Minimum wages : boon or bane? : Microeconomic evidence ...

Minimum Wages: Boon or Bane?Microeconometric Evidence from Germany

INAUGURALDISSERTATION

zur

Erlangung der Wurde

eines Doktors der

Wirtschaftswissenschaft

der

Fakultat fur Wirtschaftswissenschaft

der

Ruhr-Universitat Bochum

Kumulative Dissertation, bestehend aus 4 Beitragen

vorgelegt von

Hanna Frings, M.Sc.aus Koln

2014

Dekan: Prof. Dr. Helmut KarlReferent: Prof. Dr. Thomas K. BauerKorreferent: Prof. Dr. Christoph M. SchmidtTag der mundlichen Prufung: 23. Juli 2014

Contents

1. Introduction 1

2. Employment and Wage Effects 92.1. The Employment Effect of Industry-Specific, Collectively-Bargained Mini-

mum Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.1. Institutional Background . . . . . . . . . . . . . . . . . . . . . . . . . 132.1.2. Empirical Strategy and Data . . . . . . . . . . . . . . . . . . . . . . . 142.1.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.1.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.2. High-Impact Minimum Wages and Heterogeneous Regions . . . . . . . . . 372.2.1. Minimum Wages in Germany and Previous Evaluations . . . . . . . 412.2.2. Data Construction and Description . . . . . . . . . . . . . . . . . . . . 462.2.3. Estimation Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532.2.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 592.2.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3. Competition on labour and product markets 743.1. Monopsonistic Competition and the Minimum Wage . . . . . . . . . . . . . 74

3.1.1. Sectoral Differences in Monopsonistic Competition . . . . . . . . . . 763.1.2. Estimation Strategy and Data . . . . . . . . . . . . . . . . . . . . . . . 833.1.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 923.1.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

3.2. Minimum Wages as a Barrier to Entry . . . . . . . . . . . . . . . . . . . . . . 1043.2.1. Institutional Background . . . . . . . . . . . . . . . . . . . . . . . . . 1053.2.2. Theoretical Considerations . . . . . . . . . . . . . . . . . . . . . . . . 1083.2.3. Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . 1133.2.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1173.2.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

4. Conclusion 128

A. Supplementary Material 143A.1. Appendix Section 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143A.2. Appendix Section 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144A.3. Appendix Section 3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

B. Curriculum Vitae 153

Acknowledgements

I would like to use this opportunity to ex-press my gratitude for the support I have re-ceived in writing this dissertation.

I thank my supervisor Thomas K. Bauer,who has continuously offered his academicguidance throughout the last years. Dur-ing our discussions, I have profited enor-mously from his experience in conductingmicroeconometric research. I am also grate-ful to my second supervisor, Christoph M.Schmidt, for always challenging me to findthe most precise and convincing line of ar-gument.

My co-authors, Ronald Bachmann,Philipp vom Berge and Alfredo Paloyo havecontributed significantly to the studies con-tained in this dissertation. While workingon the individual papers, they provided mewith creative ideas, new insights and pro-found knowledge of econometric methodsand their practical applications.

Numerous discussions with colleagues,both at the RWI and the RUB, have addi-

tional helped me in writing this thesis. Toprovide an incomplete list, I would like tothank Daniel Baumgarten, Michael Kind,Anica Kramer, Sandra Schaffner, SebastianOtten, Matthias Vorell and Lina Zwick. Theresearch environment at the RWI, includingthe possibility to attend international con-ferences, participate in seminars, exchangethoughts with visiting researchers and re-ceive various trainings has always inspiredand motivated me. I also would like tothank the administrative staff at the RWI,and especially Claudia Lohkamp, who hasdone everything in her power to allow me tofocus on my research. Despite the supportI have received, all remaining shortcomingsin this dissertation are my responsibility.

Last but not least, I would like to thankmy family for their unshakable belief in me.My parents have always encouraged me tofind out where the next step takes me. Myhusband, Moritz, has taken each step withme. I dedicate all my achievements to them.

1. Introduction

Germany is one of the few industrialized countries without an explicit statutory minimum

wage. Existing minimum wages are based on collective-bargaining agreements which are

declared as general binding and cover only a handful of specific industries. For years, the

German debate focused on extending this system to additional industries, especially those

with a relatively low coverage rate of collective bargaining. However, at the beginning of

2014, the argument has been settled in favour of the supporters of a uniform wage floor:

The minister of Labour and Social Affairs, Andrea Nahles, presented a draft of a minimum

wage law to be effective as of January 2015 on April 2, 2014 [BMAS 2014]. The level is

set at e8.50 for East and West Germany alike. Exemptions will in all likelihood include

interns, apprentices, the long-term unemployed and young workers under the age of 18. A

commission consisting of six representatives of labour unions and employer associations

will provide annual advice to the government if and by how much the minimum wage

should be changed.

While the introduction of a statutory, uniform minimum wage in Germany has been de-

cided upon by policy makers, no consensus has been reached among labour economists

on the merits and threats of minimum wage laws. Quite on the contrary, the national and

international debate has experienced a revival during the last 15 years for at least two rea-

sons. First, a growing number of microeconometric studies based on natural experiments

for causal impact evaluation have found contradicting employment effects of minimum

wage introductions or increases. To name only the most prominent examples, Neumark

and Wascher [2008] conclude that the US minimum wage has led to worker displacement.

1. INTRODUCTION 2

In contrast, Card and Krueger [1994] and, more recently, Dube, Lester and Reich [2010]

provide evidence that the minimum wage did not cause lower employment levels.

Theoretical predictions are similarly ambiguous and depend on the competitive situa-

tion on the labour market. Under perfect competition, a minimum wage above the market-

clearing level will lead to higher unemployment. In a monopsonistic labour market in

contrast, a moderate minimum wage might even increase employment levels, as labour

supply increases and firms do not reduce labour demand, reducing profits instead. Given

that neither theory nor existing empirical evidence provides a clear guideline for minimum

wages in general, and for the German case in particular, additional research that helps to

understand the factors influencing the employment effects of minimum wage is needed.

Although no consensus has been reached up to date, the idea of a labour market insti-

tution that increases wages at no costs is extremely tempting to policy makers. Supporters

set high hopes in the new wage floor: It is intended to contain the low wage sector, thereby

decreasing income inequality, preventing poverty and minimizing wage subsidies by the

state (“Aufstocker”). However, if minimum wages are perceived as a complete solution to

these problems observed in the modern German labour market, they are bound to fail.

Since the mid-90’s the low wage sector (defined as workers earning less than two thirds

of the median gross wage) has indeed been growing, accompanied by growing wage in-

equality. This development was especially strong in the first years following the turn of

the century and has stabilized from 2006 onwards [Bundesagentur fur Arbeit 2011; SVR

2011]. Interestingly, the trend of increasing wage inequality has been driven by the lower

part of the wage distribution in West Germany [Dustmann, Ludsteck and Schonberg 2009;

Moller 2005; Steiner and Holzle 2000], and by the upper part of the wage distribution in

East Germany [Franz and Steiner 2000; Kohn 2006; Gernandt and Pfeiffer 2007]. The need

for a statutory minimum wage to fight increasing wage inequality can therefore only be

justified in West Germany, while the effects on wages and employment can be expected to

be much larger in East Germany where wages and productivity are lower.

Even more importantly, the link between wages and income inequality, as an indicator

1. INTRODUCTION 3

for poverty, is rather weak [Brenke and Muller 2013]. Taking into account the household

context, transfer payments and tax incentives, only 20 percent of all workers earning wages

below e8.50 are poor in terms of household income [Heumer, Lesch and Schroeder 2013].

Poverty is not mainly driven by low wages, but instead by high needs (e.g. children or

an inactive spouse living in the household) or few working hours. Brenke and Ziemen-

dorff [2008] show that full-time employed workers, who receive wage subsidies by the

state because their income is below the subsistence level, earn on average e9 per hour.

Stated differently a minimum wage of e8.50 will not prevent wage subsidies to full-time

employed workers completely, because wages are already higher than this wage floor.

This short discussion shows that minimum wages are, opposed to the impression that

is sometimes given in the public debate, no panacea to answer challenges such as poverty

or inequality in Germany. Under the assumption of neutral employment effects, minimum

wages may constitute one component of a larger policy mix targeted towards e.g. reducing

wage inequality. To the extent that minimum wages reduce low-wage employment, how-

ever, their contribution to the political aim of poverty prevention might even be negative.

Thus, the employment effects of minimum wages in Germany remain a crucial question.

This dissertation was written against the background of two developments: A resurging

academic debate on the employment effects of wage floors on the one hand, and increas-

ing political pressure to introduce a statutory minimum wage in Germany on the other

hand. Within the scientific community, the debate has shifted from an ideological discus-

sion to the more relevant question why and in which situations minimum wages might

be employment-neutral. It is crucial to understand these factors fully, since results from

other countries are not necessarily applicable to the German case due to differences in the

institutional framework governing minimum wages, the system of wage determination,

worker and industry composition, competition on the product market, macroeconomic

performance, and the general policy mix.

From a policy point of view, the results presented in the following chapters of this the-

sis are therefore of high relevance, because they, first, provide direct empirical evidence

1. INTRODUCTION 4

on the employment effects of existing sectoral minimum wages, and second, offer new in-

sights into the design of a minimum wage institution that minimizes employment losses.

In this context, I provide direct evidence that differentiation of minimum wages across

industries bears the risk of additional employment losses due to effects on the product

market. Further, I show that in addition to the wage distribution prior to the minimum

wage, the degree of monopsonistic competition at the industry level is a decisive factor for

the expected employment reaction to a uniform minimum wage.

The contribution to the academic literature of this dissertation is twofold. First, it pro-

vides more convincing identification strategies than previous studies for the employment

effect of minimum wages in general, but especially for the German institutional setting.

Second, a focus is set on under-researched topics within the minimum wage literature by

investigating the competitive situation on the labour market and the influence collectively-

bargained wage floors may have on product market competition.

This thesis consists of two chapters, each based on two papers. After this introduction,

Chapter 2 contains two studies exploiting the existing sectoral minimum wages in the con-

struction industry to analyse the employment effects in the German case. Special attention

is paid to the identification of the causal effect of the minimum wage on employment.

This requires innovative solutions since it is not possible to simply follow the international

literature due to data limitations and the institutional framework based on collectively-

bargained, industry specific minimum wages. I propose that (i) any analysis should be

carried out at the aggregate level (industry or region) in order to minimize measurement

error in hourly wages, (ii) control groups from within the same industry are a poor choice

since they are affected by the treatment as well [Kluve and Schmidt 2007], (iii) panel stud-

ies based on several variations in the minimum wage over time are superior to case studies

exploiting a single increase in the wage floor due to a higher degree of external validity,

and (iv) the employment outcome should be measured in growth rates instead of levels to

avoid serial correlation and to additionally allow for effects on job creation [Meer and West

2013].

1. INTRODUCTION 5

Section 2.1 estimates the employment effects of industry-specific, collectively-bargained

minimum wages in Germany for two occupations associated with the construction sector:

Painters and electricians. By using workers from different industries, I propose a truly ex-

ogenous control group. In contrast, the control group design of previous studies is based

on workers from the same industry earning wages above the minimum wage prior to its

introduction. While the common trend assumption appears plausible for low- and high-

wage workers in the same industry, high-wage workers are equally affected by the mini-

mum wage treatment due to potential spillover effects and substitution of low-skilled for

high-skilled labour. I therefore use workers from different industries as control groups,

where industry selection is based on a similar employment trend prior to the treatment

and on low coverage of collective bargaining.

Further, a difference-in-differences-in-differences (DDD) estimator is presented as a ro-

bustness test for industry-specific, time-varying, unobserved heterogeneity. The minimum

wage effect on employment is identified by comparing affected and unaffected occupations

(first difference) in the treatment industry to the same occupations in the control industries

(second difference), prior and after the treatment (third difference). The DDD estimator is

less sensible to the common trend assumption, but is vulnerable to spillover effects of the

wage floor to other occupations in the same industry. It additionally assumes no worker

mobility caused by the minimum wage within the same occupation, but between different

industries. Note that the two identification strategies are subject to different biases, but

both show no significantly negative employment effect, even though the minimum wage

is binding in (East) Germany.

Based on a different methodology, we address a similar question in Section 2.2 (co-

authored by Philipp vom Berge and Alfredo Paloyo), where we estimate the effects on

wage and employment growth rates of the introduction and subsequent increases of a

substantial minimum wage in the main construction industry. Using a regional dataset

constructed from individual employment histories, we exploit the spatial dimension and

border discontinuities to account for spillovers between districts and unobserved hetero-

1. INTRODUCTION 6

geneity at the local level.

Our identification strategy is again twofold. First, we estimate a panel model by regress-

ing regional employment growth on the regional minimum wage bite. This model is robust

to local spillovers, which correspond to one region’s bite affecting the employment growth

of neighbouring regions. It is, however, less robust to unobserved spatial heterogeneity

in employment growth that is independent of the minimum wage. Second, we follow a

contiguous-border approach which compares only regions to each other that share a com-

mon border. We are thus able to show that our treatment effect is not biased due to spatial

heterogeneity. It turns out that the estimated treatment effect is extremely robust to both

methods.

The results indicate that the minimum wage increased the wage growth rate for East

Germany but did not have a significant impact in West Germany. The estimated effect on

the employment growth rate reveals a contraction in the East of about 2.6 to 3.1 percent-

age points for a one-standard-deviation increase in the minimum-wage bite, amounting

to roughly half of the overall decline in the growth rate, but no significant change for the

West.

Chapter 3 consists of two studies analysing the relationship between minimum wages

and competition on labour and product markets; two topics that are less intensively re-

searched compared to the employment effects of wage floors. First, imperfect competition

on the labour market is a frequently cited theoretical explanation for non-negative employ-

ment effects of wage floors. We contribute to the literature by empirically analysing if and

to which extent the existing minimum wage industries in Germany deviate from the per-

fectly competitive model of the labour market. Second, industry-specific minimum wages

may be exploited by employers as a cost-raising strategy to deter entry, thereby decreas-

ing product market competition. We are the first to provide empirical evidence on this

hypothesis.

Section 3.1 (co-authored by Ronald Bachmann) investigates the degree of monoposony

power of German employers in different industries using a semi-structural approach based

1. INTRODUCTION 7

on a dynamic model of monopsonistic competition. As pointed out above, the degree of

monopsonistic power is crucial for the employment effects of a binding minimum wage.

We measure monopsony power by estimating the wage elasticity of labour supply to the

individual firm. High elasticities point towards perfect competition on the labour market.

Lower elasticities indicate a higher degree of monopsonistic competition. We estimate the

elasticities separately for broad industries (one-digit classification) and in particular for the

existing minimum wage industries.

The empirical analysis relies on a linked employer-employee data set which allows us

to control for heterogeneity both on the worker and on the firm side. Controlling for firm

attributes is crucial in order to ensure that the labour supply and not the labour demand

curve is identified. Our results show important differences in monopsonistic competition

between industries, and between East and West Germany. From a policy point of view,

the introduction of a uniform minimum wage may therefore lead to deviating employ-

ment reactions in industries with a similar wage structure. At the same time, existing

minimum wages at the sectoral level are apparently unrelated to the degree of monopsony

power in the respective industries. Thus, it appears unlikely that monopsony is the main

explanation for observed non-negative employment effects of the minimum wage or that

monopsony was an important selection criterion for the minimum wage industries.

Section 3.2 (co-authored by Ronald Bachmann and Thomas K. Bauer) provides an al-

ternative explanation why minimum wages were introduced in some industries, but not

in others. The existing minimum wage institution requires a consensus between unions

and employer associations on the introduction of a wage floor in a specific industry. This

naturally raises the question why employers should ever actively agree with a minimum

wage.

One answer is that employers’ support for the introduction of industry-specific mini-

mum wages may be viewed as a cost-raising strategy in order to deter market entry. Using

a unique data set consisting of 800 firms in the German service sector, we show that high-

productivity employers are more likely to support minimum wages. Low-productivity,

1. INTRODUCTION 8

labour-intensive competitors find it unprofitable to remain in the market or to enter it in

the presence of a minimum wage.

Similarly, domestic employers may support a minimum wage in order to deter market

entry from low-wage European neighbouring countries. We find some evidence that min-

imum wage support is higher in industries and regions with low barriers to entry. This

is especially true in East Germany, where the perceived threat of low-wage competition

from Central and Eastern European Countries is relatively high. In addition, firms paying

collectively agreed wages are more strongly in favour of minimum wages. This is particu-

larly relevant given the expanding Single Market of the European Union and the system of

extending collective bargaining agreements to all workers and firms in an industry, which

is present in the majority of European countries. Although theoretical work exists on this

topic, we are the first to test the implied hypotheses empirically.

The remainder of this dissertation is structured as follows. Chapter 2 provides the two

studies on employment effects of minimum wages for the construction industry using (i)

a difference-in-differences estimator with other industries as control groups (Section 2.1)

and (ii) using a spatial panel model in which the employment effect is identified through

varying treatment intensities (Section 2.2). Chapter 3 analyses the degree of competition on

labour and product markets. Section 3.1 investigates first, to which extent monopsonistic

competition is a plausible explanation for non-negative employment effects in Germany,

and second, how the employment effects of a statutory minimum wage may differ be-

tween industries and regions. Section 3.2 provides insights into the relationship between

industry-specific wage floors and product market competition. Chapter 4 offers conclud-

ing remarks with a special focus on the relevant channels of adjustment of minimum wages

in Germany and an assessment of the statutory minimum wage to be introduced in January

2015.

2. Employment and Wage Effects

This chapter is composed of two sections analysing the employment effect of minimum

wages for painters and electricians (Section 2.1) and the main construction industry (Sec-

tion 2.2) in Germany. Section 2.1 is based on “The Employment Effect of Industry-Specifc,

Collectively-Bargained Minimum Wages”, which has been published in the German Eco-

nomic Review.1 This paper has been awarded the “RWI Junior Prize for Outstanding Aca-

demic Achievement” (best paper published by a PhD student in 2013). Section 2.2 is joint

work with Alfredo Paloyo and Philipp vom Berge and based on the Ruhr Economic Papers

No. 408 “High-impact minimum wages and heterogeneous regions”. This paper has been

submitted to the Review of Economics and Statistics.

2.1. The Employment Effect of Industry-Specific,

Collectively-Bargained Minimum Wages

Germany is one of the few European countries without statutory minimum wages. For

decades this fact had remained broadly unquestioned by officials, academics and the gen-

eral public, because collective bargaining was developed to such an extent that effective

minima existed in the absence of any state regulation. However, since the beginning of the

1990’s an erosion of collective wage agreements can be observed. At the same time, the

completion of the EU’s Single Market as well as its eastward enlargement increased the

supply of low wage labour, especially in the construction industry. Based on these devel-

1. FRINGS, H. (2013): “The Employment Effect of Industry-Specifc, Collectively-Bargained MinimumWages”, German Economic Review, 14(3), 258-281.

2. EMPLOYMENT AND WAGE EFFECTS 10

opments a debate has emerged on the advantages and problems of the introduction of a

country-wide or industry-specific, statutory minimum wage.

In the beginning of the 90s, a number of empirical studies based on establishment level

data conducted in the US reported neutral or positive employment effects [e.g. Katz and

Krueger 1992; Card and Krueger 1994]. In contrast, Neumark and Wascher [2004] con-

clude, after providing an extensive survey of studies on the employment effects of mini-

mum wages exploiting the variation to be found in panel data at the state level, that the

effect on youth employment is generally negative. More recently, Dube, Lester and Reich

[2010] report neutral employment effects using panel data at the county level. The novelty

of these contributions is to control explicitly for time-varying spatial heterogeneity at the

county level, either by comparing only contiguous counties [Dube, Lester and Reich 2010]

or by incorporating region-specific trends in order to capture differences in employment

trends that are unrelated to the policy change [Addison, Blackburn and Cotti 2009; Alle-

gretto, Dube and Reich 2011]. The underlying idea is that counties with a strong minimum

wage bite are characterized by lower employment growth independently of the minimum

wage. Traditional estimates would thus be biased downwards.

Neutral or positive employment effects of a minimum wage can be explained by the no-

tion that (some) labour markets are characterized by monopsonic competition [Manning

2003a]. In the absence of collective organization, market imperfections give employers

some discretion in determining wages. Examples of such imperfections include the fact

that it is costly for employees to change jobs, that employers are imperfect substitutes for

each other, and that workers only possess imperfect information about alternative employ-

ment opportunities [Manning 2004].

Metcalf [2008] identifies alternative explanations for missing disemployment effects of

the minimum wage introduction in the UK, which are partly in line with the assumption

of perfect competition on the labour market. Examples include an adjustment of working

hours, increases in productivity, effort or education, price increases on product markets,

reduction of profits of affected firms, and incomplete compliance. Based on a survey of

firms in the US, Hirsch, Kaufman and Zelenska [2011] show that increases in labour costs


are absorbed through non-employment channels of adjustment, such as price increases,

profit reduction as well as higher performance standards. This indicates that minimum

wages may have effects on an industry, even if there are no negative employment effects.

The existing research shows that the employment effect of any minimum wage policy is

a question which cannot be answered theoretically. At the same time, the empirical results

from one country cannot necessarily be carried over to another due to differences in labour

market institutions. Boockmann [2010] performs a meta-analysis of 55 empirical studies on

the employment effects of minimum wages and shows that the interaction of the minimum

wage policy with labour market institutions is especially important. More specifically, the

benefit replacement ratio and the degree of coordination in the collective bargaining system

reduce any employment effects, while employment protection enhances such effects.

The minimum wage research in Germany has concentrated on the simulation of the em-

ployment effects of a hypothetical, statutory minimum wage, often with a focus on income

or fiscal effects [Bauer, Kluve, Schaffner and Schmidt 2009; Knabe and Schob 2009; Muller

and Steiner 2010]. Further, the Ministry of Labour and Social Affairs (BMAS) contracted the

evaluation of existing minimum wages in 2011. In general, the analyses did not find robust,

negative employment effects, despite positive wage effects in (mostly) East Germany.2

In addition to the evaluations contracted by the BMAS, Konig and Moller [2009] and

Muller [2010] analyse the effects of the minimum wage in the main construction industry.

Muller [2010] uses cross-sectional data to estimate the counterfactual wage distribution if

the minimum wage had never been introduced. A comparison of the two wage distri-

butions yields estimates for the employment effect. These are generally negative in East

Germany and insignificant in West Germany, but also very sensitive to assumptions made

on spillover effects of the minimum wage throughout the wage distribution.

In contrast, Konig and Moller [2009] investigate the minimum wage effect on employ-

ment in the main construction sector following a difference-in-differences (DiD) approach.

While Konig and Moller [2009] were the first to employ this empirical strategy using Ger-

2. A detailed review of the results of these evaluations is beyond the scope of this paper. The studiesare open to the general public and can be downloaded at: http://www.bmas.de/DE/Themen/Arbeitsrecht/Meldungen/evaluation-mindestloehne.html.


man data, their results may be biased due to the choice of the control group, which consists

of workers in the same industrial branch earning slightly more than the minimum prior

to its introduction. This approach has been criticized, noticeably by Kluve and Schmidt

[2007], because such a control group may be affected by the minimum wage treatment due

to spillover and/or output effects. Therefore, the employment levels of both groups can

be expected to react to a change in the minimum wage policy and the causal effect of the

minimum wage on employment is not identified.

The study at hand also uses the DiD approach in order to estimate the employment ef-

fect of a minimum wage introduction in Germany for two occupational groups in the con-

struction industry, namely electricians and painters. In order to prevent the bias caused

by a control group subject to the minimum wage, I propose an alternative control group

which should not be influenced by the treatment. It consists of workers belonging to other

industrial branches, in which no minimum wage exists and collective bargaining is char-

acterized by a low coverage rate. The disadvantage of such an approach is that additional

time-varying determinants of employment may exist, which affect the treatment and con-

trol group differently.

In order to test the robustness of the results towards biases resulting from dissimilarity

of the treatment and control group, a difference-in-differences-in-differences (DDD) esti-

mator is specified, which exploits the fact that minimum wages are set for individual oc-

cupations within the construction industry and thereby eliminates all industry-specific or

occupation-specific trends or shocks. While the DDD estimator is better able to deal with

unobserved, time-varying heterogeneity, it is more likely to be biased due to spillover ef-

fects and labour mobility within the occupation but across industries. To the extent that

both estimators are consistent with each other, it is possible to conclude that none of the

two possible biases distort the results.

The remainder of this paper is organized as follows. The institutional background of

minimum wages in Germany is discussed in Section 2.1.1. Section 2.1.2 describes the em-

pirical strategy and introduces the data used for estimation. The basic results are given in

Section 2.1.3. Finally, Section 2.1.4 concludes.


2.1.1. Institutional Background

In Germany, statutory, country-wide minimum wages do not exist. Instead, a few selected

industries do have collectively-bargained minimum wages. Up to date, minimum wages

have been introduced for the main construction sector (1997), electricians (1997), roofers

(1997), painters (2003), the commercial cleaning industry (2008), laundry services (2009),

miners (2009), elderly care (2010), the waste industry (2010), and security services (2011).

The legal framework for any minimum wage introduction in Germany is the Posting of

Workers Law (PWL; Arbeitnehmerentsendegesetz). The law is based on the European Union

Posted Workers Directive (96/71/EC) and aims at protecting domestic workers from in-

creasing low wage labour competition due to the completion of the EU’s Single Market by

demanding that foreign workers (i.e. posted workers) must be subject to the same work-

ing regulations and minimum standards as domestic workers. Examples of such minimum

standards include working hours, holidays and minimum wages.

In order to establish minimum wages, the PWL allows collectively bargained wage rates

to be extended to all workers and firms in an industry, independent of their membership

in trade unions or employer associations. However, at least 50 percent of all employees in

the respective industry have to be covered by the initial collective bargaining agreement

for the law to be applicable.

Posting of workers is quite common in the construction industry. Therefore, only asso-

ciated industries such as main construction, electricians, painters or roofers were initially

covered by the PWL. Because the coverage rate of collective bargaining has traditionally

been rather high in these industries, the minimum wage introduction was mainly mo-

tivated by protectionism. However, during the last decade the focus has shifted to in-

dustries in which posting of workers appears less common, such as commercial cleaning,

laundry services and elderly care3. This implies that the PWL is increasingly used to estab-

lish industry-specific, collectively-bargained minimum wage rates for domestic workers,

3. An alternative explanation for the introduction of minimum wages in these industries is the inclusion ofthe New Member States in terms of the unrestricted movement of labour in 2011. Anlong this line of argument,the more recent minimum wages are also motivated by protectionism, as firms feared increasing competitionand trade unions worrried about decreasing wage levels.


instead of preventing low-wage competition from abroad. One possible explanation for

this increased interest to establish minimum wages is a decline in union density and the

coverage rate of collective bargaining which has started in the 1990s.

Union density has been falling from 30 percent in 1990 to 21 percent in 2004 in West

Germany and from 50 percent in 1990 to 18 percent in 2001 in East Germany [Biebeler and

Lesch 2006]. Naturally, the coverage rate of collective agreements is considerably higher,

but the pattern observed for union density applies equally: coverage has been decreasing

from 69 percent in 1996 to 61 percent in 2004 in West Germany and from 56 percent to 41

percent in East Germany during the same time period [Ellguth and Kohaut 2005]. Not sur-

prisingly, the coverage rates differ significantly between industrial branches. As Figure 2.1

shows, the coverage rate of area-wide collective agreements is (and always has been) espe-

cially low for the service sector, communication and transportation, wholesale and retail-

ing, as well as for non-profit organizations. In contrast, the public sector, the banking and

insurance sector, as well as mining and energy/water provision are characterized by high

and stable coverage rates in East and West Germany alike. Finally, while area-wide col-

lective agreements have been decreasing, company-level collective agreements have been

increasing in number and importance. However, generally the proportion of workers not

covered by any agreement is growing in most industries, implying that the increase in the

latter was not large enough to outweigh the decrease in the former.4

2.1.2. Empirical Strategy and Data

Minimum wages were introduced in 1997 for electricians and in 2003 for painters respec-

tively. In addition, they were abolished again in 2003 for electricians. Thus, the observation

period (1999-2007) covers one minimum wage introduction and one minimum wage abo-

lition. Eligibility is defined by the occupation and the condition that the employer’s main

4. These numbers do not provide any information on the fraction of firms that are not officially covered bya collective agreement, but still adhere to it. Indeed, e.g. for electricians, the fraction of firms adhering to acollective agreement has actually been increasing between 2002 and 2010 [IAW 2011a].


Figure 2.1.: Collective Bargaining Coverage per Industrial Sector

(a) Proportion of Employees Covered in West Germany

(b) Proportion of Employees Covered in East Germany

Source 1998/2000: Hans Bockler Stiftung [2008]. Source 2004: Ellguth and Kohaut [2005].


business activities are within the construction industry.5

In order to estimate the employment effect of the minimum wage, a difference-in-differences

(DiD) estimator will be used. Any control group must fulfil the common trend assumption

and should not be influenced by the treatment in order for the DiD estimator to be credible.

Because minimum wages are collectively bargained at the industry level, a trade-off exists

between these two requirements. One possibility is to choose a control group within the

same industry or in neighbouring industries. In this case, comparability is high, but the

control group is likely to be affected by the minimum wage treatment due to substitution

or spillover effects. The other possibility is to choose a control group from a different in-

dustry that is not directly connected to the treatment industry along the supply chain. This

implies that the control group is clearly not subject to the treatment, but it is doubtful that

no time-varying determinants of employment exist that affect the groups differently.

To ensure that neither of the two problems biases the results, two different empirical

strategies will be followed. First, the control groups are chosen from different industries

and applied within a ‘traditional’ DiD framework . This approach appears to be preferable

insofar as it is possible to control for time-varying determinants of employment that vary

between the included groups. Second, a difference-in-differences-in-differences (DDD) es-

timator is used with control groups from neighbouring industries and occupations. The

results obtained with this estimator may be biased due to substitution and/or spillover ef-

fects. Additionally, it is not possible to determine whether the effect of the minimum wage

introduction will be overestimated or underestimated. However, if the estimated effect ob-

tained with the DDD estimator is similar to the effects obtained by the DiD estimator, the

results are robust to the biases present with both empirical strategies.

Difference in Differences

The main interest of this study is to analyse the effect of the minimum wage on aggregate

employment in the two treatment groups. Stated differently, the central research question

5. Further, apprentices are excluded in the case of electricians. Similarly, unskilled painters aged 18 years orless do not receive a minimum wage.


is whether the number of full-time employed painters and electricians changes in response

to the minimum wage. The transport and communication industry as well as the wholesale

and retailing sector are used as control groups. Similarly to the treatment groups, employ-

ment is defined as a head count of all full-time employed workers. These two industries

were selected, because they are not connected with the treatment groups along the supply

chain and they are characterized by a low coverage rate of collective bargaining (Figure

2.1). The motivation for the latter criterion is that any area-wide collective agreement with

a high coverage rate mimics a minimum wage treatment, because minimum wages in Ger-

many amount to extended collective wage agreements. Thus, the control group should

consist of workers from industrial branches with a low coverage rate of collective bargain-

ing.

In addition, the two control groups fulfil the common trends assumption (Figure 2.2).6

Note that it is a coincidence that the two treatment groups are matched by exactly two

control groups. The idea behind using more than one control group is to reduce biases

that might occur due to a violation of the common trend assumption [Meyer 1995]. If more

industries that (a) do not belong to construction or neighbouring industries, and (b) are

characterized by a low coverage rate of collective bargaining could have been identified,

more than two control groups would have been used.

Similarly, the employment of two treatment groups, one experiencing a minimum wage

introduction, the other one a minimum wage abolition, aims at estimating the average

effect of the minimum wage on employment in Germany. Clearly, the assumption is that a

minimum wage introduction and abolition have a similar effect in terms of sign and size.

In order to verify this assumption, the estimations are also run separately for each of the

treatment groups as a sensitivity check (Table A.1).

In summary, a total of four groups is used in the estimation: Painters and electricians,

which both experience a minimum wage treatment during the observation period, and

6. Placebo tests for painters confirm the visual impression that the employment trends of the treatmentand control groups are not significantly different from each other during the pre-treatment period. In EastGermany this is true for all possible and in West Germany for the majority of placebo tests. Placebo tests forelectricians are not possible due to the missing pre-treatment period. These results are not reported in thepaper, but can be obtained upon request.


Figure 2.2.: Employment Development of Treatment and Control Groups

(a) West Germany

(b) East Germany

Source: BA employment panel (Schmucker and Seth, 2009). Author’s calculations.


the transport/communication as well as the wholesale/retail industry. The employment

effect of the minimum wage introduction is estimated by a fixed effects model with time

dummies, which serves the same purpose as a DiD estimator. The industry dummies elim-

inate any time-constant differences between groups, while the time dummies capture the

effects of any exogenous variable that changes over time, but affects all groups equally. In

summary, the regression equation can be expressed as:

emplit =α + β1mwit + β2git + β3meit + β7di + β8dt + εit (2.1)

Aggregate employment at the level of the industry is represented by emplit, while mwit is

the minimum wage indicator. This indicator takes the value one for industries and time

periods in which a minimum wage exists and zero otherwise. The subscript i denotes the

four groups (two treatment groups and two control groups) and the subscript t refers to

the time periods.

The controls include industry-specific economic growth (git) as well as the proportion

of marginal employment (meit). The former is proxied by growth of revenues (treatment

groups) and growth of gross value added (control groups). These different concepts are

used due to data limitations. Therefore, this variable is only meant to measure general,

possibly deviating, trends and the estimated coefficient should not be interpreted concern-

ing its magnitude.

Marginal employment reduces the social security contributions an employer has to pay

as long as the employee does not earn more than e400 per months or a specific number of

working hours is not exceeded. Therefore, the proportion of marginal employment among

total employment is included in order to control for the possibility that regular employ-

ment is substituted for this less expensive type of workers. Insofar as these two control

variables follow a different development over time for the included groups, it is indispens-

able to include them in the regression equation.

All variables enter the regression model as growth rates. There are two main reasons for

estimating the equation in growth rates rather than in levels. First, employment expressed


as a level follows a non-stationary process over time. This is not surprising, because the

employment level is naturally determined by the level in previous periods to a large extent,

especially because there are large differences in employment levels between groups. Dif-

ferencing, or in this case calculating growth rates, is a simple and intuitive manner to solve

this problem. As the regression results will show, some autocorrelation in the error term is

still left, but serial correlation is decreased to an acceptable level. Second, the employment

growth of the treatment groups is characterized by a strong, seasonal pattern (Figure 2.2).

Using growth rates, calculated as the change in employment compared to the same quarter

in the previous year, eliminates the seasonal pattern completely.

The focus on the employment development of entire industries or occupational groups

has two advantages compared to estimation at the individual level. First, this approach

takes account of terminated as well as new employment relationships. In contrast, new em-

ployment relationships are necessarily ignored when estimating probabilities at the level

of the individual worker. Second, the data only contain average monthly wages, while

the minimum wage is defined as an hourly wage rate. In addition, information on actual

working hours is not available. Consequently, it is necessary to impose strict assumptions

on individual working hours in order to identify workers, who are affected by the mini-

mum wage when estimating the probability of continuous employment. By focusing on

the employment development of all eligible workers such individual identification is not

necessary. At the same time, it is still possible to analyse whether or not the minimum

wage has an influence on aggregate employment.

The majority of studies on the employment effect of minimum wages do not differenti-

ate between a minimum wage introduction and a (sizeable) increase in the minimum wage.

If the minimum wage is set by any outside agency, such as the government in the US or

the Low Pay Commission in the UK, both events are exogenous. The institutional context

in Germany is different. Because the social partners bargain over average wages, mini-

mum wages and employment simultaneously7, minimum wage increases are endogenous.

