Julius-Maximilians-Universität Würzburg

Wirtschaftswissenschaftliche Fakultät

Inaugural-Dissertation
zur Erlangung des akademischen Grades

doctor rerum politicarum (Dr. rer. pol.)

Essays on Intergenerational Income Mobility

in Germany and the United States

vorgelegt von

Sarah Mehringer
am 19. Januar 2018
 Erstgutachter

Prof. Dr. Norbert Berthold

Zweitgutachter

Prof. Dr. Wolfgang Dauth

Contents

List of Figures iv

List of Tables v

Danksagung vi

Deutschsprachige Zusammenfassung vii

1 Introduction 1

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Organization of this dissertation . . . . . . . . . . . . . . . . . . . . . 6

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Structure and Extent of Intergenerational Income Mobility 13

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Data and mobility measures . . . . . . . . . . . . . . . . . . . . . . . 16

2.2.1 Measurement errors and life-cycle bias . . . . . . . . . . . . . 16

2.2.2 Sample denition . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2.3 Intergenerational mobility measures . . . . . . . . . . . . . . . 20

2.3 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.3.1 Descriptive evidence . . . . . . . . . . . . . . . . . . . . . . . 27

2.3.2 Intergenerational income elasticity . . . . . . . . . . . . . . . . 31

2.3.3 Intergenerational rank and income share mobility . . . . . . . 33

2.3.4 Quantile regressions . . . . . . . . . . . . . . . . . . . . . . . . 37

2.3.5 Decomposition of intergenerational income inequality . . . . . 41

i
 Contents

2.4 Recommendations for economic policy . . . . . . . . . . . . . . . . . 42

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3 Transmission Channels of Intergenerational Income Persistence 56

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.2 Theoretical framework . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.2.1 Intergenerational income and educational persistence . . . . . 60

3.2.2 Descriptive decomposition . . . . . . . . . . . . . . . . . . . . 61

3.2.3 Structural decomposition . . . . . . . . . . . . . . . . . . . . . 64

3.3 Data and measurement issues . . . . . . . . . . . . . . . . . . . . . . 68

3.3.1 Measurement errors and life-cycle bias . . . . . . . . . . . . . 68

3.3.2 Sample denition and variables . . . . . . . . . . . . . . . . . 69

3.3.3 Descriptive evidence . . . . . . . . . . . . . . . . . . . . . . . 71

3.4 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3.4.1 Descriptive decomposition . . . . . . . . . . . . . . . . . . . . 75

3.4.2 Structural decomposition . . . . . . . . . . . . . . . . . . . . . 79

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

4 Intergenerational Income Mobility among Daughters 92

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.2 Theoretical framework . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.3 Data and measurement issues . . . . . . . . . . . . . . . . . . . . . . 97

4.3.1 Measurement errors and life-cycle bias . . . . . . . . . . . . . 98

4.3.2 Sample denition . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.4 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.4.1 Baseline estimation . . . . . . . . . . . . . . . . . . . . . . . . 103

ii
 Contents

4.4.2 Eects of assortative mating . . . . . . . . . . . . . . . . . . . 107

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

5 Conclusive remarks 117

iii
List of Figures
Figure 1.1 Development of market and net income inequality . . . . . . . . . 2

Figure 1.2 Country comparison of intergenerational income mobility . . . . . 4

Figure 2.1 Intergenerational income correlation . . . . . . . . . . . . . . . . . 20

Figure 2.2 Nonlinearities in intergenerational income elasticity . . . . . . . . 25

Figure 2.3 Income share curves . . . . . . . . . . . . . . . . . . . . . . . . . 30

Figure 2.4 Intergenerational rank mobility . . . . . . . . . . . . . . . . . . . 35

Figure 2.5 Intergenerational income share mobility . . . . . . . . . . . . . . . 36

Figure 2.6 Conditional quantile regressions . . . . . . . . . . . . . . . . . . . 39

Figure 2.7 Unconditional quantile regressions . . . . . . . . . . . . . . . . . . 40

Figure 3.1 Transmission channels of intergenerational income persistence . . 57

Figure 3.2 Intergenerational income correlation . . . . . . . . . . . . . . . . . 72

Figure 3.3 Intergenerational educational correlation . . . . . . . . . . . . . . 73

Figure 3.4 Nonlinear descriptive decomposition . . . . . . . . . . . . . . . . 78

Figure 3.5 Intergenerational persistence along the income distribution . . . . 90

Figure 4.1 Intergenerational income correlation by gender . . . . . . . . . . . 102

iv
List of Tables
Table 2.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Table 2.2 Income inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Table 2.3 Intergenerational income elasticity . . . . . . . . . . . . . . . . . . 31

Table 2.4 Intergenerational income elasticity for dierent lower income limits 32

Table 2.5 Estimated transition matrices . . . . . . . . . . . . . . . . . . . . . 34

Table 2.6 Quantile regressions . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Table 2.7 Decomposition of intergenerational income inequality . . . . . . . . 42

Table 2.8 Intergenerational income elasticity (robustness checks) . . . . . . . 55

Table 3.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 71

Table 3.2 Intergenerational income and educational persistence . . . . . . . . 74

Table 3.3 Linear descriptive decomposition . . . . . . . . . . . . . . . . . . . 75

Table 3.4 Nonlinear descriptive decomposition . . . . . . . . . . . . . . . . . 77

Table 3.5 IV estimation of intergenerational income elasticity . . . . . . . . . 81

Table 3.6 Structural decomposition I . . . . . . . . . . . . . . . . . . . . . . 82

Table 3.7 Structural decomposition II . . . . . . . . . . . . . . . . . . . . . . 83

Table 3.8 Nonlinear descriptive decomposition (detailed estimates) . . . . . . 91

Table 4.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 101

Table 4.2 Intergenerational income elasticity by gender and marital status . 104

Table 4.3 Determinants of intergenerational income mobility . . . . . . . . . 105

Table 4.4 Intergenerational income elasticities by type of income . . . . . . . 108

Table 4.5 Heckman estimation of intergenerational income elasticity . . . . . 116

v
Danksagung
Die vorliegende Dissertation entstand während meiner Tätigkeit als wissenschaftliche

Mitarbeiterin am Lehrstuhl für VWL, insb. Wirtschaftsordnung und Sozialpolitik

an der Universität Würzburg. Mein erster Dank gilt daher meinem Doktorvater

Prof. Dr. Norbert Berthold, der mir jederzeit mit Rat und Tat zur Seite gestanden,

aber auch ein hohes Maÿ an Zeit und Entscheidungsfreiheit für die Erstellung dieser

Arbeit gelassen hat. Vielen Dank für die vielen teils fachlichen, teils völlig abstrusten

Diskussionen beim Mittagessen (11:40 Mensa?). Ich werde meine Zeit am Lehrstuhl

sehr vermissen.

Mein Dank gilt auch meinem Zweitgutachter Prof. Dr. Wolfgang Dauth, an den

ich mich vor allem in den letzten Monaten vor Abgabe bei allen Fragen wenden

konnte. Ich bedanke mich bei meinen Kollegen Dr. Klaus Gründler, Dr. Sebastian

Köllner, Christian Erthle und besonderers bei Dr. Mustafa Çoban. Musti, vielen

Dank für die Zusammenarbeit mit dir, deine unermüdliche Hilfe bei allen Fragen

statistischer und (programmier-)technischer Natur und deinen guten Zuspruch, wenn

ich wieder einmal am liebsten alles hinwerfen wollte. Last but not least danke ich

Silke Amelang für ihre stetige Unterstützung in organisatorischen Fragen und das

tägliche Pläuschchen im Sekretariat.

Ich bedanke mich bei meinen Eltern und meinen Geschwistern für ihre Unterstüt-

zung während des Studiums und der Doktorandenzeit. Ich danke meinen Freunden,

die mich Abend für Abend wieder zurück ins echte Leben geholt haben. Mein

gröÿter Dank geht aber an Christian, der diese Reise  zu Beginn als mein Freund

und heute als mein Ehemann  mit mir unternommen hat. Ohne dich wäre alles nur

halb so viel Wert.

vi
Deutschsprachige Zusammenfassung
Die vorliegende Dissertation beschäftigt sich mit der intergenerativen Einkommens-

mobilität in Deutschland im Vergleich zu den USA. Im Zentrum der Analyse steht

demnach die Frage, inwiefern Einkommensunterschiede zwischen armen und reichen

Familien an die nächste Generation weitergegeben werden. Diese Thematik wird

sowohl in der Ökonomie als auch in der Soziologie seit mehreren Jahrzehnten disku-

tiert, dennoch sind  auch auf Grund der unzureichenden Datenverfügbarkeit in den

meisten Ländern  viele Fragestellungen oen. Die vorliegende Arbeit setzt an der

vorhandenen Literatur an und gliedert sich in drei Hauptkapitel, die jeweils einen

bestimmten Aspekt der intergenerativen Einkommensmobilität untersuchen und als

eigenständige Forschungsarbeiten zu betrachten sind. Der Fokus liegt dabei auf der

empirischen Analyse der jeweiligen Fragestellung mithilfe vorhandener Daten des

sozioökonomischen Panels (SOEP) für Deutschland und der Panel Study of Income

Dynamics (PSID) für die USA.

Der erste Teil der Arbeit untersucht Struktur und Ausmaÿ der intergenerativen

Einkommensmobilität in Deutschland und den USA. Dafür werden unterschiedliche

Mobilitätsmaÿe berechnet und die Ergebnisse für beide Länder miteinander vergli-

chen. Im Einklang mit der bestehenden Literatur fällt die intergenerative Einkom-

menselastizität in Deutschland geringer aus als in den USA, was für eine höhere

Mobilität in Deutschland spricht. Vergleicht man jedoch die intergenerative Rang-

mobilität, sind die Ergebnisse für beide Länder relativ ähnlich. Bei der intergenera-

tiven Einkommensanteilsmobilität bestehen dagegen stärkere Unterschiede zwischen

Deutschland und den USA. Mit jedem höheren Perzentil sinkt die Einkommensan-

teilsmobilität der Söhne im Vergleich zu ihren Vätern in Deutschland weniger stark

als in den USA. Die Regression zur Mitte ndet demnach in Deutschland langsamer

statt als in den USA, was für eine höhere Mobilität in den USA spricht. Die Ergebnis-

vii
 Deutschsprachige Zusammenfassung

se der bedingten und unbedingten Quantilsregression liefern für keines der beiden

Länder Hinweise auf Nichtlinearitäten. Eine abschlieÿende Dekomposition der in-

tergenerativen Einkommensungleichheit ergibt für Deutschland sowohl eine gröÿere

Einkommensmobilität als auch ein stärkeres progressives Einkommenswachstum als

für die USA. Insgesamt kann keine klare Rangfolge hinsichtlich der intergenerativen

Einkommensmobilität in Deutschland und den USA festgestellt werden. Abschlie-

ÿend werden mögliche Politikmaÿnahmen erläutert, die zur Erhöhung der interge-

nerativen Einkommensmobilität in Deutschland beitragen könnten.

Der zweite Teil der Dissertation beschäftigt sich mit der Frage, über welche

Transmissionskanäle das Einkommen der Eltern das Einkommen ihrer Kinder be-

einusst. Dabei sind im Wesentlichen zwei Mechanismen denkbar. Zum einen können

wohlhabende Familien mehr Geld in das Humankapital ihrer Kinder investieren, wo-

durch diese später ein höheres Einkommen auf dem Arbeitsmarkt erzielen. Dieser

Kanal wird als Investitionseekt bezeichnet und beinhaltet beispielsweise den Besuch

einer privaten Schule oder Universität oder die Finanzierung privater Nachhilfestun-

den. Zum anderen verfügen Eltern mit einem hohen Einkommen tendenziell auch

über ein höheres Humankapital, das sie auch ohne den Einsatz nanzieller Mittel an

ihre Kinder weitergeben können. Darunter fallen die genetische Weitergabe bestimm-

ter Eigenschaften, die innerfamiliär erlernten Einstellungen und Ziele, aber auch ge-

zielte nicht-monetäre Investitionen in das Humankapital der Kinder, zum Beispiel in

Form von pädagogisch hochwertiger Freizeitgestaltung oder Unterstützung bei den

Hausaufgaben. Dieser Kanal wird als Humankapitaleekt bezeichnet. Die empirische

Analyse mithilfe unterschiedlicher Dekompositionsmethoden zeigt, dass der Investi-

tionseekt und der Humankapitaleekt in Deutschland zu etwa gleichen Teilen zur

geschätzten intergenerativen Einkommenselastizität beitragen, während in den USA

der Investitionseekt vor allem in den oberen Perzentilen deutlich stärker ausgeprägt

ist. Im Hinblick auf die im Vergleich zu Deutschland deutlich höhere Privatisierung

des Bildungssektors in den USA scheint dieses Resultat plausibel. Für die Politik

in Deutschland bedeuten diese Ergebnisse, dass die bloÿe Bereitstellung nanzieller

viii
 Deutschsprachige Zusammenfassung

Mittel für Kinder aus armen Familien nicht ausreicht, um ihre Aufwärtsmobilität

zu fördern. Zusätzlich muss die fehlende direkte Weitergabe von Humankapital in-

nerhalb sozioökonomisch schwacher Familien durch staatliche Angebote substituiert

werden.

Während sich die bisherige Analyse auf Väter und ihre Söhne beschränkt, ist

das Ziel des dritten Teils der Dissertation eine Untersuchung der intergenerativen

Einkommensmobilität der Töchter. Der Hauptgrund für diese in der Literatur übli-

che Restriktion sind Probleme bei der statistischen Analyse, die sich aufgrund der

geringeren Arbeitsmarktpartizipation von Frauen im Vergleich zu Männern ergeben.

