium distribution
than in the upper part, as is reflected by the larger 50–10 gap vis-á-vis the
90–50 gap. Further, all measures of wage premia dispersion rise steadily
over time, with dispersion at the lower part of the distribution increasing
most. Between the 1994–1996 and the 2003–2009 intervals, the 90–10 gap
widens by 13.5 log points, with 8.1 log points originating in the rising 50–10
gap, but just 5.4 log points in the increasing 90–50 gap. These observations
suggest that candidate explanations for the rising wage premia dispersion
are likely to be found among those factors influencing the lower tail of the
premium distribution.

Results
OLS Wage Premium Regressions
Table 4 presents the core results of a group of OLS regressions in which we
regress the CHK plant wage effect on dummies for collective bargaining
coverage and works council existence, on the plant’s quasi rent to control
for the surplus to be split between the employer and the workers, and the
additional controls detailed above. When pooling observations for the
entire observation period 1994–2009, works council existence is associated
with a sizeable increase in the premium that is statistically significant at the
1% level.11 In plants with a works council, the wage premium is 5 log points

11
Note that plant wage premia are estimated by CHK using the universe of German plants (rather
than a mere sample). Hence, sampling error is not a concern in our application and we thus do not have
to correct the standard errors of our regression coefficients.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1135

larger, holding constant quasi rents and the other controls included. The
association between collective bargaining coverage and the wage premium
is less pronounced and only statistically significant at the 5% level. Plants
bound by a collective agreement pay an additional premium of 1.9 log
points, ceteris paribus, which is a bit more than one-third of the works coun-
cil effect.
Turning to our control for plant performance, the relation between the
wage premium and the quasi rent is statistically significant at the 1% level.
An increase in the quasi rent per worker by e 100,000 is associated with a
rise in the wage premium by 6.3 log points, which means that a rise in the
quasi rent by one standard deviation (i.e., roughly e 40,000 per worker)
comes along with an increase in the premium by just one-ninth of a stan-
dard deviation (i.e., approximately 2.5 log points). Or, to make our result
more comparable to existing studies on rent sharing, one may note that the
average quasi rent per worker is about e 22,000 so that a 1% increase in the
quasi rent is associated with an increase in the plant wage premium by
0.014% for the average worker. This rent-sharing elasticity is at the lower
end of the range of estimates of earlier studies surveyed by Card et al.
(2018). Yet, one should keep in mind that those studies analyzed wages
rather than wage premia and may thus suffer from worker sorting, thereby
producing upward-biased rent-sharing elasticities (Card et al. 2018).
Although statistically significant, we conclude that the association between
plants’ quasi rent and the plant wage premium is insignificant from an eco-
nomic point of view. Overall, whereas plant performance seems to exert just
a minor influence on the level of plant wage premia, plants’ industrial
relations regime plays a much larger role, in particular plant-level worker
codetermination through works councils.12
Since the influence of industrial relations on the average wage premium
is conditional on the plant’s quasi rent, our findings suggest that union
presence and works council existence increase workers’ bargaining power
in that they raise the share of the surplus going to workers.13 An alternative

12
As suggested by a referee, we also checked whether the association between the mean plant wage
premium and the quasi rent changes with plant’s industrial relations regime by interacting the dummies
for works council existence and collective bargaining coverage with the demeaned quasi rent. Doing so
had no impact on the estimated coefficients of the two dummies, so that the impact of industrial
relations is the same as before for plants with an average quasi rent (for results, consult Online
Appendix D). We further found that the interaction terms are negative and similar in magnitude, albeit
modest in size, suggesting that the impact of works councils and collective bargaining agreements on the
plant wage premium is somewhat smaller in plants in which there are more rents to be distributed.
Because these plants are plants with higher wage premia, this result is consistent with our later findings
that both institutions seem to ensure a minimum wage premium level in low-premium plants, as
reported in our RIF wage premium regressions section.
13
As we previously discussed, a similar explanation could be given based on efficiency wage models in
which union presence or works council existence forces employers to pay higher efficiency wages to
incentivize workers, who are thus in a better implicit bargaining position. We regard this explanation as
similar in spirit to one based on explicit bargaining between the employer and the workforce whereby
the existence of both institutions increase workers’ bargaining power directly via collective action.
1136 ILR REVIEW