7. As Layard, Nickell and Jackman [2005] point out, no empirical evidence exists that collective bargainingin Germany indeed follows the efficient bargaining model. However, given the high degree of coordination


Lemos [2005] argues that measures of the minimum wage bite, such as the Kaitz Index or

the fraction of affected workers, are generally endogenous when estimating employment

effects in the presence of collective bargaining. The reason is that these indicators depend

on the wage distribution which is set simultaneously with employment. Therefore, this

study concentrates on the minimum wage introduction, for which endogeneity is a smaller

problem compared to changes in the existing minimum wage.

It should be noted, though, that the observation period does cover several increases

in the minimum wage. A dummy that takes the value one for the treatment groups in

each period after the minimum wage introduction therefore captures the average effect of

the minimum wage introduction and all increases. Still, minimum wage increases cannot

be expected to have a significant impact on employment for at least three reasons. First,

future increases in the minimum wage are usually agreed upon for a period of two to three

years. This implies that employers often know in advance when the minimum wage will

be increased by a certain amount. Thus, there is no reason to expect a significant reaction of

employment each time the minimum wage is increased [Pinoli 2010]. Second, increases are

generally infrequent and, third, very small in magnitude.8 For these reasons the minimum

wage dummy is likely to mainly capture the effect of the minimum wage introduction,

despite the fact that the minimum wages have also been increased during the observation

period.9

Difference in Differences in Differences

The minimum wage treatment takes place for specific occupations (electricians and painters)

in a specific industry (construction). Rattenhuber [2013] proposes that this fact can be ex-

in the case of bargaining on minimum wages (compare with Section 2.1.1), it appears unlikely that the tradeunion and employers’ association ignore consequences for employment completely, even if this outcome is notconsidered explicitly.

8. Painters have only experienced one increase in the minimum wage of 15 Cents per hour during a time pe-riod of four years (2003-2007). In contrast, increases are larger in magnitude and more frequent for electricians.However, these increases are agreed upon in advance, e.g. three increases during the time period 1999-2002for electricians were already known in 1999.

9. The specifications have also been estimated for time periods without any increase in the minimum wage.The results do not change compared to estimations covering the entire observation period.


ploited by an alternative empirical design that uses two control groups: One consisting of

a different occupation in the same industry, and the other one consisting of the same occu-

pations in different (sub-) industries. More specifically, the first control group consists of

clerks in the construction industry and the second control group is composed of painters

as well as electricians in any industry, besides construction, in the manufacturing sector.

Both control groups are employed simultaneously in a difference-in-differences-in-differences

(DDD) framework. The DDD estimator has been predominantly used in contexts, in which

regional variation in treatment is present [e.g. Gruber 1994]. Hamermesh and Trejo [2000]

compare Californian men, who receive a treatment concerning overtime-pay, with Cali-

fornian women as well as men in other states. In the context at hand, variation in treat-

ment does not exist between regions, but between industries. This implies that electricians

(painters) are compared with untreated occupations within the construction industry, and

with electricians (painters) in untreated industries.

The DDD estimator requires a much weaker identification assumption than the usual

DiD estimator: in absence of the minimum wage treatment, the relative difference in the

employment growth rates of electricians (painters) and clerks in the construction industry

should be the same as the relative difference in the employment growth rates of electri-

cians (painters) and clerks in the manufacturing sector. Stated differently, all industry-

specific trends are “differenced away” as long as all occupations in that industry are af-

fected equally. Similarly, all occupation-specific trends disappear as long as occupations in

all industries are affected equally.

Under this identification assumption an unbiased estimate of the employment effect of

the minimum wage introduction β can be obtained by:

β = ∆y1,1 − ∆y0,1 − ∆y1,0 − ∆y0,0 (2.2)

= (y1,11 − y1,1

0 )− (y0,11 − y0,1

0 )− (y1,01 − y1,0

0 )− (y0,01 − y0,0

0 )

The subscript one (zero) represents periods after (before) the minimum wage introduction.

The superscript shows that the group belongs to the construction industry if the first digit


is equal to one. If the second digit is one, the group consists of electricians (painters). The

regression equation takes the following form:

emplit =α + β1occi + β2indi + β3dt+ (2.3)

β4(occi ∗ indi) + β5(occi ∗ dt) + β6(indi ∗ dt) + β7mwit + εit

The dummy occi takes the value one if the group consists of electricians (painters), while

the dummy indi takes the value one for the construction industry. These fixed effects

eliminate time-constant differences between the different industries (manufacturing and

construction) as well as the different occupations (electricians/painters and clerks). Time

dummies for each quarter are given by dt. The interaction between the industry-specific

and occupation-specific fixed effects with the time dummies eliminate any time trends or

shocks that affect all occupations within an industry or one occupation across industries

equally. Finally, the dummy mwit takes the value one for electricians (painters) in the con-

struction industry in time periods after the minimum wage introduction. The coefficient

β7 therefore gives the effect of the minimum wage introduction on employment growth

emplit. The specification is estimated separately for electricians as well as painters.

The DDD estimator ensures that industry-specific and occupation-specific trends do not

bias the coefficient of the minimum wage introduction. However, there are reasons to

believe that both control groups might be affected by the minimum wage treatment in a

similar manner as a control group consisting of higher wage workers within the same in-

dustry applied within a DiD framework. While biases might occur due to spillover effects

as well as labour mobility, the probability of their existence as well as the expected magni-

tude of such an effect is considerably smaller compared to employing solely higher-wage

workers from the same industry as a control group.

First, electricians (painters) in the manufacturing sector could be affected by the mini-

mum wage introduction in the construction industry due to spillover effects in terms of

wages within occupations, but across industries. In the presence of spillover effects, wages

for electricians (painters) in the manufacturing sector would be increased as a reaction to


the minimum wage introduction for electricians (painters) in the construction industry.

Table 2.1 shows that average wages of electricians (painters) in manufacturing are consid-

erably higher compared to average wages of electricians (painters) in construction. How-

ever, wage dispersion also appears to be higher, implying that a considerable proportion

of workers in the control group earns less than or exactly the minimum wage at the time

of introduction. Thus, spillover effects at the lower end of the wage distribution cannot be

precluded.

Additionally, the minimum wage could induce electricians (painters) in manufacturing

to switch to construction. Alternatively, workers might move from the construction indus-

try to manufacturing after losing their job due to the minimum wage introduction. While

spillover effects and labour mobility from manufacturing to construction would overesti-

mate a positive employment effect (underestimate a negative effect), the opposite is true for

labour mobility from construction to manufacturing. Thus, it is not possible to determine

whether the estimator will be biased and if this is the case, in which direction.

Finally, clerks in the construction industry could be affected by the minimum wage in-

troduction insofar as the minimum wage depresses the output of the entire industry. In

this case the reduction in output is industry-specific and will therefore (falsely) not enter

the minimum wage coefficient. However, for this problem to occur, the minimum wage

introduction in one specific occupation (e.g. electricians) would have to depress the entire

construction industry with equal employment effects for all occupations. Given that clerks

are characterized by deviating individual characteristics (Table 2.1) and perform very dif-

ferent tasks compared to electricians (painters), equal employment effects for both occupa-

tions are unlikely, even if the minimum wage introduction in one occupation depresses the

overall industry output.

Data

The data employed in the empirical analysis is the BA Employment Panel supplied by

the Research Data Centre of the German Federal Employment Agency at the Institute for


Tabl

e2.

1.:S

umm

ary

Stat

isti

cspe

rG

roup

Wes

tGer

man

y

Empl

.A

vera

geM

inim

umK

aitz

Mar

gina

lU

nski

lled

Youn

gSm

all

Blue

Wag

esW

ages

Inde

xEm

pl.(

%)

Wor

kers

(%)

Wor

kers

(%)

Firm

s(%

)C

olla

r(%

)

Who

lesa

le/R

etai

ling

3889

924

70n.

a.n.

a.18

.47

9.38

11.1

437

.99

26.8

6C

omm

unic

atio

n/Tr

ansp

ort

1860

924

57n.

a.n.

a.14

.13

13.9

28.

6627

.60

61.9

1

Elec

tric

ians

1989

2347

1417

60.9

43.

705.

6117

.25

54.0

291

.15

Pain

ters

1424

2257

1729

75.9

74.

567.

1615

.28

70.2

310

0.00

Cle

rks

1952

2055

n.a.

n.a.

19.7

94.

518.

3853

.91

2.21

Elec

tric

ians

(Man

ufac

turi

ng)

5576

2954

n.a.

n.a.

2.14

7.45

11.3

915

.18

84.6

1Pa

inte

rs(M

anuf

actu

ring

)11

9526

24n.

a.n.

a.3.

8120

.71

10.0

122

.39

100.

00C

lerk

s(M

anuf

actu

ring

)22

975

2819

n.a.

n.a.

11.5

05.

5310

.60

22.4

92.

36

East

Ger

man

y

Empl

.A

vera

geM

inim

umK

aitz

Mar

gina

lU

nski

lled

Youn

gSm

all

Blue

Wag

esW

ages

Inde

xEm

pl.(

%)

Wor

kers

(%)

Wor

kers

(%)

Firm

s(%

)C

olla

r(%

)

Who

lesa

le/R

etai

ling

7113

1863

n.a.

n.a.

15.0

03.

689.

7951

.30

32.5

2C

omm

unic

atio

n/Tr

ansp

ort

5711

2091

n.a.

n.a.

10.2

84.

765.

4128

.36

61.7

5

Elec

tric

ians

829

1664

1247

75.8

72.

541.

8113

.44

53.5

395

.35

Pain

ters

472

1657

1517

90.1

84.

923.

5018

.37

63.6

110

0.00

Cle

rks

640

1705

n.a.

n.a.

13.9

82.

187.

8857

.45

4.35

Elec

tric

ians

(Man

ufac

turi

ng)

1193

2286

n.a.

n.a.

1.68

3.03

7.03

20.4

488

.09

Pain

ters

(Man

ufac

turi

ng)

216

1876

n.a.

n.a.

3.32

6.18

12.8

131

.67

100.

00C

lerk

s(M

anuf

actu

ring

)33

0822

31n.

a.n.

a.7.

652.

358.

2830

.50

4.66

Not

es:

Val

ues

are

aver

aged

over

the

obse

rvat

ion

peri

od19

98-2

008.

Min

imum

wag

esar

eav

erag

edov

erth

eti

me

peri

ods

duri

ngw

hich

they

exis

ted.

The

yar

eex

pres

sed

asm

onth

lygr

oss

wag

es,a

ssum

ing

that

actu

alw

orki

ngho

urs

are

equa

lto

cont

ract

ualw

orki

ngho

urs.

For

pain

ters

,the

min

imum

wag

eis

calc

ulat

edas

aw

eigh

ted

aver

age

ofth

em

inim

umw

age

for

unsk

illed

and

skill

edw

orke

rs,w

ith

wei

ghts

corr

espo

ndin

gto

the

shar

eof

the

resp

ecti

vegr

oups

amon

gto

tal

empl

oym

ent.

Dev

elop

men

tof

nom

inal

min

imum

wag

es:

At

the

min

imum

wag

ein

trod

ucti

on,

unsk

illed

(ski

lled)

pain

ters

earn

ede

7(e

9.20

)in

East

Ger

man

yan

de

7.69

(e10

.53)

inW

estG

erm

any.

In20

07th

em

inim

umw

age

amou

nted

toe

7.15

(e9.

37)i

nEa

stan

de7.

85(e

10.7

3)in

Wes

tGer

man

yfo

run

skill

ed(s

kille

d)pa

inte

rs.

In19

97(m

inim

umw

age

intr

oduc

tion

)ele

ctri

cian

sin

East

Ger

man

yre

ceiv

ede

6,40

Euro

san

de

8in

Wes

tGer

man

y.A

tthe

tim

eof

the

min

imum

wag

eab

olit

ion,

the

min

imum

wag

eha

dbe

enin

crea

sed

toe

7.40

inEa

stG

erm

any

ande

8.90

inW

estG

erm

any.

The

rate

sat

the

re-i

ntro

duct

ion

in20

07w

ere

slig

htly

high

er(e

7.70

and

e9.

20).

Sour

ce:B

Aem

ploy

men

tpan

el(S

chm

ucke

ran

dSe

th,2

009)

.Aut

hor’

sca

lcul

atio

ns.


Employment Research [Schmucker and Seth 2009]. The data are at the individual level and

present a 1.92 percent sample of all employees subject to social security contributions. They

are representative for all workers subject to social security payments, which amounted to

almost 32 million individuals in 2002. The data are quarterly and cover the time period

1998 - 2007. For each individual, several personal characteristics are included, such as

gross wage, type of employment, occupation, age, nationality and educational attainment.

Further, information at the establishment level is added, e.g. the economic sector and the

composition of the workforce.

In order to aggregate the data to the level of the industry, each individual employee is

assigned to one of the four groups.10 All persons who do not belong to one of the industries

of interest are dropped from the data set. Additionally, only regular, full-time employees

are kept. This excludes part-time employees, but also apprentices, interns and marginal

employees. The proportion of marginal employment in total employment, which is in-

cluded as a control variable in the basic specifications, is calculated for each sector before

the marginal employees are dropped from the dataset. Because marginal employment is

only separately identified from June 1999 onwards, the observation period used in this

study covers the years 1999-2007. After the data set is scaled down to regular, full-time

employment, the variables are aggregated at the sectoral level.

Two important variables are not contained in the original data set and are therefore

added from an external source. First, information about minimum wage rates is taken from

the Federal Bulletin (Bundesanzeiger), where each collective bargaining agreement declared

generally binding must be published. Second, the indicator for economic growth consists

of gross value added for the control groups and revenues for the treatment groups. These

data are obtained from the Federal Statistical Office (Statistisches Bundesamt).

10. The industry classification scheme contained in the data is rather broad. As a consequence it cannot beprecluded that some workers falsely enter the treatment groups that really are under a third minimum wageregime present in the main construction sector. However, any resulting bias should be very small in magnitudebecause first, their number is likely to be low and second, the main effects on wage and employment growthof the minimum wage in the main construction sector took place prior to the start of the observation period ofthis analysis [IAB, RWI and ISG 2011].


Figure 2.3.: Bite of the Minimum Wage

(a) West Germany (b) East Germany

Source: BA employment panel (Schmucker and Seth, 2009). Author’s calculations.The figure shows the proportion of workers earning exactly/less than the minimum wage. The vertical lines depictthe points in time when minimum wages were introduced (Electricians: First line - abolishment; Second line - re-introduction).

2.1.3. Results

A prerequisite for minimum wages to have any effect on employment is that wages at the

lower end of the distribution are increased due to the minimum wage introduction. While

the focus of this paper is not to analyse whether minimum wages are an effective tool to

increase the income of low wage workers, it is necessary to gain some insight on their effect

on the bottom percentiles of the wage distribution.

Figure 2.3 shows the proportion of workers earning less than the minimum wage and the

proportion of workers earning exactly the minimum wage. Insofar as the minimum wage

is binding, the proportion of workers earning exactly the minimum wage should increase,

while the proportion of workers earning less than the minimum wage should drop to zero.

In West Germany, such a pattern cannot be observed. In comparison with East Germany

(Figure 2.3), the proportion of workers earning less than the minimum wage prior to its


introduction is generally low. Further, this low proportion of workers earning less than or

exactly the minimum wage hardly reacts at all to the minimum wage policy. In contrast,

the expected reactions occur for the treatment groups in East Germany. For electricians,

the proportion of workers earning less than the minimum wage decreases from 20 percent

to 5 percent and for painters from more than 40 percent to 15 percent. The proportion

of workers earning exactly the minimum wage simultaneously increases for all treatment

groups.

Apparently, the fraction of workers earning less than the minimum wage does not de-

crease to zero in any treatment group. This can be attributed to measurement error. First,

the minimum wage is specified as an hourly wage rate, while individual wages in the data

are provided as gross monthly earnings. When identifying those workers earning less than

or exactly the minimum wage, it is assumed that each worker supplies exactly the num-

ber of weekly working hours as stipulated in the binding collective agreement.11 Second,

some measurement error may occur when identifying those workers eligible to the mini-

mum wage. In either case, as long as this measurement error is random and the analysis is

not conducted at the level of the individual worker, the results should not be distorted.

In conclusion, Figure 2.3 presents evidence that the minimum wage has been binding

in East Germany, but not in West Germany. This is not surprising given that the Kaitz

Index, which is defined as the ratio of minimum to average wages, is much lower in West

Germany compared to East Germany (Table 2.1). This result is in line with the estimates

reported by Konig and Moller [2009], Rattenhuber [2013] and IAB, RWI and ISG [2011] for

the main construction sector.

Table 2.2 shows the results of the DiD estimator for the effect of the minimum wage

introduction on employment. After the inclusion of the control variables, the estimated

coefficient of the minimum wage dummy is negative in East and West Germany alike. Both

coefficients are not statistically significant. This result is less surprising for West Germany,

because the minimum wage hardly binds. In contrast, the minimum wage does affect the

11. The collective agreement is binding for all workers and stipulates that mean, monthly working hoursshould be equal to a specific number over the course of 12 months.


Table 2.2.: The Employment Effect of the Minimum Wage IntroductionWest Germany East Germany

Model 1 Model 2 Model 1 Model 2

Minimum wage dummy 0.0045 (0.0084) −0.0009 (0.0075) 0.0083 (0.0143) −0.0104 (0.0120)Marginal employment −0.0689∗∗∗(0.0205) −0.0155 (0.0105)Macroeconomic growth 0.3185∗∗∗(0.0996) 0.5578∗∗∗(0.1217)

Communication & Transport 0.0249∗∗∗(0.0072) 0.0211∗∗∗(0.0068) 0.0123 (0.0176) 0.0041 (0.0096)Electricians −0.0075 (0.0084) 0.0014 (0.0077) −0.0231 (0.0196) 0.0074 (0.0136)Painters −0.0220∗ (0.0125) −0.0038 (0.0111) −0.0528∗ (0.0282) 0.0145 (0.0213)

Industry dummies yes yes yes yesTime dummies yes yes yes yes

R2 0.532 0.620 0.484 0.653Observations 124 124 124 124

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Standard errors in brackets. All models are estimated by Prais-Winston regression to allow for AR(1) errors within panels(Wooldridge test for autocorrelation in panel data suggests autocorrelation of first degree). Standard errors are adjusted for het-eroskedasticity between panels (Breusch-Pagan test suggests presence of heteroskedasticity).Source: BA employment panel (Schmucker and Seth, 2009). Author’s calculations.

wages of a significant proportion of workers in East Germany. However, as Kramer [2011]

points out, the size of the estimated effect should not be disregarded completely only due to

a lack of statistical significance. While the effect in West Germany is also very low in terms

of economic significance, the coefficient in East Germany suggests that the minimum wage

might have led to a decrease in the growth rate of employment of about one percentage

point.

The industry-specific, macroeconomic growth exhibits a positive influence on the growth

rate of employment. In West Germany, the joint significance of the industry dummies

drops with the inclusion of this indicator. This observation may be interpreted as evidence

that the time-constant differences in employment growth between the industries during

the observation period can be explained by time-persisting differences in industry-specific

growth, with the construction industry performing continuously worse than the commu-

nication and transport industry or wholesale and retailing.

The coefficient of the growth rate of the proportion of marginal employment is signifi-

cant and negative in West Germany. However, it should be noted carefully that the coef-

ficient is rather small in magnitude. It amounts to 0.05, implying that an increase of one

percentage point in the growth rate of the proportion of marginal employment leads to a


Table 2.3.: Estimation Results of the DDD EstimatorWest Germany East Germany

Electricians Painters Electricians Painters

Minimum Wage −0.0063 (0.0049) −0.0027 (0.0118) 0.0060 (0.0120) 0.0421∗∗ (0.0196)Occupation 0.0010 (0.0058) −0.0321∗∗∗(0.0113) −0.0299∗∗ (0.0151) −0.0265 (0.0198)Industry −0.0039 (0.0058) 0.0014 (0.0113) −0.0369∗∗ (0.0151) −0.0216 (0.0198)Occupation*Industry 0.0165∗∗∗(0.0054) 0.0170∗ (0.0099) 0.0087 (0.0144) −0.0421∗∗ (0.0191)

Time dummies yes yes yes yesTime dummies*Occupation yes yes yes yesTime dummies*Industry yes yes yes yes

R2 0.911 0.846 0.817 0.864Observations 124 124 124 124

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Standard errors in brackets. All models are estimated by Prais-Winston regression to allow for AR(1) errors within panels(Wooldridge test for autocorrelation in panel data suggests autocorrelation of first degree). Standard errors are adjusted for het-eroskedasticity between panels (Breusch-Pagan test suggests presence of heteroskedasticity) except for electricians in East Germany.Source: BA employment panel (Schmucker and Seth, 2009). Author’s calculations.

decrease of around 0.05 percentage points in the growth rate of full-time regular employ-

ment.

Two separate interpretations may explain the observed relationship. First, as a less ex-

pensive type of employment, marginal employment may generally crowd out a fraction of

regular, full-time employment, independent of the minimum wage. Second, the negative

coefficient of marginal employment may be interpreted as a substitution of marginal for

regular employment in reaction to the minimum wage. In order to analyse the latter issue,

an interaction term between the growth rate of marginal employment and the minimum

wage dummy is included in the specification. However, as the coefficient of this interaction

term remains insignificant, the first explanation appears more likely.12

The results of the DDD estimator, which are presented in Table 2.3, are in line with the

DiD estimator. In order to make the results of the two identification strategies more com-

parable, Table A.1 in the appendix provides separate DiD estimations for each treatment

group. The separate estimations based on the DiD estimator reveal that the negative coef-

ficient of the minimum wage dummy in the joint estimations is driven by painters, as the

estimated effect for electricians is of low economic and statistical significance. The results

obtained with the DiD and the DDD estimator are therefore consistent with each other in

12. These results are not presented in any table, but can be obtained from the author upon request.


the case of electricians.

For painters in East Germany the DiD estimator suggests a neutral employment effect of

the minimum wage introduction (Table A.1), while the coefficient of the minimum wage

dummy is positive and significant at the 5percent level when employing the DDD esti-

mator (Table 2.3). One possible explanation is that other time-varying determinants of

employment growth in addition to included controls exist, which affect the treatment and

control groups differently. In this case the DiD estimator would be biased. The DDD esti-

mator, in contrast, is better able to deal with such time-varying, unobserved heterogeneity

as long as these trends are occupation-specific or industry-specific. To this extent it should

be noted that the occupation-specific time dummies as well as the industry-specific time

dummies are jointly significant at the 1 percent level in all specifications, which points

towards the existence of such trends.

Alternatively, the DDD estimator might be biased due to wage spillover effects or labour

mobility between industries (compare with Section 2.1.2). If such effects are present, the

employment level of painters in manufacturing is decreased as a reaction to the minimum

wage introduction for painters in construction. Consequently, the estimated coefficient

of the minimum wage dummy may be biased upwards. It is impossible to determine

whether the deviating results are caused by a bias in the DiD or the DDD estimator. How-

ever, the DiD and DDD estimators are consistent with each other in the remaining three

specifications. Additionally, independent of the nature of a possible bias, the explanations

discussed for the neutral employment effect apply equally well to a possibly positive effect

of the minimum wage introduction.

At least four, not mutually exclusive, explanations exist for the neutral employment ef-

fect of the minimum wage introduction in East Germany. First, the aggregate measure

of employment at the occupational level might hide substitution effects between different

types of workers. Stated differently, the employment effects could be unfavourable for the

affected workers, but neutral at the aggregate employment measure at the occupational

level. For example, if the minimum wage increases the price of unskilled labour, firms

might exchange unskilled workers for skilled ones, who become relatively less expensive.


In order to analyse this issue more deeply, all specifications are estimated separately for

skilled and unskilled workers, young and older workers, as well as workers employed in

small and large firms. The estimated coefficient of the minimum wage dummy remains

insignificant and very small in magnitude independent of the subsample used.13 Thus,

it appears unlikely that substitution effects occurred. The absence of such effects can be

explained by the very homogenous composition of the workforce in the treatment groups.

Only a maximum of 4 percent of the workers does not have any vocational training in East

Germany and the majority of the workforce is employed in small firms (Table 2.1).

Second, the neutral employment effect could be caused by monopsonic competition. As

Figure 2.1 shows, the coverage rate of collective bargaining is much lower in East Germany

compared to West Germany and has been continuously decreasing during the last decade.

This development might have increased the market power of employers, thereby giving

individual firms some discretion in setting wages. As a consequence, the equilibrium es-

tablished without minimum wages might be supply-side constrained, which implies that a

minimum wage does not necessarily reduce employment. In this context, it is imaginable

that skilled craftsmen, such as electricians or painters, move to the West German labour

market due to large wage differentials (Table 2.1), thereby causing a shortage of skilled

labour supply in East Germany. In this case, the price elasticity of labour supply, instead of

labour demand, would be relevant when predicting the employment effects of a minimum

wage introduction.

A third explanation is noncompliance with the minimum wage. To the extent that em-

ployers simply pay sub-minimum wage rates it is not surprising that the employment ef-

fect of the minimum wage is neutral. While no official statistics exist on noncompliance,

several enquiries of two smaller parties to the German government and the correspond-

ing answers suggest that controls are infrequent and penalties are hardly prohibitive.14

Still, as Figure 2.3 shows, gross monthly wages are increased with the minimum wage

13. These results are not presented in any table, but can be obtained from the author upon request.14. For example, in 2010 the left-wing party “Bundnis 90/Die Grunen” submitted an enquiry to the Ger-

man government in order to obtain more information on the results of inspections in the construction sectorconcerning minimum wages [Deutsche Bundesregierung 2010].


introduction in East Germany. Thus, it would have been necessary to increase the num-

ber of working hours significantly in order to achieve sub-minimum wage rates despite

higher gross monthly wages. Given that all workers in the sample are full-time employed,

noncompliance through a higher number of working hours might explain partly, but not

entirely, why the minimum wage has a neutral employment effect.

Finally, the minimum wage introduction might not have affected the employment growth

of domestic, but that of foreign workers. Recall from Section 2.1.1 that the initial aim of the

minimum wage introduction in Germany was to ensure that foreign (posted) workers are

subject to the same minimum standards as domestic workers. In contrast, the data set only

covers domestic workers paying social security contributions. Thus, the possibility ex-

ists that the minimum wage has been employment neutral (or even positive) for domestic

workers, while it reduced the employment of foreign workers.

The evaluations of the BMAS for electricians [IAW 2011a] and painters [IAW 2011b] reach

the same conclusions as the study at hand: they report positive effects on wages in East

Germany, but do not find any robust employment effects. In contrast, two other empirical

studies of the employment effect of minimum wages in Germany report deviating results.

First, the already mentioned study by Konig and Moller [2009] reports positive employ-

ment effects in West Germany and negative employment effects in East Germany for the

main construction industry. However, the effects in West Germany are of low statistical

significance. Compared to the study at hand, these deviating results may be caused by dif-

ferences in the control group design, different treatment groups or the deviating empirical

strategy, which focuses on the estimation of individual probabilities of continuing employ-

ment, thereby ignoring new employment relationships that started during the observation

period.

Second, Bauer et al. [2009] analyse the effect of different (hypothetical) statutory mini-

mum wage rates on aggregate employment, separately for East and West Germany. While

this study consists of a simulation analysis, it may still be classified as semi-empirical be-

cause the majority of parameters are derived from micro-data. The resulting employment

effects are clearly negative. However, Bauer et al. [2009] make the crucial assumption that a


minimum wage introduction will lead to a demand-constrained equilibrium. This assump-

tion is a direct consequence of a perfectly competitive labour market. The equilibrium prior

to the minimum wage introduction might just as well have been supply-constrained either

due to monopsonic competition or imperfect labour mobility.

2.1.4. Conclusion

This paper analyses whether the minimum wage has influenced the employment level in

two occupations belonging to the construction industry, painters and electricians. Strong

emphasis is placed on the choice of an appropriate control group as well as the choice

of a sensible minimum wage indicator. Both considerations are highly important for the

identification assumption of the DiD estimator, because minimum wages in Germany are

industry-specific and collectively bargained.

Industry-specific minimum wages imply that a fundamental trade-off exists when set-

ting up an appropriate control group design. First, each control group constructed within

the same industry is most likely subject to the minimum wage treatment as well. Sec-

ond, any control group established outside the industry in question is most likely affected

by other determinants of employment over time in addition to the minimum wage. I pro-

pose to solve this trade-off by using control groups from other industries, while controlling

for industry-specific macroeconomic growth as an additional determinant of employment

growth. The DDD estimator provides a robustness check to the results obtained with the

DiD estimator, because both estimators are subject to different biases. As long as the two

estimators yield consistent results, the results obtained with the DiD estimator appear ro-

bust to biases due to time-varying, unobserved heterogeneity.

I find no robust effect of the minimum wage introduction on employment growth in

East Germany, despite the fact that the minimum has been affecting a significant propor-

tion of the workforce. Possible explanations include substitution effects between different

types of workers, noncompliance and monopsonic competition. First, substitution effects

between skilled and unskilled workers generally seem highly likely, but can be rejected in


the case at hand due to the very homogenous nature of the treatment groups. In contrast,

substitution of foreign workers for domestic ones cannot be precluded and is a sensible ex-

planation, because the treatment groups belong to the construction industry where posting

of workers is common. To the extent that a reduced number of posted workers is the main

explanation for the neutral employment effect for domestic workers, the minimum wage

achieved its initial aim of protecting domestic workers from low-wage competition from

other EU countries. To answer the question whether or not a minimum wage should serve

such a purpose is open for debate.

Second, noncompliance with the minimum wage is a further possible explanation for the

neutral employment effect. To the extent that workers are required to supply more hours,

the true hourly wage rate might be below the minimum wage despite increasing monthly

gross wages. Even though infrequent controls make this a viable option, the analysis of the

fraction of workers affected by the minimum wage shows that a considerable proportion

of workers have experienced an increase in wages with the minimum wage introduction.

Stated differently, even if not all firms comply with the minimum wage, the majority ap-

pears to do so. Therefore, noncompliance cannot explain the overall neutral employment

effect by itself.

Last but not least, monopsonic competition may serve as an explanation for the obtained

results. In this case, the decrease of the coverage rate of collective bargaining during the last

decade has provided individual employers with some discretion in setting wages. Then,

the minimum wage simply counteracts the monopsonic power of firms, thereby taking on

the same function as collective bargaining did previously. Other non-employment chan-

nels of adjustment, such as reduction in profits or an increase in product prices, are also

imaginable if imperfect competition prevails on the labour or on the product market. To

determine whether such explanations are plausible for the case at hand is beyond the scope

of this paper. To this end, additional research on whether labour market segments with a

low coverage rate of collective bargaining in Germany are rather characterized by perfect

or monopsonic competition promises to be interesting and fruitful.

Generally, it is dangerous to infer from the presented results that minimum wages will


have no disemployment effects in Germany. Electricians and painters both belong to the

construction industry, which is special for at least three reasons. First, to the extent that

substitution of foreign for domestic workers is the main explanation for the neutral em-

ployment effect, a minimum wage introduction in other industries might well lead to neg-

ative employment effects. Second, the German minimum wage institution implies that a

self-selection into the minimum wage treatment exists, because both social partners have to

agree on the policy change. Third, the treatment groups consist predominantly of skilled,

full-time employed men.

Future research on the employment effects of minimum wages in Germany should there-

fore focus on industries not belonging to the construction sector, where minimum wages

were introduced more recently. Examples include the commercial cleaning industry (min-

imum wage introduction in 2008), laundry services (minimum wage introduction in 2009),

or elderly care (minimum wage introduction in 2010).15 In contrast to the construction sec-

tor, protectionist motives appear to be less relevant for the minimum wage introduction in

these industries. Instead, the expansion of low-wage labour markets is more important. If

these minimum wage introductions also show to be employment-neutral, other explana-

tions than substitution of domestic for foreign workers should be considered. One obvious

choice is monopsonic competition in the labour market.

15. The evaluations contracted by the BMAS are a good starting point in this context.


2.2. High-Impact Minimum Wages and Heterogeneous Regions

Over the last two decades, there has been a considerable amount of research on the eco-

nomic impacts of minimum wages. While earlier studies established a firm professional

consensus that minimum wages unambiguously increase unemployment [Minimum Wage

Study Commission 1981; Alston, Kearl and Vaughan 1992], renewed interest in the topic

was triggered by a series of papers in Industrial and Labor Relations Review [Card 1992; Neu-

mark and Wascher 1992; Katz and Krueger 1992] and especially by the influential book by

Card and Krueger [1995]. Since then, the US debate over whether minimum wages are nec-

essarily detrimental to overall employment has not subsided [Deere, Murphy and Welch

1995; Card and Krueger 2000; Neumark and Wascher 2000, 2008; Dube, Lester and Reich

2010; Neumark, Salas and Wascher 2013]. Perhaps unsurprisingly, the latest studies do not

appear to be leading toward a renewed consensus as of yet.16

Regulating the price of factors of production takes on various forms, and one can expect

the institutional framework to play a decisive role. For instance, in the UK, “wages coun-

cils” initially set minimum wages for specific industries for much of the 20th century until

they were replaced by the Low Pay Commission in 1997, which was tasked to establish

a national minimum wage [Dickens et al. 1993; Brown 2009]. Other studies outside the

US have been conducted for France [Abowd et al. 2000], Spain [Dolado, Felgueroso and

Jimeno 1997], and the Netherlands [Machin and Manning 1997, including France, Spain,

and the UK]. In the US, the federal minimum wage, currently set at USD 7.25, is a mere

44 percent of the median hourly wage of all occupations in 2011 [Bureau of Labor Statistics

2011]. This figure is much lower than the percentages seen in other OECD countries. The

literature on minimum wages in Germany, which has unique but nevertheless enlighten-

ing characteristics, is relatively recent, and a full exposition is offered in the subsequent

section.

16. In 2006, PhD-holders of the AEA were about evenly split between (1) raising the minimum wage orkeeping it at its current level and (2) decreasing it or completely eliminating it [Whaples 2006], although, tobe sure, supporting the former does not mean one has to abandon the belief that minimum wages increaseunemployment.


Methodologically, estimating the effect of minimum wages on employment has also sub-

stantially evolved, closely tracking the developments in econometric techniques and the

availability of data. Much earlier research on the topic used traditional panel-data esti-

mation methods which were common at that time (e.g., Neumark and Wascher [1992]).

In this line of research, national estimates are obtained by exploiting the cross-state varia-

tion in minimum wages over time. The “new” minimum-wage research, typified by Card

and Krueger [1994], is based on treatment group–control group comparisons: case stud-

ies comparing neighbouring geographic areas where one part is affected by a change in

the minimum wage (in a sense, a geographic difference-in-differences approach). The pre-

ferred method of the researcher seems to influence the outcome of the estimation exercise:

panel-data methods often arrived at the result that minimum wages have an overall neg-

ative effect on employment while the two-group, two-mean comparisons tended to show

either positive or neutral effects.

As noted by Dube, Lester and Reich [2010], both approaches have advantages and dis-

advantages over the other (the details of which are available in Sec. 2.2.3), particularly

in terms of controlling for spatial heterogeneity. Cognizant of this, Dube, Lester and Re-

ich [2010] offered a synthesis of the two approaches by essentially generalizing the Card

and Krueger [1994] method of comparing bordering regions with different levels of the

minimum wage. Using all counties in the US along a state border, where the two states

have different levels of the minimum wage, allowed them to make Card and Krueger-type

comparisons for many counties over time. This addresses the problems caused by spatial

heterogeneity arising from the exclusive use of either the panel-data method or the single

border-pair approach.

Using both the panel-data approach and the generalized border-pair approach, we con-

tribute to the emerging international literature that examines minimum-wage effects out-

side the well-studied US context, which is characterized by a low treatment intensity as op-

posed to the German case analysed here. Our study examines the wage and employment

effects of the introduction and subsequent increases of a minimum wage in the German

main construction sector (Bauhauptgewerbe) between 1997 and 2002. This is the first sector


where negotiations over a generally-applicable minimum wage were concluded.17 At the

time the wage floor was set, the sector was employing about 1.3 million workers, mak-

ing it the largest German industry where minimum wages apply today [Bachmann, Konig

and Schaffner 2012]. As of 2011, its share of Germany’s GDP is substantial at 4.4 percent.