Während Männer fast immer Vollzeit arbeiten, sind nach wie vor viele  insbesonde-

re verheiratete  Frauen nur in Teilzeit oder gar nicht berufstätig. Das individuelle

Einkommen der Tochter ist daher in vielen Fällen kein geeignetes Maÿ für ihren tat-

sächlichen Wohlstand. Gerade wenn assortative Paarung stattndet  also Töchter

wohlhabender Familien tendenziell auch gutverdienende Männer heiraten und sich

infolgedessen für eine geringere Anzahl an Arbeitsstunden entscheiden  kann die ge-

schätzte Einkommenselastizität verzerrt sein. Eine erste Basisregression zeigt, dass

die intergenerative Einkommenselastizität der Töchter in Deutschland höher ausfällt

als die der Söhne, während es in den USA gerade umgekehrt ist. Eine Trennung nach

Familienstand macht jedoch deutlich, dass in beiden Ländern unverheiratete Frauen

eine höhere Einkommenselastizität aufweisen als unverheiratete Männer, wohinge-

gen für verheiratete Frauen eine niedrigere Einkommenselastizität geschätzt wird

als für verheiratete Männer. Während die geringere Mobilität der unverheirateten

Töchter auf eine stärkere Humankapitaltransmission zwischen Vätern und Töchtern

im Vergleich zu den Söhnen zurückgeführt werden kann, ist die höhere Mobilität

verheirateter Frauen zum einen auf eine weniger starke Humankapitaltransmission

und zum anderen auf eine stärkere Arbeitsstundenelastizität der Töchter im Hin-

blick auf das Einkommen ihres Ehepartners zurückzuführen. Um den Eekt der

assortativen Paarung genauer zu untersuchen, werden anschlieÿend die verheirate-

ten Individuen noch einmal nach unterschiedlichen Einkommensarten untersucht.

ix
 Deutschsprachige Zusammenfassung

Dabei zeigt sich, dass die intergenerative Elastizität der Haushaltseinkommen ten-

denziell sogar gröÿer ausfällt als die der Individualeinkommen, was für eine starke

assortative Paarung spricht. Betrachtet man die Höhe des Haushaltseinkommens als

das eigentliche Wohlstandsniveau einer Person, existieren auÿerdem keine gravie-

renden Unterschiede zwischen der Einkommensmobilität von Töchtern und Söhnen.

Auch das Individualeinkommen des jeweiligen Ehepartners ist stark mit dem Ein-

kommen des Vaters korreliert, was die These der assortativen Paarung wiederum

stützt. Die intergenerative Einkommenselastizität der Schwiegersöhne im Vergleich

zu ihren Schwiegervätern fällt in Deutschland sogar gröÿer aus als die intergenerative

Einkommenselastizität der Söhne im Vergleich zu ihren eigenen Vätern.

x
Chapter 1

Introduction

1.1 Motivation

The inequality of market incomes has risen in almost all developed countries since

the 1970s. In Germany, the Gini coecient has increased from a local minimum of

0.38 in 1973 to a local maximum of 0.52 in 2014, while the Gini coecient in the

United States has risen from a local minimum of 0.42 in 1969 to a local maximum

of 0.51 in 2015 (Figure 1.1).

This development is driven by rapidly increasing incomes in the upper percentiles

of the earnings distribution and has been attributed mainly to the consequences of

globalization and technological progress. On the one hand, growing international

division of labor has led to an increase in the demand for high-skilled labor and

a decrease in the demand for low-skilled labor in developed countries (Ebenstein

et al., 2014). On the other hand, a growing number of manual tasks can today

be performed more eciently by computers and robots, whereas complementary

cognitive tasks are in high demand (Autor et al., 2003). Both developments have

led to increasing wages for high-skilled workers and constant or even decreasing

wages for low-skilled workers, resulting in an increase in the inequality of market

incomes.

From a distributive point of view, an overly high level of income inequality

seems unfair to most people. Therefore, market incomes are usually redistributed

by the government, leading to signicantly lower inequality of disposable incomes.

However, fairness is a normative rather than an economic concept, and thus the

optimal amount of redistribution remains unclear to both economists and policy

1
 Introduction

makers. In addition, a high level of redistribution is likely to create disincentives to

invest in physical and human capital and might therefore harm economic growth in

a society (Muinelo-Gallo and Roca-Sagalés, 2013).

Figure 1.1: Development of market and net income inequality

Germany United States

0.55 0.55

0.50 0.50

0.45 0.45
Gini coefficient

0.40 Gini coefficient 0.40

0.35 0.35

0.30 0.30

0.25 0.25
1960 1980 2000 2020 1960 1980 2000 2020

Market income inequality Market income inequality

Net income inequality Net income inequality

Source: Solt (2016).

However, even a high level of net income inequality is less problematic if it is

accompanied by a similarly high level of income mobility. To illustrate this, imagine

two societies, A and B. Society A is characterized by complete income immobility,

meaning that in each period, the income position of an individual is perfectly pre-

determined by their income position in the previous period. In contrast, society B

features complete income mobility, such that in each period, every individual has

an equal chance of receiving a high or low income irrespective of their income in

the previous period. In the latter case, society will be more likely to accept a high

level of inequality than in the rst case. If there exists a high level of income mo-

bility and thus economic success is directly dependent on talent, ability, and eort,

2
 Introduction

income dierences might even encourage investments in education and an extension

of working hours.

A distinction is generally made between intra- and intergenerational mobility.

While intragenerational income mobility considers the extent to which an individual

can ascend or descend the income ladder within their own working life, intergener-

ational income mobility analyzes the ascent or descent of a child relative to their

parents' position in the income distribution. Intergenerational income mobility thus

examines the question of whether and to what extent the future income of a child

is predetermined by their family background, or, as Corak (2006) puts it: Do poor

children become poor adults? This issue is closely related to the analysis of equal-

ity of opportunity and has been the subject of a broad strand of the economic and

sociological literature for several decades. However, at least partly due to a lack of

data availability in most countries, numerous questions still remain unanswered.

The present dissertation deals with intergenerational income mobility in Ger-

many as compared to the United States and is intended to be a contribution to the

currently available literature. It consists of three parts, each of which addresses a

dierent aspect of intergenerational income mobility. The overall focus is on an em-

pirical investigation using comparable data from the Socio-economic Panel (SOEP)

for Germany and the Panel Study of Income Dynamics (PSID) for the United States.

A detailed description of the structure and the main results of this dissertation is

presented in Section 1.3.

1.2 Literature review

Intergenerational income mobility has been discussed in the economic literature since

the 1970s. Early studies include Sewell and Hauser (1975), Bielby and Hauser (1977),

and Behrman and Taubman (1985). However, the results of these rst studies

are likely to be systematically biased due to measurement errors and homogenous

samples. Beginning with the seminal contributions of Solon (1989, 1992), more

3
 Introduction

recent decades have witnessed a rapid increase in the number of studies on the

transmission of income positions within families. A broad overview can, for example,

be found in Solon (1999), Björklund and Jäntti (2009), and Black and Devereux

(2011).

Figure 1.2: Country comparison of intergenerational income mobility

0.5

0.4
Intergenerational income elasticity

0.3

0.2

0.1

0.0
DK SE NO DE AU CA FI GB FR IT US

Source: Björklund and Jäntti (2009).

Notes: DK: Denmark, SE: Sweden, NO: Norway, DE: Germany, AU: Australia, CA: Canada, FI:
Finland, GB: United Kingdom, FR: France, IT: Italy, US: United States.

Intergenerational income mobility is commonly estimated by the intergenera-

tional income elasticity, which measures the share of parents' income advantage or

disadvantage that is passed on to their children. Thus, higher values for the in-

tergenerational income elasticity imply a higher persistence of income positions and

thus a lower level of intergenerational income mobility. An international comparison

of the existing literature indicates that there are major dierences among the results

for individual countries (Figure 1.2). Studies for the United States have ascertained

values ranging from 0.09 (Behrman and Taubman, 1985) to 0.61 (Mazumder, 2001).

Today the generally accepted value for the intergenerational income elasticity in the

4
 Introduction

United States is 0.4 or higher (Corak, 2006). This places the United States at the

upper end of the estimated results and makes it a country with a rather low level of

intergenerational income mobility.

Similarly low levels of intergenerational income mobility are found in the liter-

ature for France and Italy. Lefranc and Trannoy (2005) determine an intergener-

ational income elasticity of approximately 0.4 for France. Meanwhile, estimates of

around 0.5 have been established for Italy (Mocetti, 2007, Piraino, 2007). For the

United Kingdom, the ascertained values of approximately 0.3 are somewhat lower

than those for the United States. However, they are still relatively high in the

international comparison (Blanden et al., 2004, Nicoletti and Ermisch, 2007).

In contrast, the Scandinavian countries exhibit very low levels of intergenera-

tional income persistence. The estimated elasticities for Finland (Pekkarinen et al.,

2009), Norway (Nilsen et al., 2008), and Sweden (Björklund and Jäntti, 1997, Björk-

lund et al., 2012) range from 0.2 to 0.3. Hussain et al. (2009) obtain a value of only

0.14 for Denmark. Australia and Canada also exhibit comparatively low values for

the intergenerational income elasticity and thus high levels of income mobility. Leigh

(2007) nds values ranging from 0.2 to 0.3 for Australia. For Canada, Corak and

Heisz (1999) compute a value of approximately 0.2.

Germany is usually classied between the United States and the Scandinavian

countries (Schnitzlein, 2016). Existing studies, however, calculate varying results

for the intergenerational income elasticity in Germany. The estimated values range

from 0.10 (Grawe, 2004) to 0.48 (Chau, 2012). However, the majority of studies nd

values of the order of approximately 0.2 to 0.3 (Vogel, 2006, Eisenhauer and Pfeif-

fer, 2008, Schnitzlein, 2009), indicating a level of intergenerational income mobility

similar to that of the Scandinavian countries. Surprisingly often, however, studies

conducting a direct comparison with comparable data nd no signicant dierences

between Germany and the United States (Couch and Dunn, 1997, Lillard, 2001,

Couch and Lillard, 2004, Schnitzlein, 2016). The relative position of Germany in

the international comparison is thus not clearly determined.

5
 Introduction

1.3 Organization of this dissertation

This dissertation consists of three contributions. Each addresses one specic aspect

of intergenerational income mobility and is intended to be a stand-alone analysis.

All chapters use comparable data for Germany and the United States to conduct

country comparisons. As there are usually a large number of studies available for

the United States, this approach is useful for comparing the empirical results to the

existing literature. While the rst two studies are co-authored with Mustafa Çoban,

the third study is single-authored.

Structure and extent of intergenerational income mobility

Chapter 2 conducts a direct country comparison of the structure and extent of in-

tergenerational income mobility in Germany and the United States. In line with ex-

isting results, the estimated intergenerational income mobility of 0.49 in the United

States is signicantly higher than that of 0.31 in Germany. While the results for the

intergenerational rank mobility are relatively similar, the level of intergenerational

income share mobility is higher in the United States than in Germany. There are

no signicant indications of a nonlinear run of intergenerational income elasticity.

A nal decomposition of intergenerational income inequality shows both greater in-

come mobility and stronger progressive income growth for Germany compared to the

United States. Overall, we cannot identify a clear ranking of the two countries. To

conclude, several economic policy recommendations to increase intergenerational in-

come mobility in Germany are discussed. This chapter is co-authored with Mustafa

Çoban and has been published in a similar version in ORDO Yearbook of Economic

and Social Order, 67, 101-131.

6
 Introduction

Transmission channels of intergenerational income persistence

Chapter 3 examines the transmission channels of intergenerational income persis-

tence in Germany and the United States. In principle, there are two ways in which

well-o families may inuence the adult incomes of their children: rst through di-

rect investments in their children's human capital (investment eect ), and second

through the indirect transmission of human capital from parents to children (en-

dowment eect ). In order to disentangle these two eects, a descriptive as well as

a structural decomposition method are utilized. The results suggest that the in-

vestment eect and the endowment eect each account for approximately half of

the estimated intergenerational income elasticity in Germany, while the investment

eect is substantially more inuential in the United States with a share of around

70 percent. With regard to economic policy, these results imply that equality of

opportunity for children born to poor parents cannot be reached by the supply of

nancial means alone. Conversely, an ecient policy must additionally substitute

for the missing direct transmission of human capital within socio-economically weak

families. This chapter is again co-authored with Mustafa Çoban.

Intergenerational income mobility among daughters

Chapters 2 and 3 of this dissertation are restricted to the analysis of fathers and

their sons, whereas Chapter 4 explicitly focuses on the intergenerational income

mobility among daughters. The restriction to men is commonly made in the empir-

ical literature due to women's lower labor market participation. While most men

work full-time, the majority of (married) women still work only part-time or not at

all. Especially with the occurrence of assortative mating, daughters from well-o

families are likely to marry rich men and might decide to reduce their labor supply

as a result. Thus, the individual labor income of a daughter might not be a good

indicator for her actual economic status. The baseline regression analysis shows a

higher intergenerational income elasticity in Germany and a lower intergenerational

7
 Introduction

income elasticity in the United States for women as compared to men. However,

a separation by marital status reveals that in both countries unmarried women ex-

hibit a higher intergenerational income elasticity than unmarried men, while married

women feature a lower intergenerational income elasticity than married men. The

reason for the lower mobility of unmarried women turns out to be a stronger human

capital transmission from fathers to daughters than to sons. The higher mobility

of married women is driven by a weaker human capital transmission and a higher

labor supply elasticity with respect to spousal income for women as compared to

men. In order to further study the eects of assortative mating, the subsample of

married children is analyzed by dierent types of income. It shows that the esti-

mated intergenerational income elasticity of children's household incomes is even

higher than that of their individual incomes. This can be seen as an indication for

strong assortative mating. If household income is interpreted as a measure of chil-

dren's actual economic welfare, there are barely any dierences between sons and

daughters. The intergenerational income elasticity of spousal income with respect

to parental income is again relatively high, which in turn supports the hypothesis

of strong assortative mating. The elasticity of the sons-in-law with respect to their

fathers-in-law in Germany is even higher than that of the sons with respect to their

own fathers.