explanation for employer wage premia pursued in the literature are com-
pensating wage differentials, although the conventional view is that compen-
sating differentials are rather unimportant (e.g., Hornstein, Krusell, and
Violante 2011). That said, a recent study by Sorkin (2018) building on the
AKM framework to measure the employer wage components argued that
compensating wage differentials are one important source of employer
wage premia. Could our findings, then, reflect compensating differentials
rather than differences in workers’ bargaining power related to the indus-
trial relations regime?
According to the exit–voice model of unionism by Freeman and Medoff
(1984), plant-level unions and thus works councils as their German counter-
part play two distinct roles. On the one hand, they act as a collective-voice
institution that enables workers to safely express their dissatisfaction with
certain working conditions instead of quitting the job. If the employer, in
turn, listens to workers’ voice and improves on working conditions accord-
ingly, costly turnover will be avoided, worker morale will be higher, and
labor productivity may rise. On the other hand, works councils raise
workers’ bargaining power and may engage in rent-seeking activities at the
detriment of the employer.
In line with works councils’ collective-voice role, several studies have
documented that quits are lower in plants in which a works council is pres-
ent (e.g., Hirsch et al. 2010, as well as Pfeifer 2010), suggesting that working
conditions are better in these plants and that works councils positively influ-
ence workers’ job satisfaction. In light of this evidence, it seems implausible
that works councils give rise to positive compensating differentials, implying
that the positive influence of works council existence on employer wage
premia found may even understate its impact on workers’ bargaining
power. The same line of argument applies to union presence, although we
have the impression that compensating differentials are even less plausible
when it comes to collective bargaining that is typically conducted at the sec-
toral level as opposed to plant-level worker codetermination through works
councils.
In the subset of plants in our sample for which we have information on
average working hours, we also checked whether higher hours accompany
higher plant wage premia, which would then arguably compensate workers
for long working hours. In line with earlier evidence for Portugal by Card
et al. (2016), we found no significant relationship between working hours
and wage premia. We see this non-finding as further evidence that compen-
sating differentials are unlikely to be of much importance for our main
estimates, particularly because our regression controls for an extensive set
of covariates, such as sector and plant location, that are likely to pick up a
large part of the possible differences in compensating differentials across
plants.
When we run separate regressions for the three time intervals 1994–1996,
1997–2002, and 2003–2009, for which CHK provide separate plant wage
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1137

effects, we find that the impact of works council existence and collective
bargaining coverage on the plant wage premium is largest in the latest
2003–2009 interval. Specifically, the additional wage premium when work-
ing for a plant bound by a collective agreement is near zero in the earlier
intervals and amounts to 3.5 log points in this latter period, which turns out
to be statistically significant at the 1% level. Similarly, the additional wage
premium associated with a works council rises somewhat over time, as does
the relation between the plant’s quasi rent and the wage premium, which
nevertheless remains very small in magnitude. Besides, the potential impact
of the increase in the quasi rent coefficient on the evolution of premia
inequality is balanced to some extent by a drop in the dispersion of quasi
rents across plants, which is lowest (roughly 20% less compared to the inter-
mediate 1997–2002 interval) at the end of our observational window.
In summary, we find that plants with a works council and plants bound
by collective agreements pay larger wage premia conditionally on plants’
quasi rent, which suggests that these institutions raise workers’ bargaining
power thereby allowing them to extract a larger share of a given surplus.
Our results further show that the influence of industrial relations on the
average plant wage premium rises over time and is most pronounced at the
end of our observational window. In terms of the level of the plant wage
premium, where one is working thus gains in importance over time.
As plant wage premia have also become more dispersed over time, our
findings, in turn, lead to the question of whether recent trends in industrial
relations, such as the rise of the so-called codetermination-free zone
with neither collective bargaining nor works councils (for details, see
Oberfichtner and Schnabel 2019), may have contributed to the steady rise
in the dispersion of wage premia across plants observed by CHK and also
present in our data (see Tables 2 and 3). But before we turn to this ques-
tion, we have to check how industrial relations relate to the wage premia
dispersion.