Furthermore, the minimum wages for East and West Germany have been adjusted almost

annually since their introduction. This political propensity to regulate wages is unlikely to

wane,18 necessitating a careful evaluation of their effects to inform policy.

What makes the German construction case particularly interesting is the fact that the in-

dustry was subjected to a substantial minimum wage. In contrast to the minimum-wage

changes contemplated in the previous US studies, the Bauhauptgewerbe, particularly in for-

mer East Germany, experienced very high treatment intensities (e.g., the maximum re-

gional share of main construction workers earning below the minimum wage is almost

41 percent). The behaviour of the labour market in response to a slight change in the mini-

mum wage may not necessarily be similar to how it will react to a significant and sustained

increase of the wage floor. Therefore, the results here can be applied to situations where

the minimum wage to be introduced and maintained is of a significant magnitude.

Since the share of workers earning below the minimum wage in this sector varies from

one region to the next, one can exploit the spatial variation to estimate the impact of the

wage floor on various socioeconomic outcomes [Card 1992; Stewart 2002]. Our interest lies

in the effect of minimum wages on wage and employment growth rates in East and West

Germany. These two variables are expected to reflect the first-order effects of minimum

wages, and their response behaviour at the regional level has thus far not been adequately

described and quantified. The use of regional data in this case has the added benefit of

overcoming some of the problems associated with previous studies using individual-level

data in Germany, such as the difficulty in identifying who the actual recipients of the min-

17. The other sectors that have subsequently introduced minimum wages are waste removal, coal mining,roofing, electrical installation, commercial cleaning, painters and varnishers, nursing care, security services,industrial laundries, temporary work, and education and training services.

18. Indeed, US President Barack Obama called for about a 25-percent increase in federal minimum wages inhis 2013 State of the Union address. In Germany, the Greens, the Social Democrats, and now the ChristianDemocrats (the party of the chancellor, Angela Merkel) are backing an economy-wide minimum wage.


imum wage are. Hence, the identification of the appropriate control group for comparison

(either in the form of individuals above the minimum wage or industries unaffected by

the minimum wage, both of which have been employed in earlier work by others) is not a

relevant complication that can arise in our case.

Moreover, our contribution advances the literature further by, first, taking into account

potential spillover effects between regional labour markets and, second, controlling for

heterogeneity at a local level. The importance of those factors has been discussed in the

US minimum-wage literature only recently [Dube, Lester and Reich 2010; Allegretto, Dube

and Reich 2011; Kalenkoski and Lacombe 2013; Neumark, Salas and Wascher 2013], and

the appropriate techniques have thus far never been applied to a situation similar to the

German institutional setting.

Taking developments in spatial econometrics into account, we recognize that the pres-

ence of both spatial heterogeneity and spatial autocorrelation may bias traditional esti-

mates of the effect of the minimum wage. Even in the more developed US literature, spatial

spillovers have not been thoroughly addressed as much as the issue of spatial heterogene-

ity. Inadequately addressing spatial issues may explain the discrepancies in outcomes of

recent German studies (e.g., Konig and Moller [2009]; Apel et al. [2012]). It is not clear a

priori that these issues should not affect the estimated impact, especially given the nature

of the issue at hand (factor mobility, for example), so explicitly taking this into account is

an important step to undertake. By subjecting the data to a rigorous analysis that allows

for various forms of spatial effects, we attempt to rule out the possibility for those effects

to have a decisive impact on our results.

Our findings are the following. First, we conclude that the new wage floor had a negli-

gible impact on wage growth in West Germany since wages were relatively high to begin

with and the percentage of directly affected workers was therefore very small. In East Ger-

many, however, where wages in the construction sector were considerably lower, the new

minimum wage led to a significant increase in the wage growth rate. In this case, an in-

crease by one standard deviation of the percentage of affected workers is associated with

an increase of the growth rate of average wages by approximately 1.2 percentage points.


Second, while we do not find any effects on employment in West Germany, the negative ef-

fect on East German employment was quite large. Our estimates show that an increase by

one standard deviation of the percentage of affected workers is associated with a reduction

in the employment growth rate by 2.6 to 3.1 percentage points. Third, we provide evidence

that spatial spillover effects and regional heterogeneity do not alter our main results.

2.2.1. Minimum Wages in Germany and Previous Evaluations

In this section, we provide the historical and institutional background for minimum wages

in Germany, particularly for the main construction industry. We also review the existing

evaluation literature on its effects, focusing on those studies that address the issue of wage

and employment responses of the labor market. We also differentiate our approach from

the previous studies, and argue that the our methodology a more credible estimate of the

effect of minimum wages.

Institutional Details on Germany’s Minimum Wages

Minimum wages in Germany are special because they do not derive from a federal or state

law that mandates a specific wage floor.19 Unlike the US (where the majority of studies

on minimum-wage effects originates), the wage floors are set via collective bargaining be-

tween employees’ unions and employers’ associations at the industry level. The collective-

bargaining agreements (CBAs) can then be declared to be universally binding by the Fed-

eral Ministry of Labour and Social Affairs (BMAS). Once that occurs, the wage floor will

apply to all workers in that particular industry, irrespective of whether they belong to the

bargaining workers’ union or not.

One of the reasons to have a minimum wage established through a CBA is that, in com-

bination with the Posting of Workers Law (Arbeitnehmer-Entsendegesetz), it also applies to

workers sent by firms from other European Union member states and so-called “third

countries”. Therefore, in contrast to the motivation for minimum wages in other coun-

19. For more details on the German institutional setting, see IAB, RWI and ISG [2011] and the sources citedtherein.


tries, where wage regulation is typically considered an anti-poverty measure [Sabia and

Burkhauser 2010], the wage floor in Germany—at least for the construction sector—is an-

chored squarely on protectionist and anti-competitive reasons, with poverty alleviation

taking a backseat.

The evolution over time of the minimum wages established in the main construction

sector since its introduction is presented in Figure 2.4 separately for East and West Ger-

many. The differential minimum wages between East and West Germany reflect the fact

that wages are, on average, lower in the East. In general, one can observe that the nominal

minimum wage has been increasing over time except for a dip in 1998–1999. However, in

real terms, the minimum wage has remained rather stable and close to the level at which it

was first introduced, exhibiting an increase of roughly 5 and 10 percent for East and West

Germany, respectively, for the period between 1997 and 2000. Therefore, if there is any

effect on wage and employment growth rates, one can expect it to materialize in the years

immediately after its introduction.

In terms of the threats to estimating the effect of the minimum wage, there is the ques-

tion of whether the introduction of the wage floor in this industry was anticipated by the

workers and employers, and whether they had enough time to adjust their behaviour. The

negotiations that ultimately led to the introduction of the minimum wage in January 1997

were rather difficult. Consequently, it is unlikely that employers anticipated the exact date

of the minimum-wage introduction and thereby possibly mitigating the effect of the treat-

ment.

Prior to the introduction of the minimum wage in 1997, the coverage of sectoral (but not

universally binding) CBAs in German construction was already generally high. Based on

1995 firm-level data, sectoral CBAs in West Germany covered 81 percent of establishments

[Kohaut and Bellmann 1997]. In the East, the coverage rate was around 40 percent [IAB,

RWI and ISG 2011]. In a sense, therefore, the industry under investigation is not typical of

other low-wage industries where minimum wages exist. It certainly is structurally differ-

ent from the subjects of previous studies in the US, such as fast-food workers or teenage

employees.


Figure 2.4.: REAL AND NOMINAL MINIMUM WAGES, 1997–2002

(a) West Germany8

8.5

99.

510

Hou

rly W

age

Rat

e

1997 1998 1999 2000 2001 2002Year

Nominal Minimum wage Real Minimum Wage

(b) East Germany

7.8

88.

28.

48.

6H

ourly

Wag

e R

ate

1997 1998 1999 2000 2001 2002Year

Nominal Minimum wage Real Minimum Wage

Note: The nominal minimum wage has been deflated with the producer price index obtained from the Federal StatisticalOffice. The figure shows the minimum wage rates in place in January of the year in question. Note that increases usuallytake effect in September, while employment and average wages are measured on June 30th each year.Source: Own data collection and Federal Statistical Office.


Another peculiarity in the introduction of the minimum wage in Germany’s construc-

tion sector is that it came at a time of much economic contraction. Specifically, nominal

gross value added (in billion EUR) dropped from 112.97 in 1995 to 99.21 in 1998 [IAB, RWI

and ISG 2011]. The present study should therefore be taken to shed light on the question

of the effects of a minimum-wage introduction of a substantial magnitude in times of an

economic contraction in a specific industry.

Previous Studies on Wage and Employment Effects

While the effects of the minimum wage in Germany’s construction industry have already

been studied for a variety of outcomes, we review only those that specifically deal with

the effects on wages, employment, or both. In ascending chronological order, the follow-

ing studies are relevant: Bauer et al. [2009],20 Buttner and Ebertz [2009], Konig and Moller

[2009], Apel et al. [2012],21 Muller [2010], Frings [2013], Aretz, Arntz and Gregory [2013]22,

and Rattenhuber [2013]. The detailed results of these studies are more competently ad-

dressed therein, but we present the general conclusions derived from this collective body

of research and note relevant exceptions.

First, in terms of establishing whether the minimum wage in fact increased average

wages or wage growth, the consensus view is that this is indeed the case for East Ger-

many. The results for the West German wage distribution are less consistent. Buttner and

Ebertz [2009] did not explicitly examine the case of East vs. West but rather rural vs. urban

areas (“countryside” vs. “city”), and their simulation showed that wages in the countryside

will increase as a result of the introduction of minimum wages. Larger effects are generally

found in the East relative to the West, implying that the so-called “bite” (i.e., a measure of

treatment intensity explained in detail below) of the minimum wage was more intensive

in the former compared to the latter.

20. Bauer et al. [2009] did not explicitly analyze the case for the construction industry. Instead, they estimatedthe fiscal and employment effects by skill level in response to hypothetical values of the minimum wage.

21. Apel et al. [2012] is the scientific publication that arose out of the project report that is the IAB, RWI andISG [2011] study commissioned by the BMAS.

22. Aretz, Arntz and Gregory [2013], more specifically, looks only at the roofing sector.


There is more contention about the estimates on employment, where Konig and Moller

[2009], among the first few studies, stand out in claiming that, despite the minimum wage

somewhat accelerating wage growth, the employment growth rate also actually increased

in West Germany. Notably, the rest of the other studies do not support this conclusion. To

be fair, however, the estimated effect in Konig and Moller [2009] is not robust to all their

specifications, and they are careful not to over-interpret their result. The results for East

Germany are also not in accordance with each other: while Apel et al. [2012] and Frings

[2013] found neutral employment effects despite the positive effects on wages, Konig and

Moller [2009] and Muller [2010] conclude that the minimum wage had a negative impact

on employment.

With the sole exception of Buttner and Ebertz [2009], which is a simulation and does not

specifically address the minimum wage currently under investigation, none of the afore-

mentioned studies focuses on spatial effects of the minimum-wage introduction. However,

ignoring potential spillovers and unobserved spatial heterogeneity can lead both to incor-

rect coefficient estimates and to incorrectly estimated standard errors. We attempt to fill

this research gap to find out whether the previous literature is impaired by this omission.

We use spatial econometric techniques and a recently-developed approach that generalizes

regional natural experiments to control for spatial effects in our subsequent analysis.

The source of variation that we exploit in our study is based on treatment intensity,

which we defined as the minimum-wage bite and the minimum-wage gap, and the unit of

analysis is the district level. This has many advantages compared to using individual-level

data where the division between the treatment and control group is not strict, especially in

the German case where the number of hours worked is not directly recoverable from the

data. For example, in the study of Konig and Moller [2009], they use a probabilistic ap-

proach to identify whether an individual belongs to the treatment or control group under

an assumed distribution, which may or may not be correct (indeed, the size of the con-

trol group varies between 4 percent to 18.3 percent in the paper). Moreover, their use of

high-wage earners as the control group may not be valid, as spillover effects over across

the wage distribution has been empirically documented (see, for example, Stewart [2012]),


apart from the fact that the control group is rather small.

2.2.2. Data Construction and Description

This empirical study is based on administrative data that cover the entire population of

construction workers in Germany who are subject to social security contributions. The

data were drawn from the Integrated Employment Biographies (IEB, Integrierte Erwerbs-

biographien) at the Research Data Center based at the Institute for Employment Research

of the Federal Employment Agency. The dataset covers all workers that were employed

in the main construction sector at any point in time during the period 1993–2002. We are,

however, unable to consider the case of posted or foreign workers in Germany as well as

the changing situation with the number of self-employed construction workers.

The data contain sociodemographic as well as employment characteristics, including the

average daily wage.23 The analysis is limited to full-time employed men for two reasons:

(1) part-time employment is rare among blue-collar workers in the main construction sec-

tor and (2) the share of women among blue-collar workers in this sector is extremely low.

Unfortunately, no information on hours worked, which is necessary to calculate hourly

wage rates, is available. This information is crucial because the minimum wage itself is ex-

pressed in an hourly basis. IAB, RWI and ISG [2011] impute the number of hours usually

worked for full-time workers in main construction based on available information from

the census (Mikrozensus) for the years 1993–2002. We adopt their results for our calculation

of hourly wages. This involves estimating a linear model for the usual working hours as

a function of various individual, firm, and job characteristics, as well as indicators for the

federal state, available in the Mikrozensus. The estimated parameters are then used to cal-

culate the cross-sample predicted values based on data available from the IEB. Ultimately,

full-time employed workers appear, on average, to work roughly 40 hours per week irre-

23. Average daily wages are right-censored at the social security contribution limit, i.e., the wage at whichsocial security contributions no longer increase. Because the majority of construction workers earn wagesbelow this limit, any possible downward bias of average wages should be negligible, and we therefore do nottake this further into account.


spective of their individual or job characteristics.24

One advantage of using spatial variation for the identification of the minimum-wage ef-

fect is that any error in measurement of the hourly wage rates should not bias the results

as long as the error is random across individuals within regions. Stated differently, even

if wage rates are incorrect at the individual level, these measurement errors should cancel

out at the aggregate district level. In contrast, such an error is more critical when trying

to identify individuals who are (not) affected by the minimum wage, which is the strat-

egy employed in previous studies in this context. Moreover, if one supposes that worker

substitution takes place in this industry, that the level of aggregation is at the district level

implies that we are able to abstract from this issue. Therefore, even though the individual-

level error may persist, the higher-level aggregation of our data allows us to circumvent

potential problems that are present in earlier evaluations of this policy.

The IEB are spell data with specific days for the beginning and end of each spell. We

transform the data into annual observations using June 30 as the cutoff date each year.

That is, each male blue-collar worker employed in the main construction industry on that

day remains in our operational dataset.25 One advantage of the annual data is that seasonal

effects (e.g., the decline in employment in winter) become tangential for the analysis of the

employment effect of the minimum wage.

The data are regionally disaggregated down to the level of districts (Kreise und kreisfreie

Stadte, NUTS 3), and we use detailed industry classifications to define the construction

industry.26 Most subsectors of the main construction industry belong unambiguously to

the treatment group. In the rare cases in which a subsector does not clearly belong to the

24. We provide evidence of this in the supplementary material.25. For completeness, two more annual datasets were constructed by considering all spells for three- and six-

month intervals. The results based on these datasets do not change our conclusions and can be made availableupon request. Note, however, that additional difficulties are introduced concerning the calculation of wageand employment levels using these alternative datasets since an individual may change employers or mayleave the main construction industry within the extended intervals. Using one specific date circumvents theseissues.

26. We follow IAB, RWI and ISG [2011] in the choice of the relevant subsectors. These are based on theclassification scheme of 1973 and include the following economic groups (prefixed by their numeric codes):[590] general civil engineering activities, [591] building construction and civil engineering, [592] civil andunderground construction, [593] construction of chimneys and furnaces, [594] plasterers and foundry dressingshops, and [600] carpentry and timber construction.


treatment group (i.e., the subsector “scaffolding” (with a numeric code of [616]) covers the

production as well as the installation of scaffolds, while only the latter is covered by the

CBA), it is excluded from the analysis in order to ensure that the treatment effect will not

be underestimated simply because some individuals in the treatment group are not really

affected by the minimum wage.

The observation period of our operational dataset ends in 2002, which is not due to data

limitations per se, but the fact that a second, higher minimum-wage rate was introduced for

skilled workers in 2003. Unfortunately, the data do not allow us to unambiguously identify

which minimum-wage rate is applicable to which worker. In order to avoid measurement

error which cannot be eliminated via aggregation, this study therefore concentrates on the

time period from the introduction of the minimum wage in 1997 up to 2002.

The dataset is transformed from the individual to the district level. The two dependent

variables are average wage and employment growth rates in each district. The mean wage

of all construction workers eligible for the minimum wage in each district is calculated,

while employment corresponds to a head count of full-time male workers. Annual growth

rates are then computed. The minimum-wage treatment is measured by the bite, which is

defined as the share of workers earning below the minimum wage in the period prior to

its introduction or increase.

An alternative measure of the treatment intensity consists of the average wage gap,

which is defined as the nominal deviation in Euros of the daily wage27 currently earned

to the minimum wage applicable in the following year. Note that this difference is set to

zero for workers with wages exceeding the minimum wage before the average wage gap

at the district level is calculated. Results with both treatment definitions will be presented;

however, we prefer the bite over the wage gap, because the latter is more prone to mea-

surement error in wage rates at the individual level. In either case, the identification of the

minimum-wage effect is based on the differential treatment intensities across districts.

The choice of the district level as the unit of observation is motivated by two reasons.

27. Wages are measured on a daily basis in our data; we therefore calculate the wage gap at this level assum-ing 40 working hours per week.


Table 2.4.: Minimum-wage intensity, wage growth, and employment growth

YearEast Germany West Germany

Bite Wage Wage Employment Bite Wage Wage Employmentgap growth growth gap growth growth

1993 19.74 1.14 — — 3.69 0.36 — —1994 21.97 1.29 3.49 9.68 3.86 0.39 2.44 −0.851995 22.37 1.24 2.18 1.56 3.87 0.42 1.69 −3.921996 21.60 1.22 1.79 −7.64 3.95 0.45 0.08 −8.781997 11.69 0.63 1.61 −11.91 3.18 0.37 −0.68 −8.161998 11.06 0.63 −0.67 −15.21 3.18 0.37 −0.68 −8.161999 22.79 1.05 0.38 −2.83 6.24 0.73 1.46 0.372000 20.42 0.92 1.21 −12.91 6.66 0.78 0.59 −2.912001 17.79 0.90 2.06 −17.42 6.68 0.82 1.50 −8.652002 15.83 0.82 2.08 −16.21 7.22 0.90 1.34 −8.87

Note: All numbers are in percent, except for the wage gap which is expressed in Euros per day (comparewith Section 2.2.2). Growth rates are calculated annually. The bite/wage gap for the years 1993–1995 areartificial because the minimum wage was only introduced in January 1997. The artificial bite/wage gap iscalculated by deflating the minimum wage of 1997 with the average wage growth, separately for East andWest Germany.Source: Authors’ calculations based on the IEB.

First, Thompson [2009] points out that the minimum-wage bite may differ heavily between

regions. If regions used in an analysis are too large, one will estimate the average effect of

an average minimum-wage bite, which is not necessarily informative. Indeed, the mini-

mum wage does show considerably more variation at the district level compared to, for

instance, broader labor-market regions.

Figures 2.5 and 2.6 show the spatial distribution of the bite in 1996 prior to the minimum-

wage introduction for West and East Germany, respectively. The majority of neighbouring

districts is clearly characterized by different treatment intensities. In West Germany, the

bite varies between 0.45 percent and 27.02 percent, while at least 6.14 percent and at most

40.58 percent of all workers in each district are affected in East Germany. However, the

distribution is heavily skewed to the right, which implies that the bite of the minimum

wage is very low for the majority of regions, while a few regions are affected heavily. Even

though the treatment intensity is much higher in East compared to West Germany, this

variation in the bite is not exploited for the identification of the minimum-wage effect.

Instead, different treatment intensities within East and West Germany are used, especially

the variation between neighbouring districts (cf. Sec. 2.2.3).


Figure 2.5.: SPATIAL DISTRIBUTION OF THE MINIMUM-WAGE BITE IN 1996 — WEST GER-MANY

Bite 19960.45 - 2.36

2.37 - 3.81

3.82 - 5.90

5.91 - 10.09

10.10 - 27.02

Note: The bite is defined as the share of workers earning below the minimum wage in the period prior to its introductionor increase.Source: Authors’ calculations based on the IEB.


Figure 2.6.: SPATIAL DISTRIBUTION OF THE MINIMUM-WAGE BITE IN 1996 — EAST GER-MANY

Bite 19966.14 - 12.1212.13 - 17.6317.64 - 23.6023.61 - 30.8930.90 - 40.58

Note: The bite is defined as the share of workers earning below the minimum wage in the period prior to its introductionor increase.Source: Authors’ calculations based on the IEB.


Table 2.4 shows the development of the average treatment intensities as well as the mean

wage and employment growth rates over time. In East Germany, the average bite in 1996

prior to the minimum-wage introduction amounts to almost 22 percent. In 1997, it de-

creases to less than 12 percent, only to reach almost 23 percent again in 1999. The devel-

opment of the wage gap is similar: It halves from e1.22 per day in 1996 to e0.63 per day

in 1997, only to increase to e1.05 per day in 1999. This pattern mirrors the development

of the nominal minimum wage (Figure 2.4) with the only sizable increase in 2000. The

general pattern is also present in West Germany, although the average treatment intensity

is much lower in each year compared to the East. Wage and employment growth rates

appear to decrease from the mid-90s onwards. It is impossible to decide based on these

descriptive statistics whether these developments are caused by the beginning recession or

by the minimum-wage introduction.

A second advantage of using district-level data compared to more aggregated spatial

units is the identification of spatial heterogeneity in terms of average wage and employ-

ment growth rates. The mean wage growth rate over all regions and time periods amounts

to 1.1 percent with a standard deviation of 1.8; the average employment growth rate is

−5.98 percent with a standard deviation of 8.7. For wage and employment growth rates

alike, most of the variation is found over time and not between regions. Nevertheless,

possibly deviating reactions of individual districts to the minimum wage can only be mea-

sured if the analysis is carried out at this regional level.

The use of districts as observational unit also creates potential problems. It is necessary

to control for structural differences in terms of wage and employment growth rates that

are not caused by the minimum wage. To this end, two strategies are employed. First,

the average wage and employment growth rates of all other industries except construction

in each specific district are added as control variables. These indicators are based on the

weakly anonymous Sample of Integrated Labour Market Biographies (Years 1975 - 2008),

which is based on a 2-percent random sample drawn from the IEB [Dorner et al. 2010]. Sec-

ond, as we explain in Sec. 2.2.3, the classification of the Federal Institute for Research on

Building, Urban Affairs and Spatial Development (BBSR) is used for both the classification


of districts into nine different types (siedlungsstrukturelle Kreistypen28) as well as the defi-

nition of broader labour-market regions (Raumordnungsregionen). The latter is important

because districts are administrative regions that are interconnected in terms of the product

as well as the labour market. Consequently, they can be used to explicitly allow for spatial

spillover effects within these labour-market regions in the econometric specifications.

2.2.3. Estimation Strategy

In the following, we describe a statistical framework to examine the effects of the minimum-

wage bite on regional wage and employment patterns. We begin with a relatively basic

model that mimics the standard approach to analyse minimum-wage effects in a panel of

regional data. We then extend the model in various ways to more comprehensively cap-

ture spatial dependencies or heterogeneities, noting potential strengths and weaknesses of

those approaches along the way.

Basic Model

We are interested in estimating wage and employment effects of the minimum-wage intro-

duction and subsequent increases in the German construction sector using regional panel

data. Since there is no variation in nominal minimum wages (except for the difference be-

tween East and West German districts), we combine the panel approach in Neumark and

Wascher [1992] with the idea of using the level of the minimum-wage bite as in Card [1992].

Following Dolton, Bondibene and Wadsworth [2010], we separate the post-treatment ef-

fect from the more general correlation between the dependent variables and the bite by

introducing an artificial (or hypothetical) bite before the minimum-wage introduction. It

is calculated assuming that the 1997 minimum wage (adjusted for previous wage trends)

already applied.

28. The types are formed based on population density and the degree of interconnectedness with neigh-bouring districts [BBSR 2012]. These characteristics are used to proxy unobserved determinants of structuraldifferences in wage and employment growth rates.


Our initial specification is

∆ ln yit+1 = bitα + (d× bit)β + ∆ ln xitγ + µi + τt + εit, (2.4)

where ∆ ln yit+1 constitutes wage or employment growth in district i between time t and

t + 1, bit is the minimum-wage bite (or the minimum-wage gap) for district i in year t, and

d an indicator for the post-treatment period. Thus, β captures the treatment effect of the

minimum wage. The vector ∆ ln xit represents mean wage and employment growth in all

local industries except construction as additional controls to proxy for differences in local

demand shocks. The terms µi and τt represent district and time-period fixed effects. We

do not need to include the post-treatment indicator d as a separate control as long as we

include full time-period indicators. Observe that α, β and τt are vectors containing two

elements since we estimate separate effects for East and West German districts to allow

for additional flexibility regarding treatment effects. We do not run separate regressions

to ensure comparability with the later neighbourhood-effects model where splitting the

sample would mean a loss of neighbourhood information at the inner German border.29

We use growth rates as dependent variables for two reasons. First, using levels might

lead to counterintuitive correlations between the dependent variable and the bite after the

fixed-effects transformation. For example, if employers actually commit to the new mini-

mum wage, wages should stay up while the bite drops in the periods after the introduction

(see Table 2.4). The sign of the correlation might therefore change over time and complicate

the identification of a minimum-wage effect. This problem is circumvented in a specifica-

tion using growth rates. Second, this strategy helps us to avoid problems arising from serial

correlation in levels [Wooldridge 2010]. In a recent paper, Meer and West [2013] argue for

the use of growth rates as well because the effect of minimum wages on employment is

more apparent in its dynamics (i.e., growth rates) rather than in the levels (for a variety of

reasons, such as inflation erosion of the minimum wage and attenuation of the effect on

29. In any case, estimating Equation (2.4) separately for East and West does not change the results qualita-tively.


levels because of the inclusion of time trends, among others).

In Equation (2.4), α captures the correlation between the artificial or hypothetical minimum-

wage bite (or wage gap) and the wage or employment growth rate before the actual minimum-

wage introduction. If it were statistically significant, it could indicate that there are some

structural differences between regions in the pre-treatment period that cannot be ade-

quately captured by the other control variables and that are correlated with the minimum-

wage intensity. The identifying assumption for β to properly measure the treatment effect

is that the correlation between the bite (or wage gap) and the dependent variables would

have stayed constant in the absence of the minimum-wage introduction.

Note that there is one potential caveat when estimating Equation (2.4), especially with

wage growth as the dependent variable. Regional wages play a role in determining both

the degree of the minimum wage treatment intensity and the subsequent growth rate of

wages, thus violating the assumption of strict exogeneity of the bite.30 Additionally, mea-

surement error or reversion to the mean will bias the estimate of α upwards in a mechani-

cal sense.31 However, making the identifying assumption that this phenomenon does not

change over time, one can still interpret β as the unbiased treatment effect of the minimum

wage on regional wage growth. We will make that assumption in what follows.

As an alternative to Equation (2.4), we also estimate a model that allows for region-type-

specific time trends:

∆ ln yit+1 = bitα + (d× bit)β + ∆ ln xitγ + µi + τt + λr Irt + εit, (2.5)

where Ir is an indicator for region type r, which consists of nine categories between low-

density rural areas and high-density core cities, so that Irt represents differential time

trends.32 Equation (2.5) therefore allows for different patterns in wage and employment

growth rates depending on regional characteristics that might be linked to agglomeration

30. We use a weaker assumption than strict exogeneity in Sec. 2.2.3.31. Dolton, Bondibene and Wadsworth [2012] discuss this problem as well.32. We use a classification provided by the BBSR. See BBSR [2012] for more information on how the region

types are categorized.


or urbanization processes. Furthermore, population density is a crucial factor in determin-

ing the spatial wage structure in Germany [Buttner and Ebertz 2009], which indicates that

wage and employment growth rates may also be closely linked to this characteristic.

Neighbors

One might criticize the above models on the ground that they do not adequately control

for spatial spillover effects. Local characteristics might not only have effects in the home

district but also in neighbouring ones. Ignoring those effects can lead to omitted-variable

bias if local characteristics are spatially correlated. Similarly, the effect of a high bite in

a particular region might not be confined to that region. For example, while the direct

employment effect to that region might be negative, the indirect effect to neighbouring

regions might be positive if labour demand rises in those regions as a result. This could

happen if firms are forced out of business and construction orders are taken by firms from

neighbouring districts.

In contrast, if the minimum wage narrows the wage differential between districts (espe-

cially for low-skilled workers), this decreases the incentive to commute long distances to

more attractive jobs. Thus, there might be a negative effect on labour supply in the neigh-

bourhood of a high-bite district as workers decide to search for jobs closer to their homes

(and possibly displacing lower-skilled workers there).

To allow for these kinds of neighbourhood effects—both in terms of general and minimum-

wage-induced spillovers—to affect regional wage and employment growth rates, we aug-

ment the basic model as follows:

∆ ln yit+1 = bitαD + (d× bit)βD + ∆ ln xitγ

D

+ bNit αI + (d× b

Nit )βI + ∆ ln xN

it γI

+ µi + τt + λr Irt + εit.

(2.6)

Here, αD, βD and γD capture the direct effects while αI, βI and γI capture the indirect ef-

fects from neighbouring districts. The variables relevant for the indirect effects are marked


with a bar on top and are calculated as the average over all neighbours. We specify “neigh-

bourhood” in two distinct ways. In the first variant, neighbours are other districts within

a larger functional unit (Raumordnungsregion) that has been defined according to commut-

ing flows and other characteristics (cf. Sec. 2.2.2). Second, we use a contiguity matrix to

indicate districts with common borders.

Note that the model in Equation (2.6) implies that spatial spillover effects are local in na-

ture. Thus, while it allows one district to affect its direct neighbor, we rule out that this has

higher-order effects on the neighbours’ neighbours, the neighbours of those neighbours,

and so on. While this assumption restricts the way spatial effects might take hold, we

believe it is a sensible choice. Demand for construction work is relatively localized since

buildings cannot be shipped like other goods. Factors of production have to be transported

to the production site. While there are some big players that bid for contracts nationwide,

most workers are employed in small- or medium-sized firms that operate locally or region-

ally. Even large building companies often maintain local establishments to better serve lo-

cal markets. Thus, we do not expect local shocks to have ripple effects that propagate to

distant districts.33

To test whether our assumption holds, we can make use of the fact that an OLS and

a spatial error model (SEM) yield consistent estimates for Equation (2.6) if the model is

correctly specified. While OLS assumes εit to be i.i.d., the SEM allows for spatial auto-

correlation of the errors according to εit = λWεit + ξit, with ξit being i.i.d. and assuming

the spatial weighting matrix W to be known. The parameter λ and the other coefficients

are estimated using maximum-likelihood techniques.34 Pace and LeSage [2008] propose a

spatial Hausman test for model comparison. Any significant discrepancies between OLS

and SEM coefficients can be interpreted as evidence of misspecification. In that case, both

models return inconsistent results, and a more general model (probably including a spatial

lag of the dependent variable) might be more appropriate.

33. See LeSage and Pace [2009] for a discussion of local vs. global spillovers.34. See LeSage and Pace [2009] for an overview of spatial regression models, including the SEM.


Border Pairs

An approach that concentrates on problems stemming from spatial heterogeneity instead

of spatial spillovers in this context was recently proposed by Dube, Lester and Reich [2010].

It generalizes the method used by Card and Krueger [1994] to identify minimum-wage ef-

fects at state borders. The authors note that conventional panel models assume that each

region can be readily compared to all the other regions irrespective of distance. This as-

sumption is problematic if markets are localized and economic conditions in one part of

the country are quite different from the ones in another part. For example, a local demand

shock might hit adjacent regions similarly while the rest of the country remains unaffected.

In this case, it may be a superior strategy to compare regions only to their direct neighbours

and assume that those form a better comparison group. We do think that the specifics of the

construction sector make it vulnerable to the critique by Dube, Lester and Reich [2010]. We

thus redo our analysis applying their “border-pair approach” to test whether our results

are robust when using contiguous district pairs as units of comparison.

Implementing this estimation strategy requires us to change the structure of our dataset.

Instead of the usual panel, the new data consists of the universe of all district pairs in

Germany that have a common border segment. This means that each district can enter the

dataset several times depending on the number of direct neighbours it has. In our case,

this increases the number of observations more than five-fold (from 3,708 to 19,089).

Although the structure is similar to the strategy employed by Dube, Lester and Reich

[2010], there is an important difference. In the German case, there is no policy discon-

tinuity per se located at the border. The discontinuity arises because of the variation in

treatment intensity between two regions which is not the result of any difference in statu-

tory minimum wages (except between East and West Germany). Thus, the identification

of the effect comes from the variation in the minimum-wage bite between bordering pairs.

Minimum-wage effects are then estimated using the model

∆ ln yipt+1 = bitα + (d× bit)β + ∆ ln xitγ + µi + τpt + λr Irt + εipt, (2.7)


where the subscript p identifies a single pair of neighbouring districts. The term τpt is a

specific pair–period effect and treated as a nuisance parameter. Effectively, the approach

treats each district pair as a natural experiment where the difference in the continuous bite

variable proxies treatment intensity. It then pools all individual estimates to get an average

relation between the minimum-wage bite and later wage or employment growth rates.

One additional strength of the model expressed in Equation (2.7) is that it depends on an

orthogonality assumption that is considerably weaker than the strict-exogeneity assump-

tion used for fixed-effects panel estimation [Dube, Lester and Reich 2010]. We now only

need to assume that the difference in local bites is contemporaneously uncorrelated to the

difference in local residuals in wages (or employment). This is relevant since strict exo-

geneity is questionable, especially for the wage regressions where regional wages not only

enter the dependent variable but also influence the minimum-wage bite on the right-hand

side of the equation. The border-pair approach thus allows us to get an idea of whether the

inherent simultaneity in our wage equations contaminates the fixed-effects results.

One drawback of this approach is that, unlike the model outlined in Equation (2.6), we

again do not allow for effects from neighbouring districts to affect the results. If there

are strong external effects that run from one district to another, then the coefficients in

Equation (2.7) are probably biased. However, in combining the strengths and weaknesses

of the different approaches, we hope to be able to draw a consistent picture of the effect of

introducing a minimum wage in the German construction sector.

2.2.4. Results

We discuss our estimation results by presenting the estimates for the wage regressions first,

followed by those from the employment regressions.

Wage Effects

The estimates of the minimum-wage effect on wage growth rates in East and West Ger-

many resulting from the basic specification are presented in the first two columns of Ta-


ble 2.5. Using the notation in Equation (2.4), the first two rows represent the coefficient

vector α and the following two rows, the coefficient vector β. The estimated pre-treatment

correlation between the bite and wage growth is positive for both East and West Germany.

We estimate a significantly positive treatment effect in East Germany, which translates to

an increase of the regional growth rate of wages of around 1.2 percentage points if the bite

is increased from 0.22 to 0.30.35 The coefficient for West Germany shows a negative treat-

ment effect in column (1). The effect becomes insignificantly different from zero, however,

as soon as additional control variables for region-type-specific time trends are included

(column (2)). The tests for serial correlation do not indicate that this is an issue for our

sample.

Note that a negative treatment effect observed for West Germany—while counterintuitive—

is theoretically possible in districts with a high fraction of workers earning just above the

minimum wage. Setting a minimum wage can then serve as an anchor for employers, who

might perceive that super-minimum wages overcompensates their workers. In this case,

employers may either downgrade these wages or offer exactly the minimum wage to new

employees. The other possible scenario of adjustments occurring on the intensive margin

does not seem likely, as we provide evidence in the supplementary material that the usual

hours of work remained relatively stable for both East and West Germany over the sample

period.