8
 Introduction

References

Autor, D. H., Levy, F., and Murnane, R. J. (2003). The skill content of recent

technological change: An empirical exploration. Quarterly Journal of Economics,

118 (4), 12791333.

Behrman, J. and Taubman, P. (1985). Intergenerational earnings mobility in the

United States: Some estimates and a test of Becker's intergenerational endow-

ments model. Review of Economics and Statistics, 67 (1), 144151.

Bielby, W. T. and Hauser, R. M. (1977). Response error in earnings functions for

nonblack males. Sociological Methods & Research, 6 (2), 241280.

Björklund, A. and Jäntti, M. (1997). Intergenerational income mobility in Sweden

compared to the United States. American Economic Review, 87 (5), 10091018.

Björklund, A. and Jäntti, M. (2009). Intergenerational income mobility and the role

of family background. In Salverda, W., Nolan, B., and Smeeding, T. M., editors,

Oxford Handbook of Economic Inequality, 491521.

Björklund, A., Roine, J., and Waldenström, D. (2012). Intergenerational top in-

come mobility in Sweden: Capitalist dynasties in the land of equal opportunity?

Journal of Public Economics, 96 (56), 474484.

Black, S. E. and Devereux, P. J. (2011). Recent developments in intergenerational

mobility. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics,

4B, 14871541.

Blanden, J., Goodman, A., Gregg, P., and Machin, S. (2004). Changes in intergen-

erational mobility in Britain. In Corak, M., editor, Generational Income Mobility

in North America and Europe, 122146.

Chau, T. W. (2012). Intergenerational income mobility revisited: Estimation with

9
 Introduction

an income dynamic model with heterogeneous age prole. Economics Letters, 117

(3), 770773.

Corak, M. (2006). Do poor children become poor adults? Lessons from a cross-

country comparison of generational earnings mobility. In Creedy, J. and Kalb,

G., editors, Dynamics of Inequality and Poverty, 143188.

Corak, M. and Heisz, A. (1999). The intergenerational earnings and income mobility

of Canadian men: Evidence from longitudinal income tax data. Journal of Human

Resources, 34 (3), 504533.

Couch, K. A. and Dunn, T. A. (1997). Intergenerational correlations in labor market

status: A comparison of the United States and Germany. Journal of Human

Resources, 32 (1), 210232.

Couch, K. A. and Lillard, D. R. (2004). Non-linear patterns of intergenerational

mobility in Germany and the United States. In Corak, M., editor, Generational

Income Mobility in North America and Europe, 190206.

Ebenstein, A., Harrison, A., McMillan, M., and Phillips, S. (2014). Estimating the

impact of trade and oshoring on American workers using the Current Population

Surveys. Review of Economics and Statistics, 96 (4), 581595.

Eisenhauer, P. and Pfeier, F. (2008). Assessing intergenerational earnings persis-

tence among German workers. Journal for Labour Market Research, 41 (23),

119137.

Grawe, N. D. (2004). Reconsidering the use of nonlinearities in intergenerational

earnings mobility as a test for credit constraints. Journal of Human Resources,

39 (3), 813827.

Hussain, M. A., Munk, M. D., and Bonke, J. (2009). Intergenerational earnings

mobilities: How sensitive are they to income measures? Journal of Income Dis-

tribution, 18 (34), 7992.

10
 Introduction

Lefranc, A. and Trannoy, A. (2005). Intergenerational earnings mobility in France:

Is France more mobile than the US? Annals of Economics and Statistics, 78,

5777.

Leigh, A. (2007). Intergenerational mobility in Australia. B.E. Journal of Economic

Analysis & Policy, 7 (2), Article 6.

Lillard, D. R. (2001). Cross-national estimates of the intergenerational mobility in

earnings. Quarterly Journal of Economic Research, 70 (1), 5158.

Mazumder, B. (2001). Earnings mobility in the US: A new look at intergenerational

inequality. Working Paper 2001-18, Federal Reserve Bank of Chicago, Chicago,

United States.

Mocetti, S. (2007). Intergenerational earnings mobility in Italy. B.E. Journal of

Economic Analysis & Policy, 7 (2), Article 5.

Muinelo-Gallo, L. and Roca-Sagalés, O. (2013). Joint determinants of scal policy,

income inequality and economic growth. Economic Modelling, 30, 814824.

Nicoletti, C. and Ermisch, J. F. (2007). Intergenerational earnings mobility:

Changes across cohorts in Britain. B.E. Journal of Economic Analysis & Pol-

icy, 7 (2), Article 9.

Nilsen, Ø. A., Vaage, K., Aakvik, A., and Jacobsen, K. (2008). Sources of mea-

surement errors in earnings data: New estimates of intergenerational elasticities.

Working Papers in Economics 04/08, Department of Economics, University of

Bergen, Bergen, Norway.

Pekkarinen, T., Uusitalo, R., and Kerr, S. (2009). School tracking and intergenera-

tional income mobility: Evidence from the Finnish comprehensive school reform.

Journal of Public Economics, 93 (78), 965973.

Piraino, P. (2007). Comparable estimates of intergenerational income mobility in

Italy. B.E. Journal of Economic Analysis & Policy, 7 (2), Article 1.

11
 Introduction

Schnitzlein, D. D. (2009). Structure and extent of intergenerational income mobility

in Germany. Journal of Economics and Statistics, 229 (4), 450466.

Schnitzlein, D. D. (2016). A new look at intergenerational income mobility in Ger-

many compared to the U.S. Review of Income and Wealth, 62 (4), 650667.

Sewell, W. H. and Hauser, R. M. (1975). Education, Occupation and Earnings:

Achievement in the Early Career. Academic Press, New York, United States.

Solon, G. (1989). Biases in the estimation of intergenerational earnings correlations.

Review of Economics and Statistics, 71 (1), 172174.

Solon, G. (1992). Intergenerational income mobility in the United States. American

Economic Review, 82 (3), 393408.

Solon, G. (1999). Intergenerational mobility in the labor market. In Ashenfelter, O.

and Card, D., editors, Handbook of Labor Economics, 3A, 17611800.

Solt, F. (2016). The Standardized World Income Inequality Database. Social Science

Quarterly, 97 (5), 12671281.

Vogel, T. (2006). Reassessing intergenerational mobility in Germany and the United

States: The impact of dierences in lifecycle earnings patterns. SFB 649 Discus-

sion Papers 2006-055, Humbolt-Universität zu Berlin, Berlin, Germany.

12
Chapter 2

Structure and Extent of

Intergenerational Income Mobility∗

2.1 Introduction

The high level of income inequality is currently one of the most important socio-

political issues in both Germany and the United States. Closely related to income

inequality but less intensely discussed in public is the topic of income mobility. The

relationship between income inequality and income mobility can be illustrated with

the help of a simple image. If one imagines the interpersonal income distribution

as a ladder upon the rungs of which the respective income earners are located,

then income inequality determines the distance between the individual rungs. In

contrast, income mobility represents the probability of an individual to ascend or

descend from a particular rung on the ladder (Chetty et al., 2014).

Generally, one distinguishes between intra- and intergenerational income mo-

bility. While intragenerational income mobility considers the extent to which an

individual person can ascend or descend the income ladder within their own work-

ing life, intergenerational income mobility describes the ascent or descent of a child

relative to their parents' position on the income ladder. Intergenerational income

mobility thus examines the question of whether and to what extent the adult income

of a child is determined by their background. Put more simply: Do poor children

become poor adults and vice versa?

∗
This chapter is co-authored with Mustafa Çoban and has been published in a similar version
in Ordo Yearbook of Economic and Social Order, 67, 101131.

13
 Structure and Extent of Intergenerational Income Mobility

A high level of intergenerational income mobility is desirable from both a dis-

tributive and an allocative point of view. On the one hand, society perceives it

as unfair when the income of a child is determined to a large extent by that of

their parents, and thus the future prospects of children from low-income families

are largely eliminated. On the other hand, if one assumes that initial abilities are

equally distributed among income classes, but children from low-income families are

unable to obtain well-paid employment reective of their abilities, then society is

failing to use its resources in an ecient manner (Schnitzlein, 2008).

However, high intergenerational income persistence cannot generally be inter-

preted as a lack of equal opportunity. Instead, children from wealthier households

may on average demonstrate a stronger preference for human capital investments

than children from poorer households. Reasons for this may be the intergenerational

transfer of aspirations, skills, and occupational choices within the family. Inequal-

ities that are driven by these determinants are accepted within a market economy

(Roemer, 2004). In contrast, high intergenerational income persistence is indicative

of a lack of equal opportunity if it is exogenously inuenced by institutional condi-

tions, credit market constraints for poor households, or other social factors that are

beyond the control of the individual.

Intergenerational income mobility is commonly estimated by the intergenera-

tional income elasticity (Section 2.2.3), where, for example, a value of 0.3 means

that 30 percent of the income advantage or disadvantage of the parents is passed on

to their children. Thus, if a father's income is 10 percent higher than the average

income in the parents' generation, the expected income of his children is 3 percent

higher than the average income in the children's generation. Higher values for the

intergenerational income elasticity therefore imply a higher persistence of income

positions and thus a lower level of intergenerational income mobility. A comparison

of the existing literature on intergenerational income mobility shows that there are

considerable dierences between individual countries (Solon, 1999, Björklund and

Jäntti, 2009, Black and Devereux, 2011). The consensus estimate for the intergen-

14
 Structure and Extent of Intergenerational Income Mobility

erational income elasticity in the United States lies between 0.4 and 0.5 (Corak,

2006). Thus, in the international comparison, the United States is located at the

upper end of the ascertained values and is therefore a country with a rather low

level of intergenerational income mobility. The Scandinavian countries, in contrast,

exhibit very low levels of intergenerational income elasticity with values estimated

at around 0.2 (Nilsen et al., 2008, Hussain et al., 2009, Pekkarinen et al., 2009,

Björklund et al., 2012). Germany is generally classied between the United States

and the Scandinavian countries. The estimates obtained are of the order of approx-

imately 0.2 to 0.3 (Vogel, 2006, Eisenhauer and Pfeier, 2008, Schnitzlein, 2009).

However, Schnitzlein (2016) nds no signicant dierences between the intergener-

ational income elasticities in Germany and the United States.

This chapter conducts a direct comparison of the structure and extent of inter-

generational income mobility in Germany and the United States. Consistent with

existing results, the intergenerational income elasticity in the United States is found

to be higher than in Germany. When comparing intergenerational rank mobility,

however, the results for the two countries are relatively similar. In terms of intergen-

erational income share mobility, greater dierences exist between Germany and the

United States. With each higher percentile, the income share mobility of the sons

in the United States drops by a higher amount when compared to their fathers than

in Germany. For both countries, the results of the quantile regressions provide no

evidence of nonlinearities. The nal decomposition of intergenerational income in-

equality shows both greater income mobility and stronger progressive income growth

for Germany than for the United States. Section 2.2 subsequently describes the data

used and provides an overview of the various mobility measures. The results of the

estimates are presented in Section 2.3. Finally, Section 2.4 includes several economic

policy recommendations to increase intergenerational income mobility in Germany

and is followed by a brief summary in Section 2.5.

15
 Structure and Extent of Intergenerational Income Mobility

2.2 Data and mobility measures

In order to examine intergenerational income mobility empirically, individual data

are required for at least two generations. Long-term panel surveys of households that

start capturing information on children while they are still living with their parents

and follow them into the older adult years are suitable for this purpose (Corak,

2006). In addition, in order to conduct a country comparison, it is necessary that

the data used are highly comparable regarding the survey design, the survey method,

and the survey period. In this study, we use the Socio-economic Panel (SOEP) for

Germany and the Panel Study of Income Dynamics (PSID) for the United States.

Both records collect information on all adult persons of a household and survey

them repeatedly in the subsequent years. Thus, children who leave their parents'

homes and establish their own households can continue to be covered over time.

Both surveys are part of the Cross-National Equivalent File (CNEF) project, which

oers a harmonized panel data set of the underlying national household surveys

(Frick et al., 2007). In particular, it provides a reliable data basis for international

comparisons of income, taxes, and transfers. The individual annual labor income in

the CNEF used in this study includes wages and salaries from both paid employment

and self-employment as well as bonus payments, income from overtime, and prot

sharing (Lillard, 2013, Grabka, 2014).

2.2.1 Measurement errors and life-cycle bias

In order to measure lifetime income, all of a respondent's income statements over

1
the entire working life would be required. In the case of an academic, for example,

income observations over the course of 35 to 40 years would need to be available

(Schnitzlein, 2009). However, with such a long survey period, the number of people

1 The lifetime income of a person generally includes both labor and capital income. Since in
surveys the collection of capital income is linked to problems, here the concept of income refers to
the labor income of a person.

16
 Structure and Extent of Intergenerational Income Mobility

who continue to participate in the survey is reduced. This so-called panel mor-

tality can correlate with certain characteristics of the respondents (e.g., income or

education), resulting in a relatively homogeneous longitudinal sample (Fitzgerald

et al., 1998). This circumstance can lead to substantial distortions of the estimation

parameters (panel attrition bias ) (Solon, 1989, 1992).