RIF Wage Premium Regressions

To investigate the influence of industrial relations on the dispersion of wage
premia across plants, we run RIF regressions (Firpo et al. 2009) for the vari-
ance, the median, and the first and ninth decile of the wage premium distri-
bution including the same set of regressors as in the OLS regressions. Table
5 presents the core results of a group of RIF regressions when pooling
observations for 1994–2009 as well as when running separate regressions for
the three time intervals 1994–1996, 1997–2002, and 2003–2009.
Our overall finding is that collective bargaining coverage and works coun-
cil existence differ in their influence on the dispersion of plant wage
premia. In the pooled sample comprising the years 1994–2009, works coun-
cil existence is accompanied by a larger variance of wage premia, which is
statistically significant at the 1% level. The presence of works councils is
1138 ILR REVIEW

Table 5. Wage Premium RIF Regressions

(1) (2) (3) (4)

1994–2009 1994–1996 1997–2002 2003–2009

Variance
Collective wage agreement –0.013*** –0.015 0.001 –0.014*
(0.005) (0.014) (0.007) (0.008)
Works council 0.018*** 0.003 0.020** 0.016*
(0.005) (0.011) (0.009) (0.008)
Quasi rent per worker (in e 100,000) –0.010** –0.007 –0.009** –0.013
(0.004) (0.008) (0.004) (0.010)
First decile
Collective wage agreement 0.051** 0.004 0.010 0.071**
(0.024) (0.068) (0.050) (0.035)
Works council 0.007 0.022 –0.017 0.051*
(0.019) (0.048) (0.040) (0.027)
Quasi rent per worker (in e 100,000) 0.069*** 0.057 0.067 0.081**
(0.023) (0.056) (0.048) (0.035)
Median
Collective wage agreement 0.020** –0.002 0.018 0.037***
(0.008) (0.023) (0.012) (0.011)
Works council 0.071*** 0.056** 0.060*** 0.077***
(0.011) (0.027) (0.017) (0.013)
Quasi rent per worker (in e 100,000) 0.057*** 0.054*** 0.054*** 0.071***
(0.010) (0.018) (0.015) (0.013)
Ninth decile
Collective wage agreement –0.004 –0.037 0.012 0.012
(0.008) (0.028) (0.010) (0.011)
Works council 0.076*** 0.029 0.056*** 0.070***
(0.015) (0.032) (0.019) (0.023)
Quasi rent per worker (in e 100,000) 0.052*** 0.015 0.022** 0.072***
(0.011) (0.017) (0.010) (0.017)
Observations 3,829,441 495,880 1,508,950 1,824,611
Plants (unweighted) 9,054 1,273 4,977 6,079

Notes: IAB Establishment Panel, 1994–2009, West Germany, and CHK plant wage effects provided by
Card, Heining, and Kline (2015). Weighted using sample weights. The regressand is the respective
parameter of the distribution of the CHK plant wage effects. The control variables consist of 32 district
dummies (Regierungsbezirke), 62 two-digit sector dummies, four plant-size dummies, four plant-age
dummies, a dummy for a single plant (as opposed to a plant belonging to a multi-branch company),
and the percentages of women as well as part-time workers in the plant’s workforce. Standard errors (in
parentheses) come from a block bootstrap at plant level with 500 replications. RIF, recentered
influence function.
***; **; * indicate statistical significance at the 1; 5; 10% level.

associated with an increase in the ninth decile (median) of wage premia by

7.6 (7.1) log points, which is statistically significant at the 1% level, whereas
there exists no such association for the first decile. As a result, the wage pre-
mium distribution is wider if works councils are present, ceteris paribus,
with the middle and upper parts of the distribution being lifted. Running
separate RIF regressions for the three time intervals 1994–1996, 1997–2002,
and 2003–2009, we find that this pattern of a differential influence of works
council existence on the lower and the upper part of the premium distribu-
tion is least pronounced in the early 1994–1996 period.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1139