To investigate the effect of spillovers from neighbouring districts, the last two columns

in Table 2.5 report estimates for the models outlined in Equation (2.6). Column (3) de-

fines close neighbours as those districts that lie within one labour-market region while

Column (4) computes neighbourhood averages over all districts that have a contiguous

border with the observational unit. The results prove to be very robust to the inclusion of

local spillover effects. This holds irrespective of what spatial structure is assumed. Allow-

ing for indirect effects from neighbouring districts does not alter our previous conclusion.

Indirect treatment effects are both small and statistically insignificant. The results based on

35. Using the distribution of regional bites in East Germany in 1996, this represents an increase of the bite byapproximately one standard deviation.


Table 2.5.: The mininimum wage bite’s effect on mean wage growth

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

Artificial bite (West) 0.333∗∗∗ 0.331∗∗∗ 0.386∗∗∗ 0.328∗∗∗

(0.043) (0.044) (0.053) (0.044)Artificial bite (East) 0.125∗∗∗ 0.131∗∗∗ 0.149∗∗∗ 0.128∗∗∗

(0.024) (0.024) (0.030) (0.024)Treatment effect (West) −0.054∗∗ −0.041 −0.072∗ −0.021

(0.024) (0.027) (0.037) (0.029)Treatment effect (East) 0.148∗∗∗ 0.147∗∗∗ 0.144∗∗∗ 0.156∗∗∗

(0.022) (0.021) (0.025) (0.022)Wage growth (other industries) 0.005 0.004 −0.001 0.006

(0.016) (0.016) (0.018) (0.017)Employment growth (other industries) −0.009∗∗ −0.009∗∗ −0.009∗∗ −0.009∗∗

(0.004) (0.004) (0.004) (0.004)Artificial bite (West), neighbours −0.173∗∗ −0.051

(0.078) (0.049)Artificial bite (East), neighbours −0.053 −0.006

(0.040) (0.033)Treatment effect (West), neighbours 0.080 −0.042

(0.064) (0.037)Treatment effect (East), neighbours 0.011 −0.032

(0.036) (0.029)Wage growth (other industries), neighbours 0.034 −0.014

(0.033) (0.032)Employment growth (other industries), neighbours 0.001 −0.002

(0.011) (0.010)

District fixed effects Yes Yes Yes YesYear indicators Yes Yes Yes YesDistrict-type-specific trends No Yes Yes Yes

Wooldridge test for serial correlation (p-value) 0.434 0.409 0.412 0.475Within R2 0.395 0.399 0.401 0.400Observations 3708 3708 3708 3708

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Notes: Column (3) defines neighbours as being in the same labour-market region and Column (4) defines neighbours assharing a common border (cf. Sec. 2.2.3). Standard errors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.


the wage gap as a measure for the minimum wage treatment intensity (Table 2.6) are sim-

ilar to the results obtained with the minimum wage bite. There are no robust, statistically

significant effects on wage growth in West Germany. In contrast, an increase in the wage

gap by e0.50 per day36 leads to an increase in wage growth of 0.75 percentage points in

East Germany.

To check whether a specification including global spillover effects might be more appro-

priate, we perform a spatial Hausman test for differences between OLS and SEM coeffi-

cients (cf. Sec. 2.2.3). To illustrate the procedure, columns (1) and (2) of Table A.2 show

the OLS and SEM estimates of a clearly misspecified model for wage growth, where we

assume a row-standardized contiguity matrix for W in the SEM. The p-value of the spa-

tial Hausman test clearly rejects the null hypothesis that the coefficients of the OLS and

the SEM model are equal. Thus, both models are potentially misspecified and bias drives

the coefficient estimates apart. We perform the same test for the much richer models in

columns (3) and (4), which derive from Equation (2.6) with neighbours again defined as

sharing a contiguous border (see also column (4) from Table 2.5). This time, we cannot

reject the null hypothesis, which constitutes a necessary condition for the model to be con-

sistent. We conclude that our models in Table 2.5 describe the data reasonably well, making

a spatial model including global spillovers unnecessary.37

While the spatial Hausman test does not rule out the possibility of omitted-variable bias,

the results are reassuring. The discrepancy between OLS and SEM estimates occurs in the

presence of spatial dependence, and the bite variable shows a considerable amount of spa-

tial correlation. One might expect that omitted variables confounding the treatment effect

would also be spatially correlated, but this does not seem to be the case. The amount of

residual spatial correlation, represented by the λ parameter in Table A.2, is rather modest.

Finally, Table 2.7 depicts the main coefficients using the border-pair approach and the

36. Using the distribution of regional wage gaps in East Germany in 1996, this represents an increase of thewage gap by approximately one standard deviation.

37. Observe that column (3) in Table 2.5 is identical to a spatial Durbin model (SDM) including a spatiallag in both dependent and independent variables already, which is due to the special structure of the spatialweighting matrix [Gibbons and Overman 2012]. The direct and indirect effects from estimating column (4)using an SDM are very close to the ones depicted in Table 2.5. Results will be provided upon request.


Table 2.6.: The wage gap’s effect on mean wage growth

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

Artificial wage gap (West) 0.022∗∗∗ 0.022∗∗∗ 0.027∗∗∗ 0.022∗∗∗

(0.003) (0.003) (0.005) (0.004)Artificial wage gap (East) 0.011∗∗∗ 0.013∗∗∗ 0.017∗∗∗ 0.013∗∗∗

(0.003) (0.003) (0.004) (0.004)Treatment effect (West) −0.005∗∗ −0.004 −0.007∗ −0.002


(0.005) (0.004) (0.004) (0.005)Wage growth (other industries) 0.003 0.003 −0.008 0.006

(0.017) (0.017) (0.018) (0.017)Employment growth (other industries) −0.010∗∗ −0.009∗∗ −0.010∗∗ −0.009∗∗

(0.004) (0.004) (0.004) (0.004)Artificial wage gap (West), neighbours −0.015∗∗ −0.013∗∗

(0.006) (0.006)Artificial wage gap (East), neighbours −0.014∗∗ −0.006

(0.007) (0.005)Treatment effect (West), neighbours 0.008 −0.000

(0.005) (0.004)Treatment effect (East), neighbours 0.002 0.003

(0.009) (0.006)Wage growth (other industries), neighbours 0.057∗ 0.002

(0.032) (0.033)Employment growth (other industries), neighbours 0.006 0.002

(0.011) (0.010)



Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Column (3) defines neighbours as being in the same labour-market region and Column (4) defines neighbours assharing a common border (cf. Sec. 2.2.3). Standard errors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.


Table 2.7.: Border-approach models for wages

(1) (2)

Artificial bite (West) 0.065∗∗∗ 0.323∗∗∗

(0.024) (0.034)Artificial bite (East) 0.054∗∗∗ 0.148∗∗∗

(0.011) (0.020)Treatment effect (West) −0.004 −0.019

(0.022) (0.031)Treatment effect (East) 0.031∗∗ 0.107∗∗∗

(0.013) (0.014)Wage growth (other industries) 0.031∗∗ 0.025

(0.016) (0.015)Employment growth (other industries) −0.006 −0.010∗∗

(0.004) (0.004)

Pair–period fixed effects Yes YesDistrict fixed effects No Yes

Adjusted R2 0.374 0.501Observations 19089 19089

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Notes: Standard errors are enclosed in parentheses and clustered at the dis-trict level. Column (2) employs the user-written routine reg2hdfe for Stataby Guimaraes and Portugal [2010].Source: Authors’ calculations based on the IEB.

transformed data set. As an intermediate step, column (1) only absorbs pair–period fixed

effects from the data. In line with the previous results, we do not find any significant

effect for West Germany but a significantly positive effect in East Germany, albeit weaker

than before: The estimated treatment effect equals 0.031 based on the border pair approach

(Table 2.7) and varies between 0.147 to 0.156 in our basic model (Table 2.5). Column (2)

recognizes the fact that while adding pair–period effects controls for spatial heterogeneity

at a very low level, there might still be heterogeneity that is unique to a single district.

Additionally absorbing those district fixed effects does not change the treatment effect for

West Germany but increases it considerably in East Germany. While still somewhat lower

than in Table 2.5, it now lies in close proximity to the earlier results.

Overall, the results for the minimum-wage effect on wage growth at the regional level

are very robust.38 The effect of the introduction and subsequent increases of the mini-

38. Indeed, the results are robust to the exclusion of regions that belong to the top and bottom 5 percent of theminimum-wage bite as well as to estimating the model using labor-market regions (Raumordnungsregionen) asthe observational unit. Those results can be provided upon request. Further, Table A.5 in the supplementary


mum wage in West Germany is statistically indistinguishable from zero in most cases while

there is a pronounced positive effect in East Germany. This is congruent to the descriptive

statistics for East and West German districts before the minimum-wage introduction (cf.

Figures 2.5 and 2.6). In West Germany, the bite is quite low on average throughout the ob-

servational period. There were probably very few firms in each district that had to adjust

wages for a significant fraction of their workforce. If there were only a few workers who

experienced wage increases due to the new wage floor, those changes will not be visible in

district-level aggregated data.

Conversely, there are strong differences in East Germany, where the minimum wage

does pose a significant hurdle. Here, a relatively large fraction of all construction workers

received wage increases, which led to a statistically and economically significant effect on

regional wage growth. With significant wage effects being confined to East Germany, we

expect employment effects—if there are any—to be found only there, too.

Employment Effects

Table 2.8 mirrors the analyses displayed in Table 2.5 but now uses district-wise employ-

ment growth rate as the dependent variable. In this case, we do not find a significant corre-

lation between the minimum-wage bite and employment growth rate in the pre-treatment

period as captured by the artificial bite. This substantiates our hypothesis that the strongly

positive coefficients for α in Table 2.5 are not driven by structural differences but rather by

a simultaneity bias.

Again, we find no significant treatment effect for West German districts. While the coeffi-

cient in column (1) is still rather large (but statistically insignificant), it becomes negligible

as soon as we add additional controls. In contrast, the estimated employment effect in East

Germany is negative, large, and highly statistically significant. It starts off with a coefficient

of −0.39 in the basic specification and falls slightly when adding controls for region-type

time trends. The effect is even smaller when we look at our two specifications that add

material provides the results of regressions using a shorter time interval (1994–1998).


Table 2.8.: The mininimum wage bite’s effect on employment growth

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

Artificial bite (West) −0.152 −0.105 −0.261 −0.177(0.150) (0.151) (0.187) (0.154)

Artificial bite (East) 0.012 −0.026 −0.048 −0.041(0.111) (0.110) (0.138) (0.111)

Treatment effect (West) 0.130 0.005 0.044 0.078(0.131) (0.130) (0.192) (0.150)

Treatment effect (East) −0.385∗∗∗ −0.368∗∗∗ −0.319∗∗∗ −0.340∗∗∗

(0.093) (0.093) (0.112) (0.102)Wage growth (other industries) −0.023 −0.017 0.044 −0.025

(0.066) (0.065) (0.080) (0.065)Employment growth (other industries) 0.030 0.028 0.009 0.028

(0.020) (0.020) (0.022) (0.020)Artificial bite (West), neighbours 0.595∗ 0.560∗

(0.361) (0.300)Artificial bite (East), neighbours 0.074 0.143

(0.198) (0.179)Treatment effect (West), neighbours −0.156 −0.369

(0.294) (0.251)Treatment effect (East), neighbours −0.129 −0.085

(0.188) (0.160)Wage growth (other industries), neighbours −0.318∗ 0.035

(0.167) (0.185)Employment growth (other industries), neighbours 0.089∗ 0.027

(0.049) (0.047)



Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Notes: Column (3) defines neighbours as being in the same labour-market region and Column (4) defines neighbours assharing a common border (cf. Sec. 2.2.3). Standard errors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.


local spillover effects (−0.32 and −0.34).

Nevertheless, this still implies a large disemployment effect of the minimum wage in

East German districts. Increasing the bite from 0.22 to 0.30 is associated with a decline of

the regional growth rate of employment in construction by 2.6 to 3.1 percentage points,

depending on the particular specification.39 The results using the wage gap as measure for

the minimum wage treatment intensity (Table 2.9) again confirm the results based on the

bite: There is no correlation between the average wage gap at the district level and em-

ployment growth in the pre-treatment period. In the post-treatment period, the wage gap

is not associated with a change in employment growth in West Germany, while districts

with a higher wage gap display lower employment growth rates in East Germany. More

specifically, an increase in the wage gap by e0.50 per day leads to a decline in employment

growth of 1.8 to 2.1 percentage points.

While the estimated disemployment effect seems rather large at first glance, note that the

construction industry experienced a deep recession during the observation period, starting

in the mid-90s. As a back-of-the-envelope calculation, consider that the average growth

rate of employment in East Germany was approximately −12 percent between 1996 and

1997. Setting the coefficient of the treatment effect to −0.35—which is in the middle of the

range of our estimates—and observing that the average bite was around 20 percent in 1996

yields a treatment effect of 7 percentage points. Thus, while employment contracted in all

East German districts between 1996 and 1997, our estimates suggest that the minimum-

wage introduction caused more than half of the overall decline. In this light, minimum

wages may be especially dangerous in times of an economic downturn if they are set too

high.40 Indeed, Dolton and Bondibene [2012] provide evidence that at least the negative

effect on youth unemployment is aggravated by minimum wages during a downturn.

While allowing for spillover effects leads to slightly lower estimates of the direct effect

as measured by βD, we again have imprecise estimates of the indirect effects βI. Both

39. Again, this shift in the bite represents an increase by roughly one standard deviation.40. Neumark, Salas and Wascher [2013] note that potential differential effects of a recession across regions

may introduce a bias to the estimated minimum-wage effect in models with linear state-specific trends. Ourresults are robust with respect to the inclusion of region-type-specific trends (see Table 2.8) and also state-specific trends (shown in the supplementary material).


Table 2.9.: The wage gap’s effect on employment growth

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

Artificial wage gap (West) −0.002 0.003 −0.010 −0.007(0.010) (0.010) (0.014) (0.011)

Artificial wage gap (East) 0.012 0.005 −0.006 0.001(0.017) (0.017) (0.021) (0.017)

Treatment effect (West) 0.002 −0.009 −0.002 0.000(0.008) (0.008) (0.013) (0.012)

Treatment effect (East) −0.042∗∗∗ −0.042∗∗∗ −0.036∗∗ −0.037∗∗

(0.015) (0.015) (0.017) (0.015)Wage growth (other industries) −0.026 −0.017 0.051 −0.029

(0.066) (0.065) (0.079) (0.065)Employment growth (other industries) 0.030 0.027 0.010 0.028

(0.020) (0.020) (0.022) (0.020)Artificial wage gap (West), neighbours 0.038 0.050∗∗

(0.027) (0.029)Artificial wage gap (East), neighbours 0.037 0.038∗∗

(0.029) (0.023)Treatment effect (West), neighbours −0.019 −0.037

(0.022) (0.028)Treatment effect (East), neighbours −0.024 −0.008

(0.032) (0.028)Wage growth (other industries), neighbours −0.339∗∗ 0.030

(0.167) (0.184)Employment growth (other industries), neighbours 0.086∗∗ 0.020

(0.048) (0.047)



Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Column (3) defines neighbours as being in the same labour-market region and Column (4) defines neighbours assharing a common border (cf. Sec. 2.2.3). Standard errors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.


Table 2.10.: Border-approach models for employment

(1) (2)

Artificial bite (West) −0.089 −0.177(0.103) (0.140)

Artificial bite (East) 0.155∗∗∗ −0.017(0.053) (0.085)

Treatment effect (West) −0.000 0.088(0.116) (0.122)

Treatment effect (East) −0.378∗∗∗ −0.388∗∗∗

(0.063) (0.065)Wage growth (other industries) 0.003 −0.026

(0.062) (0.066)Employment growth (other industries) 0.043∗∗ 0.026

(0.018) (0.019)

Pair–period fixed effects Yes YesDistrict fixed effects No Yes

Adjusted R2 0.439 0.531Observations 19089 19089

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Standard errors are enclosed in parentheses and clustered at the dis-trict level. Column (2) employs the user-written routine reg2hdfe for Stataby Guimaraes and Portugal [2010].Source: Authors’ calculations based on the IEB.

columns (3) and (4) show no statistically significant indications of negative spillover effects

of a larger minimum-wage bite on neighbouring counties.

The results for the spatial Hausman specification test are depicted in Table A.3. Again,

testing an obviously underspecified model leads to significant differences between OLS

and SEM estimates. However, the coefficients of the richer specification are very close to

each other. We therefore omit more general spatial formulations from our discussion.

To analyse whether the above results are still contaminated by spatial heterogeneity at

the local level, columns (1) and (2) of Table 2.10 present estimates using the border-pair

sample and using employment growth rate as the dependent variable. The estimates point

toward minimum-wage effects of the same magnitude as in the basic model.

Analogous to the previous section, the coefficient of the treatment effect of the minimum-

wage bite proves to be very robust with respect to different modelling approaches. We

find no effect on employment growth at the district level in West Germany. This matches

our findings for wage effects, since we do not expect to find employment effects without


any measurable wage changes. However, we do find pronounced negative employment

effects in East Germany, where a higher proportion of workers are directly affected by the

minimum wage.

2.2.5. Conclusion

This paper analyses the effect of a minimum wage in the German main construction sector

on wage and employment growth rates. This is enriching to the literature because of the

unique characteristics of the minimum wage in this sector. The most important elements

of these are the fact that the minimum wage introduced was of a substantial magnitude,

and that it was introduced during a period of economic contraction, particularly in the

former East Germany. Much of the previous research on the impacts of minimum wages

has provided evidence of modest changes in the minimum wage during less turbulent

periods of the economy. The evidence presented here is consistent with the view that a

moderate minimum wage might have negligible effects, but that this can easily change if

it is allowed to cut too deeply into the wage distribution. In this case, it will benefit some

workers, but this comes at the cost of making other workers (the displaced ones and those

who are unable to find employment) worse off.

We focus our attention on the minimum-wage effects in regional labour markets. We

control for spatial spillovers and regional heterogeneity, two aspects which previous re-

search has largely ignored. Our results indicate that wage growth in East Germany was

positively affected by the minimum wage while the West German wage growth rate was

not affected at all. In terms of employment growth rates, we do not find any adverse ef-

fect in West Germany. The contraction in the employment growth rate, however, is rather

stark in the East, where the bite of the minimum wage was relatively high. The results are

consistent beyond our baseline models, and are robust to various alternative specifications.

While some of our results are consistent with previous work on minimum wages in

Germany, there are also important differences. Contrary to Apel et al. [2012], we do find

negative employment effects in East Germany. We also note that the previous result which


generated the most controversy—that of a positive employment effect in West Germany

found by Konig and Moller [2009]—cannot be supported by the present study.

A number of advances in the approach taken in this study may explain the difference

in our conclusions compared to earlier work. First, the higher level of aggregation that

we use in our regional data circumvents much of the measurement errors that plague

other studies. More explicitly, the difficulties associated with the identification of treat-

ment and control groups elsewhere do not materialize here. Second, worker substitution

taking place at the individual level is also a problem that we are able to sidestep with our

approach. Third, focusing on local labour markets allows us to extend the analysis beyond

job destruction and gain insights about the overall effect of the sectoral minimum wage

including job creation.

Our analysis is limited by the fact that we are unable to take into account the presence of

posted (i.e., foreign) workers in Germany as well as the self-employed. Self-employment in

construction rose substantially during the observation period in East Germany despite the

strong decline in overall employment.41 One might speculate that part of this increase was

driven by former employees who registered themselves as self-employed to avoid compli-

ance with the minimum wage. While this is possible, we present some rough calculations

in the supplementary material showing that this is unlikely to be the principal driver of the

observed employment decline.42 While posted workers do not enter the analysis, the over-

all employment effect is likely to be more negative if they were included. Posted workers

usually received lower wages than native workers before the new minimum wage eroded

at least a significant part of that price advantage.

Moreover, we have examined the “raw” effect of the minimum wage on employment

and wage growth rates but have not taken into account other channels of adjustment, par-

ticularly employment turnover. A decrease in turnover might indicate that firms are in-

vesting more in their employees as a result of the minimum wage, and such an investment

41. Between 1995 and 2000, the number of proprietors increased by nearly 70 percent [ELVIRA 2013].42. Even under generous assumptions about the amount of employment replaced by self-employment, self-

employment could only account for about 6 percent of the total employment losses observed during this pe-riod.


can have a profound impact on employment stability or the health of the labour market

itself [Gittings and Schmutte 2013]. Bachmann, Konig and Schaffner [2012] study the issue

of turnover in the German construction sector. They find that both accessions and separa-

tions increased in East Germany due to the minimum wage, but that job-to-job transitions

declined, which they attribute to the resulting wage compression. Our results indicate that

the effect on separations must have been most important, since we find net disemployment

effects.

Two other caveats are in order. First, we have not considered the demand elasticity for

the products and services of the construction sector in this study. Depending on this elas-

ticity, a change in the minimum wage may manifest itself in higher output prices or lower

profit margins.43 However, we note that the economies of West and East Germany were

in differing states during the period of analysis. East Germany was in relative decline,

and as such, we expect that the mitigating effects that the product market may have on the

negative impacts of the minimum wage did not materialize. It would seem that the ineffec-

tiveness of this transmission channel may have exacerbated the impact for East Germany.

Second, we have not examined both the mobility of construction firms and the changes

in the number of firms. If these firms are sufficiently mobile, they may adjust by moving

their operations to regions which are less affected by the minimum wage. While we do not

expect this mobility to be too important due to the nature of the market for products of the

construction sector, this is another channel of adjustment that is left for further research.

Although the original motivation for a minimum wage in Germany’s construction sector

was anti-competitive instead of anti-poverty, more recent political discussions about this

issue in general have made references to subsistence wages and social safety nets. While we

advise against directly carrying over our results to the discussion about adopting a national

minimum wage spanning all sectors—construction is just one sector, a rather special one,

and the circumstances during the study period were unusual—we still see our findings as

a cautionary tale, reminding us that minimum-wage legislation has the inherent potential

43. The effect of minimum wages on competition, prices, and profit margins is a less-studied area within theminimum-wage literature. Bachmann, Bauer and Frings [2014] deal with these issues for the German case,although not specifically for the main construction industry.


to backfire. Indeed, in the present case, we find that a strongly binding minimum wage

in East Germany led to rather large negative effects on employment growth in that region.

While we did not observe a similar effect in the West, this is probably because the minimum

wage there was mostly not or only slightly binding.

3. Competition on labour and product

markets

This chapter consists of two sections studying the relationship between mininmum wages

and competition on the labour and product market. Section 3.1 is co-authored by Ronald

Bachmann and unpublished at this point in time. Section 3.2 is joint work with Ronald

Bachmann and Thomas Bauer and based on “Minimum Wages as a Barrier to Entry: Ev-

idence from Germany”, which is forthcoming in LABOUR.Review of Labour Economics

and Industrial Relations.

3.1. Monopsonistic Competition and the Minimum Wage

Germany is special in not having an explicit minimum wage at the national level. Instead,

minimum wages exist for a number of industries, which are based on collective bargaining

agreements declared generally binding. The main institutional framework for minimum

wages in Germany, the Posting of Workers Law (‘Arbeitnehmerentsendegesetz’), stipulates

that unions and employer associations have to apply jointly for a minimum wage intro-

duction in their industry. Thus, the social partners not only determine the level of the

wage floor, but also influence the decision whether it should exist in the first place. Politi-

cians also have an important role to play in this context, because they decide about the

introduction of minimum wages in an industry through its inclusion into the Posting of

Workers Law.

The German minimum wage institution will change drastically as of January 1st, 2015.

3. COMPETITION ON LABOUR AND PRODUCT MARKETS 75

At this date, a country-wide minimum wage will be introduced at a level of e8.50 for most

workers, with exemptions for specific industries until 2017. The minimum wage will affect

15 percent of all West German employees and 27 percent of all East German employees

[Brenke and Muller 2013], and will be relatively high compared to other industrialized

countries [Kluve 2013]. Consequently, there are wide-spread fears that many jobs are at

risk of getting destroyed.

In the context of the changing minimum wage institution, we analyze if and to which ex-

tent industries are characterized by deviating degrees of monopsonistic competition. First,

the higher the variation in monopsony power among low-wage industries, the more di-

verse are the expected employment effects of a uniform minimum wage. Second, an in-

vestigation of monopsony power in the existing minimum wage industries allows us to

provide new insights into the mechanism underlying the introduction of minimum wages

at the sectoral level. Third – and to the best of our knowledge – we are the first to study

sectoral differences in monopsonistic competition, which is of primary interest because

wage-setting takes place at the sectoral and regional level to a large extent.

The employment effects of minimum wages depend on two main factors. On the one

hand, the absolute level of the minimum wage, be it at the national or at the sectoral level,

plays a crucial role. On the other hand, given a binding minimum wage, the structure of

the labour market is an important determinant of employment effects. In a neo-classical

labour market, the wage elasticity of labour supply to the firm is infinite, the wage equals

the marginal product of labour, and an increase of the wage therefore unambiguously leads

to an increase in unemployment [Neumark and Wascher 2008]. In a monopsonistic labour

market, by contrast, mobility of workers is limited, and the elasticity of labour supply

is relatively low. As a consequence, firms can use their market power to set the wage

below a worker’s productivity [Manning 2003a]. Minimum wages may therefore lead to a

reduction in firms’ profits, without a corresponding increase in unemployment.

Against this background, it is of great importance to analyse the degree of monopsony

power in the German labour market at the sectoral level for two reasons. First, the structure

of the labour market may have played a role for the introduction of minimum wages at


the sectoral level, which started in 1997. If the main aim of the introduction of minimum

wages was to raise wages at the bottom of the wage distribution while avoiding large

displacement effects, minimum wages should have been introduced mainly in industries

characterised by low wages and a high degree of monopsony power. Second, for the new

statutory minimum wage, estimates of monopsony power at the sectoral and regional level

yield insights into the likely employment effects of the minimum wage introduction in

2015.

In this paper, we therefore investigate the role of monopsony power of German employ-

ers following a semi-structural approach based on the dynamic model of monopsonistic

competition proposed by Manning [2003a]. We do so using a unique linked employer-

employee data set for Germany which allows to control for worker heterogeneity, firm

heterogeneity and demand side effects. Our analysis yields estimates of the wage elasticity

of labour supply, which provides a measure of monopsony power, separately for differ-

ent industries and for East and West Germany. Furthermore, we investigate the role that

the composition of the workforce may play for monopsony power in an industry. Finally,

we examine how our estimates are correlated with the vacancy rate at the industry level,

which is a key outcome of the monopsonistic model of the labour market.

The paper is structured as follows. In the next section, we discuss the theoretical back-

ground of our estimation strategy and provide an overview of the existing empirical re-

search. Section 3.1.2 presents details of the empirical approach and describes the data set.

Section 3.1.3 discusses the empirical results and the final section concludes.

3.1.1. Sectoral Differences in Monopsonistic Competition

The early versions of monopsonistic labour market models, epitomized by the textbook

version of a Robinsonian one-firm monopsony model, are clearly theoretical artifacts and

an unlikely characteristic of labour markets of industrialised countries such as Germany.

The reason for this is that monoposony power in this model derives from the existence of a

single employer of labour in a market. In contrast, the source of monopsony power in dy-


namic models of monopsonistic competition is not employer concentration.1 Instead, the

source of monopsony power lies in search frictions as well as heterogeneous preferences

over non-wage employer characteristics [Bhaskar, Manning and To 2002; Boal and Ransom

1997].

Search frictions constitute any factor that lengthens the time firms and workers need

to find each other. Popular examples include limited mobility of workers, or information

asymmetries between firms, unemployed job searchers, and workers who search on the

job. Non-wage employer characteristics cover, among others, flexible working-time ar-

rangements, commuting time, training and career opportunities, or the general working

atmosphere. Workers differ in their preferences and in the extent to which they face search

frictions. Firms differ in their non-wage characteristics. Consequently, workers who are

exactly equal to each other, but differ in their preferences over non-wage employer charac-

teristics or the degree of search frictions they face, can earn different wages at any point in

time.

Independently of the exact cause of monopsony power, the single most important result

is that the labour supply to the firm is not perfectly elastic. Figure 3.1 shows the situation

of the individual firm in a monopsonistic labour market facing an upward sloping labour

supply curve. In contrast to the competitive model of the labour market, firms are wage

setters and can choose any wage-employment combination on the labour supply curve.

Intuitively, this means that some - but not all - workers will leave the firm if the wage

is reduced by a small amount. Consequently, the only possibility for a firm to increase

its employment level is to offer a higher wage rate. Therefore, the employer-size wage

effect, i.e. large employers paying a higher wage, is a natural outcome in this framework

[Manning 2011].

While workers with the same productivity may earn deviating wages in different firms,

the monopsony model assumes that equal workers in terms of observable characteristics

1. Manning [2003b] and Hirsch, Konig and Moller [2013] do propose models of geographic oligopsony,in which a combination of regional employer concentration and limited mobility of workers are the sourcesof monopsony power. However, in the majority of modern monopsony models, employer concentration isirrelevant.


Figure 3.1.: The firm in a monopsonistic labour market

Employment

Wage

W2

W1

E1 E2

D=MRP

MC

S=AC

E3

Legend: S = Labour supply; AC=Average cost of labour; MC= Marginal cost of labour; D=Demand for labour;MRP=Marginal revenue product of labour.Source: Own illustration, based on [Manning 2003a].

receive the same wage rates within one firm. Consequently, if a firm wants to increase

its employment level, the higher wage has to be paid not only to the additional worker,

but also to all existing employees of the same type. Stated differently, the marginal cost of

labour includes the wage paid to the new employee as well as the wage increases of the

workers already employed. Therefore, the marginal cost (MC) of labour exceeds the aver-

age cost (AC) of labour. A profit-maximizing firm will choose its employment level such

that marginal costs are equal to the marginal revenue product (MRP) of labour [Manning

2003a]. Thus, the firm depicted in Figure 3.1 will choose employment level E1. The wage

that needs to be paid to obtain this employment level equals W1.

This stylized description of a monopsonistic labour market has several important impli-

cations. First, wages W1 and employment E1 are lower in the monopsonistic equilibrium

compared to the equilibrium under perfect competition (W2 and E2). Second, workers

earn less than their marginal product, since the marginal cost exceeds the average cost of

labour. Third, the firm operates with a constant amount of vacancies, i.e. at the going wage


rate W1 the firm would like to employ workers up to E3. Stated differently, the equilibrium

is supply-side constrained.

Finally, a moderate minimum wage that is slightly above the going wage rate could in-

crease wages and at the same time increase employment, while decreasing firms’ profits.

For example, exogeneously increasing the wage rate slightly above W1, implies moving

along the labour supply curve. However, this relationship only holds until labour supply

equals labour demand. At higher wage rates, labour demand is the decisive factor in de-

termining the employment level. Thus, a minimum wage exceeding the level W2 would

lead, exactly as under perfect competition, to employment losses. Therefore, the effects of

a minimum wage depend, among other things, on its level as well as the degree of monop-

sonistic competition in the labour market. The degree of monopsonistic competititon is

defined by the wage elasticty of labour supply or, stated differently, the slope of the labour

supply curve. The flatter this curve, i.e. the higher the wage elasticitiy of labour supply,

the more competitive the labour market.

Against this background, a crucial question for the minimum wage debate in Germany

is whether industries are characterized by different degrees of monopsonistic competition.

The industry dimension is particularly important in the German context for several rea-

sons. First, the existing minimum wages are determined at the industry level. The ques-

tion why minimum wages have been introduced in some industries but not in others has

not been answered so far. If minimum wages had been introduced in industries with the

highest degree of monopsony power, this would be an economic justification for why min-

imum wages were introduced in the chosen industries, but not in others. Additionally, if

minimum wages had been introduced in monopsonistic industries, this could explain the

largely non-negative employment effects which were found in a large-scale evaluation of

the existing industry-specific minimum wages by the Ministry of Labour and Social Af-

fairs.2

Second, differences in monopsonistic competition across the German industries are rele-

2. The reports containing detailed results can be downloaded at: http://www.bmas.de/DE/Themen/

Arbeitsrecht/Meldungen/evaluation-mindestloehne.html.


vant for the employment effects of the introduction of a statutory minimum wage of e8.50

in 2015. Wage-setting takes place at the sectoral and regional level to a large degree, re-

sulting in inter-industry wage differentials. This will lead to large differences in the bite

of the minimum wage across industries and regions. All else equal, industries and regions

with lower average wages can be expected to show a stronger reaction to the uniform min-

imum wage in terms of employment. However, the employment effects of the minimum

wage will also depend on the degree of monopsony power in the different industries and

regions. For example, if all low-wage industries were characterized by a relatively high de-

gree of monopsonistic competition, the overall employment effect of the minimum wage

would be negligible. If the opposite was the case, i.e. if monopsony power was relatively

low in low-wage industries, one would expect large employment effects. Finally, if the

picture was more diverse, i.e. if there were large differences in monopsonistic competition

among low-wage industries, this could explain deviating employment reactions in these

industries, despite similar wage levels prior to the introduction of the statutory minimum

wage.

Previous Empirical Studies

The single most important test for monopsony power in the labour market is to estimate

the labour supply elasticity to the individual firm. If labour supply is rather elastic, perfect

competition is a more appropriate model to describe the functioning of the labour market

than monopsonistic competition and vice versa. At first sight such an estimation appears to

be straightforward and involves regressing the firm’s employment level on the wage paid.

However, such a regression would be endogenous as the firm decides simultaneously on

wages and employment.

The existing empirical literature can be divided into two methodological strands, de-

pending on how the endogeneity problem is solved. The first set of studies uses exoge-

nous wage variations, which should not affect all firms in the market. In such a situation,

the unaffected firms constitute the control group that is needed to identify the effect of in-


creasing the wage rate on the firm’s employment level. From a theoretical point of view,

the exogenous change should not affect all firms because competition between firms for

workers is not altered in this case. Stated differently, if the wage distribution over firms

is simply shifted to the right, the relative position in the wage distribution of each indi-

vidual worker is not changed. Since such wage variations are extremely rare, the second

strand of empirical studies follow a semi-structural approach based on the dynamic model

of monopsonistic competition proposed by Manning [2003a], which is explained in more

detail in Section 3.1.2.

As for the first strand of the literature, two studies, Falch [2010] as well as Staiger, Spetz

and Phibbs [2010], estimate a static model by regressing the employment level on the wage

rate. In order to solve the endogeneity problem, both studies exploit an exogenous vari-

ation in wages, in the first case for school teachers in Norway and in the second case for

nurses employed in Veteran Hospitals in the US. The estimated labour supply elasticities

are low at 1.4 for school teachers in Norway and 0.1 for nurses in the US. In a sense, these

studies are in the tradition of the Robinsonian monopsony model, as employer concentra-

tion is an important source of monopsony power, and focus on very special labour markets.

Therefore, the external validity is low and the degree of inference that can be drawn for the

more general functioning of the labour market is limited.

While Ransom and Oaxaca [2010] as well as Ransom and Sims [2010] also concentrate on

specific labour market segments, namely the grocery retail industry and school teachers,

their empirical specification is based on Manning’s model of dynamic monopsony. This

eliminates the need for an exogenous variation in wages, but still limits the studies’ trans-

ferability to other labour market segments. To some degree this is less true for the analysis

by Ransom and Oaxaca (2010), since instead of employer concentration, search frictions

or heterogeneous preferences are more likely reasons for monopsony power in the retail

grocery industry compared to school teachers. The estimated labour supply elasticities are

in the range of 1.4 – 3.0 for the grocery retail industry in the US, and 3.7 for school teachers

in the US.