Therefore, lifetime incomes are approximated by means of annual income obser-

vations. These income statements consist of a permanent as well as a uctuating

component, where the second causes lifetime income to be determined with mea-

surement errors (Solon, 1989, 1992, Zimmerman, 1992). Thus, if parental income is

approximated by income data from only one particular point in time, the classical

errors-in-variables problem occurs (Wooldridge, 2010). This leads to a system-

atic downward bias of the estimated intergenerational income elasticity (attenuation

bias ). Solon (1992) proposes to form an average of ve valid annual income obser-

vations for the parental generation in order to reduce the variance of the uctuating

component. This procedure does not completely eliminate the bias, but it can sig-

nicantly reduce it. Since the direction of the bias is known, an estimate of the

intergenerational income elasticity can be interpreted as a lower bound for the true

estimation parameter. In the approximation of children's lifetime income, measure-

ment errors only lead to higher standard errors.

In addition, Haider and Solon (2006) point out that the approximation of chil-

dren's lifetime income depends on the chosen stage of life. On the one hand, in-

dividual income during the working life assumes a hump-shaped run, so that the

income at the beginning of the working life is lower and thus the lifetime income of

a person is underestimated. On the other hand, dierences in income between high-

and low-skilled workers are smaller at the beginning of their working lives and only

increase over time. If incomes are thus observed at the beginning of the working

life, this leads to an underestimation of intergenerational income elasticity (life-cycle

bias ). This circumstance is veried by Böhlmark and Lindquist (2006) for Sweden

and Brenner (2010) for Germany. For the United States, Haider and Solon (2006)

17
 Structure and Extent of Intergenerational Income Mobility

show that for the sons, the age range between the mid-30s and the mid-40s produces

a good approximation of the lifetime income. Schnitzlein (2016) uses the income of

sons between 35 and 42 years of age for Germany.

2.2.2 Sample denition

The selected baseline samples from the SOEP and the PSID are dened congruently

so as to ensure reliable comparability of the results. The analysis is based on data

from the years of 1984 to 2013. The individual annual labor income is used. We

2
exclude imputed income data from the SOEP sample. All income statements are

3
deated to the year 2010. In order to be able to compare the results with the

existing literature, annual real incomes of less than 1,200 Euro/US dollar are not

included in the baseline samples. To avoid a bias due to wage developments in East

Germany after reunication, the analysis for Germany is limited to the persons who

lived in West Germany in 1989 (Schnitzlein, 2009).

The generation of the parents is restricted to the income observations of the

4
fathers and the generation of the children to the income observation of the sons.

Fathers' incomes are drawn from the period of 1984 to 1993, from which at least ve

valid income observations must be available. The lifetime income of the fathers is

approximated by the formation of the average of the annual incomes. Only income

observations from the ages of 30 to 55 are considered. Thus, the fathers belong to

the birth cohorts of the period from 1933 to 1959. The income observations of the

sons are drawn from the years of 2003 to 2013, during which time period at least

one valid income observation must be available. Again, the lifetime income of the

2 Missing income statements are estimated in the SOEP with the help of personal and household
characteristics as well as past income data (Frick et al., 2012). The CNEF-PSID features no im-
puted income data.
Consumer Price Index and, for the PSID, the Consumer Price Index of
3 For the SOEP, the
All Urban Consumers and All Items based on the recommendation of Grieger et al. (2009) are
utilized.
4 This limitation is due to the divergent labor market participation of women in both countries,
which can lead to a bias of dierences in intergenerational income elasticity. While in the United
States female labor market participation was on average at 54.2 percent in the 1980s and at 59.5
percent in the 2000s, Germany features values of 41.4 percent and 50.8 percent, respectively (World
Bank, 2017).

18
 Structure and Extent of Intergenerational Income Mobility

sons is approximated by the formation of the average of the annual incomes. Only

incomes from the age of 35 to 42 years are taken into account. Thus, the sons belong

to the birth cohorts from 1961 to 1978, which do not overlap with the cohorts of

their fathers.

Table 2.1: Descriptive statistics

Fathers Sons
Mean Std. Dev. Mean Std. Dev.

SOEP
Income 40,441.96 19,611.62 46,868.19 27,724.28

Age 46.84 4.53 38.13 1.80

Father-son pairs 354

PSID
Income 64,070.24 59,633.14 67,199.29 69,580.75

Age 43.76 5.42 37.89 1.88

Father-son pairs 601

Source: SOEP (1984-2013), PSID (1984-2013).

A total of 354 father-son pairs are thus recorded in the SOEP and 601 father-son

pairs in the PSID (Table 2.1). On average, the sons earn more than their fathers

in both countries. The average income of the sons is 15.9 percent higher than the

average income of the fathers in Germany, while it is only 4.9 percent higher in the

United States. The average age of the fathers is mid-40s in both countries, older

than the sons, whose average age is late-30s. The younger age of the sons might also

determine the observed higher variance in incomes.

The logarithmized incomes of the fathers and sons exhibit a positive correlation

(Figure 2.1). The slope of the line of best t from the bivariate ordinary least squares

(OLS) regression is higher for the United States than for Germany. However, it also

becomes clear that the income data points in both countries are heavily scattered

around the regression line. In order to examine the simple linear relationship more

closely, the course of the bivariate Nadaraya-Watson (NW) estimation is additionally

19
 Structure and Extent of Intergenerational Income Mobility

depicted. Both countries show deviations compared to the OLS estimation. How-

ever, the 95 percent condence intervals include the OLS regression line over nearly

the entire distribution of paternal income and deviations on the upper and lower

ends might be caused by inuential outliers. Thus, it cannot be concluded from the

bivariate evidence that the intergenerational income elasticity changes signicantly

along the income distribution of the fathers.

Figure 2.1: Intergenerational income correlation

Germany United States

13 14

12
12
Log. income of son

Log. income of son

11

10
10

9 8

6
9 10 11 12 9 10 11 12 13 14
Log. income of father Log. income of father

Income data OLS Income data OLS

NW 95% confidence interval NW 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a range based on
Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson.

2.2.3 Intergenerational mobility measures

Intergenerational income elasticity

The theoretical basis for the relationship between the income of parents and the

income of their children is expressed by the model of Becker and Tomes (1979,

20
 Structure and Extent of Intergenerational Income Mobility

1986). The starting point is a family comprising two generations which maximizes

its utility by dividing its disposable income between consumption and investment

in the human capital of its children. Solon (2004) simplies this approach in order

to rationalize the intergenerational income elasticity usually estimated in empirical

studies by

log(yis ) = β0 + β1 log(yif ) + γx0i + εsi . (2.1)

For each family i, the lifetime income of the son yis and the lifetime income of

the father yif are logarithmized. The intercept β0 can then be interpreted as the

average logarithmized lifetime income in the generation of the son, and the slope β1
is the searched-for intergenerational income elasticity. It states that an increase in

the father's lifetime income by 1 percent increases the son's lifetime income by an

average of β1 percent. If β1 = 0, the lifetime income of the son is independent of

the father's lifetime income and assumes the average value of the son's generation.

The higher the value of β1 , the stronger the link between the lifetime income of

a father and his son is, and consequently, the lower the intergenerational income

mobility. If the variance of the logarithmized lifetime incomes of the fathers and

sons is approximately equal, β1 can also be interpreted as the correlation between the
logarithmized lifetime incomes of the two generations (Solon, 2004). In the selected

samples, the income observations of the sons and the fathers are sometimes measured

at dierent times of their lives. Moreover, the number of valid observations varies

between respondents. Thus, the vector xi includes polynomials of the average age

of the father and the son, respectively, as well as the number of valid observations

5
of the son (Schnitzlein, 2016). Deviations from the predicted value due to factors

orthogonal to the income of the father are summarized in the idiosyncratic error

term εsi .

5 If the generation of the fathers and the generation of the sons exhibit dierent age-income
proles, the estimated intergenerational income elasticity might be biased (Fertig, 2003). How-
ever, the bootstrapped Hausman test for the intergenerational income elasticity with commonly
estimated age-income proles and separately estimated age-income proles yields no signicant
dierence for Germany (p = 0.6460) and the United States (p = 0.1672), respectively.

21
 Structure and Extent of Intergenerational Income Mobility

Intergenerational transition matrices

While the intergenerational income elasticity is a useful summary measure of relative

intergenerational mobility, it has some limitations. For example, it is not informative

regarding dierences between upward and downward mobility and does not consider

nonlinearities along the income distribution (Bratberg et al., 2017). As a starting

point to overcome these issues, estimated transition matrices provide the possibility

of an illustrative representation. Here, the position of the son in the children's in-

come distribution is conditioned to the position of the father in the parents' income

distribution. More specically, each cell cjk of the estimated transition matrix can

be interpreted as the probability that a son born to a father from quintile j reaches

quintile k. Again, the vector xi includes polynomials of the average age of the father

and the son as well as the number of valid observations of the son (Fertig, 2003).

cjk = P(qis = k | qif = j, xi ), j, k = 1, ..., 5 (2.2)

Intergenerational transition matrices provide detailed information about the up-

ward and downward mobility at certain income quintiles. They thus supplement

the intergenerational income elasticity by determining to where sons from dierent

backgrounds migrate within the income distribution.

Intergenerational rank mobility

Since estimated transition matrices cannot illustrate movements of the sons within

the respective quintiles, intergenerational rank mobility provides another way to

determine upward and downward mobility in more detail (Aaberge and Mogstad,

2014, Bratberg et al., 2017). Intergenerational rank mobility (RM) measures the

expected dierence between the percentile of a son psi and the percentile of his

22
 Structure and Extent of Intergenerational Income Mobility

father pfi conditioned to the percentile aliation of the father:

RM(p) = E(psi − pfi | pfi = p), p = 1, ..., 100 (2.3)

It thus provides additional information on how the mobility of the sons varies along

the income distribution of the fathers (Bhattacharya and Mazumder, 2011, Chetty

et al., 2014, Mazumder, 2014). A further advantage of intergenerational rank mobil-

ity in comparison to intergenerational income elasticity is that it is relatively robust

6
to measurement issues and life-cycle bias (Nybom and Stuhler, 2016).

Countries with similarly high levels of intergenerational income elasticity may ex-

hibit dierent levels of intergenerational rank mobility if they dier greatly in terms

of income inequality. This is due to the fact that for income recipients in countries

with higher income inequality, it is more dicult to reach higher ranks, because the

absolute income limits of the percentiles are further apart from one another than

in a country with lower income inequality. Thus, combining intergenerational rank

mobility with income inequality allows further conclusions for a country comparison.

Intergenerational income share mobility

While intergenerational rank mobility measures relative positional movements, it

does not consider the distance between individual ranks in terms of absolute income

dierences. In contrast, intergenerational income share mobility provides a hybrid

measure containing aspects of both absolute and relative mobility. In addition, it

also allows us to compare absolute income changes that are measured using dierent

currencies (Bratberg et al., 2017).

Intergenerational income share mobility (IS) is dened as the expected dierence

between a child's income relative to their generation's average income and their

parents' income relative to the parental generation's average income conditioned to

6 According to Chetty et al. (2014), rank persistence stabilizes at the age of 30. Mazumder
(2014) shows that by the age of 40, the rank persistence in the PSID no longer exhibits a downward
bias. Thus, by limiting our sample, we meet both requirements.

23
 Structure and Extent of Intergenerational Income Mobility

the percentile aliation of the father:

!
yis yif
IS(p) =E s
− f
pfi =p , p = 1, ..., 100 (2.4)
E(yi ) E(yi )

As we use a balanced panel of families in each generation, this measure is equal

to the change in a family's share of their generation's total income scaled by the

population of the generation. The estimation of the intergenerational income rank

mobility and the intergenerational income share mobility is carried out with the aid

of nonparametric mobility curves with the respective OLS estimator being used as

a benchmark (Aaberge and Mogstad, 2014).

Quantile regressions

Until now, it has been assumed that the relationship between the logarithmized

income of fathers and their sons is linear, i.e., that the intergenerational income

elasticity is constant along the entire income distribution. However, Becker and

Tomes (1986) already pointed out that the intergenerational income relationship

can assume a concave run when poor families experience credit market constraints

that do not apply for rich families. Rich families will then invest in the human

capital of their children until the marginal costs equal the marginal rate of return.

Therefore, the expected relation between earnings of parents and children in rich

families depends solely on the expected relation between their endowments λ. In

contrast, credit-constrained families might be forced to invest less than the optimal

amount in their children's human capital. This means that a small increase in a

poor father's income will increase his child's income by more than λ. The intergen-

erational income persistence will then be more pronounced for poor families than

for rich families, creating a concave intergenerational earnings relationship (Figure

2.2.a).

24
 Structure and Extent of Intergenerational Income Mobility

Figure 2.2: Nonlinearities in intergenerational income elasticity

Log. income of son (a) Concavity (a) Convexity

Log. income of son

Log. income of father Log. income of father

Source: Bratsberg et al. (2007).

However, a concave run of the intergenerational income elasticity neither needs

to follow from credit market constraints, nor is market failure implied by concavity.

If the income of a father correlates with the unobservable talent of his son, credit

market constraints do not necessarily imply a concave relationship. In this case,

poor fathersregardless of whether credit market constraints existwill reduce in-

vestments in the human capital of their sons as a result of a lower expected rate

of return. Likewise, a concave run is not a clear indication for credit market con-

straints. The relationship might be triggered by institutional, social, or unobservable

circumstances which inuence poor and rich families in dierent ways (Grawe, 2004).

On the other hand, a convex run of the intergenerational income elasticity can

be observed if educational policy is designed in such a way as to ensure a basic level

of human capital for all sons, regardless of their fathers' income. Then, particularly

at the bottom of the parents' earnings distribution, the slope of the regression line

is equal to λ. Beyond this socially guaranteed level, all families experience credit

25
 Structure and Extent of Intergenerational Income Mobility

market constraints, such that the total amount of human capital investment in the

son is dependent on paternal income, i.e., the slope of the regression line is again

7
higher than λ (Bratsberg et al., 2007). Consequently, the intergenerational income

persistence among poor families will be lower than among rich families, resulting in

a convex run of the intergenerational income elasticity (Figure 2.2.b).