In interpreting these findings, it is important to bear in mind that the

RIF regressions provide the impact of works council existence on the
respective unconditional quantile of the wage premium distribution that
combines both between-group and within-group effects (Firpo et al. 2009).
To learn more about the factors that likely contribute to our findings, we
run conditional quantile regressions (with the same covariates) that capture
the within-group effect and compare those estimates to the unconditional
quantile regressions from Table 5 (for the results, consult Online Appendix
D). We find a positive association between works councils and wage premia,
that is, a positive within-group effect, at all quantiles considered and that
the within-group effect is largest at the ninth decile, which suggests that the
total (unconditional) effect is partly driven by a larger conditional effect at
the top of the distribution.
Together, these results are in line with the notion that works councils use
their local bargaining power to capture additional rents in plants that pay
nontrivial wage premia and are increasingly able to do so more recently,
likely because of rising decentralization in the wage-formation process.
Given our previous discussion, our reading of this result is that workforces
who enjoy pronounced bargaining power from the outset and who are thus
working for employers that pay substantial wage premia, use plant-level
codetermination as a means to foster rent extraction. What is more, our
result of a positive within-group effect at the first decile also lends support
to the possibility that workers who face a powerful employer that pays a low
premium use works councils to ensure a minimum premium level. That
being said, the fall in plant-level worker codetermination in recent years
(see Ellguth and Kohaut 2016; Oberfichtner and Schnabel 2019) is ruled
out as a plausible explanation for rising premium dispersion because falling
prevalence of works councils is expected to narrow the premium distribu-
tion rather than to widen it.
Our results for worker codetermination through works councils contrast
with those for collective bargaining coverage. In the pooled sample, cover-
age by a collective agreement is associated with a lower variance of wage
premia, which is statistically significant at the 1% level. Whereas collective
agreements are associated with a 5.1 (2.0) log points larger first decile
(median) of wage premia, which is statistically significant at the 5% level,
there is no such association with the ninth decile. As a result, bargaining
coverage compresses the wage premium distribution from below.
Conditional quantile regressions show that this finding is driven by a
within-group effect of collective agreements that is largest at the first decile
of wage premia and virtually zero at the ninth decile (see Online Appendix
D). Hence, our results are clearly in line with collective agreements that set-
tle minimum terms that are more binding for low-premium plants and less
so for high-premium plants, which may even exploit collective agreements
1140 ILR REVIEW

to ‘‘hide’’ behind the collective wage.14 Running separate RIF regressions

for the three time intervals 1994–1996, 1997–2002, and 2003–2009, we see
that this pattern of a differential influence of bargaining coverage on the
first and ninth deciles of the plant wage premium is most pronounced in
the most recent 2003–2009 period.
Turning to our control for plant performance, a larger quasi rent is asso-
ciated with a lower premia variance in the 1994–2009 pooled sample, which
is statistically significant at the 5% level, and a lower interdecile range of
wage premia. The latter finding is visible from the somewhat larger influ-
ence of the quasi rent on the first decile compared to the ninth decile of
the premium distribution. An increase in the quasi rent per worker by
e 100,000 is associated with a rise in the first (ninth) decile of the premium
distribution by 6.9 (5.2) log points, which is statistically significant at the 1%
level.15 Running separate RIF regressions for the three time intervals, we
see no clear changes in the influence of quasi rents on wage premia
dispersion.
In short, the RIF regressions show that the level of quasi rents has almost
no influence on the dispersion of wage premia. What is more, because of
rents’ minor influence on the mean premium found in the OLS
regressions, changes in their variance offer no plausible point of departure
for explaining the rise in the premia dispersion that we observe.

Decomposition of the Rise in the Wage Premia Variance

Together, our findings make clear that falling collective bargaining cover-
age in recent years (see Ellguth and Kohaut 2016; Oberfichtner and
Schnabel 2019) offers one plausible explanation for the rise in the disper-
sion of wage premia across plants, as also suggested by CHK, and why the
premia dispersion increased most at the bottom of the premium distribu-
tion (see Table 3). This result additionally aligns with existing evidence for
wages (rather than wage premia) by Dustmann et al. (2009), Biewen and
Seckler (2019), and Baumgarten et al. (2020) that the fall in collective
bargaining coverage can account for a substantial part of the rise in
German wage inequality.
To assess to what extent the declining prevalence of collective
agreements and works councils has contributed to the rise in wage premia
dispersion, we finally decompose the increase in the premia variance into
two parts: one part that is explained by changes in any of the plant
characteristics included in our previous regressions and one part that is

14
Our finding of a positive within-group effect of both collective agreements and works councils at the
first decile of the wage premium distribution aligns with our earlier results that suggest mean wage
premia are less influenced by these institutions in plants with more rents to be distributed. If unions and
works councils ensure a minimum wage premium level, we expect them to have a larger impact in plants
with small rents that, in turn, are paying low wage premia.
15
Matano and Naticchioni (2017) found a decreasing rent-sharing intensity along the wage distribution
in Italy.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1141