A semi-structural investigation of monopsony power for an entire labour market is pre-


sented by Hirsch, Schank and Schnabel [2010a]. Using linked employer-employee data for

Germany, the authors provide separate estimations for men and women, showing that the

labour supply elasticity to the firm lies in the range of 1.9 – 3.7, and that this elasticity is

lower for women.3 A back of the envelope calculation demonstrates that this difference is

large enough to account for the unexplained part of the gender pay gap in Germany. Ac-

cording to the underlying model of dynamic monopsony, the inverse of the labour supply

elasticity gives the proportional deviation of actual wages to the marginal revenue product

of labour. As women show lower labour supply elasticities than men, the actual wages

paid to women are further away from the competitive level than wages paid to men. The

magnitude of this difference fits the unexplained gender wage gap as estimated by the au-

thors. This result lends credibility to the theoretical framework, because a prediction of

the model explains an observed empirical puzzle. Hirsch, Schank and Schnabel [2010a]

further show that it is crucial to take heterogeneity at the establishment level into account

in order to ensure that demand-side effects are adequately controlled for. Only if this is the

case, the empirical investigation identifies supply-side effects, which can be interpreted as

monopsony power.

Hirsch and Jahn [2012] use exactly the same set-up as Hirsch, Schank and Schnabel

[2010a], but focus on differences between natives and immigrants in Germany. They es-

timate labour supply elasticities of 1.64 – 2.6, and show that the labour supply elasticity of

migrants is low enough to account for the unexplained pay gap between migrants and na-

tives. Sulis [2011] confirms the result that women have lower labour supply elasticity than

men for Italy. The elasticities are below one for both men and women. Booth and Katic

[2011] also find evidence for monopsonistic competition for the entire Australian labour

market using individual level data. The estimated labour supply elasticity is 0.71.

Dube, Lester and Reich [2013] is the only study that explicitly links minimum wages

to monopsonistic competition in the labour market by exploiting discontinuities at state

borders in federal minimum wage rates in the US to estimate wage elasticities of accession

3. Note that the labour supply elasticity of women to the market is actually more elastic than that of men.Stated differently, women react more strongly than men to the offered wage when deciding how many hoursto supply. At the same time, men react more strongly to wages when moving between potential employers.


and separation rates. The minimum wage elasticities of the separation rate are small, with

an increase of 1 percent in the minimum wage leading to a decrease in separations of−0.24

percent for teenage workers in the entire economy, and of −0.32 percent for restaurant

workers. Based on these wage elasticities, the remaining parameters of the Burdett and

Mortensen [1998] equilibrium search model are estimated. The results point towards a

significant degree of search frictions in the low-wage labour market in the US, which Dube,

Lester and Reich [2013] interpret as an explanation for non-negative employment effects of

the minimum wage.

3.1.2. Estimation Strategy and Data

The aim of this study is to analyse the degree of monopsonistic competition in Germany

across sectors in general and for the minimum wage industries in particular. Ideally, each

industry would have an exogenous wage variation that could be used to estimate the

labour supply elasticity to the individual firm. At first sight, minimum wages or collective

bargaining agreements appear to offer such a variation at the industry level in Germany.

Unfortunately, all firms are equally affected by this wage increase, which implies that the

wage distribution over firms and workers is just shifted to the right or compressed from

below. Since no convincing exogenous wage change exists that only affects some firms in

a specific industry, we follow the semi-structural approach proposed by Manning [2003a].

The following paragraphs will give a very short (and stylized) overview of the dynamic

model of monopsonistic competition [Manning 2003a], which in turn heavily draws from

the Burdett and Mortensen [1998] equilibrium search model. The underlying idea is that

a stable equilibrium distribution of wages exists, both over workers and over firms. Each

worker receives job offers at an exogenously determined job offer rate. If the offered wage

is higher than the wage paid in the current job, the worker accepts and moves up the

job ladder. This implies that firms have a constant flow of hirings and separations. The

separation rate s(wt) depends negatively on the wage, simply because there are fewer firms

that will make a better wage offer in comparison to the current wage paid. The opposite is


true for the number of recruits R(wt). The number of workers in a firm Nt can be expressed

as the sum of workers who were already employed in the firm in the previous period Nt−1

and the number of recruits in period t. The number of separations s(wt)Nt−1 has to be

subtracted.

Nt = [1− s(wt)]Nt−1 + R(wt) (3.1)

Note that both, the separation rate s(wt) and the number of recruits R(wt) depends on

the wage rate offered by the firm. In the steady state, firm size should be constant, which

means that the number of separations should be equal to the number of recruits:

N(w) = R(w)/s(w) (3.2)

This implies that the long-term elasticity of labour supply to the individual firm εNw can

be expressed as:

εNw = εRw − εsw (3.3)

Thus, in order to estimate the labour supply elasticity, it is sufficient to estimate the re-

cruitment elasticity as well as the separation rate elasticity. Under the assumption that

recruitment from and separations to non-employment are wage inelastic, only the sepa-

ration rate elasticity of job-to-job transitions has to be estimated.4 The reason is that in

this case, the recruit of one firm must be a separation to another firm, which implies that

εsw = −εRw. The long-term elasticity of labour supply can then be expressed as:

εNw = −2εsw (3.4)

4. Clearly, the assumption that separations to non-employment are wage inelastic may not be true for allworkers. Manning’s (2003a) model of monopsonistic competition relaxes this assumption at a later point.However, since the estimation is considerably complicated by relaxing this assumption and the empiricalliterature shows that differences in the estimated labour supply elasticities are small [Hirsch, Schank andSchnabel 2010a], we only present results based on the assumption that separations to non-employment arewage inelastic.


Estimating the wage elasticity of labour supply to the individual firm therefore amounts

to estimating the wage elasticity of job-to-job transitions. The focus on job-to-job transi-

tions has the additional advantage that the majority of job-to-job transitions is voluntary

from the point of view of the worker, i.e. they are mostly supply-side driven; by con-

trast, many transitions to non-employment are due to dismissals and thus involuntary, i.e.

they are more likely to be due to demand-side factors. This is crucial because we aim at

identifying the labour supply, not the labour demand curve of the individual firm. We

additionally control for firm characteristics to ensure that demand-side shocks do not bias

the results. This is especially important for our comparison of different industries as the

macroeconomic situation may vary between industries.

We model the instantaneous separation rate of employment spell i in firm j at duration

time t as:

si(xi(t), zj(t)) = h0 exp(xi(t)′β + zj(t)′γ) (3.5)

where s is a dummy variable which takes the value 1 if a separation takes place and 0 oth-

erwise. Thus, the instantaneous separation rate depends on a constant baseline hazard h0

as well as worker characteristics xi(t) and firm attributes zj(t)that shift the baseline haz-

ard. Worker characteristics include sex, age, educational attainment and the current wage.

On the firm side, we control for the profitability of the firm, whether re-organisation or

outsourcing takes place, and for the share of women and temporary workers among total

employment. We allow for time-variant control variables by splitting spells into episodes

at the end of the year. This is especially relevant on the firm side, since the firm’s situation

may change over time. Thus, profitability is measured each year for the last financial year,

outsourcing activities refer to the 12 months preceding the interview and re-organisation

activities may have taken place during the last two years prior to the interview. The re-

maining indicators on the firm side as well as the worker characteristics are measured at

the timing of the interview. Furthermore, the regression equation includes year dummies

to control for aggregate year-specific effects, such as business cycle conditions. All estima-

tions are carried out separately for East and West Germany as well as specific industries


(one digit industry classification and minimum wage industries5).

The wage rate is specified in logs, which enables us to interpret the coefficient directly as

the wage elasticity of job-to-job transitions. The absolute value of the separation elasticity

multiplied by two equals the wage elasticity of labour supply to the individual firm. We

opt for the exponential model with a constant baseline hazard, because we explicitly do

not want to control for tenure. In the model of monopsonistic competition, higher wages

induce lower separation rates, thereby increasing tenure. Thus, including tenure would

take away variation from wages and therefore bias the estimated wage elasticity [Hirsch,

Schank and Schnabel 2010a; Booth and Katic 2011].

An alternative estimation approach to the hazard model presented above consists in es-

timating a logit or linear probability model. However, this method does not allow one to

deal with delayed entry and correcting for length-biased sampling [Cameron and Trivedi

2005]. As Section 3.1.2 shows, our data is based on a stock sample with a rather short ob-

servation period. The stock sample itself leads to an oversampling of short employment

spells, which might be selective in terms of transition probabilities. Fortunately, we do

have information on the original start date of the spell, which enables us to correct the

spell’s contribution to the likelihood function for delayed entry, i.e. the fact that no transi-

tion took place until the spell comes under observation. Additionally, duration analysis is

able to explicitly deal with right-censoring of spells, which occurs frequently.

In a last step, we test whether the estimated labour supply elasticities are in line with

the theoretical framework of monopsonistic competition. This is done by correlating the

industry-specific labour supply elasticities with (i) indicators of worker composition and

(ii) the amount of vacancies. Note that a high share of vacancies is a direct prediction of

the theoretical model, while our hypothesis in terms of worker composition are based on

existing empirical studies.

In terms of worker composition, existing empirical studies show that women and mi-

grants are subject to a higher degree of monopsonistic competition compared to men and

5. The minimum wage industries are identified based on the five digit industry classification as suggestedin the individual reports of the large scale evaluation of minimum wages by the BMAS.


natives, which is discussed in detail in Section 3.1.1. No studies exist for differences in the

degree of monopsony power faced by workers belonging to different skill groups. Never-

theless, it seems plausible that low-skilled workers face higher information asymmetries

or higher mobility costs. This is in line with low-skilled workers generally featuring lower

transition rates in the German labour market [Bachmann 2005]. We therefore expect the

degree of monopsonistic competition in an industry to be higher with increasing shares of

women, migrants or low skilled workers of the workforce.

Note that we already control for sex, nationality and highest educational attainment at

the individual level. While this implies allowing for differences in the separation proba-

bility, the wage elasticity of the separation rate is still assumed to be homogenous across

individuals. To the extent that e.g. high-skilled individuals do not only make more transi-

tions per se, but are also more sensitive to the wage in their decision, the average estimated

wage elasticity will be higher with increasing shares of high-skilled labour at the industry

level.

As for the role of vacancies, one of the key predictions of the monopsonisticc model of

the labour market is that in equilibrium, firms are supply-side constrained, and therefore

operate with a constant amount of vacancies. Given this theoretical prediction, the exis-

tence of vacancies in an industry can be viewed as a potential indicator for the existence of

monopsony power. We therefore expect to find higher degrees of monopsonistic competi-

tion in industries with a larger share of vacancies among total employment.

Data

The following analysis uses the LIAB, a linked employer-employee data set for the German

labour market.6 The basis of the data set is the Employment Statistics Register, an administra-

tive panel data set of the employment history of all individuals in Germany who worked

in an employment covered by social security between 1975 and 2010. The basis of the em-

6. The LIAB is described in Alda, Bender and Gartner [2005]. Detailed information on the data on individualworkers and on the firm side contained in the LIAB, the IAB Establishment Panel, is provided by Klosterhuber,Heining and Seth [2013] and Ellguth, Kohaut and Moller [2014], respectively.


ployee history is the integrated notification procedure for health insurance, the statutory

pension scheme, and unemployment insurance. At the beginning and at the end of any

employment spell, employers have to notify the social security agencies. This information

is exact to the day. For spells spanning more than one calendar year, an annual report for

each employee registered within the social insurance system is compulsory, and provides

an update on, for example, the wage and the current occupation of the employee. Further

worker characteristics included are the year of birth, sex, and nationality.

The LIAB combines this information on workers’ employment and unemployment his-

tory with plant-level information from the IAB Establishment Panel, an annual represen-

tative survey of German establishments that employ at least one worker who pays social

security contributions. Starting in 1993, the establishments covered by the survey were

questioned each year about various issues, such as the number of employees, the composi-

tion of the workforce, sales and investments. Using the unique establishment identification

number, one can match the information on workers with the establishment panel, and ob-

tain a linked employer-employee data set providing detailed information on individual

and establishment characteristics.

In order to follow firms and workers over time and thereby to control for heterogeneity

at both levels, we use a longitudinal version of the LIAB (“LIAB LM2”).7 This data set is

constructed as follows. First, establishments who participated in the IAB Establishment

Panel between 2000 and 2002 are selected.8 In a second step, the Employment Statistics

Register is used to link the sample of establishments with the employee history informa-

tion for all individuals who worked at least one day in one of the selected establishments

between 1997 and 2003. At the individual level, the information is updated at least once

a year when the annual notification is supplied by the employer. At the establishment

level, a new wave is provided each year as of June 30. We are thus able for time-varying

covariates in our analysis.

7. The longitudinal LIAB versions ”LM3” as well as the LM9310 both offer data for more recent years;however, in these versions the matching between firms and workers is poor (i.e. a significant share of workersis matched to the wrong establishment.)

8. To be exact, establishment that participate in the time period 1999-2001 or 2000-2002 are selected. Becauseweights are only available for the second group, we restrict our analysis to these establishments.


In order to compute separation elasticities from the LIAB, we need to identify labour

market states and direct job-to-job transitions at an individual level, as well as workers’

wages. We can derive three labour market states at each point in time: employment (E)

covered by social security, unemployment (U), if the worker is receiving transfer payments,

and non-participation (N). Non-participants are those individuals not recorded in the data

sets. Therefore, this state includes those workers out of the labour market, as well as work-

ers not covered by social security legislation, e.g. civil servants and self-employed workers.

As the distinction between unemployment and non-participation is not relevant for our

analysis, we consider these two labour market states jointly as non-employment. Because

we are only interested in job-to-job transitions, we drop all employment spells that result

in non-employment.

Because of the way the data are collected, both firms’ reports of a new employee and

individuals’ notifications of moving into or out of unemployment are not exactly consistent

with the actual change of labour market state. For example, workers might report to the

unemployment office only a few days after having been laid off. In order to deal with these

potential measurement errors, we proceed as follows. If a worker makes a transition from

one firm to another (according to the establishment identification number), we consider

this to be a direct job-to-job transition if the two employment records are less than 8 days

apart, and as a transition from employment to non-employment otherwise. In order to deal

with recalls, if the time lag between two employment notifications at the same firm does not

exceed 120 days, it is defined as one single employment spell. If the non-employment spell

is equal to or larger than 120 days, we drop this observation, as in this case a distinction

between a transition from employment to non-employment and a continuous employment

would be arbitrary. Additionally, all employment spells that are shorter than three days are

dropped, as are individuals with more than 300 employment spells.

The data provide precise information on the daily wage of every spell. However, no

information on working hours is provided. To ensure comparability between daily wage

rates, we restrict our analysis to regular, full-time employees. Workers in vocational train-

ing, marginal employees and part-time workers are thus excluded. Furthermore, all em-


ployment spells with wages in the bottom one percent of the wage distribution are ex-

cluded. This procedure is not sufficient for the upper end of the wage distribution, because

wages are right-censored at the social security contribution limit. To avoid possible biases

in the estimated wage elasticity of labour supply, all workers whose wages are at this limit

at least once during the observation period are dropped. Finally, in order to exclude tran-

sitions to non-employment due to (early) retirement, only individuals aged 16 to 55 on 1

January 2000, the beginning of our observation period, are included in the analysis.

Table 3.1 describes the resulting sample, separately for East and West Germany. We

observe a total of 727,610 (241,664) employment spells in West Germany (East Germany),

of which 112,000 (39,816) end in a job-to-job transition. The remaining spells are right-

censored. The annual transition probability is similar in East and West Germany and equals

six percent. Note that the number of workers is only slightly below the number of spells.

At first sight this seems odd, since a job-to-job transition would result in at least two em-

ployment spells per worker. This is however not entirely true for our sample, because we

only fully observe the subsequent employment spell if the establishment also participates

in the IAB Establishment Panel, which is rarely the case.

The descriptive evidence on our main explanatory variables is in line with expectations

– where it should be taken into account that our sample is conditioned on individuals in

employment who do not make a transition to non-employment. Not surprisingly, the aver-

age daily wage is at e99.75 higher in West Germany compared to East Germany (e73.83).

Interesingly, the average educational attainment is higher in East compared to West Ger-

many, which may be partly explained by focusing on employment spells ending in job-to-

job transitions. On the firm-side, 19 percent of all firms report a low profitability during

the last year in West Germany, while this figure only amounts to 15 percent in East Ger-

many. This surprising difference can be explained by the observation that 40 percent of

all establishments do not answer this question in East Germany. We therefore control for

non-response by including a dummy in the regression analysis.


Table 3.1.: Sample Description

East Germany West Germany

Mean sd Mean sd

Daily wage 73.38 23.19 99.75 27.38Log(Daily wage) 4.24 0.35 4.56 0.30Age 40.93 8.89 38.74 9.18Non-German 0.0065 0.0805 0.0950 0.2932Female 0.4416 0.4966 0.2697 0.4438Educational attainment: School degree 0.0393 0.1943 0.1640 0.3703Educational attainment: University degree 0.1692 0.3749 0.1178 0.3224

Firm profitability: Low 0.1518 0.3006 0.1908 0.3190Firm profitability: High 0.2278 0.3614 0.3266 0.3812Firm profitability: Non-response 0.0352 0.1516 0.0538 0.2001Firm profitability: Not applicable 0.4005 0.4726 0.1828 0.3730Reorganisation: yes 0.2710 0.3138 0.4063 0.3146Reorganisation: Non-response 0.3828 0.2540 0.3774 0.2375Outsourcing 0.1272 0.2574 0.1453 0.2888Share of women 46.18 27.27 32.63 23.81Share of temp. workers 10.91 21.09 5.34 9.02

Workers’ council: yes 0.8446 0.3623 0.9409 0.2359Workers’ council: Non-response 0.0266 0.1309 0.0176 0.1092Collective bargaining: Industry level 0.6365 0.4531 0.8186 0.3636Collective bargaining: Firm level 0.1556 0.3278 0.1021 0.2845Collective bargaining: Non-response 0.0051 0.0457 0.0020 0.0314

Spell duration 2,318 1,343 2,328 1,331Transition probability 0.0681 0.0626

Observation numbersJob-to-job transitions 39,816 112,058Employment spells 241,664 727,610Workers 239,689 721,415Firms 3,693 4,529

Notes: The unit of observation are continuous employment spells that do no result in non-employment. Note that the reference category is ommitted in the case of dummy variables. The ref-erence catogories include “vocational degree”, “normal firm profitability”, “no reorganisation”, “noworkers’ council” and “no collective bargaining agreement”.Source: LIAB, version “LM2”. Authors’ calculations.


3.1.3. Results

The Wage Elasticity of Labour Supply

In order to estimate the labour supply elasticity, we use the exponential model for job-to-

job transitions described in Section 3.1.2. The results of our baseline specification, which

includes all industries, are presented in Table 3.2 for East Germany and in Table 3.3 for

West Germany. We present four different models: Model 1 only contains industry and year

dummies in addition to log wages, Model 2 adds individual-level controls and Model 3

also includes controls at the establishment level in order to account for demand-side ef-

fects. Model 4 additionally controls for the existence of a workers’ council and collective

bargaining coverage.

The coefficients of the control variables do not differ qualitatively in East and West Ger-

many. Women are less likely than men to change employers. The transition probability

also decreases with age, but at a diminishing rate as workers get older. In contrast, em-

ployees with a university degree are more likely and workers holding a schooling degree

as highest educational attainment are less likely to make a job-to-job transition compared to

individuals who received vocational training. Non-Germans also show a lower separation

probability, although this relationship is only statistically significant in West Germany. The

low statistical significance of this correlation in East Germany is due to the fact that only

very few workers in East Germany are non-Germans (Table 3.1). Overall, these estima-

tion results are in line with the existing literature on labour market transitions in Germany

[Bachmann 2005; Kluve, Schaffner and Schmidt 2009].

Turning to the establishment-level controls, workers in firms with a low profitability in

the previous year or in firms pursuing outsourcing have a higher separation probability

(Model 3 in Tables 3.2 and 3.3). This shows the importance of controlling for demand-side

factors: Some workers change employers with an increasing threat of job loss. This deci-

sion is most probably independent of the wage. Reorganisation within the establishment

has, in contrast, no influence on the likelihood to change employers.9 The presence of a

9. The coefficient of the dummy for non-response to the question on re-organisation is actually negative and


Table 3.2.: Separation rate to employment in East Germany


Log(Daily wage) −1.383∗∗∗(0.106) −1.325∗∗∗(0.138) −1.146∗∗∗(0.118) −1.090∗∗∗(0.113)Female −0.153∗∗ (0.076) −0.134∗∗∗(0.036) −0.130∗∗∗(0.036)Age −0.131∗∗∗(0.013) −0.120∗∗∗(0.015) −0.121∗∗∗(0.015)Age2 0.001∗∗∗(0.000) 0.001∗∗∗(0.000) 0.001∗∗∗(0.000)Educational attainment:

School degree −0.023 (0.075) −0.104 (0.079) −0.101 (0.078)Vocational training 0.407∗∗∗(0.096) 0.356∗∗∗(0.088) 0.349∗∗∗(0.085)

Non-German 0.094 (0.093) 0.026 (0.093) 0.036 (0.094)

Profitability:Low 0.306∗∗ (0.120) 0.310∗∗∗(0.119)High 0.021 (0.135) 0.019 (0.134)Non-response 0.955∗∗∗(0.335) 0.952∗∗∗(0.332)Not applicable −0.134 (0.175) −0.121 (0.181)

Reorganisation:yes −0.059 (0.124) −0.050 (0.125)Non-response −1.222∗∗∗(0.261) −1.124∗∗∗(0.297)

Outsourcing 0.613∗∗∗(0.156) 0.616∗∗∗(0.157)Share of women −0.001 (0.003) −0.001 (0.003)Share of temp. workers 0.011∗∗∗(0.003) 0.010∗∗∗(0.004)

Workers’ council:yes −0.155 (0.112)Non-response −0.333∗∗ (0.170)

Collective bargaining:Industry level 0.029 (0.109)Firm level 0.103 (0.137)Non-response −0.257 (0.665)

Industry dummies yes yes yes yesYear dummies yes yes yes yes

LogLikelihood -102,941 -93,046 -90,156 -90,089Observations 610,635 574,750 573,175 573,175

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Clustered standard errors at the establishment level in parentheses.Source: LIAB, version “LM2”. Authors’ calculations.


Table 3.3.: Separation rate to employment in West Germany


Log(Daily wage) −1.271∗∗∗(0.149) −1.377∗∗∗(0.154) −1.463∗∗∗(0.089) −1.387∗∗∗(0.081)Female −0.065 (0.050) −0.100∗∗∗(0.033) −0.095∗∗∗(0.033)Age −0.082∗∗∗(0.012) −0.076∗∗∗(0.009) −0.079∗∗∗(0.009)Age2 0.001∗∗∗(0.000) 0.001∗∗∗(0.000) 0.001∗∗∗(0.000)Educational attainment:

School degree −0.060 (0.047) −0.098∗∗ (0.045) −0.094∗∗ (0.045)Vocational training 0.631∗∗∗(0.083) 0.545∗∗∗(0.051) 0.526∗∗∗(0.050)

Non-German −0.272∗∗∗(0.056) −0.266∗∗∗(0.058) −0.260∗∗∗(0.058)

Profitability:Low 0.189 (0.121) 0.182 (0.121)High −0.243∗ (0.126) −0.243∗ (0.126)Non-response 0.540∗ (0.296) 0.562∗ (0.299)Not applicable −0.435∗∗∗(0.160) −0.379∗∗ (0.166)

Reorganisation:Yes −0.147 (0.099) −0.134 (0.100)Non-response −0.371 (0.245) −0.330 (0.247)

Outsourcing 0.405∗∗∗(0.131) 0.420∗∗∗(0.135)Share of women 0.001 (0.003) 0.001 (0.003)Share of temp. workers 0.010∗∗∗(0.004) 0.010∗∗∗(0.004)

Workers’ council:Yes −0.177∗ (0.105)Non-response −0.386∗∗∗(0.149)

Collective bargaining:Industry level −0.166∗ (0.101)Firm level −0.097 (0.144)Non-response −0.704 (0.637)

Industry dummies yes yes yes yesYear dummies yes yes yes yes

LogLikelihood -304,360 -282,286 -264,823 -264,452Observations 1,885,004 1,819,537 1,760,060 1,760,060

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Clustered standard errors at the establishment level in parentheses.Source: LIAB, version “LM2”. Authors’ calculations.


workers’ council and collective bargaining coverage are not correlated with the job-to-job

transition probability in East Germany and are only significant at the 10 percent level in

West Germany. The point estimate of the presence of a workers’ council is, though not

statistically significant in East Germany, negative in East and West Germany alike, which

is in line with the literature on workers’ voice and the labour turnover rate. For example,

Hirsch, Schank and Schnabel [2010b] show for Germany that the separation rate is reduced

by the presence of a workers’ council through voice, monopoly (i.e. wage) and insurance

effects.

The coefficient of interest is the one on the (log) daily wage. It can directly be interpreted

as the wage elasticity of the separation rate of job-to-job transitions (cf. Section 3.1.2). The

estimation results show that if the wage increases by one per cent, the probability to make

a separation, conditional on job survival until time t, decreases by 1.09–1.15 per cent in East

Germany and by 1.39–1.46 per cent in West Germany. Assuming that separations to non-

employment are wage inelastic, the labour supply elasticity to the individual firm is twice

the wage elasticity of separations to employment (cf. Equation 3.4). Thus, taking values of

2.18–2.3 in East Germany and 2.78–2.92 in West Germany, the average labour supply elas-

ticity to the individual firm is considerably lower as suggested by the neoclassical model

of the labour market, which assumes that labour supply to the individual firm is perfectly

elastic. Note that this result is in line with other estimates of the labour supply elasticity

[Ransom and Oaxaca 2010; Hirsch, Schank and Schnabel 2010a].

In order to obtain the wage elasticity of labour supply by industry, we now estimate the

baseline specification separately for each industry, and East and West Germany. We do

so using Model 3, which is our preferred specification for several reasons. First, it con-

trols for worker-level heterogeneity. Second, demand-side factors are taken into account

through the inclusion of firm-level variables. Third, it is unclear whether the existence of

a workers’ council and coverage by a collective bargaining agreement should be included

as additional control variables at the establishment level, as is done in Model 4. On the

highly significant in East Germany. One possible explanation is that firms that undertake restructuring tendnot to provide an answer.


one hand, workers’ councils and union coverage tend to increase wages and reduce sep-

aration rates [Hirsch, Schank and Schnabel 2010b]. On the other hand, the role of unions

is not well grounded in the theoretical framework of monopsonistic competition. Further,

Model 3 provides more conservative estimates than Model 4, since the inclusion of col-

lective bargaining coverage and the existence of a workers’ council further reduces the

estimated wage elasticities of labour supply.

Before turning to the labour supply elasticities, we provide a descriptive overview of the

average daily wage and the job-to-job transition probability by industry (Tables 3.4 and 3.5).

Even though the industry classification is rather broad, significant differences exist along

both dimensions.

One may be tempted to describe a labour market segment that is characterized by monop-

sonistic competition as one in which wages are low and job-to-job transitions are rare.

However, the wage level itself is mainly influenced by composition effects of the work-

force leading to productivity differences, and the degree of observed transition dynamics

may be equally affected by third factors, such as the share of temporary workers. This am-

biguity in the direct relationship between wages and job-to-job transitions is also present

in Tables 3.4 and 3.5. In East Germany for example, mining is characterized by high wages

(e80.40) and a high annual job-to-job transition probability (10.6 percent). Along the same

line, public and private services have low average wages of e54 and a low separation rate

of 6.3 percent. Similar examples can be found for West Germany: Electricity and water

supply is characterized by high wages (e107.3) and a high job-to job transition probability

(11.5 percent), and public administration has low wages (e87.5) and low separation rates

(4.6 percent). However, in other industries the relation between the average wage level

and the job-to-job transition rate is reversed: Financial services show high average wages

and a low transition probability in West and East Germany alike, while the opposite is true

for the hotel and restaurant industry.

From a theoretical point of view, however, it is not the level of job separations that char-

acterises the degree of monopsony power in a market, but its sensitivity to the wage. Recall

from Section 3.1.1 that monopsonistic competition is defined as a situation in which work-


Table 3.4.: Wage elasticity in individual industries - East Germany

Industry averagesElasticity oflabour supply

LogLLObservations

Wages Trans.Spells

Transi-tionsMean sd prob.

Agriculture 41.47 14.89 0.0999 1.537∗∗ (0.309) -1,435 7,560 673Mining 80.40 16.59 0.1060 7.181∗∗∗(1.154) -704 8,801 1,597Manufacturing 67.03 23.93 0.0503 3.024∗∗∗(0.256) -21,143 156,212 8,088Electricity and water supply 86.15 18.17 0.0896 4.598∗∗∗(0.616) -1,812 19,965 1,869Construction 60.47 19.95 0.0984 2.219∗∗∗(0.249) -5,915 29,219 3,134Wholesale and repairs 59.24 21.09 0.0565 1.481∗ (0.403) -1,631 11,136 649Retailing 52.30 22.37 0.0811 −0.380 (0.283) -1,251 6,698 531Hotels and restaurants 40.90 15.41 0.1721 −0.430 (0.344) -361 1,113 259Transportation and communication 73.11 17.85 0.0894 1.491∗ (0.432) -4,138 32,689 3,105Financial services 86.91 20.69 0.0622 4.313∗∗∗(0.277) -2,541 18,005 1,086Business services 60.76 30.38 0.1187 1.712∗∗∗(0.298) -5,465 22,797 2,791Public administration 73.22 20.37 0.0460 1.512∗ (0.392) -15,902 130,188 5,960Education 64.27 30.81 0.1308 3.172∗∗∗(0.250) -9,027 38,580 4,815Health 69.75 22.60 0.0595 1.744∗∗∗(0.241) -9,306 64,733 3,674Public and private services 53.93 26.09 0.0627 3.345∗∗∗(0.244) -3,806 25,470 1,585

Minimum wage industriesElectricians 57.15 16.09 0.1284 −0.810 (0.438) -479 1,906 271Main construction 63.23 20.24 0.0942 1.797∗∗ (0.358) -4,057 21,301 2,198Waste removal 50.08 17.66 0.0907 1.873∗∗∗(0.355) -1,086 6,569 637Elderly care 60.15 20.19 0.0702 2.850∗∗∗(0.458) -783 6,577 454Cleaning 35.33 14.67 0.0934 1.558∗∗∗(0.301) -456 2,158 304

All industries 66.91 24.84 0.0681 2.292∗∗∗(0.118) -90,156 573,175 39,816

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Clustered standard errors at the establishment level in parentheses. Each row shows the coefficient of log(daily wage) mulit-plied by −2. The estimations are run separately by industry and based on Model 3 (cf. Section 3.1.3). LogLL=LogLikelihood of themodel. Additionally, the average wage and the annual transition probability at the industry level are provided. The average wagelevel is calculated based on a different sample, i.e. before employment spells resulting in non-employment are excluded.Source: LIAB, version “LM2”. Authors’ calculations.

ers do not change employers necessarily if they could earn a higher wage in another job.

What is therefore needed for an assessment of the degree of monopsonistic competition in

an industry is a connection between worker mobility and wages at the individual level, i.e.

the labour supply elasticity of the separation rate.

The estimation results of the labour supply elasticities for East and West Germany are

also presented in Tables 3.4 and 3.5, respectively, both for the broad industry classification

as well as for the minimum wage industries. These results reveal considerable differences

between industries, ranging from zero (retailing, hotels and restaurants) to 7.2 (mining) in

East Germany and from zero (hotels and restaurants) to 4.2 (transportation and commu-

nication) in West Germany. Industries with especially low labour supply elasticities, and

consequently a higher degree of monopsonistic competition, include construction, whole-


Table 3.5.: Wage elasticity in individual industries - West Germany

Industry averagesElasticity oflabour supply

LogLLObservations

Wages Trans.Spells

Transi-tionsMean sd prob.

Agriculture 60.37 23.22 0.0947 3.441∗∗∗ (0.493) -400 1,725 155Mining 85.67 15.67 0.1614 4.031∗∗∗ (0.313) -4,025 23,470 3,582Manufacturing 99.79 27.78 0.0545 2.958∗∗∗ (0.175) -118,957 922,142 52,381Electricity and water supply 107.29 26.08 0.1152 3.907∗∗∗ (0.355) -4,247 23,208 2,585Construction 88.80 24.18 0.0886 1.516∗∗∗ (0.177) -7,440 40,270 3,669Wholesale and repairs 92.03 31.75 0.0813 1.804∗∗∗ (0.313) -9,350 48,343 3,810Retailing 73.00 25.94 0.1075 0.628∗∗ (0.149) -5,724 25,075 2,790Hotels and restaurants 55.52 22.11 0.1820 0.702 (0.223) -1,460 4,110 757Transportation and communication 92.51 23.74 0.0464 4.149∗∗∗ (0.367) -12,742 105,782 4,901Financial services 110.04 28.28 0.0653 2.883∗∗∗ (0.306) -23,599 149,905 9,596Business services 82.86 39.60 0.1319 2.979∗∗∗ (0.213) -18,419 66,962 9,064Public administration 87.50 23.39 0.0456 3.259∗∗∗ (0.194) -20,918 158,978 7,123Education 85.97 30.00 0.0815 2.797∗∗∗ (0.388) -4,545 26,336 2,054Health 83.83 27.22 0.0647 2.593∗∗∗ (0.101) -20,839 122,339 7,588Public and private services 90.75 33.24 0.0432 2.502∗∗∗ (0.157) -5,251 41,415 2,003

Minimum wage industriesPainters 76.00 20.28 0.0984 3.687∗∗∗ (0.364) -206 960 108Electricians 87.16 29.88 0.1379 0.456 (0.409) -489 3,988 546Main construction 90.83 23.07 0.0824 1.751∗∗∗ (0.208) -5,233 30,449 2,591Waste removal 97.26 20.45 0.0293 2.928∗∗∗ (0.354) -1,133 14,355 428Elderly care 71.84 25.29 0.0929 2.437∗∗∗ (0.275) -2,057 8,966 773Cleaning 46.64 24.72 0.1398 1.177∗∗∗ (0.226) -771 2,318 673

All industries 95.12 29.36 0.0626 2.925∗∗∗ (0.089) -264,823 1,760,060 112,058

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Clustered standard errors at the establishment level in parentheses. Each row shows the coefficient of log(daily wage) mulitpliedby −2. The estimations are run separately by industry and based on Model 3 (cf. Section 3.1.3). LogLL=LogLikelihood of the model.Additionally, the average wage and the annual transition probability at the industry level are provided. The average wage level iscalculated based on a different sample, i.e. before employment spells resulting in non-employment are excluded.Source: LIAB, version “LM2”. Authors’ calculations.

sale, retailing, and hotels and restaurants. In contrast, mining, manufacturing, electricity

and water supply, financial services, and education are characterized by relatively high

labour supply elasticities in East and West Germany alike.

The source of these differences in the degree of monopsonistic competition across indus-

tries lies in the behaviour of workers, who do not change jobs to obtain a higher wage.

Possible reasons include non-wage preferences, imperfect mobility or incomplete infor-

mation. Previous empirical literature shows that women, non-Germans and low-skilled

workers are characterized by lower separation rate elasticities due to one of these three

reasons (cf. Section 3.1.1). Therefore, we expect to find a higher degree of monopsonistic

competition in industries with a high share of these worker groups. Indeed, as Table 3.6

shows, the correlation coefficients all have the expected sign, although they are statisti-


Table 3.6.: Correlation of labour supply elasticities with industry-level indicators


Correlation p-value Correlation p-value

Vacancies −0.187 0.431 −0.692 0.001Share of women −0.138 0.561 −0.335 0.137Share of non-Germans −0.250 0.288 −0.340 0.131Share of low-skilled −0.356 0.123 0.123 0.596Share of high-skilled 0.412 0.071 0.147 0.525

Notes: Correlation coefficient of the industry-specific labour supply elasticities (cf. Ta-bles 3.4 and 3.5) and industry-level indicators. Note that the share of vacancies amongtotal employment is calculated based on the firm-side information (‘Betriebspanel’) inour data.Source: LIAB, version “LM2”. Authors’ calculations.

cally insignificant except for the share of high-skilled workers in East Germany. Note that

not the individual worker, but the 1-digit-industries are the unit of analysis here. Thus, the

number of observations is extremely low and statistical significance is hard to achieve. Still,

the worker composition does explain differences in the estimated labour supply elasticities

across industries.