According to this reasoning, countries with a largely public education system

will likely exhibit a convex run of the intergenerational income elasticity, while

in countries with a high level of privatization of the education system, a concave

run is assumed. In 2013, the share of private spending in the German education

system amounted to 13.5 percent, whereas the United States exhibited a share of

31.8 percent (OECD, 2016). The curve of intergenerational income elasticity is thus

assumed to feature a rather convex run in Germany and a rather concave run in the

United States.

Decomposition of intergenerational income inequality

If the observed fathers and sons are interpreted as representatives of their respective

families at two dierent points in time, income inequality and intergenerational

income mobility can be considered together. Jenkins and Van Kerm (2006) provide

an analytical framework within which changes in income inequality G(ν) over time

can be additively decomposed into a progressivity component P(ν) and a mobility

component R(ν):

∆G(ν) = R(ν) − P(ν), (2.5)

8
where ν represents the inequality aversion of the society. While Jenkins and Van

Kerm (2006) utilize intragenerational income mobility to decompose income inequal-

ity, we transfer this method to intergenerational income mobility. We thus interpret

the progressivity component P(ν) as the change in income inequality when relative

7 This situation can be accounted for by the fact that the optimal human capital investment
of the fathers grows with the increasing talent of the sons (Han and Mulligan, 2001, Grawe and
Mulligan, 2002).
8 The conventional Gini coecient is obtained with ν = 2.

26
 Structure and Extent of Intergenerational Income Mobility

incomes between families change, but all sons take on the respective income ranks of

their fathers. If income growth is more pronounced among the lower income quan-

tiles, i.e., income growth is progressive (pro-poor ), P(ν) > 0 and therefore leads to a
reduction of income inequality. If, in contrast, income growth is concentrated among

the upper income quantiles, i.e., income growth is regressive (pro-rich ), P(ν) <0
and thus reinforces income inequality. In the same manner, the mobility component

R(ν) is interpreted as the change in income inequality when the income ranks of

the sons in comparison to those of their fathers change but the relative incomes of

the sons equal the relative incomes of their fathers. When there is no reranking,

R(ν) = 0, and otherwise, R(ν) > 0. Thus, for a given level of P(ν), a higher R(ν)

will lead to a rise in income inequality.

In conclusion, Equation (2.5) states that income inequality is reduced by pro-

gressive income growth unless more than oset by concomitant income mobility.

These mutually compensatory eects can also explain the paradox that increasing

income inequality is compatible with progressive income growth. If poor families

benet relatively more from income growth, they move upward within the income

distribution and the income gap between poor and rich families diminishes such that

the overall income inequality declines. However, some of the initially poor families

might not only be able to catch up relative to richer families, but also overtake some

of them. This situation counteracts the reduction in income inequality.

2.3 Empirical results

2.3.1 Descriptive evidence

Comparing income inequality in the fathers' and the sons' generations in Germany

and the United States, respectively, it can be observed that the Gini coecient is

lower in Germany in both generations but has increased in both countries over time

(Table 2.2). While in the United States income inequality rose by 7.84 Gini points

27
 Structure and Extent of Intergenerational Income Mobility

(23.42 percent) from an initial value of 33.48 Gini points to a nal value of 41.32

Gini points, it increased by 9.11 Gini points (42.49 percent) from an initial value of

21.44 Gini points to a nal value of 30.55 Gini points in Germany. Thus, the level of

income inequality has increased more sharply in Germany both in terms of absolute

and relative values.

Table 2.2: Income inequality

Generation of the fathers Generation of the sons

(1984-1993) (2003-2013)
Germany 21.44 30.55

United States 33.48 41.32

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Income inequality is estimated based on unweighted samples. A comparison with
weighted values shows that the eects of panel mortality and selection bias are minor. The de-
marcations from Section 2.2.2 were applied, but the sample was not limited to father-son pairs.

Figure 2.3 shows that there were both winners and losers as a result of this devel-

opment. The quantile curves for Germany and the United States show the share of

the total income covered by the respective percentile of the income distribution (left

partial gure). Percentiles with values smaller than one claim a disproportionately

low share of the total income for themselves, while percentiles with values greater

than one claim a disproportionately high share. The income share curves assume

a slightly s-shaped run in Germany, while the United States exhibits rather convex

curves. In Germany, fathers from the 65th and sons from the 62nd percentile on

possess a disproportionate share of total income, while in the United States fathers

from the 61st and sons from the 67th percentile on receive a disproportionate share

of total income. Thus, the two countries do not dier much with regard to the

proportionality limit. However, the empirical picture changes when considering the

top income percentile. The top 1 percent of income earners in Germany receive 3.7

percent of the total income in the fathers' generation and 4.8 percent in the sons'

generation, while in the United States these values are found to be 6.0 percent for

28
 Structure and Extent of Intergenerational Income Mobility

9
the fathers and 9.1 percent for the sons. The quantile curves of the two generations

intersect at the 53rd percentile in Germany and at the 83rd percentile in the United

States. This means that in Germany, just over half of the sons' generation is poorer

compared to their fathers' generation. In the United States, this share reaches four-

fths of the sons' generation (right partial gure). Measured in percentage points,

the lower percentiles of the sons' generation must accept greater losses in Germany

than in the United States. The percentile with the greatest loss in Germany loses

0.32 percentage points (66.65 percent) in comparison to the percentile of the fathers'

generation, while the maximum loss in the United States is 0.12 percentage points

(43.42 percent). Thus, on the one hand, the drop at the lower end of the income

distribution in Germany is stronger than in the United States. On the other hand,

the share of losers in the total population in the United States is greater than in

Germany.

9 The analysis of the top incomes in the SOEP and the PSID should be treated with caution,
since high-income earners are systematically less likely to provide information about their income.
The values can thus be biased downwards and are to be regarded as a lower limit for the true
parameter.

29
 Structure and Extent of Intergenerational Income Mobility

Figure 2.3: Income share curves

Germany

5 3

Intergenerational difference (in % points)

4
Income share (in %)

1
2

1
0

0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Income percentile Income percentile

Generation of fathers Income share difference

Generation of sons 95% confidence interval

United States

10 6
Intergenerational difference (in % points)

8
Income share (in %)

4
6

4
2

0 0

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Income percentile Income percentile

Generation of fathers Income share difference

Generation of sons 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Condence intervals were calculated using paired bootstrap resampling with 1,000 repli-
cations.

30
 Structure and Extent of Intergenerational Income Mobility

2.3.2 Intergenerational income elasticity

If the samples of the two countries are limited to the observed father-son pairs, the

simple intergenerational income elasticity can be determined using OLS estimations

(Table 2.3).

Table 2.3: Intergenerational income elasticity

Germany United States

Dependent variable: Log. income (son)

Log. income (father) 0.3114*** 0.3180*** 0.4929*** 0.4639***

(0.0801) (0.0777) (0.0722) (0.0723)

Age (son) 0.3025 -1.2207

(1.2841) (1.2906)

2
Age (son) -0.0039 0.0163

(0.0167) (0.0168)

Age (father) -0.0347 0.0208

(0.1258) (0.1070)

2
Age (father) 0.0005 -0.0002

(0.0014) (0.0012)

Observations (son) 0.0291 0.1336**

(0.0280) (0.0544)

Observations 354 354 601 601

2
R 0.0448 0.0722 0.1080 0.1397

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Estimations for the SOEP are based on non-imputed income data. The intergenerational
income elasticities have been determined for an annual lower income limit of 1,200 Euro/US dol-
lar. Standard errors are clustered at the family level and were calculated using paired bootstrap
resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

For Germany, a value of 0.3114 is obtained, while in the United States, the

value is 0.4929. According to this, 31 percent of the father's income advantage or

disadvantage is passed on to his son in Germany and 49 percent of the father's income

advantage or disadvantage is passed on to his son in the United States. Including

polynomials of the average age of the father and the son as well as the number

of valid observations of the son, the estimates change only slightly. Therefore, we

31
 Structure and Extent of Intergenerational Income Mobility

can assume that the selected age limits are chosen correctly. Thus, at rst glance,

intergenerational income elasticity is higher in the United States than in Germany.

Table 2.4: Intergenerational income elasticity for dierent lower income limits

Germany United States

Without imputed With imputed

income data income data

Income > 1,200 Euro/US dollar

IIE 0.3180*** 0.2889*** 0.4639***

(0.0777) (0.0809) (0.0723)

Observations 354 392 601

2
R 0.0722 0.0743 0.1397

Income > 6,000 Euro/US dollar

IIE 0.3238*** 0.3360*** 0.4600***

(0.0711) (0.0739) (0.0622)

Observations 348 387 583

2
R 0.1032 0.1199 0.1688

Income > 12,000 Euro/US dollar

IIE 0.3591*** 0.3666*** 0.4187***

(0.0704) (0.0667) (0.0606)

Observations 337 376 557

2
R 0.1299 0.1515 0.1533

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Other control variables include polynomials of the father's and the son's age as well as
the number of valid observations of the son. Standard errors are clustered at the family level
and were calculated using paired bootstrap resampling with 1,000 replications. *** p < 0.01, **
p < 0.05, * p < 0.1. IIE: Intergenerational income elasticity.

The baseline estimations include observations with earned incomes of at least

1,200 Euro/US dollar per year. However, such a low income is not sucient for the

survival of a single individual in either country without additional income sources

or social transfers. Thus, the estimates are repeated for lower income limits of 6,000

Euro/US dollar and 12,000 Euro/US dollar per year (Table 2.4). For Germany,

the estimates were conducted both with and without imputed income data. The

two countries show dierent developments of intergenerational income elasticity.

32
 Structure and Extent of Intergenerational Income Mobility

The intergenerational income elasticity in the United States decreases with a rising

lower income limit from an estimated value of 0.4639 to an estimated value of 0.4187.

The intergenerational income elasticity in Germany increases from 0.3180 to 0.3591

without imputed incomes and from 0.2889 to 0.3666 with imputed incomes. Thus,

the gap between the United States and Germany is shortened by an increase in the

income limit, even though the United States exhibits higher intergenerational income

elasticities across all lower income limits. Since with a rising lower income limit, an

increasingly larger piece is cut o at the left-hand side of the income distribution,

the estimates provide evidence that the intergenerational income elasticity might

10
dier along the income distribution.

2.3.3 Intergenerational rank and income share mobility

As a starting point, estimated transition matrices oer the possibility to further

examine intergenerational income mobility by providing information about nonlin-

earities along the income distribution and dierences between upward and downward

mobility. Here, the position of the son in the children's income distribution is con-

ditioned to the position of his father in the parents' income distribution. More

specically, each value indicates the probability of a son to reach a certain quintile

depending on his father's quintile aliation. Thus, in a completely mobile society,

all cells should assume a value of 0.2. The income position of the son is then inde-

pendent of the income position of his father. In a completely immobile society, on

the other hand, the main diagonal assumes a value of one with a value of zero being

assigned to all remaining cells. In this case, the income position of the son can be

perfectly predicted from the income position of his father.

10 Considering the birth cohorts of the fathers and sons as well as including periods of unem-
ployment, we nd no major dierences in the intergenerational income elasticity in Germany and
the United States after accounting for inuential observations according to Belsley et al. (1980)
(see Table 2.8 in the Appendix).

33
 Structure and Extent of Intergenerational Income Mobility

Table 2.5: Estimated transition matrices

Germany
Income quintile (son)

Income quintile (father) 1 2 3 4 5

1 27.31 24.52 21.87 16.19 10.11

2 26.53 24.31 22.08 16.59 10.48

3 15.28 18.88 23.24 23.52 19.08

4 11.57 15.85 22.11 25.95 24.51

5 7.66 11.73 19.13 27.70 33.78

United States
Income quintile (son)

Income quintile (father) 1 2 3 4 5

1 39.16 21.79 18.33 11.49 9.23

2 27.89 20.52 21.26 15.82 14.51

3 17.12 16.30 21.70 20.74 24.14

4 13.20 13.81 20.52 22.31 30.15

5 8.29 9.77 17.02 22.92 42.01

Source: SOEP (1984-2013), PSID (1984-2013).
Notes: The income positions of fathers and sons are based on the unweighted income distribu-
tion of their respective generation. Other control variables include polynomials of the father's
and the son's age as well as the number of valid observations of the son.

Along the main diagonal, the results for Germany and the United States dier

11
strongly from one another only in the lowest and the highest quintiles (Table 2.5).

In the United States, the probability of a son whose father is located in the lowest

quintile remaining in that quintile is 39.16 percent, whereas the probability is 27.31

percent in Germany. Likewise, the probability of a son whose father is located in the

highest quintile remaining in that quintile is 42.01 percent in the United States and

33.78 percent in Germany, respectively. However, the upward mobility of sons from

the higher quintiles is more pronounced in the United States. Consequently, the

downward mobility of sons from the higher quintiles is slightly higher in Germany.

Overall, intergenerational persistence at the bottom and the top of the income dis-

11 Since income quintiles are an ordinal variable, the transition probabilities of the sons are
estimated using ordered logistic regressions (Fertig, 2003, Schnitzlein, 2009). Subsequently, the
estimated transition probabilities are averaged over the entire sample.

34
 Structure and Extent of Intergenerational Income Mobility

tribution appears to be stronger in the United States than in Germany, although

the dierences are not very pronounced.

Figure 2.4: Intergenerational rank mobility

Germany United States

100 100

80 80
Income rank mobility of son (in percentiles)

Income rank mobility of son (in percentiles)

60 60

40 40

20 20

0 0

−20 −20

−40 −40

−60 −60

−80 −80

−100 −100
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Income percentile of father Income percentile of father

Bin OLS Bin OLS

NW 95% confidence interval NW 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a bandwidth ac-
cording to Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson.