unexplained. The Oaxaca–Blinder decomposition is based on two separate

RIF regressions for the premium variance in the 1994–1996 and 2003–2009
intervals, in which the later period serves as the reference category in the
decomposition. Yet, we do not perform the Oaxaca–Blinder decomposition
directly on the RIFs of the variances in the two time intervals. The reason
for this approach is that the linear approximation involved in using the
RIFs is only locally valid and may thus result in a poor approximation and a
large specification error in our application when we observe large changes
in the distribution of the covariates over time. Instead, we follow Firpo,
Fortin, and Lemieux’s (2018) suggestion and perform the RIF decomposi-
tion after reweighting the distribution of covariates in the earlier period to
that of the later period.16 To arrive at this counterfactual sample, we apply
the reweighting approach by DiNardo, Fortin, and Lemieux (1996), which
involves estimating the propensity score that we obtain from a logit model
with the same covariates as in our RIF regressions and additional interaction
terms with the plant’s industrial relations regime.
Table 6 shows that roughly one-third of the rise in the premium variance
of 0.028 is attributable to changing plant characteristics. In other words, the
increase in the premium variance would have been 0.0182 instead of 0.0280
if all the plant characteristics had remained unchanged. The decline in
bargaining coverage contributes 40% of the explained variance change and
thus accounts for 14% of the overall variance increase. As we expected from
the previous results in Table 5, declining works council prevalence
contributes negatively to the explained variance change. Yet, its contribu-
tion is small and statistically insignificant, which is readily explained by the
rather modest decline in works council prevalence in absolute terms (see
Table 1). (Choosing the 1994–1996 interval as the reference category yields
similar results and thus does not change any of our conclusions.) Finally,
the specification error stemming from the local linearity of the RIF
regressions is negligible as is the reweighting error, pointing at a good qual-
ity of the first-step reweighting.

Conclusions
Linking employer survey data to administrative data for West Germany for
the years 1994–2009, we have investigated how the level and the dispersion
of wage premia across plants depend on their industrial relations regime,
conditional on plant performance. To measure wage premia, we used the
plant wage effects of CHK that stem from an AKM-type two-way fixed-effects
decomposition of individual workers’ log wages. Hence, our wage premium
measure gives the wage premium enjoyed by every worker employed by a
certain plant and accounts for worker sorting on unobservable worker

16
We thank one of our referees for pointing out this problem and for suggesting the two-step
approach by Firpo et al. (2018) as a remedy.
1142 ILR REVIEW

Table 6. Oaxaca–Blinder Decomposition of the Change in the Wage Premia

Variance between the 1994–1996 and 2003–2009 Intervals

Variance Percentage of the raw differential

Variance 2003–2009 0.0687***

(0.0036)
Variance 1994–1996 0.0407***
(0.0054)
Raw differential in variances 0.0280*** 100
(0.0064)
Explained part of the differential 0.0098** 35
(0.0045)
Unexplained part of the differential 0.0192** 68
(0.0078)
Specification error –0.0005 –2
(0.0027)
Reweighting error –0.0004 –1
(0.0023)
Contribution to the explained part . . .
Collective wage agreement 0.0040* 14
(0.0022)
Works council –0.0008 –3
(0.0007)
Quasi rent per worker (in e 100,000) 0.0002 1
(0.0004)
Observations 2,320,491
Plants (unweighted) 6,958

Notes: IAB Establishment Panel, 1994–2009, West Germany, and CHK plant wage effects provided by
Card, Heining, and Kline (2015). Weighted using sample weights. Oaxaca–Blinder decomposition
based on recentered influence function (RIF) regressions of the wage effects variance with 2003–2009
as the reference period, following Firpo, Fortin, and Lemieux’s (2018) two-step approach that performs
the RIF decomposition after reweighting the distribution of covariates in the earlier period to that of
the later period. The control variables consist of 32 district dummies (Regierungsbezirke), 62 two-digit
sector dummies, four plant-size dummies, four plant-age dummies, a dummy for a single plant (as
opposed to a plant belonging to a multi-branch company), and the percentages of women as well as
part-time workers in the plant’s workforce. The estimated propensity score for the reweighting is
obtained from a logit model including all these covariates as well as additional interactions of these with
the industrial relations dummies. Standard errors (in parentheses) come from a block bootstrap at
plant level with 500 replications.
***; **; * indicate statistical significance at the 1; 5; 10% level.

characteristics into plants that may have contaminated prior studies based
on workers’ individual wages rather than plant wage premia. In our econo-
metric analysis, we regressed the plant wage premium on dummies for the
existence of collective agreements and a works council, which mirror the
plant’s industrial relations regime, the plant’s quasi rent per worker as a
performance measure, and a rich set of control variables. As we did for our
wage premium measure, we assessed plant performance such that worker
sorting on unobservable worker characteristics into plants is accounted for
in that we based the quasi rent per worker on the outside options available
to a plant’s workforce making use of both the plant and the worker wage
effects of CHK.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1143