Further, one of the key predictions of the monopsony model is that firms operate with

a constant amount of vacancies. Therefore, those industries with high point estimates of

the labour supply elasticity should be characterized by few vacancies and vice versa. The

amount of vacancies in an industry is measured as the number of vacancies a firm of-

fers divided by the current number of employee, given that a firm has vacancies. This

information is derived from the establishment-side contained in the data. The expected

negative correlation between the estimated labour supply elasticities and the amount of

vacancies prevails in East and West Germany (Table 3.6). However, it is much stronger in

West Germany, with a correlation coefficient of -0.69 that is significant at the 1 percent level

(Table 3.6).

Discussion

The results presented above have important implications for the existing industry-specific

minimum wages. Overall, the minimum wage industries show labour supply elastici-

ties that lie in the middle of our range of estimates. Electricians are the only exception


with a labour supply elasticity that is very small in magnitude and not statistically differ-

ent from zero.10 The labour supply elasticities of the commercial cleaning industry, main

construction, and waste removal (only East Germany) are smaller than two. Waste re-

moval (only West Germany) and elderly care show elasticities between two and three (Ta-

bles 3.4 and 3.5).

Thus, the fact that the existing empirical literature generally found no disemployment

effects of industry-specific minimum wages in Germany [Apel et al. 2012; Boockmann et al.

2013; Frings 2013]11 may partly be explained by the degree of monopsonistic competition

in the labour market. While the minimum wage industries are not characterized by espe-

cially low labour supply elasticities in comparison to other industries, the results do also

not point towards a perfectly elastic labour supply. Of course, other explanations exist that

are at least equally important. These include a low bite of the sectoral minimum wages (es-

pecially in West Germany), substitution of high-skilled for low-skilled workers, decreasing

profits of firms, higher prices on the product market, increasing working hours and, espe-

cially for the construction sector, potential disemployment effects for posted workers as

well as increasing occurrence of undeclared work [IAB, RWI and ISG 2011; IAW 2011a,b].

Concerning the question why minimum wages were introduced in some, but not in other

industries, our results suggest that the degree of monopsony power in an industry did not

play a role. A well-informed, welfare-oriented politician would introduce minimum wages

in industries with a high degree of monopsonistic competition, because the risk of job

destruction due to increasing wages would be minimized. However, our estimation results

do not provide evidence for this. For example, the industries with the lowest labour supply

elasticities, such as retailing as well as the hotel and restaurant industry, have no minimum

wage. The strongest economic justification exists for the commercial cleaning industries,

where average wages are the lowest in our sample and the labour supply elasticity is just

above one in East and West Germany alike (Tables 3.4 and 3.5). In general, however, the

10. To be exact, the labour supply elasticity for painters in West Germany is rather high at 3.7. However,this estimate should be interpreted with care due to the low number of observations, especially in terms ofjob-to-job transitions.

11. Aretz, Arntz and Gregory [2013] find that the roofing sector is an exception in this respect.


chosen industries were neither those with the lowest wage, nor those with the strongest

monopsony power.

Given the institutional framework governing sector-specific minimum wages, the selec-

tion criterion for the minimum wage introduction in specific industries appears thus more

politico-economic in nature. Recall that after a specific industry is added to the Posting

of Workers Law (by politicians), the social partners bargain over the minimum wage rate

and jointly apply for an extension to all workers and firms in that industry. The needed

consensus between the social partners can be explained by possible effects of sectoral mini-

mum wages on product market competition. Haucap, Pauly and Wey [2001] proposes that

industry-specific minimum wages may be used as a cost-raising strategy of firms to deter

market entry. Bachmann, Bauer and Frings [2014] provide some empirical evidence on this

theory for Germany. Indeed, the first minimum wage introductions in Germany during the

late 1990s were clearly motivated by preventing low-wage competition from abroad [IAB,

RWI and ISG 2011].

In terms of the introduction of a uniform minimum wage of e8.50 in 2015, one can first

note that industries with the highest labour supply elasticities are generally characterized

by lower average wages and vice versa (Tables 3.4 and 3.5). Still, there are exceptions to this

pattern. Agriculture in West Germany shows one of the lowest average wage rates and a

rather high wage elasticity of 3.4 at the same time. The same is true for public and private

services in East Germany. Stated differently, negative employment effects of a uniform

minimum wage of e8.50 might be mitigated in some, but not in all low-wage industries.

Brautzsch and Schulz [2013] calculate the bite of the minimum wage for specific sectors,

i.e. the share of workers currently earning less than the level ofe8.50 at which the statutory

minimum wage will be introduced. The bite is strongest in agriculture, retailing, the hotel

and restaurant industry as well as transportation and communication, ranging between

26.2–67.1 percent in East Germany and between 14.1–41.1 percent in West Germany. At the

same time, the estimated labour supply elasticity is rather high in agriculture as well as

transportation and communication in West Germany. The combination of high labour sup-

ply elasticities and high-impact minimum wages implies that employment losses appear


inevitable, not only in East but also in West Germany.

3.1.4. Conclusion

In this paper, we have analysed the degree of monopsony power of German employers

following a semi-structural approach based on the dynamic model of monopsonistic com-

petition proposed by Manning [2003a]. In doing so, we compute the degree of monopsony

power for different industries, and separately for East and West Germany. Using a unique

linked employer-employee data set for Germany allows to control for heterogeneity of both

firms and workers, and for demand side effects.

Our findings are first, that the labour supply elasticity to the individual firm is consid-

erably lower than suggested by the neoclassical model of the labour market, which is in

line with existing estimates. Second, we find important differences in labour supply elas-

ticities between industries. Therefore, the labour markets of different industries seem to

be characterised by differing degrees of monopsony power. Finally, we showed that the

estimated labour supply elasticities are negatively correlated with the amount of vacancies

at the sectoral level. This is consistent with the monopsonistic model of the labour market,

which predicts the existence of a positive stock of vacancies.

As the degree of monopsony power is one important determinant of the employment

effects of minimum wages, our results have important policy implications. First, it be-

comes obvious that the industries where, starting in the late 1990s, minimum wages were

introduced at a sectoral level, were generally not characterised by a high degree of monop-

sony power. Therefore, avoiding negative employment effects does not seem to have been

an important criterion when introducing minimum wages at the sectoral level. This is

likely to be due to the institutional framework governing sectoral minimum wages in Ger-

many, where trade unions and employer associations have to agree on collective bargain-

ing agreements with respect to a minimum wage which can then – under certain conditions

– be declared generally binding.

Our findings are also relevant for the expected effects of the introduction of a statutory


minimum wage in Germany on January 1st, 2015. On the one hand, given our finding of

monopsony power on the German labour market, the negative employment effects of the

minimum wage introduction may be less severe than suggested by a neoclassical model

of the labour market, although they may still be sizeable, especially because the minimum

wage will be introduced at a relatively high level. On the other hand, given large inter-

industry differences in monopsony power between East and West Germany and between

industries, the employment effects of the minimum wage introduction are likely to be very

unevenly distributed in the German labour market. This calls for a very close monitoring

and evaluation of employment effects, as well as – if necessary – swift political action.


3.2. Minimum Wages as a Barrier to Entry

The economic effects of minimum wages on employment have been a very active field of

research for at least the last two decades, from both a theoretical and an empirical per-

spective (see Neumark and Wascher 2008, for an overview). Other important economic

consequences of minimum wages, however, have been rather neglected. This is particu-

larly the case for the possibility that minimum wages might be used to influence the degree

of competition on the product market, which has been shown to have played a role in both

the U.S. and Germany (cf. Williamson 1968; Heitzler and Wey 2010). In particular, firms

may be able to increase their profits by improving their competitive position through a

minimum wage, thus effectively forming a cartel.

The possibility of raising rivals’ costs through minimum wages critically hinges upon

the condition that all firms can be forced to pay the higher, entry-preventing wage. In

this regard, the German labour market offers an interesting opportunity to directly study

the link between minimum wages and product market competition, because minimum

wages in Germany are introduced at the industry level at the initiative of employers and

trade unions by declaring collective bargaining agreements as generally binding. This way

of introducing minimum wages also exists in a number of other industrialized countries

such as France, the Netherlands and Portugal. The results of our study are thus of general

interest for the analysis of cost-raising strategies to influence the degree of competition in

an industry.

To the best of our knowledge, this is the first empirical analysis of employers’ attitude

towards the introduction of minimum wages. We use a unique data set covering 800 firms

from eight different service sectors in Germany, where a minimum wage introduction was

being discussed at the time of the survey. Our analysis explores the determinants of sup-

porting or opposing the introduction of a minimum wage paying particular attention to the

role of product market competition. Furthermore, we scrutinize the institutional features

of the labour market that are associated with firms’ support of minimum wages.

The results of our analysis have several important implications. First, a cartel in favour


of minimum wages may not only reduce the number of firms operating in the market,

but also the number of employees. This can be seen as an alternative mechanism through

which minimum wages are potentially harmful to employment. Second, the monopolistic

rents achieved on the product market by the surviving firms are paid by the consumers,

which potentially reduces social welfare. Third, the incentive of firms’ and workers’ as-

sociations in the more prosperous countries of the European Union to collude by intro-

ducing minimum wages may increase further as service provision from poorer (mainly

Eastern European) EU Member States becomes more important. Firms’ support of mini-

mum wages can thus be viewed as a showcase example for how social policies in Western

Europe may act as protectionist instruments against competition from low-wage countries.

The remainder of this paper is structured as follows. Section 3.2.1 briefly reviews the

institutional background with respect to minimum wages in Germany. We present the rel-

evant economic theory in Section 3.2.2. The empirical strategy and the data are described

in Section 3.2.3. The empirical results are presented in section 3.2.4. The final section sum-

marizes and concludes the discussion.

3.2.1. Institutional Background

Germany is one of the few European countries without statutory minimum wages. This

has remained broadly unquestioned for several decades, because high coverage rates of

collective bargaining provided an effective floor for wages. However, since the beginning

of the 1990s, union density as well as coverage have been decreasing continuously [Kohaut

and Ellguth 2008]. This development coincided with an increase in wage inequality, espe-

cially at the bottom of the wage distribution [Dustmann, Ludsteck and Schonberg 2009],

although it is not clear whether the decline in collective bargaining has been a causal factor

in this context [Antonczyk, Fitzenberger and Sommerfeld 2010].

At the same time, the completion of the EU’s Single Market progressed in terms of both

increased intra-EU trade and labour mobility. This led to a rise in the number of posted,

low-wage workers in the German construction industry during the 1990s, which was per-


ceived as a threat to the employment prospects of German workers and eventually led to

the introduction of minimum wages in this industry in 1997. Thus, the main motive for the

introduction of minimum wages in this sector was protectionist, i.e. the aim was to shield

the German construction industry from low-wage competition [Woolfson and Sommers

2006].

The statutory framework of the existing minimum wages in Germany is the “Posting of

Workers Law”, which is based on the European Union’s directive on posted workers (Di-

rective 96/71/EG). Based on this law, industry-specific and collectively bargained wage

rates may be extended to all workers and employers, independently of their membership

in trade unions or employer associations. However, there are strict requirements for a col-

lective agreement to be declared generally binding. First, the initial collective agreement

must be representative, implying that no additional collective agreement exists in the re-

spective industry that covers more workers or union members. Second, the extension of

the collective agreement should be in the interest of the general public. And third, the

social partners need to apply jointly for an extension, which requires a high degree of con-

sensus. If these conditions are met, the Federal Ministry of Labour and Social Affairs usu-

ally declares the collective agreement generally binding without consulting any additional

governmental bodies or institutions. Only when the application is filed for the first time, a

committee consisting of three representatives of the respective trade union and employer

association has to give its consent [Bilous 2010; Eurofound 2010].

If an industry has introduced a minimum wage based on the Posting of Workers Law,

this minimum wage does not only apply to domestic, but also to posted (i.e. foreign)

workers. This attribute of the German minimum wage institution leads to an interesting

interaction with the European’s Union Service Directive, which was passed in 2006. The

Service Directive, also referred to as ‘Bolkestein Directive’, aimed at increasing competition

within the European Union by enhancing the free movement of services already agreed

upon in the Treaty of Rome as one of the four freedoms of the EU’s Single Market [Menz

2010]. The key instrument for achieving this goal is to lower regulatory barriers between

countries which prevent the provision of services in another EU Member State. Indeed,


Kox and Lejour [2006] argue that policy heterogeneity is the main obstacle to intra-EU

trade in services and to the possibility to open an establishment in another EU country.

The intense public discussion of the directive’s first draft may have influenced the per-

ception of low-wage, competitive threats from the new Member States. The Service Direc-

tive’s first draft intended to apply the ‘country of origin principle’ (COP) to the temporary

provision of services abroad, which specifies that workers in the host country are subject to

the laws and regulations of the home country. This applies to social policy and labour mar-

ket regulation, including collectively bargained wage rates and minimum wages. Saint-

Paul [2007] argues that the COP makes personal services tradable in the sense that the

service can be bought in any EU Member State and the travelling cost for the worker per-

forming the service constitutes a special type of transportation cost. The COP has been

heavily criticized in high-wage countries such as Germany by both employer associations

and trade unions, as it was perceived as a threat to employment, wages and working con-

ditions of German workers. As a consequence, the social partners as well as the German

government have agreed upon extending minimum wages beyond the main construction

industry [Menz 2010].

Interestingly, two EU directives with opposing aims now compete at the level of the

Member States. On the one hand, the Bolkestein Directive intends to facilitate the provision

of services in another EU member state by specifying that the rules and regulations of the

home country apply. On the other hand, the Posting of Workers Law clearly states that all

workers have to be employed under the minimum working standards of the host country.

This conflict is resolved as the Posting of Workers Law takes precedence over the COP.

Stated differently, in an industry with minimum wages, foreign and domestic workers

must both be paid accordingly. In contrast, in industries without minimum wages, the

rules and regulations of the home country, and not of the host country, apply to foreign

workers. Therefore, the introduction of minimum wages in a specific industry does not

only raise labour costs to domestic, but also to foreign employers.

Currently, minimum wages do not exist in the majority of sectors in Germany.12 How-

12. In December 2012, minimum wages existed for the waste disposal industry, main construction, mining,


ever, against the background of the decline in union density and coverage, the increase

in wage inequality, and the introduction of the Service Directive, the political pressure to

extend minimum wage legislation has grown further during the last few years. Given that

minimum wages in Germany are introduced at the industry level, it is conceivable that

they may be used as an instrument to influence the degree of competition in an industry.

Note that extensions of collective agreements, partly in addition to minimum wages, are

not uncommon in the European Union. According to Kerckhofs [2011] such extensions

are frequently used in Belgium, the Czech Republic, Finland, France, the Netherlands, and

Portugal, although the exact mechanisms differ. In Spain, collectively agreed wage rates

automatically apply to all workers in an industry by law. Therefore, our analysis does not

only apply to Germany, but to all countries with a tradition in the extension of collective

agreements.

3.2.2. Theoretical Considerations

Traditional labour market theory, such as the Marshallian or the monopsonistic models of

the labour market, does not offer an explanation for the observation that some firms are

in favour of minimum wages. By contrast, the industrial organization literature explicitly

models cost raising strategies in order to deter entry and/or push existing competitors out

of the market [Salop and Scheffman 1983, 1987]. There, the focus is on activities by indi-

vidual firms such as inducing suppliers to discriminate against competitors, controlling

exclusive distribution channels, lobbying for product standards or government regulation,

as well as advertising and R&D races. However, cost-raising strategies can also become

effective through the labour market.

The first type of model in this vein is based on the insider-outsider theory in terms of

wage setting (Gollier 1991, Ishiguro and Zhao 2009). If outsiders, i.e. unemployed workers,

are not unionized, they can be hired at a lower cost compared to insiders. This will stimu-

late low-cost firms to enter the market. Incumbent employers may therefore be interested

roofers, electricians, commercial cleaning, painters and varnishers, elderly care, security services, laundryservices, further education, as well as temporary employment.


in raising the wages of outsiders, thereby increasing product prices and restricting indus-

try output. Insiders are also motivated to increase wages of outsiders in order to secure

themselves against becoming unemployed. Furthermore, the increased price on the prod-

uct market legitimates insiders’ wage premium. Thus, (unionized) insiders and employers

form a coalition to increase wages for outsiders. This behaviour discourages market entry

of new firms. Chappell, Kimenyi and Mayer [1992] deliver one of the few empirical studies

on this topic. Using US data, they show that a higher degree of unionization in an industry

is indeed associated with entry deterrence.

Another cost-raising strategy that uses a labour market mechanism consists in minimum

wage legislation. In the U.S., this was made clear by the Supreme Court decision “United

Mine Workers vs. Pennington”, stating that the trade union had violated antitrust laws

when agreeing with one employer on relatively high wages that were binding for the entire

industry [Williamson 1968]. In Williamson’s model, the large-scale, capital-intensive firm

is more productive than the small-scale, labour-intensive firm. The former can therefore

afford to pay wages in excess of the competitive rate, which pushes small-scale firms out

of the market. This has important effects for productivity, product variety and prices, as

can be inferred from the model by Braun [2011], who analyses the consequences of sector-

level bargaining – which effectively amounts to a minimum wage – within the context of

the heterogeneous firm model by Melitz and Ottaviano [2008].

In Germany, the notion that high-productivity firms may use minimum wages to im-

prove their competitive position has been distinctly demonstrated when minimum wages

were introduced in the postal sector in 2008. After liberalization of the postal service in-

dustry, the former state monopolist ‘Deutsche Post’ strongly supported the introduction of

minimum wages. Due to its large-scale logistic infrastructure and the high market share,

Deutsche Post could afford to pay considerably higher wages compared to new entrants

(Heitzler and Wey 2010).

Independently of the exact mechanism that is used to increase labour costs, the exist-

ing theoretical models share two assumptions which are crucial for our empirical analysis.

First, the union wage or the minimum wage must apply to all firms and workers alike.


This implies either that unionization is in practice mandatory or that union wage rates

are extended to the entire industry. Second, the good should not be tradable and no close

substitutes should exist, because these models require firms to be able to pass on increas-

ing labour costs to consumers through higher prices. Both conditions are fulfilled for the

industries that we investigate in our analysis. First, the Posting of Workers Law ensures

that union wage rates can be declared collectively binding. Second, the industries anal-

ysed belong to the service sector (e.g. hairdressers, florists, motor mechanics, restaurants),

which means that the services cannot be imported from abroad but have to be delivered

domestically.

Foreign competition in these sectors arises nevertheless if foreign workers and firms are

able to offer their services in Germany, which has become a more important phenomenon

with the European Union’s Service Directive (cf. Section 3.2.1). Therefore, minimum wages

cannot only be used as a barrier to entry towards domestic competitors, but also against

low-wage competition from the new Member States. A survey of German firms, which

focuses on the perception of the Service Directive, confirms the impression that some firms

support minimum wages in order to avoid low-wage competition from the new Member

States [Kiessl, Pohl and Schmalholz 2006; Nerb 2006]. Furthermore, industries with skill-

intensive production technologies are likely to benefit from trade liberalization through the

Service Directive, while the opposite is true for industries with a high share of low-skilled

workers [Nerb 2006]. Security services, retailing, and restaurants are specifically named as

low-skilled industries in which competition is expected to increase significantly.

The correlation between the degree of existing competition and the support of a mini-

mum wage is not clear ex ante. On the one hand, the collusion of the social partners may

be facilitated if the existing degree of competition is rather low. In that case, the degree of

competition should have a positive impact on the likelihood to support a minimum wage

introduction. On the other hand, the correlation may be negative. This could be caused

by a simultaneous occurrence of a low degree of product market competition and a high

degree of union power [Stewart 1990].13 In the presence of strong unions, i.e. a small non-

13. This relation is confirmed in the data by the highly negative and significant correlation between the HHI


union sector, there is no need to collude in order to establish an additional barrier to entry.

Thus, weaker competition on the product market may be negatively correlated with mini-

mum wage support. Therefore, the correlation between the degree of existing competition

and the probability to support a minimum wage cannot be determined theoretically and

boils down to an empirical question.

In addition to the existing degree of competition, the perceived threat of an increase in

competition in the future may be important for employers’ support of minimum wages.

The public discussion on the ‘Country of Origin Principle’ of the Service Directive influ-

enced employers’ perception in this regard. A sharp increase in low-wage competition

from abroad appears to be especially likely in industries with low barriers to entry and

in regions close to the Central and Eastern European Member States. First, the threat of

low-wage competition should be especially high in those industries that are characterized

by low existing barriers to entry, either because of low capital intensity or few regulatory

requirements. Both requirements are met by the industries covered in this study, such as

personal services (e.g. hairdressers) or retailing (e.g. florists). We therefore expect lower

barriers to entry in a sector to be positively correlated with a firm’s likelihood to support

a minimum wage introduction. Second, the threat of foreign competition is likely to be

perceived as being stronger in regions close to Germany’s Eastern border, because com-

petition in the sectors considered in this paper takes place at a regional level. To take an

example, a Polish hairdresser is much more likely to offer his services close to the Polish

border than further away from it. This is in line with results reported by Kiessl, Pohl and

Schmalholz [2006], who show that firms in East Germany are considerably more likely to

perceive the Bolkestein Directive as a threat, and not as an opportunity, compared to firms

in West Germany.

As pointed out above, Williamson [1968] postulated a strong link between productiv-

ity and entry deterrence. In this context, firm size is proposed as an indicator for labour

productivity. Whether this is a good proxy depends, however, on the economic sector con-

sidered. Williamson [1968] developed his model with the mining sector or more generally

and the industry union differential, see Section 3.2.4.


manufacturing in mind and equates productivity with firm size and age. By contrast, for

the service sector no explicitly positive correlation exists between productivity and firm

size [Oi and Idson 1999; Brown and Medoff 2003; Bayard and Troske 1999]. In addition,

the models based on the insider-outsider theory do not assume that productivity differ-

ences between incumbents and entrants exist necessarily. Hence, the relationship between

labour productivity and the likelihood to support a minimum wage is not clear-cut from a

theoretical perspective.

The cost structure of a firm is another factor determining an employer’s attitude towards

minimum wages. Ceteris paribus, a firm that already pays relatively high wages will not

be faced with a (strong) increase in labour costs when a minimum wage is introduced.

Firms that are covered by a collective bargaining agreement and that have a low share of

unskilled workers among their employees, usually pay higher wages. Therefore, collective

bargaining coverage and a low share of unskilled workers are expected to be positively

correlated with minimum wage support.

Finally, collusion between trade unions and employers to raise wages is only likely if the

non-union sector is large. In the model by Haucap, Pauly and Wey [2001], trade unions

have the option to maximize members’ wage revenue by discriminating wages between

firms according to productivity. Whether wage discrimination is superior from the union’s

perspective compared to a single entry-preventing minimum wage depends on two fac-

tors: Unions will prefer minimum wages over wage discrimination as long as the differ-

ence between the union and the competitive wage rate is large (i.e. the union mark-up is

high), and if a considerable proportion of workers is not unionized (i.e. the non-coverage

rate is high).

The expected sign of the correlation between the size of the non-union sector and the

support for minimum wages depends on the collective bargaining coverage of the indi-

vidual firm. On the one hand, firms that are covered should be more likely to support a

minimum wage with an increasing size of the non-union sector. While these firms would

not be directly affected by the minimum wage, the likelihood of market exit for their uncov-

ered competitors increases with a higher non-union sector. On the other hand, uncovered


firms’ probability to oppose a minimum wage should increase with a higher non-union

sector, simply because the average increase in the wage bill would be larger.

Our empirical strategy, which is described in detail in the next section, aims at analysing

which factors play a role for employers’ attitudes towards minimum wages. Against

the background of the theoretical considerations above, the following hypotheses can be

stated. We expect high wages at the firm level and being part of a collective bargaining

agreement to be positively correlated with minimum wage support, while a high share of

low-skilled workers is likely to be negatively correlated. The latter may also be interpreted

as a proxy for labour productivity, while we do not expect firm size to be a good indicator

in this context. We nevertheless follow the literature in including firm size as an explana-

tory variable, but expect insignificant correlations. Concerning the size of the non-union

sector, we expect a positive correlation for firms covered by collective agreements, and a

negative relationship for those who are not. Due to the intense public discussion of the

Bolkestein Directive prior to our survey, firms with a high proximity to Eastern Europe

and low barriers to entry should be more inclined to support minimum wages. Finally, for

the existing degree of competition, theory does not deliver clear-cut predictions, making

an empirical analysis all the more relevant.

3.2.3. Data and Empirical Strategy

The data used in this study come from a survey of firms in eight industries in Germany

which were likely candidates for the introduction of a minimum wage at the time of the

survey.14 However, minimum wages were not yet introduced in any of these industries.

They belong to trade or services and include hardware stores (representing wholesale

trade), men’s outfitters and florists (both retailing), motor mechanics (repairs), restau-

14. The data were collected by means of telephone interviews. The response rate amounted to 39 percentand was achieved by contacting individual firms up to eight times. A total of 800 interviews were completedwithin six weeks in February and March 2008. In order to ensure a sufficient number of observations foreach industry/region combination, observations from specific industries as well as from East Germany wereoversampled. See http://fdz.rwi-essen.de/UnternehmensBefragung.html for a description of the data set.


rants (the hotel and restaurant industry), hairdressers (personal services), security firms

(business services), and plumbing (construction). Note that production in these sectors

is relatively labour-intensive and non-tradable, which is a fundamental condition for a

cost-raising strategy to be effective. Among other things, the survey includes questions

concerning the attitude of the managers of the firms towards minimum wages, the skill

level of employees, and the institutional background in terms of coverage and degree of

collective bargaining.

Managers were asked whether they think that a minimum wage of e7.50 is too low,

appropriate or too high. This amount has often been proposed in the public debate on

statutory minimum wages in Germany. In a strict sense, this questions aims at the height

of a minimum wage given that it would be introduced. However, the response should

nevertheless be a highly relevant proxy in terms of minimum wage support. Managers

who answer that the amount is too low are therefore classified as being in favour of a

minimum wage introduction, while the opposite is true for managers who think that the

amount is too high. In order to take into account the categorical nature of the dependent

variable, we estimate an ordered logit model using maximum likelihood. The basic model

specification takes the following form:

P(MW = j|xik) = G(β0 + xikβ)

= G(β0 + β1PMCik + β2Eastik + β3PMCik ∗ Eastik

+β4Sik + β5ULik + β6CBik)

MW represents the attitude of firm i in industry k towards the minimum wage, with

MW = 0 indicating that the firm disapproves an introduction, MW = 1 signifying inter-

mediate support towards an introduction, and MW = 2 showing that the firm supports an

introduction. As a measure for the existing degree of product market competition (PMC)

the normalized Hirschman-Herfindahl Index (HHI) is added to the model. The HHI is

calculated as the sum of squared market shares, which are defined in terms of turnover. It

can take values between 1/n and 1, where n is the number of firms operating in a market.


Increasing values of the HHI imply higher market concentration, and a value of one rep-

resents a perfect monopoly with one firm holding a market share of 100 percent (Martin

2002).

As an alternative measure of competition, which takes into account more dynamic as-

pects, we include the share of new firms in the model. More specifically, a higher share

of new firms, defined as the ratio of new firms to existing firms within the same year, in-

dicates lower barriers to entry. Both indicators, the HHI and the share of new firms are

calculated separately for each included sector at the level of districts (Kreise). This regional

disaggregation is important because competition only takes place at the regional level in

the industries considered in the following analysis.15 All specifications are estimated with

both competition measures, because the motivation for including each of them differs. In

order to investigate whether the increased pressure from the New Member States increases

minimum wage support in regions close to Eastern Europe, we include an interaction term

of the share of new firms with the dummy for East Germany (East).

The basic specification additionally includes a dummy for small firms, Sik, which takes

the value one if a company’s annual turnover does not exceed e250,000. This definition

is data-driven in the sense that the limit was set in such a way that about 40 percent of all

firms are small.16 As discussed in Section 3.2.2, firm size may not be an appropriate pro-

ductivity measure for the service sector. As an alternative we therefore consider the share

of unskilled labour ULik, because the lower human capital endowment of unskilled work-

ers decreases their labour productivity. Furthermore, as pointed out in Section 3.2.2, the

share of unskilled labour also has an impact on the cost structure of the firm. This measure

seems particularly important because the industries we analyse are generally characterized

by high labour intensity.

Finally, the basic specification includes a dummy CBik, which indicates whether the firm

15. The HHI as well as the number of new and existing firms are calculated and provided by the German Fed-eral Statistical Office based on an official statistic on value-added taxes (Umsatzsteuerstatistik). Both indicatorsare calculated for 2008 - the same year in which the survey took place.

16. As a robustness test, we also included a dummy for young firms and the indicator for relative produc-tivity as described in Appendix A instead of the dummy for small firms. However, no significant relationbetween these productivity measures and the likelihood to support a minimum wage could be established.


pays wages according to collectively agreed rates. Because minimum wages in Germany

are set in line with collective bargaining, firms already paying collectively agreed rates

should be more inclined to be in favour of a minimum wage introduction as their labour

costs would not be affected.

In addition to the basic specification, we estimate two extended models. First, a measure

for the size of the non-union sector, the ’industry union differential’ (IUD) is added. The

IUD is the product of the ‘union mark-up’ (relative deviation of union wages to free market

wages in percent) and the ‘non-coverage rate’ (share of workers not covered by a collec-

tive agreement in percent).17 The relative deviation of union wages to free market wages

is a relevant indicator because existing minimum wages in Germany are collectively bar-

gained. Thus, if a minimum wage was to be introduced in one of the analysed industries,

its level would be close to the collectively agreed wage rate. Overall, the IUD provides

a general measure of the potential impact of a minimum wage introduction in a specific

industry.18 In particular, it quantifies the magnitude of the effect of a potential minimum

wage introduction by taking into account both, how many people would be affected and

how significant the average wage increase would be. Note that the IUD has been divided

by 100 to ensure comparability in terms of size with the non-coverage rate and the union

mark-up. As spelt out in Section 3.2.2, the effect of the size of the non-union sector is

likely to differ between firms who are or who are not covered by a collective bargaining

agreement. Therefore, we include an interaction term between the IUD and the dummy

‘collective bargaining’.

Note that the IUD is computed at the industry level and is therefore only characterized

by variation between industries. This calls for clustered standard errors at the same level

[Moulton 1986]. However, with only eight groups the clustered standard errors might be

equally biased [Angrist and Pischke 2008]. We therefore estimate all our empirical mod-

els using both heteroscedasticity robust and clustered standard errors and carefully note

17. As a robustness test, we included both indicators separately instead of the IUD. The results are the samein terms of statistical significance and the sign of the correlation.

18. Appendix A contains a complete description of the construction of the union mark-up and the non-coverage rate.


Table 3.7.: Attitude towards a minimum wage


- 0 + N - 0 + N

Hairdresser 56.00 40.00 4.00 50 15.79 57.89 26.32 57Retailing 16.67 60.00 23.33 30 5.45 72.73 21.82 55Security Firms 66.67 28.57 4.76 21 21.54 64.62 13.85 65Motor Mechanics 17.78 51.11 31.11 45 10.53 36.84 52.63 57Wholesale 18.18 54.55 27.27 33 28.57 44.90 26.53 49Florists 53.49 37.21 9.30 43 19.05 61.90 19.05 63Plumbing 19.51 53.66 26.83 41 7.02 50.88 42.11 57Restaurants 32.00 62.00 6.00 50 13.33 68.33 18.33 60

All industries 33.36 53.09 13.55 313 15.55 58.99 25.46 463

Observations (N) 108 154 51 313 70 267 126 463

Notes: Shown are the response rates (in %) by industry to the question if a minimum wage of7.50 Euro would be too high (“-”), appropriate (“0”) or too low (“+”).Source: Own data collection. For a detailed description of the survey, see Section 3.2.3.

differences.

In the second extended model, we include an alternative measure of the non-union sec-

tor instead of the IUD as a control at the firm level. The average wage differential, i.e. the

difference between the firm wage and the average industry wage in percent, is meant to

capture the bite of a possible minimum wage for a specific firm in an industry. Unfortu-

nately, the response rate for the wage items of the survey is relatively low, which implies

that the number of observations is decreased when this indicator is included.

3.2.4. Results

The first important result from the survey is that the majority of firms state that a minimum

wage of 7.50 Euros would be appropriate (Table 3.7). In plumbing, retailing, and motor

mechanics, firms state more often that a minimum wage of e7.50 is too low. The opposite

is true for hairdressing, security services, florists, and restaurants. On average, the support

for a minimum wage is much stronger in West compared to East Germany, which is in all

likelihood due to the higher wage levels in West Germany.

Competition, as measured by the HHI, is especially intense for restaurants and plumb-

ing, and relatively weak for wholesale and security services (see Table 3.8). While a theo-


Table 3.8.: Summary statistics - Overall and by industryHHI Share of Share of Unskilled Collective IUD

new firms small firms Workers bargaining I II III

Hairdresser 24.5 (4.6) 12.3 (19.3) 72.7 (44.8) 2.5 (8.4) 80.9 (39.5) 8.1 49.6 4.00Retailing 29.6 (13.7) 10.4 (3.6) 40.0 (49.3) 14.0 (27.5) 67.0 (47.3) 6.8 20.3 1.38Security Service 47.2 (16.2) 13.3 (16.0) 38.3 (48.9) 35.2 (42.2) 89.7 (30.6) 0.6 4.1 0.02Motor Mechanics 24.1 (8.5) 9.7 (3.7) 46.1 (50.1) 6.2 (18.0) 65.7 (47.7) 20.1 54.5 10.94Wholesale 40.1 (12.9) 15.2 (17.1) 10.4 (30.7) 12.1 (24.1) 71.6 (45.4) 18.5 18.7 3.45Florists 26.0 (6.1) 13.0 (12.3) 67.8 (47.0) 11.0 (23.9) 63.9 (48.3) 47.3 29.4 13.92Plumbing 23.1 (5.7) 8.4 (11.3) 39.3 (49.1) 4.3 (16.3) 76.8 (42.4) 16.7 −3.0 −0.50Restaurants 19.8 (3.0) 19.9 (23.4) 69.1 (46.5) 37.9 (41.9) 52.3 (50.2) 15.6 41.7 6.50

All industries 27.9 (12.2) 12.8 (15.6) 49.0 (50.0) 14.4 (29.4) 70.5 (45.6) 17.3 28.3 5.26

Notes: Standard deviation in parentheses. All figures are reported in percent, except for the IUD (III) that is defined asthe product of the non-coverage rate (I) and the union mark-up (II), divided by 100). There is no variation at the industrylevel for the IUD and its components.Source: Own data collection, except for HHI and the share of new firms (both German Statistical Office). For a detaileddescription of the survey, see Section 3.2.3.

retical connection exists between low barriers to entry, implying a high share of new firms,

and a high existing degree of competition, measured by a low HHI, no significant corre-

lation exists between the two indicators. In addition, Table 3.9 shows that it is difficult

to establish a relation between the degree of competition and the attitude towards min-

imum wages without controlling for composition effects. In some industries, minimum

wage supporters face a more competitive environment, while the opposite is true in other

sectors.

Table 3.9 suggests that across all industries no significant correlation exists between firm

size and minimum wage support. However, in the majority of industries minimum wage

supporters are smaller than opponents. Retailing and restaurants constitute an exception

insofar as the share of small firms among minimum wage opponents is higher than among

minimum wage supporters. Unskilled labour is clearly more common in some industries

than in others. The share of unskilled labour is negligibly low in traditional craft occupa-

tions, such as hairdressers (3 percent) and plumbing (4 percent), and relatively common

in security services (35 percent) as well as the restaurant industry (38 percent) (Table 3.8).

Especially in industries relying heavily on unskilled labour, minimum wage opponents

are generally characterized by an even higher proportion of unskilled labour compared

to minimum wage supporters (Table 3.9). Throughout all industries, firms paying collec-


tively agreed wages are more likely to support a minimum wage introduction. The average

wage differential confirms this result: Minimum wage opponents generally pay wages be-

low and supporters above the industry’s average (Table 3.9).