The analysis of intergenerational mobility curves renes the picture of upward

and downward mobility along the income distribution. The intergenerational rank

mobility measures by how many percentiles the son is expected to ascend or de-

scend dependent on the income position of his father. The estimated curves show

a negative slope in both countries, with an OLS estimate of -0.7167 for Germany

and -0.5873 for the United States (Figure 2.4). Thus, if the father's income posi-

tion increases by one percentile, the absolute rank mobility of the son is reduced by

0.72 percentiles in Germany and 0.59 percentiles in the United States. Sons whose

fathers rank in the lowest ve percentiles ascend on average by 33-36 percentiles in

Germany and by 28-30 percentiles in the United States. Sons whose fathers rank in

35
 Structure and Extent of Intergenerational Income Mobility

the highest ve percentiles descend on average by 32-35 percentiles in Germany and

by 25-28 percentiles in the United States. Thus, the upward and downward mobility

of the sons located at the bottom and the top of the paternal income distribution is

again more pronounced in Germany than in the United States. Comparing the OLS

estimates with the results of the Bin estimation and the Nadaraya-Watson estima-

tion, it can be concluded for both countries that there is no evidence of nonlinearities

in the development of intergenerational rank mobility along the income distribution

of the fathers.

Figure 2.5: Intergenerational income share mobility

Germany United States

2 2
Income share mobility of son (in % points)

Income share mobility of son (in % points)

1 1

0 0

−1 −1

−2 −2
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Income percentile of father Income percentile of father

Bin OLS Bin OLS

NW 95% confidence interval NW 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a bandwidth ac-
cording to Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson.

In contrast, the intergenerational income share mobility measures the expected

change in a family's share of the total income over two generations dependent on

12
the income position of the father. Similar to the nding of mean reversion in

12 Incomes of sons with the same father were averaged to ensure a family comparison.

36
 Structure and Extent of Intergenerational Income Mobility

ranks, there is also mean reversion in income shares. Families that start at higher

percentiles in the income distribution experience a smaller increase in income share

than families that start at lower percentiles. The OLS estimation yields values of

-0.0091 in Germany and -0.0121 in the United States, respectively (Figure 2.5).

Thus, if the father's income position increases by one percentile, the income share

mobility of the son is reduced by 0.0091 percentage points in Germany and 0.0121

percentage points in the United States. The income share of the sons whose fathers

rank in the lowest ve percentiles increases on average by 0.38-0.42 percentage points

in Germany and by 0.58-0.62 percentage points in the United States. The income

share of the sons whose fathers rank in the highest ve percentiles decreases on

average by 0.45-0.49 percentage points in Germany and 0.53-0.57 percentage points

in the United States. However, the income drop in the United States is not statisti-

cally signicant. Therefore, regarding intergenerational income share mobility, the

United States is more mobile than Germany. Unlike intergenerational rank mobility,

intergenerational income share mobility tends to exhibit nonlinearities in Germany.

In particular, the sons located at the top of the paternal income distribution experi-

ence an abrupt reduction in income share. Thus, the empirical picture suggests that

the OLS estimator actually overestimates intergenerational income share mobility

due to outliers at the upper end of the fathers' income distribution.

2.3.4 Quantile regressions

For a valid assessment of nonlinearities in the relationship between the incomes of

13
fathers and sons, estimates along the income distribution of the sons are necessary.

For this purpose, the intergenerational income elasticity is estimated using condi-

tional and unconditional quantile regressions at selected percentiles of the income

distribution of the sons (Table 2.6).

13 The empirical picture is mixed for both Germany and the United States. Lillard (2001) and
Couch and Lillard (2004) nd evidence of a nonlinear run of intergenerational income elasticity for
both countries. Bratsberg et al. (2007) determine a more or less linear relationship for the United
States. Schnitzlein (2009, 2016) also nds no signicant dierence along the conditional income
distribution in Germany.

37
 Structure and Extent of Intergenerational Income Mobility

Table 2.6: Quantile regressions

Germany United States

CQR UQR CQR UQR

20th percentile
IIE 0.3114*** 0.2483** 0.4493*** 0.4268***

(0.1036) (0.1034) (0.0868) (0.1044)

2
Pseudo R 0.0476 0.0351 0.0701 0.0583

40th percentile
IIE 0.3270*** 0.4178*** 0.3923*** 0.3765***

(0.0891) (0.0862) (0.0908) (0.0702)

2
Pseudo R 0.0623 0.1049 0.0634 0.0925

50th percentile
IIE 0.3586*** 0.4093*** 0.3613*** 0.3935***

(0.0968) (0.0809) (0.0798) (0.0657)

2
Pseudo R 0.0719 0.1221 0.0665 0.1067

60th percentile
IIE 0.4173*** 0.4068*** 0.3881*** 0.4000***

(0.0934) (0.0876) (0.0614) (0.0666)

2
Pseudo R 0.0731 0.1132 0.0737 0.0961

80th percentile
IIE 0.3782*** 0.4420*** 0.5101*** 0.4782***

(0.0716) (0.1106) (0.0707) (0.0926)

2
Pseudo R 0.0856 0.0768 0.0883 0.0924

Observations 354 354 601 601

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Other control variables include polynomials of the father's and the son's age as well as
the number of valid observations of the son. Standard errors are clustered at the family level
and were calculated using paired bootstrap resampling with 1,000 replications. *** p < 0.01,
** p < 0.05, * p < 0.1. IIE: Intergenerational income elasticity, CQR: Conditional quantile
regression, UQR: Unconditional quantile regression.

38
 Structure and Extent of Intergenerational Income Mobility

Figure 2.6: Conditional quantile regressions

Germany United States

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Conditional income percentile Conditional income percentile

CQR CQR
OLS OLS
95% confidence interval 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. CQR: Conditional quantile regression, OLS: Ordinary
least squares.

The conditional quantile regressions show a slightly hump-shaped run for Ger-

many and a u-shaped curve for the United States over the conditional income quan-

14
tiles of the sons. Using conditional quantile regressions, however, statements about

a nonlinear run of the intergenerational income elasticity can only be made when

the monotonicity of the estimation parameter along the income distribution is un-

15
ambiguous. Likewise, for both Germany and the United States, the 95 percent

condence interval completely covers the OLS estimator of intergenerational income

elasticity (Figure 2.6). Thus, neither a concave nor a convex run of the intergener-

ational income elasticity in Germany and the United States can be veried.

14 The conditional quantile regression denes the income quantile of the son conditional on
the income of his father and estimates the intergenerational income elasticity on the conditional
quantile of the income distribution of the son (Koenker and Bassett, 1978, Koenker, 2005).
15 Simple Wald tests show that the estimates do not dier signicantly across the percentiles
either for Germany (p = 0.7857) or for the United States (p = 0.1793).

39
 Structure and Extent of Intergenerational Income Mobility

Figure 2.7: Unconditional quantile regressions

Germany United States

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Unconditional income percentile Unconditional income percentile

UQR UQR
OLS OLS
95% confidence interval 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. UQR: Unconditional quantile regression, OLS: Ordi-
nary least squares.

Using conditional quantile regressions, insights into how strong the eect of

parental income is for the sons at the selected quantile of the marginal income distri-

bution cannot be obtained. For such questions, the unconditional quantile regression

or RIF regression is suitable (Firpo et al., 2009). In Germany, the intergenerational

income elasticity assumes an s-shaped curve along the ascending quantiles. Between

the 40th and the 60th percentile, it is relatively constant at about 0.4, whereas it

is lower for the 20th percentile and higher for the 80th percentile. The intergenera-

tional income mobility is therefore higher at the lower end of the income distribution

of the sons and slightly decreases when moving upward through the quantiles. In

the United States, the development of the estimation parameters across the quan-

tiles takes on a slightly u-shaped form. According to this, intergenerational income

mobility is higher in the middle range of the income distribution of the sons than at

40
 Structure and Extent of Intergenerational Income Mobility

the lower and upper end of the income distribution. While the curve for Germany

indicates a convex development, the United States displays an initially concave and

then a convex course. However, the deviations from proportionality must be inter-

preted with caution since the condence bands for both countries are relatively large

and always contain the respective OLS estimator (Figure 2.7). Overall, the results

of the conditional and unconditional quantile regressions provide no clear indication

of nonlinearities in the development of intergenerational income elasticity over the

income distribution of the sons either for Germany or for the United States.

2.3.5 Decomposition of intergenerational income inequality

Viewing the fathers and the sons as representatives of their families at dierent

points in time, the income inequality between families for two generations can be

16
measured. In Germany, income inequality has risen by 3.11 Gini points (13.54

percent) from an initial value of 22.94 Gini points to a nal value of 26.05 Gini

points. In the United States, income inequality has increased by 7.35 Gini points

(21.98 percent) from an initial value of 33.43 Gini points to a nal value of 40.77

17
Gini points (Table 2.7). Thus, in both countries, income inequality has increased

over time, but the increase was stronger in the United States than in Germany.

In principle, the smaller rise in income inequality in Germany could reect a

pattern of either progressive income growth being oset by signicant reranking

or simply fewer changes overall. Our results show that the former was the case:

income growth was more pro-poor in Germany than in the United States. The

decomposition according to Jenkins and Van Kerm (2006) shows that progressive

income growth has reduced income inequality by 14.71 Gini points (64.14 percent)

in Germany and by 15.43 Gini points (46.16 percent) in the United States. Thus,

in the case of unchanged income positions of the families in the second generation,

16 Since a father can have several economically active sons, the incomes of the sons of a family
were averaged in the calculations.
17 Note that the obtained Gini coecients dier from those presented in Section 2.3.1 because
in-sample rather than overall observations are used.

41
 Structure and Extent of Intergenerational Income Mobility

there should have been a strong reduction in income inequality. However, income

mobility in both countries overcompensates for progressive income growth, such

that income inequality between families ultimately increases. In Germany, income

mobility raises income inequality by 17.82 Gini points (77.68 percent), whereas in the

United States, income inequality is increased by 22.78 Gini points (68.14 percent).

Thus, Germany exhibits both more progressive income growth and higher income

mobility as measured by percentage of the initial Gini coecients in comparison to

the United States.

Table 2.7: Decomposition of intergenerational income inequality

Germany United States

Initial Gini of the fathers 22.94 33.43

Final Gini of the sons 26.05 40.77

Sizes in Gini points

∆ Gini 3.11 7.35

Mobility 17.82 22.78

Progressive income growth 14.71 15.43

Sizes in percent of initial Gini

∆ Gini 13.54 21.98

Mobility 77.68 68.14

Progressive income growth 64.14 46.16

Source: SOEP (1984-2013), PSID (1984-2013).

2.4 Recommendations for economic policy

Our results suggest that paternal income has a strong inuence on the future in-

come of the sons both in Germany and the United States. Although there are no

indications of nonlinearities which might be caused by credit market constraints,

the substantially lower intergenerational income elasticity in, e.g., the Scandinavian

countries indicates the additional inuence of exogenous determinants on the success

of children from poor households. Thus, measures to mitigate these exogenous in-

42
 Structure and Extent of Intergenerational Income Mobility

uences can reduce intergenerational income elasticity and facilitate a more ecient

use of the human capital in society.

However, stronger redistribution of income via the tax and transfer system does

not necessarily have a positive eect on the level of social mobility. Although the

disposable incomes of poor and rich families converge as a result of more redistri-

bution, a more progressive tax and transfer system leads to a declining return to

human capital in the labor market, and thus to a reduction in the incentive to invest

in education. While this is true for all families, it aects poor households relatively

more strongly than it does rich households. In sum, a higher level of redistribu-

tion could even reduce intergenerational income mobility. The method of choice

should therefore be an improvement in the institutional design of the preschool and

school system to increase equality of opportunity without severe distortion of market

processes.

Early childhood education

The barriers to the later income of relatively poor children are not found in the late

stages of education, but rather in early childhood care. Stimulation that children

experience in the early stages of brain development greatly inuences the limits

of future mental capability. A stimulating environment thus results in improved

cognitive development, better social skills, and better health (Knudsen et al., 2006).

While children whose families have above-average incomes and human capital

are able to receive this stimulation at home, this support often falls by the way-

side in less well-o families. Lee and Burkam (2002) show that there are already

severe dierences in education between children from dierent social backgrounds

at the beginning of preschool. As these dierences are expected to grow over the

course of the children's education, this means that early childhood care is of great

importance. Thus, for children from socio-economically weak households, incen-

tives and opportunities must be created for their earlier attendance of public or

private childcare facilities where they can be supported according to their abilities.

43
 Structure and Extent of Intergenerational Income Mobility

This particularly applies to those children with an immigration background who

rst come into contact with the German language at day care centers or in kinder-

garten. In 2016, however, only 21 percent of children under 3 years of age with

a migrant background visited day care, while 38 percent of under-3-year-olds with

no migration background did so (Federal Statistical Oce, 2017). An expansion

of childcare facilities especially for children under 3 years of age as well as a good

sta-to-student ratio with well-trained educators would therefore be conductive to

higher intergenerational income mobility. The German Betreuungsgeld, a childcare

subsidy for parents who raise their under-3-year-olds at home, is obviously not.

Desegregation

Another starting point is the pronounced segregation of children according to their

social background. This problem is particularly evident in the strong heterogeneity

of the quality of schools in Germany. The variation in the 2009 PISA scores between

schools is 68 percent, which is well above the average of 42 percent for the OECD

countries. At the same time, the variance in the results within the individual schools

is only 45 percent, which is considerably below the OECD average of 65 percent

(OECD, 2012). Thus, pupils at the respective schools are at a comparable level,

while the variation between the performance of pupils in good and bad schools is

substantial.