In OLS regressions that pool observations for our entire observational

window, we found that the level of the plant wage premium is only margin-
ally influenced by plant performance, with a rent-sharing elasticity of 0.014,
that is, a 0.014% increase in the premium for a 1% rise in the quasi rent
per worker. Hence, differences in the surplus to be split between workers
and the employer across plants exert only a minor influence on relative
wage premia, although this small degree of responsiveness does not mean
that workers’ share in the surplus has to be small in general. Workers’ share
in the surplus is in fact shaped by a plant’s industrial relations regime.
Plant-level codetermination through a works council is associated with an
additional plant wage premium of 5.0 log points whereas collective
bargaining coverage is accompanied by an extra premium of 1.9 log points.
Together with our finding that collective agreements and works councils
are associated with an increase in the wage premium in plants that pay low
premia and have smaller quasi rents, the overall picture is that workers use
these institutions to bargain over a minimum level of wages rather than over
their share in the plant’s quasi rent. Running separate OLS regressions for
the three time intervals within our observational window, for which CHK
provide separate plant wage premia, we found that all these drivers of the
level of the plant wage premium gained in importance over time.
Since the additional wage premium associated with a plant’s industrial
relations regime is conditional on plant performance and since our wage
premium measure accounts for worker sorting on observable and unobserv-
able worker characteristics, our findings suggest that collective bargaining
coverage and, even more so, works council existence enable workers to
obtain a larger share of the surplus. Our reading is that both institutions of
worker representation, in particular plant-level codetermination through
works councils, increase workers’ bargaining power and allow them to nego-
tiate a larger wage premium. By contrast, an alternative explanation of wage
premia rooted in compensating differentials seems implausible given that
previous research has documented nontrivial voice effects of these
institutions that would suggest negative (if any) compensating differentials.
We also saw the influence of both institutions on wage premia rising over
time. Panel attrition in our data, however, prevented us from testing in a
convincing way whether this observed increase in rent extraction is driven
by a changing composition of plants with collective agreements or a works
council, or rather by workers’ changing bargaining power within existing
plants over time. Discriminating between these two mechanisms is a promis-
ing avenue for future research.
In RIF regressions for the variance, the median, and the first and the
ninth decile of the wage premium distribution, we observed that the quasi
rent has a negligible impact on the dispersion of wage premia whereas col-
lective bargaining coverage is associated with less dispersed premia and the
opposite holds for works council existence. Specifically, we found that col-
lective bargaining compresses the premium distribution from below and
1144 ILR REVIEW

works council existence widens the distribution at the top. Our reading of
these findings is that collective agreements are primarily used to settle mini-
mum terms and thus yield additional premia in low-premium plants,
whereas workers in plants that pay nontrivial wage premia, who arguably
possess high bargaining power from the outset, use plant-level codetermina-
tion through works councils to foster rent extraction.
Our findings suggest that the widening of the plant wage premium distri-
bution over time observed in our data cannot plausibly be related to
changes in the dispersion of performance across plants or to the drop in
works council prevalence over time. Among the factors we consider in this
study, only the fall in collective bargaining coverage contributes to
explaining the rise in the wage premium dispersion across plants. Our
decomposition analysis documented that falling collective bargaining cover-
age can account for 40% of the part of the rising variance in wage premia
attributable to changing plant characteristics and thus explains 14% of ris-
ing premia dispersion. This finding lends support to the view expressed in
the earlier discussion on the drivers of rising wage inequality that changes
in labor market institutions may be an important part of the story (e.g.,
DiNardo et al. 1996; Card and DiNardo 2002). Yet, we also saw that the
larger part of the rise in the wage premia variance remains unaccounted for
despite the detailed set of explanatory variables in our regressions. Clearly,
further research is needed to shed more light on the likely reasons for the
rising dispersion of wage premia across employers.