Tables 3.10 and 3.11 show the estimation results from the ordered logit models discussed

in Section 3.2.3 including the HHI and the share of new firms, respectively. Both tables re-

port the results in terms of odds ratios. This implies that any value below one signals a

negative correlation of the minimum wage support and the explanatory variable in ques-

tion. The opposite is true for an odds ratio larger than one. The motivation of reporting

odds ratios instead of marginal effects on probabilities is that they are constant over all val-

ues of the respective variable and independent of the covariates. Note that, following the

discussion in Section 3.2.3, heteroskedasticity robust as well as clustered standard errors

are reported because some of the variables only vary at the level of the industry. Finally,

the specifications were also estimated as logit models as a robustness test, where minimum

wage support was defined as stating that a minimum wage of e7.50 would be appropri-

ate or too low, and as linear probability models using OLS. The results are very similar

compared to those reported in Tables 3.10 and 3.11.

Table 3.10 shows that the HHI is of low economic and statistical significance in West

Germany. The odds ratio of 0.995 implies that a one unit increase in the HHI is associated

with a decrease in the odds to support the minimum wage of 0.5 percent. In contrast, this

correlation is considerably larger in East Germany, although the sign changes: A one unit

increase in the HHI leads to an increase in the odds to support the minimum wage of 2.1

(= 0.995 ∗ 1.026) percent19. Given a standard deviation of the HHI of 8 percentage points

in East Germany20, this effect is non-negligible.

In order to illustrate the estimation results for the HHI, Figure 3.2 shows the effect of

the HHI on the likelihood to support a minimum wage in terms of predicted probabilities,

separately for each outcome as well as East and West Germany. The predicted probability

19. It should be noted that this estimate constitutes an upper bound as the relationship is only statisticallysignificant when employing clustered standard errors.

20. The standard deviation in West Germany amounts to 13.7 percentage points. As Table 3.8 shows, theoverall standard deviation of the HHI equals 12.2 percentage points.


Table 3.9.: Summary statistics by minimum wage support and industry

HHI Share of Share ofnew firms small firms

- 0 + - 0 + - 0 +

Hairdresser 23.1 24.8 25.5 22.2 8.3 7.9 58.1 80.5 85.7Retailing 27.5 32.3 23.1 9.3 10.4 10.7 50.0 39.2 35.7Security Firms 43.8 48.1 50.8 18.9 10.3 25.0 38.5 37.5 42.9Motor Mechanics 22.3 24.1 23.9 9.2 10.1 9.6 30.8 50.0 47.2Wholesale 45.7 37.0 38.0 17.5 12.9 15.2 0.0 16.2 11.1Florists 24.2 26.7 27.3 12.2 12.6 17.4 70.4 63.8 81.8Plumbing 20.5 23.1 24.3 6.7 10.2 6.6 41.7 33.3 44.4Restaurants 20.2 19.7 19.4 19.5 20.4 19.5 70.0 68.3 66.7

All industries 27.9 28.2 26.6 15.9 12.6 11.0 47.4 49.6 48.9

Unskilled Collective Average wageworkers bargaining differential

- 0 + - 0 + - 0 +

Hairdresser 1.7 2.6 4.9 72.2 83.0 93.8 -13.6 9.2 12.6Retailing 37.3 14.1 3.3 37.5 73.7 57.9 -18.8 2.2 0.4Security Firms 39.2 38.1 10.3 85.7 89.6 100.0 -7.8 0.7 15.4Motor Mechanics 9.2 6.8 4.7 35.7 67.4 72.1 -13.5 1.3 4.7Wholesale 24.9 8.5 7.6 52.6 73.7 85.0 -3.7 0.6 3.3Florists 15.3 7.7 13.3 51.4 66.7 81.3 -8.2 1.0 12.4Plumbing 0.0 5.5 4.2 66.7 74.5 82.4 6.8 -5.9 7.8Restaurants 31.1 40.2 46.4 33.3 54.2 76.9 -4.6 -2.0 22.0

All industries 18.5 15.7 8.4 58.0 71.9 78.9 -8.2 0.8 7.7

Notes: - “Too high”; 0 “Appropriate”; + “Too low”. The table reports mean characteristics of mini-mum wage supporters and opponents. All figures are reported in percent.Source: Own data collection, except for HHI and the share of new firms (both German StatisticalOffice). For a detailed description of the survey, see Section 3.2.3.


Table 3.10.: Estimation results for minimum wage support - HHIModel 1 Model 2 Model 3 Model 4

Herfindahl Index (HHI) 0.995 (0.011) 0.995 (0.011) 0.995 (0.011) 0.990 (0.011)[0.006] [0.006] [0.006] [0.012]

East Germany 0.257 (0.180)∗ 0.257 (0.180)∗ 0.268 (0.189)∗ 0.443 (0.329)[0.114]∗∗∗ [0.114]∗∗∗ [0.118]∗∗∗ [0.300]

HHI*East Germany 1.026 (0.024) 1.026 (0.024) 1.025 (0.024) 1.011 (0.025)[0.010]∗∗ [0.010]∗∗ [0.011]∗∗ [0.016]

Small 1.086 (0.263) 1.068 (0.263) 1.089 (0.264) 1.033 (0.266)[0.300] [0.300] [0.298] [0.273]

Unskilled labour 0.994 (0.004) 0.994 (0.004) 0.994 (0.005) 0.994 (0.006)[0.004] [0.004] [0.004] [0.00]∗

Collective bargaining 1.979 (0.692)∗ 1.979 (0.692)∗ 1.458 (0.680) 2.423 (0.877)∗∗

[0.405]∗∗∗ [0.004]∗∗∗ [0.631] [0.940]∗∗∗

Industry union differential (IUD) 0.906 (0.051)∗ 0.863 (0.073)∗

[0.018]∗∗∗ [0.053]∗∗

IUD*Collective bargaining 1.064 (0.083)[0.061]

Average wage differential 1.004 (0.005)[0.003]

Industry Dummies yes yes yes yes

F 21.315 21.315 56.635 39.098Observations 515 515 515 441

Legend: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Odds ratios from the ordered logit model. Robust standard errors in parentheses. Clustered standard errors at the industrylevel in brackets.Source: Own data collection, except for HHI (German Statistical Office). For a detailed description of the survey, see Section 3.2.3.

Figure 3.2.: Effect of the HHI on the predicted probabilities

(a) Minimum wage is too high

-.2

0.2

.4.6

Pre

dict

ed p

roba

bilit

y

15 20 25 30 35 40 45 50 55 60 65 70 75 80HHI

West Germany East Germany

(b) Minimum wage is too low

-.2

0.2

.4.6

.8P

redi

cted

pro

babi

lity

15 20 25 30 35 40 45 50 55 60 65 70 75 80HHI


(c) Minimum wage is appropriate

.4.5

.6.7

.8P

redi

cted

pro

babi

lity

15 20 25 30 35 40 45 50 55 60 65 70 75 80HHI


Notes: The predicted probability is calculated separately for each observation in the sample and then averaged over allobservations. HHI expressed in percent.Source: Own data collection, except for the HHI (German Statistical Office). For a detailed description of the survey,see Section 3.2.3.


hardly changes for West Germany over the different values of the HHI, which reflects the

low economic and statistical significance in West Germany. In East Germany, the predicted

probability to state that a minimum wage of e7.50 is too high decreases on average from

almost 40 percent to about 15 percent as the HHI increases from its minimum of 15 percent

to its maximum of 80 percent. The opposite is true for the probability to propose that such

a minimum wage rate would be too low, which increases from 10 to almost 30 percent as

the HHI increases.

The results imply that weaker competition on the product market is associated with

higher support of the minimum wage in East Germany. This can be interpreted as evidence

that a low degree of existing competition indeed facilitates the collusion of firms by using

a minimum wage as a cost raising strategy, at least in East Germany.

According to Table 3.11, the share of new firms is not correlated with the likelihood to

support minimum wages in neither West nor East Germany. The odds ratio is only slightly

above one and statistically indifferent from one. The only exception is Model (4), which

controls for the average wage differential, i.e. the wage level of the individual firm in com-

parison to all other firms in the same industry. While the statistical significance remains

low, the size of the odds ratio is non-negligible. The results suggest that a one percentage

point increase in the share of new firms is associated with a 0.6 percent decrease in the

odds to support a minimum wage in West Germany. In comparison, the odds increase

by 0.9 (= 0.994 ∗ 1.016) percent with each percentage point increase in the share of new

firms in East Germany. The standard deviation of the share of new firms in East Germany

amounts to 26 percentage points, implying that this effect is large.

As discussed in Section 3.2.2, one hypothesis is that many firms in the sample fear an in-

crease in competition because of the eastward enlargement of the European Union. In this

context, a minimum wage could constitute a very effective barrier to entry. However, min-

imum wages are a relatively blunt instrument to reduce competition, and, given that they

potentially increase wages, may have major other consequences at the firm level. In this

context, it is interesting that we only find a noteworthy - though not statistically significant

- correlation of the share of new firms and minimum wage support when controlling for


the average wage differential at the firm level (Model 4). Stated differently, using minimum

wages as a cost raising strategy to prevent market entry of foreign firms only appears to

be a valid approach for those firms that already pay high wages relative to other domestic

firms. This mechanism is more important in East Germany, because the threat of foreign

competition is more credible in comparison to West Germany.

Firms paying wages according to or in excess of the collectively bargained rate are in

general about twice as likely to support the minimum wage (Table 3.10). The economic

and statistical significance is lower in the specification including the share of new firms

instead of the HHI. In contrast, the average wage differential is only positively significant

in specifications including the share of new firms (Table 3.11). The positive link between the

average wage differential as well as the dummy for collective bargaining and the tendency

to support a minimum wage is likely to be due to the same effect: Firms not covered

by collective bargaining agreements are likely to pay lower wages and therefore oppose

minimum wages because this would lead to an increase in their wage costs. By contrast,

firms which are covered by such agreements are unlikely to face increased wage costs and

therefore support a minimum wage introduction.

The relationship between the size of the non-union sector, measured by the industry

union differential (IUD), and the probability of individual firms to support minimum wages

also depends on whether the firm is covered by a collective agreement. For uncovered

firms the odds to support the minimum wage decrease by 14 percent with each unit in-

crease of the IUD (Table 3.10). In comparison to uncovered firms, the odds to support

a minimum wage increase by 6 percent with each unit increase in the IUD for covered

firms, although this correlation is not statistically significant. In contrast, the relationship

is stronger in terms of economic and statistical significance in the models including the

share of new firms instead of the HHI (Table 3.11).

The deviating reaction of covered and uncovered firms to the IUD does, however, not

imply that a higher IUD increases the probability of covered firms to support a minimum

wage. Instead, as Figure 3.3 shows, the relation between the IUD and minimum wage

support is only less negative for covered compared to uncovered firms. Thus, a higher IUD


Table 3.11.: Estimation results for minimum wage support - Share of new firmsModel 1 Model 2 Model 3 Model 4

Share of new firms (SNF) 1.006 (0.035) 1.006 (0.035) 1.008 (0.036) 0.994 (0.031)[0.024] [0.024] [0.025] [0.017]

East Germany 0.374 (0.184)∗∗ 0.374 (0.184)∗∗ 0.408 (0.205)∗ 0.386 (0.188)∗

[0.160]∗∗ [0.160]∗∗ [0.175]∗ [0.137]∗∗

SNF*East Germany 1.001 (0.036) 1.001 (0.036) 0.999 (0.036) 1.016 (0.034)[0.024] [0.024] [0.025] [0.018]

Small 1.171 (0.316) 1.171 (0.316) 1.164 (0.315) 1.270 (0.378)[0.427] [0.427] [0.414] [0.501]

Unskilled labour 0.996 (0.005) 0.996 (0.005) 0.996 (0.005) 0.996 (0.007)[0.003] [0.003] [0.003] [0.002]

Collective bargaining 1.470 (0.562) 1.470 (0.562) 0.755 (0.370) 1.961 (0.796)∗

[0.327] [0.327] [0.164] [0.923]Industry union differential (IUD) 0.876 (0.056)∗∗ 0.790 (0.073)∗∗

[0.024]∗∗∗ [0.045]∗∗∗

IUD*Collective bargaining 1.138 (0.091)[0.046]∗∗∗

Average wage differential 1.004 (0.005)[0.002]∗

Industry Dummies yes yes yes yes

F 28.561 28.561 9.324 13.068Observations 416 416 416 349

Legend: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Odds ratios from the ordered logit model. Robust standard errors in parentheses. Clustered standard errors at the industrylevel in brackets.Source: Own data collection, except for the share of new firms (German Statistical Office). For a detailed description of the survey,see Section 3.2.3.

Figure 3.3.: Effect of the IUD on the predicted probabilities

(a) Minimum wage is too high

-.5

0.5

11.

5P

redi

cted

pro

babi

lity

0 1 2 3 4 5 6 7 8 9 10 11 12 13IUD

Not covered by a collective agreement Covered by a collective agreement

(b) Minimum wage is too low

0.2

.4.6

Pre

dict

ed p

roba

bilit

y

0 1 2 3 4 5 6 7 8 9 10 11 12 13IUD


(c) Minimum wage is appropriate

-.5

0.5

1P

redi

cted

pro

babi

lity

0 1 2 3 4 5 6 7 8 9 10 11 12 13IUD


Notes: The predicted probability is calculated separately for each observation in the sample and then averaged over allobservations.Source: Own data collection, except for the IUD (German Statistical Office). For a detailed description of the survey,see Section 3.2.3.


decreases the odds to support a minimum wage for all firms, but uncovered firms react

more strongly. At first sight, this result seems to contradict the theoretical model proposed

by Haucap, Pauly and Wey [2001], where a larger non-union sector facilitates the collusion

on minimum wages to unions and employers. However, this line of argument mostly

applies to unions while other factors, such as productivity differences and the existing

degree of competition, may be more important to employers who are at the centre of this

study.

Finally, firm size and the share of unskilled workers in a firm’s labour force are analysed.

The odds ratio of the dummy variable for small firms is larger than one, but statistically

insignificant in all specifications. In addition, the magnitude of the relation is very small

with an odds ratio of 1.089, which implies that the odds to support minimum wages are 8.9

percent higher for small compared to large firms (Table 3.10). This result is in contrast to

the theoretical analysis by Williamson [1968] predicting the opposite for the manufacturing

sector. However, our results may be interpreted as an indication that firm size is not a good

measure for labour productivity in the selected industries, which belong to the service

sector as opposed to manufacturing.

The share of unskilled workers in a firm’s labour force, by contrast, is negatively corre-

lated with being in favour of a minimum wage. While the statistical significance is low in

the majority of specifications, the economic relevance is rather high with a one percentage

point increase in the share of unskilled workers leading to a 0.5 percent decrease in the

odds to support a minimum wage (Table 3.10 or Table 3.11). Given a standard deviation

of 36 percentage points in the share of unskilled workers, this effect is large. On the one

hand, this could be explained by the fact that a higher share of unskilled labour implies

lower labour productivity. This interpretation is consistent with higher-productivity firms

being in favour of minimum wages, as the latter can improve their competitive position.

On the other hand, the “bite” of the minimum wage would be relatively high in a firm with

many unskilled workers. Therefore, the introduction of a minimum wage would have a

strong impact for these firms, either through an increased wage bill or through the need to

reorganise the workforce, which would in all likelihood imply laying off workers and thus


a reduction in output and profits.

3.2.5. Conclusion

This paper analyses the determinants of employers’ attitude towards minimum wages us-

ing a unique data set on 800 firms in eight service sector industries in Germany. One

important finding of this survey is that a majority of firms in these industries supports the

introduction of minimum wages. The main question of the paper is whether this finding

can be explained by employers trying to use minimum wages in order to raise barriers

to entry and reduce the competitive pressure in their industry, both from domestic com-

petitors and from abroad. We find evidence that weaker competition, as measured by the

Hirschman-Herfindahl-Index, is associated with stronger support for minimum wages by

firms in East Germany.

Further, we find weak evidence that the share of new firms in an industry and region is

related to the support of minimum wages in East Germany. This can be explained by the

fact that a high share of new firms indicates low existing barriers to entry, which makes

incumbent firms vulnerable to new competitors. Thus, minimum wages are only an at-

tractive instrument if other barriers to entry in the industry, e.g. entry regulations such as

the requirement of mandatory educational standards, are low. Given the proximity of East

Germany to the low-wage countries of Eastern Europe, the (perceived) threat of competi-

tion from abroad seems to be the main driving factor here.

Our analysis also investigates the role of institutional features for the support of mini-

mum wages. The estimation results clearly show that firms paying according to collective

agreements are in favour of a minimum wage. This observation supports the theoretical

model by Haucap, Pauly and Wey [2001] and implies that these employers try to use the

minimum wage to increase the labour costs of their rivals, thereby increasing their costs.

Our results have several important implications. First, the German case shows that

firms’ attitude towards minimum wages seems to be influenced by the potential effects

of minimum wages on competition. While this is naturally not the only possible expla-


nation for employers’ support of minimum wages, it is one that should be considered by

policy makers when designing minimum wage institutions. Second, this result implies that

minimum wages may have an effect on both labour market outcomes and product markets

(especially prices) through the impact of minimum wages on the level of competition in an

industry. From an economic policy point of view, this makes minimum wages a subject for

anti-trust authorities. Finally, the case of minimum wages in Germany highlights the role

social policies may play as protectionist policy instruments in the wealthier EU Member

States, especially with regard to the low-wage countries of Eastern Europe.

Therefore, further research clearly seems warranted along at least two lines. On the one

hand, it is important to identify the social policies which are most vulnerable to lobby

groups, as these policies are most likely to be abused as protectionist policy instruments.

On the other hand, more effort should be made in the empirical analysis of attitudes to-

wards social policies and lobby efforts in this context (e.g. Afonso 2011). Up to now, this

type of research is unfortunately hampered by a lack of adequate data.

4. Conclusion

This dissertation has provided empirical evidence on minimum wages in Germany by (i)

analysing the employment effects of minimum wages in the construction industry (Chap-

ter 2), by (ii) studying the competitive situation on the labour market as a possible expla-

nation for non-negative employment effects, and by (iii) focusing on the potential role of

minimum wages for product market competition as a largely neglected, but highly relevant

outcome of minimum wage laws (Chapter 3).

One can view Chapter 2 as a reflection of the minimum wage literature on employment

effects: We find both, negative and neutral employment effects. One reason behind these

deviating results is the source of variation used to identify the employment effect. In the

main construction industry, we compare highly affected to moderately affected regions and

conclude that an increase of one standard deviation in the minimum wage bite decreases

employment growth by 2.6 to 3.1 percentage points (Section 2.2). For painters and electri-

cians, the minimum wage effect is identified based on a comparison of two occupational

groups with minimum wages to different industries that are not exposed to such a regu-

lation. The results do not suggest any impact on employment growth (Section 2.1). The

observation that employment growth on average is not influenced by the minimum wage

does not, however, preclude that employment growth is reduced for highly affected firms

or regions.

In addition, the macroeconomic environment was quite different when the minimum

wage was introduced in the main construction industry and for painters, respectively. The

entire construction industry experienced a recession in East Germany during the late 90’s.

4. CONCLUSION 129

The introduction of the wage floor coincided with this recession in the main construction

industry, while the minimum wage for painters was only introduced in 2003. Since a reces-

sionary environment increases the possibility of negative employment effects of minimum

wages [Addison, Blackburn and Cotti 2013], the timing of the minimum wage introduc-

tion offers an additional explanation for the diverging results. Further, non-compliance

may have been higher for painters and electricians compared to main construction. Un-

fortunately, no possibility exists to investigate this formally, but one argument in favour

of this hypothesis is that opposed to painters and electricians a few extremely large firms

exist in the main construction industry that employ a significant share of workers. These

firms may be subject to more frequent controls.

Additional adjustment channels of minimum wages include a reduction in working

hours, fewer non-wage benefits, less employer-provided training, substitution of high

skilled for low skilled workers, wage compression from above, reduction in profits or

higher prices [Schmitt 2013; Hirsch, Kaufman and Zelenska 2011]. The composition and

strength of each of these channels may be different in each minimum wage industry. The

same is true for the degree of monopsonistic competition in the labour market. However,

as Section 3.1 shows, monopsony is an unlikely explanation for possible non-negative em-

ployment effects in the minimum wage industries. While the labour supply elasticities

are generally low in the majority of industries, and monopsony power of employers is

consequently comparably high, the minimum wage industries are not characterized by

especially low labour supply elasticities. The only exception is commercial cleaning.

The introduction of a statutory minimum wage in 2015 is a vast political experiment,

affecting millions of workers in Germany. The consequences of this introduction cannot be

determined ex ante, but past experience can offer some guidelines about the likely effects.

In conclusion, I therefore offer an assessment of the planned statutory minimum wages

in terms of (i) its level, (ii) regional and sectoral differentiation and (iii) the institutional

framework governing increases in the wage floor.

First, the existing evidence, presented in this dissertation and available in the interna-

4. CONCLUSION 130

tional minimum wage literature, clearly shows that neutral employment effects can only

be hoped for in the presence of a moderate wage floor. Thus, with respect to the introduc-

tion of a statutory minimum wage in Germany, the question arises whether the proposed

level of e8.50 is appropriate or too high. Kluve [2013] compares the relation of the mini-

mum to the median wage (Kaitz Index) of full-time employees in Germany to other OECD

countries in 2011. The estimates range from 50.4 to 52.4 percent, which is considerably

higher than the Kaitz Index in the UK at 46.7 percent or the US at 38.3 percent.

Another measure of the minimum wage treatment intensity is the bite, defined as the

share of workers currently earning below the planned minimum wage. Brenke [2014]

shows that 15 percent of all employees would be affected by a minimum wage of e8.50.

However, the bite amounts to 23 percent in East Germany, and is therefore considerably

higher than in West Germany, where it is 14 percent. Full-time employees are much less af-

fected (8 percent) than workers in marginal employment (58 percent) [Brenke 2014]. Some

industries will be influenced more strongly than others. For East Germany, Brautzsch and

Schulz [2013] calculate a bite of 58 percent in agriculture, 52 percent in retailing and 67 per-

cent in the hotel and restaurant industry. Given that the bite of the UK’s minimum wage

amounted to 5.2 percent when it was introduced in 1999 [Falck et al. 2013] and that the bite

in the US equalled 2.7 percent in 2009 [Schmitt 2013], the treatment intensity of the German

statutory minimum wage appears dangerously high.

This is especially true in the context of the results presented in Section 2.2 which show

that high-impact minimum wages are unambiguously harmful to employment. At the

regional level, the highest observed bite for the construction industry in East Germany

equalled 40 percent in 1996. This figure is considerably lower than the estimated minimum

wage bites for entire industries in East Germany. Combined with the observation that the

degree of monopsonistic competition is not necessarily highest in the industries with the

lowest average wage (Section 3.1), employment losses appear inevitable.

Second, the results from Chapter 3 show that a uniform wage floor is preferable to dif-

ferentiation across industries because the risk that the minimum wage is abused as a cost-

4. CONCLUSION 131

raising strategy to deter entry is reduced (cf. Section 3.2). However, a less known facet

of Andrea Nahles’ draft of the minimum wage law consists of extending the legal frame-

work by which sectoral minimum wages have been introduced so far (“Posting of Workers

Law”) to all industries [BMAS 2014]. This would enable trade unions and employer as-

sociations that only represent a minority of workers and firms to set wages for the entire

industry. Further, the administration and enforcement of such a complex system of sectoral

minimum wage rates is bound to fail [Konig et al. 2012]. In summary, the introduction of

industry-specific minimum wage rates on top of a unitary minimum wage appears dan-

gerous from an economic point of view.

Third, an independent commission should uprate the minimum wage in order to pre-

vent that the wage floor is misused for election campaign purposes. The draft of the min-

imum wage law does propose such a commission, but the institutional design is faulty.

The commission only advices the government; the final decision is still left to politicians.

Insofar as the commission’s advice is adhered to, as has always been the case in the UK,

this problem might be negligible. However, the danger exists that the commission’s rec-

ommendations are not taken for granted. Equally important, unlike Britain’s Low Pay

Commission, there is little room for a systematic evaluation on the impacts of the wage

floor [Lesch, Mayer and Schmid 2014; Arni et al. 2014]. The first evaluation is planned in

2018, three years after the introduction of the wage floor (and in the next legislative period).

Finally, the commission consists of three representatives of trade unions and employer as-

sociations, respectively. Both parties appoint one researcher to offer advice, but without

voting power. As the British experience shows, a commission consisting of social partners

and researcher, all with equal shares and voting rights, appears to be more successful in

setting a moderate minimum wage.

In conclusion, the risk of worker displacement due to the proposed uniform minimum

wage could be reduced along three dimensions. First, the level is clearly too high. In this

context, a differentiation between East and West Germany is advisable. If separate mini-

mum wage rates are not acceptable for political reasons, the minimum wage rate should

4. CONCLUSION 132

be set at a level that leads to a moderate impact in East Germany. A non-binding minimum

wage in West Germany has to be accepted in this case. Second, the Posting of Workers Law

should by no means be extended to industries with a low coverage rate of collective bar-

gaining. Employer collusion reducing output and employment, while increasing product

prices is a likely outcome in this case. And third, the minimum wage commission has to

be completely separated from politics and should base its decisions on regular, scientific

evaluations of the minimum wage.

Bibliography

Abowd, John M., Francis Kramarz, Thomas Lemieux and David N. Margolis. 2000. Min-imum Wages and Youth Employment in France and the United States. In Youth Employ-ment and Joblessness in Advanced Countries, ed. David G. Blanchflower and Richard B.Freeman. University of Chicago Press pp. 427–472.

Addison, John T, McKinley L Blackburn and Chad D Cotti. 2013. “Minimum wage in-creases in a recessionary environment.” Labour Economics 23:30–39.

Addison, J.T., M.K.L. Blackburn and C.D. Cotti. 2009. “Do minimum wages raise employ-ment? Evidence from the US retail-trade sector.” Labour Economics 16(4):397–408.

Afonso, Alexandre. 2011. “Employer strategies, cross-class coalitions and the free move-ment of labour in the enlarged European Union.” Socio-Economic Review 10:705–730.

Alda, Holger, Stefan Bender and Hermann Gartner. 2005. “The linked employer-employeedataset created from the IAB establishment panel and the process-produced data of theIAB (LIAB).” Schmoller Jahrbuch 125:327–336.

Allegretto, Sylvia A., Arindrajit Dube and Michael Reich. 2011. “Do Minimum WagesReally Reduce Teen Employment? Accounting for Heterogeneity and Selectivity in StatePanel Data.” Industrial Relations 50(2):205–240.

Alston, Richard M., J.R. Kearl and Michael B. Vaughan. 1992. “Is There a Consensus AmongEconomists in the 1990’s?” The American Economic Review 82(2):203–209.

Angrist, J.D. and J.S. Pischke. 2008. Mostly harmless econometrics: An empiricist’s companion.Princeton University Press.

Antonczyk, Dirk, Bernd Fitzenberger and Katrin Sommerfeld. 2010. “Rising wage inequal-ity, the decline of collective bargaining, and the gender wage gap.” Labour Economics17(5):835–847.

Apel, Helmut, Ronald Bachmann, Stefan Bender, Philipp vom Berge, Michael Fertig,Hanna Frings, Marion Konig, Joachim Moller, Alfredo Paloyo, Sandra Schaffner, Mar-cus Tamm, Matthias Umkehrer and Stefanie Wolter. 2012. “Arbeitsmarktwirkungen derMindestlohneinfuhrung im Bauhauptgewerbe (Labour market effects of the minimumwage introduction in the main construction sector).” Journal for Labour Market Research45(3/4):257–277.

Aretz, Bodo, Melanie Arntz and Terry Gregory. 2013. “The Minimum Wage Affects ThemAll: Evidence on Employment Spillovers in the Roofing Sector.” German Economic Review14(3):282–315.

BIBLIOGRAPHY 134

Arni, Patrick, Werner Eichhorst, Nico Pestel, Alexander Spermann and Klaus F. Zimmer-mann. 2014. “Kein Mindestlohn ohne unabhangige wirtschaftliche Evaluation.” IZAStandpunkte (65).

Bachmann, Ronald. 2005. Labour Market Dynamics in Germany: Hirings, Separations, andJob-to-Job Transitions over the Business Cycle. SFB 649 Discussion Papers No. 045.

Bachmann, Ronald, Marion Konig and Sandra Schaffner. 2012. “Lost in Transition? – Min-imum Wage Effects on German Construction Workers.” Ruhr Economic Paper No. 358.

Bachmann, Ronald, Thomas K. Bauer and Hanna Frings. 2014. “Minimum Wages as aBarrier to Entry – Evidence from Germany.” LABOUR - Review of Labour Economics andIndustrial Relations 28(3):338–357.

Bauer, Thomas K., Jochen Kluve, Sandra Schaffner and Christoph M. Schmidt. 2009. “Fis-cal Effects of Minimum Wages: An Analysis for Germany.” German Economic Review10(2):224–242.

Bayard, Kimberly and Kenneth R. Troske. 1999. “Examining the Employer-Size WagePremium in the Manufacturing, Retail Trade, and Service Industries Using Employer-Employee Matched Data.” The American Economic Review 89(2):99–103.

BBSR. 2012. “Raumabgrenzungen: Referenzdateien und Karten.” http://goo.gl/CxgkN.Accessed 14 February 2013.

Bhaskar, Venkataraman, Alan Manning and Ted To. 2002. “Oligopsony and monopsonisticcompetition in labor markets.” The Journal of Economic Perspectives 16(2):155–174.

Biebeler, H. and H. Lesch. 2006. “Mitgliederstruktur der Gewerkschaften in Deutschland.”IW Trends 33(4/2006):1–16.

Bilous, Alexandre. 2010. “Die Entsendung von Arbeitnehmern und die Durchfuhrung derRichtlinie.” http://www.eurofound.europa.eu/eiro/1999/09/study/tn9909249s.

htm. Accessed 16 December 2013.

BMAS. 2014. “Entwurf eines Gesetzes zur Strkung der Tarifautonomie.” http://www.

bmas.de/DE/Service/Presse/Pressemitteilungen/mindestlohn-kommt.html. Ac-cessed 04 April 2014.

Boal, William M and Michael R Ransom. 1997. “Monopsony in the labor market.” Journalof Economic Literature 35(1):86–112.

Boockmann, Bernhard. 2010. “The combined employment effects of minimumwages and labor market regulation – A meta-analysis.” Applied Economics Quarterly61(Supplement):167–188.

Boockmann, Bernhard, Raimund Krumm, Michael Neumann and Pia Rattenhuber. 2013.“Turning the Switch: An Evaluation of the Minimum Wage in the German ElectricalTrade Using Repeated Natural Experiments.” German Economic Review 14(3):316–348.

Booth, Alison L and Pamela Katic. 2011. “Estimating the Wage Elasticity of Labour Supplyto a Firm: What Evidence is there for Monopsony?” Economic Record 87(278):359–369.

BIBLIOGRAPHY 135

Braun, Sebastian. 2011. “Unionisation structures, productivity and firm performance: Newinsights from a heterogeneous firm model.” Labour Economics 18(1):120–129.

Brautzsch, Hans-Ulrich and Birgit Schulz. 2013. “Mindestlohn von 8,50 Euro: Wie vieleverdienen weniger, und in welchen Branchen arbeiten sie?” IWH, Wirtschaft im Wandel19(3):53–56.

Brenke, Karl. 2014. “Mindestlohn: Zahl der anspruchberechtigten Arbeitnehmer wird weitunter fnf Millionen liegen.” DIW Wochenbereicht (5):71–77.

Brenke, Karl and Johannes Ziemendorff. 2008. “Hilfebedurftig trotz Arbeit?: KeinMassenphanomen in Deutschland.” DIW Wochenbericht 75(4):33–40.

Brenke, Karl and Kai-Uwe Muller. 2013. “Gesetzlicher Mindestlohn: kein verteilungspoli-tisches Allheilmittel.” DIW-Wochenbericht 80(39):3–17.

Brown, Charles and James L. Medoff. 2003. “Firm Age and Wages.” Journal of Labor Eco-nomics 21(3):677–698.

Brown, William. 2009. “The Process of Fixing the British National Minimum Wage, 1997–2007.” British Journal of Industrial Relations 47(2):429–443.

Bundesagentur fur Arbeit. 2011. Beschaftigungsstatistik, SozialversicherungspflichtigeBruttoarbeitsentgelte (Entgeltstatistik). Technical report Bundesagentur fur Arbeit.

Burdett, Kenneth and Dale T Mortensen. 1998. “Wage differentials, employer size, andunemployment.” International Economic Review 32(2):257–273.

Bureau of Labor Statistics. 2011. “May 2011 National Occupational Employment and WageEstimates, United States.” http://goo.gl/imNv0. Accessed 08 March 2013.

Buttner, Thiess and Alexander Ebertz. 2009. “Spatial Implications of Minimum Wages.”Jahrbucher fur Nationalokonomie und Statistik 229(2):292–312.

Cameron, A Colin and Pravin K Trivedi. 2005. Microeconometrics: methods and applications.Cambridge university press.

Card, David. 1992. “Using Regional Variation in Wages to Measure the Effects of the Fed-eral Minimum Wage.” Industrial and Labor Relations Review 46(1):22–37.

Card, David and Alan B Krueger. 1994. “Minimum Wages and Employment: A Case Studyof the Fast-Food Industry in New Jersey and Pennsylvania.” American Economic Review84(4):772–93.

Card, David and Alan B Krueger. 1995. Myth and Measurement: The New Economics of theMinimum Wage. Princeton: Princeton University Press.

Card, David and Alan B. Krueger. 2000. “Minimum Wages and Employment: A CaseStudy of the Fast-Food Industry in New Jersey and Pennsylvania: Reply.” The AmericanEconomic Review 90(5):1397–1420.

BIBLIOGRAPHY 136

Chappell, W.F., M.S. Kimenyi and W.J. Mayer. 1992. “The Impact of Unionization on theEntry of Firms: Evidence from US Industries.” Journal of Labor Research 13(3):273–283.

Deere, Donald, Kevin M Murphy and Finis Welch. 1995. “Employment and the 1990-1991Minimum-Wage Hike.” The American Economic Review 85(2):232–237.

Deutsche Bundesregierung. 2010. Durchsetzung von Mindestlohnen. Drucksache 17/2282Deutscher Bundestag 17. Wahlperiode.

Dickens, Richard, Paul Gregg, Stephen Machin, Alan Manning and Jonathan Wadsworth.1993. “Wage Councils: Was There a Case for Abolition?” British Journal of IndustrialRelations 31(4):515–529.

Dolado, Juan J., Florentino Felgueroso and Juan F. Jimeno. 1997. “The effects of minimumbargained wages on earnings: Evidence from Spain.” European Economic Review 41(3–5):713–721.

Dolton, Peter and Chiara Rosazza Bondibene. 2012. “The international experience of min-imum wages in an economic downturn.” Economic Policy 27(69):99–142.

Dolton, Peter, Chiara Rosazza Bondibene and Jonathan Wadsworth. 2010. “The UK Na-tional Minimum Wage in Retrospect.” Fiscal Studies 31(4):509–534.

Dolton, Peter, Chiara Rosazza Bondibene and Jonathan Wadsworth. 2012. “Employment,Inequality and the UK National Minimum Wage over the Medium-Term.” Oxford Bulletinof Economics and Statistics 74(1):78–106.

Dorner, Matthias, Jorg Heining, Peter Jacobebbinghaus and Stefan Seth. 2010. “Sampleof Integrated Labour Market Biographies (SIAB) 1975-2008.” FDZ data report, 01/2010(en), Nuremberg.

Dube, Arindrajit, T. William Lester and Michael Reich. 2010. “Minimum wage effects acrossstate borders: Estimates using contiguous counties.” The Review of Economics and Statistics92(4):945–964.

Dube, Arindrajit, T William Lester and Michael Reich. 2013. “Do frictions matter in thelabor market? Accessions, separations and minimum wage effects.” IZA Discussion Paper5811:1–39.

Dustmann, Christian, Johannes Ludsteck and Uta Schonberg. 2009. “Revisiting the Ger-man Wage Structure.” The Quarterly Journal of Economics 124(2):843–881.