Musset (2012) illustrates that a large part of educational segregation can be

traced back to local segregation. On the one hand, families with a lower educational

level spend less time choosing a school for their children and often suer from a con-

siderable information decit with regard to the educational system and the quality

of schools (Hastings et al., 2005). Thus, families with a weaker socio-economic sta-

tus tend to send children to the locally nearest school, while wealthy families choose

the subjectively best school for their children and tend to avoid schools with a high

number of children from socially vulnerable families (Schneider and Buckley, 2002,

Raveaud and Zanten, 2007). On the other hand, a strong variation in school quality

44
 Structure and Extent of Intergenerational Income Mobility

means that the demand for spots at good schools exceeds the existing capacities. In

such cases, the risk of so-called cream skimming, i.e., the selection of subjectively

better pupils, is high (Lubienski, 2006). Here, the location of children's homes is

an indicator of their social background, which can be used by schools as a basis

for the selection of pupils. Thus, if cities become increasingly segregated by social

background, this intensies the problem of intergenerational income persistence.

To increase intergenerational income mobility, investments in the education sys-

tem must therefore primarily promote equal opportunity and desegregation. A sim-

ple enhancement in educational spending is not an adequate means of increasing

social mobility: the higher the average level of human capital, the more dicult the

process of catching up is for pupils from disadvantaged families (Hanushek, 2003).

A so-called formula funding based on the Dutch model could help to decrease cream

skimming and reduce the segregation of children according to social status. Here, a

weight is assigned to each student and the nancial resources allocated to a certain

school are calculated based on the sum of the weights of its students. If pupils

from disadvantaged families are assigned a higher weight, there is an incentive for

schools to accept these pupils. This also takes account of the fact that due to the

more intensive support they require, the admission of disadvantaged children may

be more cost-intensive in some circumstances.

Secondary school tracking

Another issue often discussed in politics is the division of pupils into various sec-

ondary school tracks after only four years of elementary school in Germany. Thus,

while the median age of rst formal selection is 15 years in the OECD countries,

selection in Germany takes place when students are only 10 years old (OECD, 2012).

As a consequence, the decision as to whether a child apprentices to learn a trade

or attends university is made very early in most cases. However, children's level

of education is one of the most important determinants for their adult income and

heavily inuences the probability of becoming unemployed during their working life.

45
 Structure and Extent of Intergenerational Income Mobility

In 2015, the unemployment rate of persons aged between 15 and 74 who earned a de-

gree below the secondary education level was 11.2 percent in Germany. In contrast,

the possession of a secondary (4.3 percent) or tertiary education level (2.3 percent)

leads to a signicantly lower probability of unemployment (European Commission,

2017).

However, the decision to attend a particular type of secondary school depends

heavily on the education level of the parents. While 43.8 percent of the parents of

children at the German Hauptschule also attended this institution, only 7.2 percent

of the parents of pupils at the Gymnasium did so. Similarly, 62.5 percent of the

parents of children at the Gymnasium achieved a high school diploma, while only

14.5 percent of the parents of children at the Hauptschule have (Federal Statistical

Oce, 2017). Therefore, later secondary school tracking, e.g., at the age of 12

instead of 10, as a measure to support equality in the schooling system has been

discussed for some time. A similar school reform in Finland has led to a reduction of

intergenerational income elasticity by 23 percent (Pekkarinen et al., 2009). Hanushek

and Wöÿmann (2006) conrm that early tracking is associated with a signicantly

larger inequality of performance between pupils, while there are no signicant eects

on the overall performance. In contexts where there is reluctance to delay early

tracking in the short term, the negative eects could be lessened by an improvement

of the selection methods for the dierent tracks, a limitation of grouping to specic

subjects, and an increase in the exibility to change tracks.

2.5 Conclusion

The present study examines the structure and extent of intergenerational income

mobility in Germany and the United States with the help of dierent statistical

concepts. In line with existing results, intergenerational income elasticity in the

United States is higher than in Germany. While the results for intergenerational

rank mobility are relatively similar, the level of intergenerational income share mo-

46
 Structure and Extent of Intergenerational Income Mobility

bility is higher in the United States than in Germany. There are no indications

of a nonlinear run of the intergenerational income elasticity. The decomposition of

intergenerational income inequality shows both higher income mobility and stronger

progressive income growth for Germany compared to the United States. Overall, we

cannot identify a clear ranking of the two countries. In order to increase the level of

social mobility, policy needs to focus on equality of opportunity in the educational

system. This solution is more incentive-compatible in the long run than a policy of

pure redistribution.

47
 Structure and Extent of Intergenerational Income Mobility

References

Aaberge, R. and Mogstad, M. (2014). Income mobility as an equalizer of permanent

income. Discussion Papers 769, Statistics Norway, Research Department, Oslo,

Norway.

Becker, G. S. and Tomes, N. (1979). An equilibrium theory of the distribution

of income and intergenerational mobility. Journal of Political Economy, 87 (6),

11531189.

Becker, G. S. and Tomes, N. (1986). Human capital and the rise and fall of families.

Journal of Labor Economics, 4 (3), S1S39.

Belsley, D. A., Kuh, E., and Welsch, R. E. (1980). Regression Diagnostics: Indenti-

fying Inuential Data and Sources of Collinearity. John Wiley & Sons, Hoboken,

United States.

Bhattacharya, D. and Mazumder, B. (2011). A nonparametric analysis of black-

white dierences in intergenerational income mobility in the United States. Quan-

titative Economics, 2 (3), 335379.

Björklund, A. and Jäntti, M. (2009). Intergenerational income mobility and the role

of family background. In Salverda, W., Nolan, B., and Smeeding, T. M., editors,

Oxford Handbook of Economic Inequality, 491521.

Björklund, A., Roine, J., and Waldenström, D. (2012). Intergenerational top in-

come mobility in Sweden: Capitalist dynasties in the land of equal opportunity?

Journal of Public Economics, 96 (56), 474484.

Black, S. E. and Devereux, P. J. (2011). Recent developments in intergenerational

mobility. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics,

4B, 14871541.

48
 Structure and Extent of Intergenerational Income Mobility

Böhlmark, A. and Lindquist, M. J. (2006). Life-cycle variations in the associa-

tion between current and lifetime income: Replication and extension for Sweden.

Journal of Labor Economics, 24 (4), 879896.

Bratberg, E., Davis, J., Mazumder, B., Nybom, M., Schnitzlein, D. D., and Vaage,

K. (2017). A comparison of intergenerational mobility curves in Germany, Norway,

Sweden, and the US. Scandinavian Journal of Economics, 119 (1), 72101.

Bratsberg, B., Røed, K., Raaum, O., Naylor, R., Jännti, M., Eriksson, T., and

Österbacka, E. (2007). Nonlinearities in intergenerational earnings mobility: Con-

sequences for cross-country comparisons. Economic Journal, 117 (519), C72C92.

Brenner, J. (2010). Life-cycle variations in the association between current and

lifetime earnings: Evidence for German natives and guest workers. Labour Eco-

nomics, 17 (2), 392406.

Chetty, R., Hendren, N., Kline, P., Saez, E., and Turner, N. (2014). Is the United

States still a land of opportunity? Recent trends in intergenerational mobility.

American Economic Review: Papers & Proceedings, 104 (5), 141147.

Corak, M. (2006). Do poor children become poor adults? Lessons from a cross-

country comparison of generational earnings mobility. In Creedy, J. and Kalb,

G., editors, Dynamics of Inequality and Poverty, 143188.

Couch, K. A. and Lillard, D. R. (2004). Non-linear patterns of intergenerational

mobility in Germany and the United States. In Corak, M., editor, Generational

Income Mobility in North America and Europe, 190206.

Eisenhauer, P. and Pfeier, F. (2008). Assessing intergenerational earnings persis-

tence among German workers. Journal for Labour Market Research, 41 (23),

119137.

European Commission (2017). Eurostat database. Luxembourg.

Federal Statistical Oce (2017). Genesis database. Wiesbaden, Germany.

49
 Structure and Extent of Intergenerational Income Mobility

Fertig, A. R. (2003). Trends in intergenerational earnings mobility in the U.S.

Journal of Income Distribution, 12 (34), 108130.

Firpo, S., Fortin, N. M., and Lemieux, T. (2009). Unconditional quantile regressions.

Econometrica, 77 (3), 953973.

Fitzgerald, J., Gottschalk, P., and Mott, R. (1998). An analysis of sample attrition

in panel data: The Michigan Panel Study of Income Dynamics. Journal of Human

Resources, 33 (2), 251299.

Frick, J. R., Grabka, M. M., and Groh-Samberg, O. (2012). Dealing with incomplete

household panel data in inequality research. Sociological Methods & Research, 41

(1), 89123.

Frick, J. R., Jenkins, S. P., Lillard, D. R., Lipps, O., and Wooden, M. (2007). The

Cross-National Equivalent File (CNEF) and its member country household panel

studies. Journal of Contextual Economics, 127 (4), 627654.

Grabka, M. M. (2014). Codebook for the $PEQUIV File 1984-2013: CNEF vari-

ables with extended income information for the SOEP. Data Documentation 74,

German Institute for Economic Research, Berlin, Germany.

Grawe, N. D. (2004). Reconsidering the use of nonlinearities in intergenerational

earnings mobility as a test for credit constraints. Journal of Human Resources,

39 (3), 813827.

Grawe, N. D. and Mulligan, C. B. (2002). Economic interpretations of intergenera-

tional correlations. Journal of Economic Perspectives, 16 (3), 4558.

Grieger, L. D., Danziger, S., and Schoeni, R. F. (2009). Accurately measuring the

trend in poverty in the United States using the Panel Study of Income Dynamics.

Journal of Economic and Social Measurement, 34 (23), 105117.

Haider, S. and Solon, G. (2006). Life-cycle variation in the association between

current and lifetime earnings. American Economic Review, 96 (4), 13081320.

50
 Structure and Extent of Intergenerational Income Mobility

Han, S. and Mulligan, C. B. (2001). Human capital, heterogeneity and estimated

degrees of intergenerational mobility. Economic Journal, 111 (470), 207243.

Hanushek, E. A. (2003). The failure of input-based schooling policies. Economic

Journal, 113 (485), F64F98.

Hanushek, E. A. and Wöÿmann, L. (2006). Does education tracking aect perfor-

mance and inequality? Dierences-in-dierences evidence across countries. Eco-

nomic Journal, 116 (510), C63C76.

Hastings, J. S., Kane, T. J., and Staiger, D. O. (2005). Parental preferences and

school competition: Evidence from a public school choice program. NBER Work-

ing Paper 11805, National Bureau of Economic Research, Cambridge, United

States.

Hussain, M. A., Munk, M. D., and Bonke, J. (2009). Intergenerational earnings

mobilities: How sensitive are they to income measures? Journal of Income Dis-

tribution, 18 (34), 7992.

Jenkins, S. P. and Van Kerm, P. (2006). Trends in income inequality, pro-poor

income growth, and income mobility. Oxford Economic Papers, 58 (3), 531548.

Knudsen, E. I., Heckman, J. J., Cameron, J. L., and Shonko, J. P. (2006). Eco-

nomic, neurobiological, and behavioral perspectives on building America's future

workforce. Proceedings of the National Academy of Sciences, 103 (27), 10155

10162.

Koenker, R. (2005). Quantile Regression. Cambridge University Press, Cambridge,

United Kingdom.

Koenker, R. and Bassett, G. (1978). Regression quantiles. Econometrica, 46 (1),

3350.

51
 Structure and Extent of Intergenerational Income Mobility

Lee, V. E. and Burkam, D. T. (2002). Inequality at the Starting Gate: Social Back-

ground Dierences in Achievement as Children Begin School. Economic Policy

Institute, Washington, D.C., United States.

Lillard, D. R. (2001). Cross-national estimates of the intergenerational mobility in

earnings. Quarterly Journal of Economic Research, 70 (1), 5158.

Lillard, D. R. (2013). Codebook for the Cross-National Equivalent File 1970-2013:

BHPS  HILDA  KLIPS  PSID  RLMS-HSE  SHP  SLID  SOEP. Colum-

bus, United States.

Lubienski, C. (2006). School diversication in second-best education markets. Ed-

ucational Policy, 20 (2), 323344.

Mazumder, B. (2014). Black-white dierences in intergenerational economic mobility

in the United States. Economic Perspectives, 38 (1), 118.

Musset, P. (2012). School choice and equity: Current policies in OECD countries

and a literature review. OECD Education Working Papers 66, OECD, Paris,

France.

Nilsen, Ø. A., Vaage, K., Aakvik, A., and Jacobsen, K. (2008). Sources of mea-

surement errors in earnings data: New estimates of intergenerational elasticities.

Working Papers in Economics 04/08, Department of Economics, University of

Bergen, Bergen, Norway.

Nybom, M. and Stuhler, J. (2016). Heterogeneous income proles and lifecycle bias

in intergenerational mobility estimation. Journal of Human Resources, 51 (1),

239268.

OECD (2012). Equity and Quality in Education: Supporting Disadvantaged Students

and Schools. Paris, France.

OECD (2016). Education at a Glance 2016. Paris, France.

52
 Structure and Extent of Intergenerational Income Mobility

Pekkarinen, T., Uusitalo, R., and Kerr, S. (2009). School tracking and intergenera-

tional income mobility: Evidence from the Finnish comprehensive school reform.

Journal of Public Economics, 93 (78), 965973.

Raveaud, M. and Zanten, A. V. (2007). Choosing the local middle school: Middle

class parents' values and social and ethnic mix in London and Paris. Journal of

Education Policy, 22 (1), 107124.

Roemer, J. E. (2004). Equal opportunity and intergenerational mobility: Going

beyond intergenerational income transition matrices. In Corak, M., editor, Gen-

erational Income Mobility in North America and Europe, 4857.

Schneider, M. and Buckley, J. (2002). What do parents want from schools? Evidence

from the internet. Educational Evaluation and Policy Analysis, 24 (2), 133144.