References
Abowd, John M., and Laurence Allain. 1996. Compensation structure and product market
competition. Annales d’Économie et de Statistique 11(41/42): 207–17.
Abowd, John M., Francis Kramarz, and David N. Margolis. 1999. High wage workers and high
wage firms. Econometrica 67(2): 251–333.
Abowd, John M., and Thomas Lemieux. 1993. The effects of product market competition on
collective bargaining agreements: The case of foreign competition in Canada. Quarterly
Journal of Economics 108(4): 983–1014.
Abowd, John M., Paul A. Lengermann, and Kevin L. McKinney. 2003. The measurement of
human capital in the U.S. economy. LEHD Program Technical Paper No. TP-2002-09.
Suitland, MD: U.S. Census Bureau.
Addison, John T. 2009. The Economics of Codetermination: Lessons from the German Experience.
New York: Palgrave Macmillan.
Addison, John T., Claus Schnabel, and Joachim Wagner. 2001. Works councils in Germany:
Their effects on establishment performance. Oxford Economic Papers 53(4): 659–94.
Addison, John T., Paulino Teixeira, and Thomas Zwick. 2010. German works councils and
the anatomy of wages. Industrial and Labor Relations Review 63(2): 247–70.
Alda, Holga, Stefan Bender, and Hermann Gartner. 2005. The linked employer–employee
dataset created from the IAB Establishment Panel and the process-produced data of the
IAB (LIAB). Schmollers Jahrbuch 125(2): 327–36.
Arrowsmith, James, Paul Marginson, and Keith Sisson. 2003. Externalisation and
internalisation of collective bargaining in Europe: Variation in the role of large compa-
nies. Industrielle Beziehungen 10(3): 363–92.
 FIRM WAGE PREMIA, IR, AND RENT SHARING IN GERMANY 1145

Baumgarten, Daniel, Gabriel Felbermayr, and Sybille Lehwald. 2020. Dissecting between-
plant and within-plant wage dispersion — Evidence from Germany. Industrial Relations,
forthcoming.
Behrens, Martin. 2009. Still married after all these years? Union organizing and the role of
works councils in German industrial relations. Industrial and Labor Relations Review 62(3):
275–93.
———. 2013. Arbeitgeberverbände – auf dem Weg in den Dualismus? WSI-Mitteilungen
66(7): 473–81.
Biewen, Martin, and Matthias Seckler. 2019. Unions, internationalization, tasks, firms, and
worker characteristics: A detailed decomposition analysis of rising wage inequality in
Germany. Journal of Economic Inequality 17(4): 461–98.
Bryson, Alex. 2014. Union wage effects: What are the economic implications of union wage
bargaining for workers, firms, and society? IZA World of Labor 1(35): 1–10.
Card, David, Ana Rute Cardoso, Jörg Heining, and Patrick Kline. 2018. Firms and labor mar-
ket inequality: Evidence and some theory. Journal of Labor Economics 36(S1): S13–S70.
Card, David, Ana Rute Cardoso, and Patrick Kline. 2016. Bargaining, sorting, and the gender
wage gap: Quantifying the impact of firms on the relative pay of women. Quarterly Journal
of Economics 131(2): 633–86.
Card, David, Francesco Devicienti, and Agata Maida. 2014. Rent-sharing, holdup, and wages:
Evidence from matched panel data. Review of Economic Studies 81(1): 84–111.
Card, David, and John E. DiNardo. 2002. Skill-biased technological change and rising wage
inequality: Some problems and puzzles. Journal of Labor Economics 20(4): 733–83.
Card, David, Jörg Heining, and Patrick Kline. 2013. Workplace heterogeneity and the rise of
West German wage inequality. Quarterly Journal of Economics 128(3): 967–1015.
———. 2015. CHK effects: Version 2. FDZ-Methodenreport 06/2015. Nuremberg: Institute
for Employment Research.
DiNardo, John E., Nicole M. Fortin, and Thomas Lemieux. 1996. Labor market institutions
and the distribution of wages, 1973–1993: A semiparametric approach. Econometrica 64(5):
1001–44.
Dustmann, Christian, Johannes Ludsteck, and Uta Schönberg. 2009. Revisiting the German
wage structure. Quarterly Journal of Economics 124(2): 843–81.
Dustmann, Christian, Bernd Fitzenberger, Uta Schönberg, and Alexandra Spitz-Oener. 2014.
From sick man of Europe to economic superstar: Germany’s resurgent economy. Journal
of Economic Perspectives 28(1): 167–88.
Eberle, Johanna, Peter Jacobebbinghaus, Johannes Ludsteck, and Julia Witter. 2011. Genera-
tion of time-consistent industry codes in the face of classification changes: Simple heuris-
tic based on the Establishment History Panel (BHP). FDZ-Methodenreport 05/2011.
Nuremberg: Institute for Employment Research.
Ellguth, Peter, and Susanne Kohaut. 2016. Tarifbindung und betriebliche Interessenver-
tretung: Ergebnisse aus dem IAB-Betriebspanel 2015. WSI-Mitteilungen 69(4): 283–91.
Ellguth, Peter, Susanne Kohaut, and Iris Möller. 2014. The IAB Establishment Panel – Meth-
odological essentials and data quality. Journal for Labour Market Research 47(1–2): 27–41.
Firpo, Sergio, Nicole M. Fortin, and Thomas Lemieux. 2009. Unconditional quantile
regressions. Econometrica 77(3): 953–73.
———. 2018. Decomposing wage differentials using recentered influence function
regressions. Econometrics 6(2): 68–107.
Freeman, Richard B., and James L. Medoff. 1984. What Do Unions Do? New York: Basic Books.
Goldschmidt, Deborah, and Johannes F. Schmieder. 2017. The rise of domestic outsourcing
and the evolution of the German wage structure. Quarterly Journal of Economics 132(3):
1165–217.
Gürtzgen, Nicole. 2009. Rent-sharing and collective bargaining coverage: Evidence from
linked employer–employee data. Scandinavian Journal of Economics 111(2): 323–49.
Hirsch, Boris, and Claus Schnabel. 2014. What can we learn from bargaining models about
union power? The decline in union power in Germany, 1992–2009. Manchester School
82(3): 347–62.
1146 ILR REVIEW