Ellguth, P. and S. Kohaut. 2005. “Tarifbindung und betriebliche Interessenvertretung: Ak-tuelle Ergebnisse aus dem IAB-Betriebspanel.” WSI Mitteilungen 58(7):398–403.

Ellguth, Peter, Susanne Kohaut and Iris Moller. 2014. “The IAB Establishment Panel -methodological essentials and data quality.” Journal for Labour Market Research 47:27–41.

ELVIRA. 2013. “Statistical Database of the Central Federation of the German ConstructionIndustry.” http://www.bauindustrie.de/zahlen-fakten/datenbank-elvira/. Ac-cessed 14 February 2013.

BIBLIOGRAPHY 137

Eurofound. 2010. “Posted Workers in the European Union.” http://www.eurofound.

europa.eu/docs/eiro/tn0908038s/tn0908038s.pdf. Accessed 16 December 2013.

Falch, Torberg. 2010. “The elasticity of labor supply at the establishment level.” Journal ofLabor Economics 28(2):237–266.

Falck, Oliver, Andreas Knabe, Andreas Mazat and Simon Wiederhold. 2013. “Mindestlohnin Deutschland: Wie viele sind betroffen?” ifo Schnelldienst 24:68–73.

Franz, Wolfgang and Viktor Steiner. 2000. “Wages in the East German transition process:Facts and explanations.” German Economic Review 1(3):241–269.

Frings, Hanna. 2013. “The Employment Effect of Industry-Specific, Collectively BargainedMinimum Wages.” German Economic Review 14(3):258–281.

Garnsey, Elizabeth. 1998. “A Theory of the Early Growth of the Firm.” Industrial and Cor-porate Change 7(3):523.

Gernandt, Johannes and Friedhelm Pfeiffer. 2007. “Rising Wage Inequality in Germany.”Jahrbucher fur Nationalokonomie & Statistik 227(4).

Gibbons, Stephen and Henry G. Overman. 2012. “Mostly Pointless Spatial Econometrics?”Journal of Regional Science 52(2):172–191.

Gittings, R. Kaj and Ian M. Schmutte. 2013. “Getting Handcuffs on an Octopus: MinimumWages, Employment, and Turnover.” Unpublished.

Gollier, C. 1991. “Wage Differentials, the Insider–Outsider Dilemma, and Entry-Deterrence.” Oxford Economic Papers 43(3):391–408.

Gruber, Jonathan. 1994. “The incidence of mandated maternity benefits.” The AmericanEconomic Review 84(3):622–641.

Guimaraes, Paolo and Pedro Portugal. 2010. “A Simple Feasible Alternative Procedure toEstimate Models with High-Dimensional Fixed Effects.” Stata Journal 10(4):628–649.

Hamermesh, D.S. and S.J. Trejo. 2000. “The demand for hours of labor: Direct evidencefrom California.” Review of Economics and Statistics 82(1):38–47.

Hans Bockler Stiftung. 2008. “WSI Tarifarchiv.” http://www.boeckler.de.

Haucap, Justus, Uwe Pauly and Christian Wey. 2001. “Collective Wage Setting when Wagesare Generally Binding - An Antitrust Perspective.” International Review of Law and Eco-nomics 21(3):287–307.

Heitzler, S. and C. Wey. 2010. “Raising Rivals’ Fixed (Labor) Costs: The Deutsche PostCase.” DIW Discussion Paper 1008.

Heumer, Moritz, Hagen Lesch and Christoph Schroeder. 2013. “Mindestlohn, Einkom-mensverteilung und Armutsrisiko.” IW Trends, Januar 1.

Hirsch, Barry T., Bruce E. Kaufman and Tetyana Zelenska. 2011. “Minimum Wage Chan-nels of Adjustment.” IZA Discussion Paper 6132.

BIBLIOGRAPHY 138

Hirsch, Boris and Elke Jahn. 2012. Is there monopsonistic discrimination against immi-grants? First evidence from linked employer-employee data. Discussion Paper No. 79Friedrich-Alexander Universitat Erlangen-Nurnberg.

Hirsch, Boris, Marion Konig and Joachim Moller. 2013. “Is There a Gap in the Gap? Re-gional Differences in the Gender Pay Gap.” Scottish Journal of Political Economy 60(4):412–439.

Hirsch, Boris, Thorsten Schank and Claus Schnabel. 2010a. “Differences in Labor Supplyto Monopsonistic Firms and the Gender Pay Gap: An Empirical Analysis Using LinkedEmployer-Employee Data from Germany.” Journal of Labor Economics 28(2):291–330.

Hirsch, Boris, Thorsten Schank and Claus Schnabel. 2010b. “Works councils and separa-tions: Voice, monopoly, and insurance effects.” Industrial Relations: A Journal of Economyand Society 49(4):566–592.

IAB, RWI and ISG. 2011. “Evaluation bestehender gesetzlicher Mindestlohnregelun-gen – Branche: Bauhauptgewerbe.” http://www.rwi-essen.de/publikationen/

rwi-projektberichte. Accessed on 14 March 2013.

IAW. 2011a. “Evaluation bestehender gesetzlicher Mindestlohnregelungen. Branche:Elektrohandwerk.” http://www.iaw.edu/iaw/De:Publikationen:Abgeschlossene\

_IAW-Gutachten. Accessed on 14 March 2013.

IAW. 2011b. “Evaluation bestehender gesetzlicher Mindestlohnregelungen. Branche:Maler- und Lackiererhandwerk.” http://www.iaw.edu/iaw/De:Publikationen:

Abgeschlossene\_IAW-Gutachten. Accessed on 14 March 2013.

Ishiguro, S. and L. Zhao. 2009. “Raising Wages to Deter Entry into Unionized Markets.”Japanese Economic Review 60(4):435–445.

Kalenkoski, Charlene M. and Donald J. Lacombe. 2013. “Minimum wages and teen em-ployment: A spatial panel approach.” Papers in Regional Science 92(2):407–417.

Katz, Lawrence F. and Alan B. Krueger. 1992. “The Effect of the Minimum Wage on theFast-Food Industry.” Industrial and Labor Relations Review 46(1):6–21.

Kerckhofs, P. 2011. “Extension of Collective Bargaining agreements in the EU.” EurofoundBackground Paper pp. 1–12.

Kiessl, T., C. Pohl and H. Schmalholz. 2006. “Ostdeutsche Unternehmen betrachten EU-Dienstleistungsrichtlinie mit Skepsis.” ifo Dresden berichtet 13(01):15–26.

Klosterhuber, Wolfram, Jorg Heining and Stefan Seth. 2013. Linked Employer-Employee-Daten des IAB: LIAB-Langsschnittmodell 1993-2010 (LIAB LM 1993-2010). FDZ Daten-report 08 Institute for the Study of Labor (IZA).

Kluve, J. and C. Schmidt. 2007. “Mindestlohne ohne Reue - eine aussichtsreiche Option furDeutschland?” RWI: Positionen 22.

Kluve, Jochen. 2013. So hoch wie moglich, so niedrig wie notig: Was ist der optimaleMindestlohn? RWI Position No. 53.

BIBLIOGRAPHY 139

Kluve, Jochen, Sandra Schaffner and Christoph M. Schmidt. 2009. Labor Force Status Dy-namics in the German Labor Market - Individual Heterogeneity and Cyclical Sensitivity.Ruhr Economic Papers No. 139.

Knabe, Andreas and Ronnie Schob. 2009. “Minimum wage incidence: The case for Ger-many.” Public Finance Analysis 65(4):403–441.

Kohaut, Susanne and Lutz Bellmann. 1997. “Betriebliche Determinanten der Tarif-bindung: Eine empirische Analyse auf der Basis des IAB-Betriebspanels 1995.” Indus-trielle Beziehungen 4(4):317–334.

Kohaut, Susanne and Peter Ellguth. 2008. “Neu gegrundete Betriebe sind seltener tarifge-bunden.” Nurnberg: IAB-Kurzbericht 16.

Kohn, Karsten. 2006. Rising wage dispersion, after all! The German wage structure at theturn of the century. Technical report ZEW Discussion Papers.

Konig, M. and J. Moller. 2009. “Impacts of minimum wages: A micro data analysis for theGerman construction sector.” International Journal of Manpower 30(7):716–741.

Konig, Marion, Joachim Moller, Micheal Stops and Ulrich Walwei. 2012. “Zur Einfhrungvon gesetzlichen oder branchenspezifischen Mindestlohnen.” IAB Stellungnahme 2.

Kox, H. and A. Lejour. 2006. “The Effects of the Services Directive on Intra-EU Trade andFDI.” Revue economique 57(4):747–769.

Kramer, Walter. 2011. “The cult of statistical significance - What economists should andshould not do to make their data talk.” Schmollers Jahrbuch 131:455–468.

Layard, R., S. Nickell and R. Jackman. 2005. Unemployment: Macroeconomic performance andthe labour market. Oxford University Press, USA.

Lemos, Sara. 2005. “Political variables as instruments for the minimum wage.” The B.E.Journal of Economic Analysis & Policy 4(1):Article 16.

LeSage, James P. and R. Kelley Pace. 2009. Introduction to Spatial Econometrics. New York,NY: Taylor and Francis/CRC Press.

Lesch, Hagen, Alexander Mayer and Lisa Schmid. 2014. “Das deutsche Mindestlohngesetz:Eine erste konomische Bewertung.” IW policy paper 4.

Machin, Stephen and Alan Manning. 1997. “Minimum wages and economic outcomes inEurope.” European Economic Review 41(3–5):733–742.

Manning, A. 2003a. Monopsony in motion: Imperfect competition in labor markets. PrincetonUniversity Press.

Manning, A. 2004. “Monopsony and the efficiency of labour market interventions.” LabourEconomics 11(2):145–163.

Manning, Alan. 2003b. “The real thin theory: monopsony in modern labour markets.”Labour Economics 10(2):105–131.

BIBLIOGRAPHY 140

Manning, Alan. 2011. “Imperfect competition in the labor market.” Handbook of labor eco-nomics 4:973–1041.

Martin, Stephen. 2002. Advanced Industrial Economics. 2 ed. Oxford, UK: Blackwell Publish-ing.

Meer, Jonathan and Jeremy West. 2013. “Effects of the minimum wage on employmentdynamics.” NBER Working Paper No. 19262.

Melitz, M.J. and G.I.P. Ottaviano. 2008. “Unionisation structures, productivity and firmperformance: New insights from a heterogeneous firm model.” Review of Economic Stud-ies 75(1):295–316.

Menz, G. 2010. “Are you Being Served? Europeanizing and Re-regulating the Single Mar-ket in Services.” Journal of European Public Policy 17(7):971–987.

Metcalf, D. 2008. “Why has the British national minimum wage had little or no impact onemployment?” Journal of Industrial Relations 50(3):489–512.

Meyer, Bruce D. 1995. “Natural and quasi-experiments in economics.” Journal of Businessand Economic Statistics 13(2):151–161.

Minimum Wage Study Commission. 1981. Report of the Minimum Wage Study Commission.Washington, D.C.: U.S. Government Printing Office.

Moller, Joachim. 2005. “Die Entwicklung der Lohnspreizung in West-und Ostdeutsch-land.” Institutionen, Lohne und Beschaftigung, Beitrage zur Arbeitsmarkt-und Berufsforschung294:47–63.

Moulton, Brent R. 1986. “Random group effects and the precision of regression estimates.”Journal of Econometrics 32(3):385 – 397.

Muller, Kai-Uwe. 2010. “Employment Effects of a Sectoral Minimum Wage in Germany:Semi-Parametric Estimations from Cross-Sectional Data.” DIW Discussion Paper 1061.

Muller, Kai-Uwe and Viktor Steiner. 2010. “Labor market and income effects of a legalminimum wage in Germany.” IZA Discussion Paper 4929.

Nerb, G. 2006. “Mogliche Auswirkungen der geplanten EU-Dienstleistungsrichtlinie aufdie Wirtschaft Deutschlands.” Ifo Schnelldienst 59(05):19–24.

Neumark, David, J.M. Ian Salas and William Wascher. 2013. “Revisiting the MinimumWage–Employment Debate: Throwing Out the Baby with the Bathwater?” NBER Work-ing Paper No. 18681.

Neumark, David and William Wascher. 1992. “Employment effects of minimum and sub-minimum wages: panel data on state minimum wage laws.” Industrial and Labor RelationsReview 46(1):55–81.

Neumark, David and William Wascher. 2000. “Minimum Wages and Employment: A CaseStudy of the Fast-Food Industry in New Jersey and Pennsylvania: Comment.” The Amer-ican Economic Review 90(5):1362–1396.

BIBLIOGRAPHY 141

Neumark, David and William Wascher. 2004. “Minimum wages, labor market institutions,and youth employment: A cross-national analysis.” Industrial and Labor Relations Review57(2):223–248.

Neumark, David and William Wascher. 2008. Minimum Wages. Vol. 1 1 ed. Cambridge,Massachusetts: The MIT Press.

Oi, Walter Y. and Todd L. Idson. 1999. Firm Size and Wages. In Handbook of Labor Economics,ed. O. Ashenfelter and D. Card. Vol. 3 of Handbook of Labor Economics Elsevier chapter 33,pp. 2165–2214.

Pace, R. Kelley and James P. LeSage. 2008. “A spatial Hausman test.” Economics Letters101:282–284.

Pinoli, Sara. 2010. “Rational expectations and the puzzling no-effect of the minimumwage.” IZA Discussion Paper 4933.

Ransom, Michael R and David P Sims. 2010. “Estimating the firm’s labor supply curve ina new monopsony framework: Schoolteachers in Missouri.” Journal of Labor Economics28(2):331–355.

Ransom, Michael R and Ronald L Oaxaca. 2010. “New market power models and sexdifferences in pay.” Journal of Labor Economics 28(2):267–289.

Rattenhuber, Pia. 2013. “Building the minimum wage: The distributional impact of Ger-many’s first sectoral minimum wage on wages and hours across different wage bargain-ing regimes.” Empirical Economics . forthcoming.

Sabia, Joseph J. and Richard V. Burkhauser. 2010. “Minimum Wages and Poverty: Will a$9.50 Federal Minimum Wage Really Help the Working Poor?” Southern Economic Journal76(3):592–623.

Saint-Paul, G. 2007. “Making Sense of Bolkestein-bashing: Trade Liberalization under Seg-mented Labor Markets.” Journal of International Economics 73(1):152–174.

Salop, S.C. and D.T. Scheffman. 1983. “Raising Rivals’ Costs.” The American Economic Re-view 73(2):267–271.

Salop, S.C. and D.T. Scheffman. 1987. “Cost-raising Strategies.” The Journal of IndustrialEconomics 36(1):19–34.

Schmitt, John. 2013. Why Does the Minimum Wage Have No Discernible Effect on Em-ployment? Technical report Center for Economic and Policy Research (CEPR).

Schmucker, Alexandra and Stefan Seth. 2009. BA-Beschaftigtenpanel Codebuch, . FDZ Daten-report 1/2009.

Staiger, Douglas O, Joanne Spetz and Ciaran S Phibbs. 2010. “Is there monopsony in thelabor market? Evidence from a natural experiment.” Journal of Labor Economics 28(2):211–236.

BIBLIOGRAPHY 142

Steiner, V. and T. Holzle. 2000. The Development of Wages in Germany in the 1990s De-scriptions and Explanations. In The Personal Distribution of Income in an International Per-spective, ed. R. Hauser und I. Becker. The Personal Distribution of Income in an Interna-tional Perspective Springer pp. 7–30.

Stewart, Mark B. 2012. “Wage inequality, minimum wage effects, and spillovers.” OxfordEconomic Papers 64(4):616–634.

Stewart, M.B. 1990. “Union Wage Differentials, Product Market Influences and the Divisionof Rents.” The Economic Journal 100(403):1122–1137.

Stewart, M.B. 2002. “Estimating the Impact of the Minimum Wage Using GeographicalWage Variation.” Oxford Bulletin of Economics and Statistics 64(supplement):583–605.

Sulis, Giovanni. 2011. “What can monopsony explain of the gender wage differential inItaly?” International Journal of Manpower 32(4):446–470.

SVR. 2011. Verantwortung fur Europa wahrnehmen. Jahresgutachten 20011/12Sachverstandigenrat zur Begutachtung der gesamtwirtschaftlichen Entwicklung.

Thompson, Jeffrey P. 2009. “Using Local Labor Market Data to Re-examine the Employ-ment Effects of the Minimum Wage.” Industrial and Labor Relations Review 62(3):343–366.

Whaples, Robert. 2006. “Do Economists Agree on Anything? Yes!” The Economists’ View3(9).

Williamson, Oliver E. 1968. “Wage Rates as a Barrier to Entry: The Pennington Case inPerspective.” The Quarterly Journal of Economics 82(1):85–116.

Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and Panel Data. 2 ed.Cambridge, MA: MIT Press.

Woolfson, C. and J. Sommers. 2006. “Labour Mobility in Construction: European Implica-tions of the Laval un Partneri Dispute with Swedish Labour.” European Journal of Indus-trial Relations 12(1):49.

A. Supplementary Material

A.1. Appendix Section 2.1

Table A.1.: Separate Estimations of the DiD EstimatorWest Germany East Germany

Electricians Painters Electricians Painters

Minimum wage dummy −0.0039 (0.0075) 0.0194 (0.0131) 0.0033 (0.0108) −0.0130 (0.0251)Marginal employment −0.0262 (0.0199) −0.0953∗∗∗(0.0290) −0.0286∗∗∗(0.0100) −0.0085 (0.0163)Macroeconomic growth 0.0886 (0.0765) 0.3044∗∗∗(0.1084) 0.1861∗ (0.1034) 0.7205∗∗∗(0.1614)

Communication & Transport 0.0241∗∗∗(0.0044) 0.0215∗∗∗(0.0057) 0.0106∗ (0.0059) 0.0015 (0.0103)Electricians −0.0006 (0.0077) −0.0120 (0.0110)Painters −0.0129 (0.0114) 0.0318 (0.0297)

Industry dummies yes yes yes yesTime dummies yes yes yes yes

R2 0.671 0.674 0.665 0.668Observations 93 93 93 93

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Notes: Standard errors in brackets. All models are estimated by Prais-Winston regression to allow for AR(1) errors within panels(Wooldridge test for autocorrelation in panel data suggests autocorrelation of first degree). Standard errors are adjusted for het-eroskedasticity between panels (Breusch-Pagan test suggests presence of heteroskedasticity).Source: BA employment panel (Schmucker and Seth, 2009). Author’s calculations.

A. SUPPLEMENTARY MATERIAL 144


Table A.2.: Spatial error models for wages(1) OLS (2) SEM (3) OLS (4) SEM

Artificial bite (West) 0.221∗∗∗ 0.279∗∗∗ 0.328∗∗∗ 0.330∗∗∗

(0.033) (0.032) (0.032) (0.031)Artificial bite (East) 0.140∗∗∗ 0.144∗∗∗ 0.128∗∗∗ 0.129∗∗∗

(0.011) (0.012) (0.015) (0.014)Treatment effect (West) 0.026 0.014 −0.021 −0.022


(0.008) (0.010) (0.018) (0.017)Wage growth (other industries) 0.006 0.006

(0.013) (0.012)Employment growth (other industries) −0.009∗∗ −0.009∗∗

(0.004) (0.004)Artificial bite (West), neighbors −0.051 −0.046

(0.050) (0.048)Artificial bite (East), neighbors −0.006 −0.007

(0.024) (0.023)Treatment effect (West), neighbors −0.042 −0.046

(0.041) (0.039)Treatment effect (East), neighbors −0.032 −0.030

(0.027) (0.026)Wage growth (other industries), neighbors −0.014 −0.017

(0.027) (0.026)Employment growth (other industries), neighbors −0.002 −0.002

(0.008) (0.008)Post-treatment dummy −0.011∗∗∗ −0.013∗∗∗

(0.001) (0.001)Spatial autocorrelation (λ) 0.474∗∗∗ 0.094∗∗∗

(0.01) (0.025)

District fixed effects Yes Yes Yes YesYear indicators No No Yes YesDistrict-type-specific trends No No Yes Yes

Spatial Hausman test (p-value) 0.000 1.000Within R2 0.167 0.163 0.400 0.400Observations 3708 3708 3708 3708

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Notes: Neighbors are defined as sharing a common border. Standard errors are enclosed in parentheses. The SEM modelsare estimated employing the user-written routine xsmle for Stata.Source: Authors’ calculations based on the IEB.


Table A.3.: Spatial error models for employment(1) OLS (2) SEM (3) OLS (4) SEM

Artificial bite (West) −0.767∗∗∗ −0.504∗∗∗ −0.177 −0.182(0.156) (0.150) (0.156) (0.147)

Artificial bite (East) 0.027 −0.006 −0.041 −0.038(0.053) (0.057) (0.070) (0.066)

Treatment effect (West) 0.397∗∗∗ 0.285∗∗ 0.078 0.076(0.134) (0.131) (0.138) (0.130)

Treatment effect (East) −0.586∗∗∗ −0.525∗∗∗ −0.340∗∗∗ −0.346∗∗∗

(0.039) (0.048) (0.086) (0.080)Wage growth (other industries) −0.025 −0.017

(0.064) (0.060)Employment growth (other industries) 0.028 0.027

(0.019) (0.017)Artificial bite (West), neighbors 0.560∗∗ 0.535∗∗

(0.240) (0.229)Artificial bite (East), neighbors 0.143 0.161

(0.118) (0.112)Treatment effect (West), neighbors −0.369∗ −0.353∗

(0.195) (0.186)Treatment effect (East), neighbors −0.085 −0.085

(0.129) (0.123)Wage growth (other industries), neighbors 0.035 0.048

(0.130) (0.124)Employment growth (other industries), neighbors 0.027 0.021

(0.039) (0.037)Post-treatment dummy −0.027∗∗∗ −0.026∗∗∗

(0.006) (0.007)Spatial autocorrelation (λ) 0.460∗∗∗ 0.090∗∗∗

(0.018) (0.025)

District fixed effects Yes Yes Yes YesYear indicators No No Yes YesDistrict-type-specific trends No No Yes Yes

Spatial Hausman test (p-value) 0.036 1.000Within R2 0.159 0.158 0.384 0.384Observations 3708 3708 3708 3708

Legend: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Notes: Neighbors are defined as sharing a common border. Standard errors are enclosed in parentheses. The SEM modelsare estimated employing the user-written routine xsmle for Stata.Source: Authors’ calculations based on the IEB.


Robustness to inclusion of different trends

As discussed in Section 2.2.3, our identification strategy rests on controlling for time-varying

spatial heterogeneity. One solution consists in including region-type-specific trends; i.e. we

allow that each region type (recall that each of districts belongs to exactly one type) shows

a different linear trend in wage and employment growth, independent of the minimum

wage. However, the districts within a specific type are expected to follow the same trend.

In order to relax this assumption, we instead include trends at the state level and for each

labour market region (ROR, cf. Section 2.2.2). As Table A.4 shows, the results are hardly

influenced by this choice.

Table A.4.: Baseline specification with state and labour-market-region trendsWages Employment

(1) State (2) ROR 1) State (2) ROR

Artificial bite (West) 0.365∗∗∗ 0.387∗∗∗ −0.140 −0.028(0.048) (0.047) (0.159) (0.164)

Artificial bite (East) 0.120∗∗∗ 0.143∗∗∗ 0.029 −0.060(0.026) (0.025) (0.129) (0.133)

Treatment effect (West) −0.064∗∗ −0.045 0.096 −0.168(0.027) (0.032) (0.144) (0.159)

Treatment effect (East) 0.188∗∗∗ 0.174∗∗∗ −0.435∗∗∗ −0.387∗∗∗

(0.023) (0.023) (0.107) (0.112)Wage growth (other industries) 0.005 0.002 −0.023 −0.010

(0.016) (0.017) (0.065) (0.065)Employment growth (other industries) −0.009∗∗ −0.009∗∗ 0.029 0.023

(0.004) (0.004) (0.020) (0.020)

District fixed effects Yes Yes Yes YesYear indicators Yes Yes Yes YesDistrict-type-specific trends No No No NoState-specific trends Yes No Yes NoROR-specific trends No Yes No Yes


Notes: The table shows the specification in column 2 of Table 2.5 and 2.8. Instead of region-type-specific trends, weinclude state-specific trends or labour-market-region-specific trends. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Standarderrors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.

Length of post-treatment period

The post-treatment period of all estimations presented in the paper starts in 1997 and ends

in 2002. A valid concern consists in the notion that this post-treatment period includes

further increases in the minimum wage. This implies that the bite is not only influenced by


the nominal minimum wage in the next period, but also by the effect of the minimum wage

on employment and wages in previous periods. Only the first year of the post treatment

period 1997 is by definition not subject to this problem.

Therefore, Table A.5 shows the baseline specification (column 2 in Table 2.5 and 2.8)

with different post-treatment periods. In the first column the post-treatment period con-

sists only of the year 1997. In each of the following columns, the post-treatment period

is increased by one year. The probability that the results are biased is higher with each

additional year. However, the treatment effect is extremely robust to the different post-

treatment periods. Additionally, no pattern in terms of an increasing or decreasing magni-

tude of the treatment effect can be observed.

Working hours

In Table A.6, we present the mean and standard deviation of the usual working hours based

on the German census (Mikrozensus). Based on the Scientific Use File of the Mikrozensus,

this is identified as the variable EF131, and corresponds to the response to the question,

”Wie viele Stunden arbeiten Sie normalerweise pro Woche?” (How many hours do you usually

work per week?)

Note that there is very little variation over time and between East and West Germany.

We do not think that much of the adjustment in the labor market occured in the intensive

margin (i.e., a reduction in the number of working hours).

Although mandatory working time accounts do exist in the construction industry, not

all firms implement it (only about two thirds, according to the Zentralverband Deutsches

Baugewerbe). With respect to overtime, these are typically recorded as part of the worker’s

working hours [IAB, RWI and ISG 2011].


Table A.5.: Baseline specification with different post-treatment periodsWage Growth

1997 1998 1999 2000 2001 2002

Artificial bite (West) 0.462∗∗∗ 0.398∗∗∗ 0.407∗∗∗ 0.352∗∗∗ 0.346∗∗∗ 0.331∗∗∗

(0.056) (0.054) (0.056) (0.051) (0.050) (0.044)Artificial bite (East) 0.285∗∗∗ 0.185∗∗∗ 0.133∗∗∗ 0.139∗∗∗ 0.137∗∗∗ 0.131∗∗∗

(0.036) (0.029) (0.029) (0.026) (0.025) (0.024)Treatment effect (West) −0.046 0.066 0.036 −0.003 −0.022 −0.041

(0.057) (0.048) (0.040) (0.034) (0.030) (0.027)Treatment effect (East) 0.141∗∗∗ 0.162∗∗∗ 0.169∗∗∗ 0.162∗∗∗ 0.151∗∗∗ 0.147∗∗∗

(0.028) (0.028) (0.027) (0.023) (0.021) (0.020)Wage growth (other industries) −0.017 0.003 −0.009 −0.004 0.004 0.004

(0.026) (0.023) (0.019) (0.018) (0.018) (0.016)Employment growth (other industries) 0.003 0.001 −0.010∗ −0.011∗∗ −0.009∗∗ −0.009∗∗

(0.008) (0.007) (0.006) (0.004) (0.004) (0.004)

District fixed effects Yes Yes Yes Yes Yes YesYear indicators Yes Yes Yes Yes Yes YesDistrict-type-specific trends Yes Yes Yes Yes Yes Yes

Wooldridge test for serial correlation (p-value) 0.846 0.907 0.794 0.516 0.578 0.411Within R2 0.560 0.534 0.479 0.446 0.422 0.399Observations 1648 2060 2472 2884 3296 3708

Employment Growth

1997 1998 1999 2000 2001 2002

Artificial bite (West) −0.073 −0.037 −0.237 −0.087 −0.076 −0.104(0.282) (0.249) (0.219) (0.183) (0.162) (0.151)

Artificial bite (East) −0.615∗∗∗ −0.221 −0.066 −0.042 −0.021 −0.026(0.218) (0.147) (0.141) (0.126) (0.121) (0.110)

Treatment effect (West) −0.295 −0.378∗∗ −0.214 −0.025 −0.037 0.004(0.208) (0.188) (0.160) (0.162) (0.146) (0.130)

Treatment effect (East) −0.325∗∗ −0.400∗∗∗ −0.393∗∗∗ −0.379∗∗∗ −0.409∗∗∗ −0.368∗∗∗

(0.152) (0.136) (0.120) (0.102) (0.097) (0.093)Wage growth (other industries) −0.016 0.014 0.047 0.012 −0.034 −0.016

(0.151) (0.127) (0.090) (0.083) (0.075) (0.065)Employment growth (other industries) −0.027 0.002 −0.018 0.011 0.027 0.028

(0.052) (0.042) (0.030) (0.023) (0.022) (0.020)

District fixed effects Yes Yes Yes Yes Yes YesYear indicators Yes Yes Yes Yes Yes YesDistrict-type-specific trends Yes Yes Yes Yes Yes Yes

Wooldridge test for serial correlation (p-value) 0.902 0.645 0.404 0.488 0.312 0.452Within R2 0.412 0.421 0.402 0.366 0.382 0.382Observations 1648 2060 2472 2884 3296 3708

Notes: The table shows the specification in column 2 of Table 2.5 and 2.8. The post-treatment period is increased in steps of one year, startingwith only 1997 as post-treatment period and ending with our original period from 1997 until 2002. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.Standard errors are enclosed in parentheses and clustered at the district level.Source: Authors’ calculations based on the IEB.


Table A.6.: Hours worked

YearEast Germany West Germany

Mean Std. Dev. Obs. Mean Std. Dev. Obs.

1993 — — — 39.95 3.07 6, 8611995 — — — 39.74 3.17 6, 3281996 40.08 2.25 2, 255 39.95 3.47 4, 0621997 40.14 2.15 984 40.05 3.25 4, 8501998 40.10 1.96 901 40.16 3.40 4, 2871999 40.18 3.05 787 40.26 3.58 4, 0202000 40.36 3.26 1, 599 40.39 4.03 3, 3872001 40.23 3.22 610 40.22 3.40 3, 7512002 40.09 2.23 598 40.15 3.10 3, 545All 40.17 2.62 7, 734 40.06 3.36 41, 091

Source: Authors’ calculations based on the German microcensus.

Table A.7.: Estimated employment losses due to transition to self-employmentYear Observed Predicted level Jobs lost Level of Change in Maximum share of

employment without the due to the self-employment self-employment of employment losseslevel minimum wage minimum wage due to self-employment

1993 337094 — — — — —1994 374205 — — — — —1995 383872 — — 10148 — —1996 351712 — — 11169 1021 0.001997 311465 336663 25198 13048 1879 7.461998 265309 276819 11510 14862 1814 15.761999 255806 268179 12373 16020 1158 9.362000 221684 243168 21484 17032 1012 4.712001 186049 199034 12985 16282 −750 0.002002 158524 167410 8886 16425 143 1.61

Notes: The calculation here takes the treatment effect to be 0.35 percentage points, which lies roughly in the middle of the various estimates.Moreover, we use the observed employment level in the previous period to predict the number of jobs lost due to the minimum wage. Alternatively,one could use the predicted employment level in the previous period net of the minimum wage effect, but this would widen the gap betweenthe observed and predicted employment levels even further (i.e., the share of employment losses that can be accounted for by increasing self-employment would even be lower). The above calculations are more generous in allowing self-employment to take a larger share of the totalemployment losses.Source: Authors’ calculations based on ELVIRA [2013].

Self-employed

Although we are unable to formally take into account the dynamics underlying the number

of self-employed construction workers, we present some evidence which indicates that

the decrease in employment observed in the data is largely due to the minimum wage.

Assuming a treatment effect of −0.35 percentage points for East Germany, we are able to

construct Table A.7 based on our sample.

For the whole period 1996–2002, the total job losses due to the minimum wage is roughly

calculated to be equal to about 92,435 while self-employment increased by 5,256 between

1995 and 2002. Even if we are generous with the employment loss due to the movement


from employed to self-employed, this can only account for roughly 6 percent of total em-

ployment losses we attribute to the imposition of minimum wages for the whole analysis

period. It can be as high as about 15.76 percent (1998), but this still means that the substan-

tial part of the effect is due to the minimum wage.



Construction of variables

The IUD consists of the product of the union mark-up and the non-coverage rate, divided

by 100. These indicators are derived from the survey as follows. The union mark-up is de-

fined as the relative deviation of average wages paid by collectively organized employers

to average wages paid by firms not covered by any collective agreement. The necessary

indicators for wages and collective bargaining are derived from two items in the survey.

First, respondents state whether they pay wages according to, in excess of, or below the

collectively bargained rate applicable to their industry. This item is used as a proxy for the

firm’s membership in an employer association. Second, we use the hourly wage of skilled

labour with a minimum of three years of work experience. The survey asks specifically for

the wage rates of skilled and unskilled labour, with and without work experience. Clearly,

in the context of minimum wages, the wage rate for unskilled workers appears to be most

relevant. However, the wage rate for skilled labour with work experience is the only one

with a sensible number of observations. Assuming that within each firm, wage gaps be-

tween different skill groups are proportionally equal in size, it should not matter which

wage rate is used. The non-coverage rate amounts to the ratio of workers not covered by

collective agreements to all workers in the industry, and quantifies the difference between

the union and the non-union sector in terms of the number of workers.

Robustness tests

In the empirical analysis, we included the following additional/alternative variables which

did not alter the results significantly and are therefore not included in the final specifica-

tion:

• An indicator variable for young firms, which is defined as a dummy variable which

takes the value one if the firm is younger than or equal to six years. The specific


threshold is based upon the work by Garnsey [1998], who shows that 60 percent of

all new companies fail within the first six years.

• Annual turnover per employee as a measure of productivity. As annual turnover

varies strongly between the different industries, we used its percentage deviation

from the industry’s mean in order to be able to compare this variable across indus-

tries.

B. Curriculum Vitae

Personal

Name Hanna Frings

Date of Birth 11 May 1984

Citizenship German

Education

2009 - 2014 Doctoral student in economics, Ruhr-Universitat Bochum, Germany

2007 - 2009 Master of Science in International Economic Studies, Maastricht Univer-

sity, Netherlands

2006 Study abroad, Universidad Malaga, Spain

2003-2007 Bachelor of Science in International Business, Maastricht University,

Netherlands

Career

2009 - present RWI, Researcher, Research Division “Labor Markets, Education, Popu-

lation”, Germany

2007 - 2008 Maastricht University, Student assistant, International Relations Office,

Netherlands

2007 Eurostat, European Commission, Intern, Unit D2 “Regional Indicators

and Geographic Indicators”, Luxembourg

B. CURRICULUM VITAE 154

Publications

Articles in Refereed Journals

Bachmann, R., Th.K. Bauer and H. Frings (2014), Minimum Wages as a Barrier to Entry

Evidence from Germany. Labour: Review of Labour Economics and Industrial Relations 28(3):

338-357.

Frings, H. (2013), The employment effect of industry-specific, collectively-bargained

minimum wages. German Economic Review 14 (3): 258-281.

Apel, H., R. Bachmann, S. Bender, M. Fertig, H. Frings, M. Konig, J. Moller, A. Paloyo,

S. Schaffner, M. Tamm, S. Wolter, M. Umkehrer and P. vom Berge (2012), Arbeitsmarkt-

wirkungen der Mindestlohneinfuhrung im Bauhauptgewerbe. Zeitschrift fur Arbeitsmarkt-

forschung 45(3): 257-277.

Discussion Papers

vom Berge, P., H. Frings und A. Paloyo (2013), High-Impact Minimum Wages and Het-

erogeneous Regions. Ruhr Economic Papers #408. RWI, RUB.

Kroger, H. und S. Schaffner (2011), The intensive and extensive margin of European

Labour Supply. Ruhr Economic Papers #291. RWI, RUB.

Date post:	13-Apr-2022
Category:	Documents
Upload:	others
View:	11 times
Download:	0 times

Minimum wages : boon or bane? : Microeconomic evidence ...

Documents