Hirsch, Boris, Thorsten Schank, and Claus Schnabel. 2010. Works councils and separations:
Voice, monopoly, and insurance effects. Industrial Relations 49(4): 566–92.
Hornstein, Andreas, Per Krusell, and Giovanni L. Violante. 2011. Frictional wage dispersion
in search models: A quantitative assessment. American Economic Review 101(7): 2873–98.
Jung, Sven, and Claus Schnabel. 2011. Paying more than necessary? The wage cushion in
Germany. Labour 25(2): 182–97.
Kohaut, Susanne, and Claus Schnabel. 2003. Zur Erosion des Flächentarifvertrags: Ausmaß,
Einflussfaktoren und Gegenmaßnahmen. Industrielle Beziehungen 10(2): 193–219.
Macis, Mario, and Fabiano Schivardi. 2016. Exports and wages: Rent sharing, workforce com-
position, or returns to skills? Journal of Labor Economics 34(4): 945–78.
Matano, Alessia, and Paolo Naticchioni. 2017. The extent of rent sharing along the wage dis-
tribution. British Journal of Industrial Relations 55(4): 751–77.
Mueller, Steffen. 2008. Capital stock approximation using firm level panel data. Jahrbücher für
Nationalökonomie und Statistik 228(4): 357–71.
———. 2011. Works councils and firm profits revisited. British Journal of Industrial Relations
49(s1): s27–s43.
———. 2012. Works councils and establishment productivity. ILR Review 65(4): 880–98.
Mueller, Steffen, and Jens Stegmaier. 2017. The dynamic effects of works councils on plant
productivity: First evidence from panel data. British Journal of Industrial Relations 55(2):
372–95.
———. 2020. Why is there resistance to works councils in Germany? An economic perspec-
tive. Economic and Industrial Democracy, forthcoming.
Oberfichtner, Michael, and Claus Schnabel. 2019. The German model of industrial relations:
(Where) does it still exist? Jahrbücher für Nationalökonomie und Statistik 239(1): 5–37.
Pfeifer, Christian. 2010. Works councils, union bargaining and quits in Germany. Economic
and Industrial Democracy 32(2): 243–60.
Schank, Thorsten, Claus Schnabel, and Joachim Wagner. 2007. Do exporters really pay
higher wages? First evidence from linked employer–employee data. Journal of International
Economics 72(1): 52–74.
Shapiro, Carl, and Joseph E. Stiglitz. 1984. Equilibrium unemployment as a worker discipline
device. American Economic Review 74(3): 433–44.
Sorkin, Isaac. 2018. Ranking firms using revealed preferences. Quarterly Journal of Economics
133(3): 1331–93.
Van Reenen, John. 1996. The creation and capture of rents: Wages and innovation in a
panel of U.K. companies. Quarterly Journal of Economics 111(1): 195–226.