Schnitzlein, D. D. (2008). Verbunden über Generationen: Struktur und Ausmaÿ der

intergenerationalen Einkommensmobilität in Deutschland. SOEPpapers on Mul-

tidisciplinary Panel Data Research 80, German Institute for Economic Research,

Berlin, Germany.

Schnitzlein, D. D. (2009). Structure and extent of intergenerational income mobility

in Germany. Journal of Economics and Statistics, 229 (4), 450466.

Schnitzlein, D. D. (2016). A new look at intergenerational income mobility in Ger-

many compared to the U.S. Review of Income and Wealth, 62 (4), 650667.

Solon, G. (1989). Biases in the estimation of intergenerational earnings correlations.

Review of Economics and Statistics, 71 (1), 172174.

Solon, G. (1992). Intergenerational income mobility in the United States. American

Economic Review, 82 (3), 393408.

Solon, G. (1999). Intergenerational mobility in the labor market. In Ashenfelter, O.

and Card, D., editors, Handbook of Labor Economics, 3A, 17611800.

53
 Structure and Extent of Intergenerational Income Mobility

Solon, G. (2004). A model of intergenerational mobility variation over time and

place. In Corak, M., editor, Generational Income Mobility in North America and

Europe, 3847.

Vogel, T. (2006). Reassessing intergenerational mobility in Germany and the United

States: The impact of dierences in lifecycle earnings patterns. SFB 649 Discus-

sion Papers 2006-055, Humbolt-Universität zu Berlin, Berlin, Germany.

Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data.

MIT Press, Cambridge, United States.

World Bank (2017). World Development Indicators database. Washington, D.C.,

United States.

Zimmerman, D. J. (1992). Regression toward mediocrity in economic stature. Amer-

ican Economic Review, 82 (3), 409492.

54
 Structure and Extent of Intergenerational Income Mobility

Appendix

Table 2.8: Intergenerational income elasticity (robustness checks)

Germany United States

Without With Without With
unemployment unemployment unemployment unemployment
periods periods periods periods

Dependent variable: Log. income (son)

Log. income (father) 0.3428*** 0.3283*** 0.4613*** 0.4433***

(0.0776) (0.0790) (0.0731) (0.0657)

Age (son) -0.0747 1.1732 -1.3013 0.6606

(1.2985) (1.1114) (1.2942) (1.0109)

2
Age (son) 0.0003 -0.0157 0.0175 -0.0087

(0.0168) (0.0143) (0.0169) (0.0132)

Age (father) -0.0729 -0.0151 0.0317 0.0354

(0.1365) (0.1310) (0.1081) (0.0975)

2
Age (father) 0.0003 -0.0003 0.0000 0.0003

(0.0013) (0.0013) (0.0013) (0.0012)

Observations (son) 0.0437 0.0047 0.1368** -0.0114

(0.0297) (0.0238) (0.0539) (0.0450)

Birth cohort (son) -0.0296* -0.0230 0.0063 -0.0166

(0.0160) (0.0159) (0.0187) (0.0156)

Birth cohort (father) -0.0436 -0.0433 0.0236 0.04944

(0.0479) (0.0465) (0.0547) (0.0495)

Observations 354 353 601 597

2
R 0.0847 0.0788 0.1402 0.1222

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Estimations for the SOEP are based on non-imputed income data. The intergenerational
income elasticities have been determined for an annual lower income limit of 1,200 Euro/US dol-
lar. Standard errors are clustered at the family level and were calculated using paired bootstrap
resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

55
Chapter 3

Transmission Channels of
Intergenerational Income
Persistence†

3.1 Introduction

Intergenerational income persistencethe fact that children from rich families tend

to have higher adult incomes themselves than children from poor familieshas been

18
extensively discussed in the economic literature since the 1980s. A wide range

of studies analyze the extent of intergenerational income persistence (Solon, 1992,

Björklund and Jäntti, 1997, Schnitzlein, 2016), the development of intergenerational

income persistence over time (Fertig, 2003, Lee and Solon, 2009, Chetty et al., 2014),

and dierences in the intergenerational income persistence between individual coun-

tries (Corak, 2006, Corak et al., 2014, Bratberg et al., 2017). The measure commonly

estimated in order to quantify intergenerational income persistence is the intergener-

ational income elasticity, where, for example, a value of 0.3 means that 30 percent of

the parents' income advantage or disadvantage is passed on to their children. Thus,

if a family's income is 10 percent higher than the average income in the parental gen-

eration, the expected income of their children is 3 percent higher than the average

19
income in the lial generation.

†
This chapter is co-authored with Mustafa Çoban.
18 For a broad literature review, see Solon (1999), Björklund and Jäntti (2009), and Black and
Devereux (2011).
19 The intergenerational income elasticity is described in detail in Section 3.2.1.

56
 Transmission Channels of Intergenerational Income Persistence

Figure 3.1: Transmission channels of intergenerational income persistence

higher human capital

of the father

higher income higher human capital higher income

of the father of the son of the son

So far, however, little is known about the underlying transmission channels of

intergenerational income persistence. Why exactly do children born to wealthy fam-

ilies earn more than less fortunate children? Essentially, there are two conceivable

mechanisms (Figure 3.1). On the one hand, well-o families can use their nancial

resources to invest in the education of their ospring, which is then reected in their

children's higher human capital and thus higher income in adulthood (investment

eect ). This includes, for example, the attendance of private schools and universi-

ties as well as additional private lessons, which might not be aordable for children

from poor families. On the other hand, if the higher parental income is at least

partially determined by a higher parental human capital, more auent families may

in addition directly pass on this human capital to their children (endowment eect ).

Possible examples are the genetic transmission of certain traits, the intergenera-

tional transfer of aspirations and skills, and at-home nonnancial investments such

as reading books or assisting with children's homework.

The economic literature is limited to very few studies that explicitly analyze

the transmission channels described above. Blanden (2013) proposes a descriptive

decomposition to estimate (i) the extent of intergenerational income persistence if

57
 Transmission Channels of Intergenerational Income Persistence

intergenerational educational persistence were the only determinant, (ii) the impact

of inequalities in parental income on lial income within education groups, and (iii)

the cross-eect between parental education and children's residual earnings. She

concludes that the majority of the dierences in intergenerational income persis-

tence between the United Kingdom and the United States are due to the second

eect. Lefgren et al. (2012) use a structural decomposition to establish an upper

and lower bound for the investment and endowment eects using data from a 35

percent sample of Swedish sons and their fathers. They show that only a minority

of the intergenerational income elasticity can be plausibly attributed to the causal

eect of fathers' nancial resources. Cardak et al. (2013) use stochastic properties

of the intergenerational income elasticity to decompose the estimate for the United

States into the investment and endowment eects without the need for additional

data. They nd an investment eect of approximately one third and an endowment

eect of approximately two thirds.

A complementary strand of the economic and sociological literature deals with

the intergenerational transmission of certain characteristics that might help to ex-

plain the transmission channels of intergenerational income persistence. For exam-

ple, Behrman and Rosenzweig (2002), Oreopoulos et al. (2006), and Holmlund et al.

(2011) study the intergenerational transmission of education, Hauser and Logan

(1992) discuss the intergenerational transmission of occupational status, and Blau

(1999), Blanden and Gregg (2004), and Dahl and Lochner (2012) analyze the causal

eects of parental income on children's educational achievement.

This chapter builds on the existing literature and seeks to determine the ex-

tent to which the investment and endowment eects contribute to the estimates of

intergenerational income persistence in Germany and the United States. We use

a linear and a nonlinear version of the Blanden (2013) descriptive decomposition

method as well as a structural decomposition method as presented in Lefgren et al.

(2012). Overall, we nd that while the investment and the endowment eects in

Germany contribute more or less equally to the estimated intergenerational income

58
 Transmission Channels of Intergenerational Income Persistence

persistence, the investment eect is more pronounced in the United States. In light

of the higher level of privatization in the education system of the United States, this

result seems reasonable. Furthermore, we use unconditional quantile regressions in

order to reveal nonlinearities in the transmission mechanisms along the income dis-

tribution. While there is a mild but steady downward trend of the endowment eect

along the increasing percentiles in the United States, no clear trend can be observed

in Germany. Section 3.2 subsequently presents the theoretical background which

establishes the link between income and human capital. Section 3.3 describes the

data used and discusses possible measurement issues. The results of the estimations

are presented in Section 3.4. To conclude, Section 3.5 includes a brief summary as

well as several economic policy recommendations, which can be derived from our

results.

3.2 Theoretical framework

Our decomposition methods are based on the theoretical framework of Becker and

Tomes (1979, 1986), wherein each family maximizes a utility function dependent

on the parents' consumption and their children's future income. Children's income

is raised when they receive investments in human capital from their parents. In

addition, children's income is inuenced by a variety of inherited endowments in-

cluding race, ability, and other characteristics, family reputation and connections,

and knowledge, skills, and goals provided by their family environment. However, en-

dowments and investments in human capital are not independent from one another,

as children who receive more parental endowments have a higher return to human

capital than those who receive less and therefore the incentive to invest in their

human capital is higher. The equilibrium income of children is thus determined by

the income and endowment of their parents as well as by their fortuitous endowment

and their luck on the labor market.

59
 Transmission Channels of Intergenerational Income Persistence

3.2.1 Intergenerational income and educational persistence

In the empirical literature, particular attention has been given to the intergener-

ational persistence of income and education. Intergenerational income persistence

measures the inuence of parents' income on the adult income of their children. In

contrast, intergenerational educational persistence analyzes, how strongly the edu-

cational success of children depends on their parents' degree of education. These two

measures are commonly considered separately from one another, though they are

indeed closely related. The standard approach in order to measure intergenerational

income persistence is based on the estimation of a log-linear equation in the form of

log(yis ) = α1 + β log(yif ) + us1i , (3.1)

20
where yis is the lifetime income of the son and yif is the lifetime income of the father.

The intercept α1 represents the average lifetime income in the son's generation, and

the slope β is the searched-for intergenerational income elasticity. It states that an

increase in a father's lifetime income by 1 percent increases the expected lifetime

income of his son by β percent. If β = 0, sons' lifetime incomes are independent

of their fathers' lifetime incomes. In this case, a society has complete intergenera-

tional income mobility. In contrast, the higher the value of β, the stronger the link

between the lifetime income of a father and his son is, and consequently, the lower

the intergenerational income mobility. Deviations from the expected income of the

son due to factors orthogonal to the income of the father are summarized in the

idiosyncratic error term us1i .

The estimation of the intergenerational educational persistence provides the ad-

vantage that data on education are usually more easily available and constant over

an adult's lifetime. The intergenerational educational persistence is measured, just

like the intergenerational income persistence, by estimating a linear equation in the

20 Since the analyses in this chapter are limited to father-son pairs, the explanations refer to
the eect of the father's lifetime income on the son's lifetime income. In principle, the subsequent
relationships apply to any parent-child pair.

60
 Transmission Channels of Intergenerational Income Persistence

form of

Edsi = α2 + γEdfi + us2i , (3.2)

where Edsi and Edfi correspond to the son's and the father's education level, respec-

tively. The slope γ is the searched-for intergenerational educational persistence and

can be interpreted in such a way that an increase in the father's education by 1 unit

raises the expected education of his son by γ units. Again, the residual term us2i
captures all deviations from the expected education level of the son orthogonal to

his father's education.

3.2.2 Descriptive decomposition

Linear descriptive decomposition

It is a widely accepted fact that the education level is one of the most important

determinants of a person's lifetime income. The relationship between education and

income can be estimated for the fathers by

log(yif ) = θf + δ f Edfi + νif (3.3)

and for the sons by

log(yis ) = θs + δ s Edsi + νis , (3.4)

where δf and δs correspond to the rate of return to education for the generation of

the fathers and the sons, respectively. In contrast, νif and νis capture income varia-

tions that are due to a father's or son's fortune in the labor market. This includes,

for example, benets from a generous union contract, unusually good or bad job

matches, or working in a rm that goes out of business (Lefgren et al., 2012). Blan-

den (2013) shows that in order to decompose the intergenerational income elasticity

β, the simple Mincer equations (3.3) and (3.4) can be combined with the mobility

61
 Transmission Channels of Intergenerational Income Persistence

measure equations (3.1) and (3.2) to obtain

δs s
), ν f ) s
, Edf ) 2
 
2 Cov(log(y 2 1 Cov(ν
β= γ REd f + (1 − REd f) + REdf , (3.5)
δf f
Var(ν ) δf Var(Edf )

2 s
where REd f is given by Equation (3.3), Cov(log(y ), ν f )/Var(ν f ) is the estimated

coecient from a regression of the son's lifetime income log(yis ) on his father's income
due to luck in the labor market νif , and Cov(ν
s
, Edf )/Var(Edf ) is the estimated

coecient from a regression of the luck component of son's income νis on the father's

education Edfi . The rst term of Equation (3.5) can thus be interpreted as the

magnitude of the intergenerational income elasticity if educational persistence were

the only transmission channel, therefore capturing the endowment eect described

in Section 3.1. Holding the intergenerational transmission of education constant, the

endowment eect increases if the relation between the rates of return to education

in both generations rises or if the relationship between education and income in

the fathers' generation is more pronounced. The second term of Equation (3.5)

measures the impact of the association between the son's lifetime income and the

within-education group inequalities in paternal incomes and can thus be interpreted

as the investment eect described in Section 3.1. The investment eect increases

if the within-education group income inequality increases, which might be due to

divergent rates of return to education between individual occupations with the same

amount of human capital or a strong regional variation in the quality of schools and

universities (Blanden, 2013). Finally, the third term of Equation (3.5) yields the

cross-eect between paternal education and the residual income of the son.

Nonlinear descriptive decomposition

The threefold decomposition of Blanden (2013) implicitly assumes that the relation-

ship between fathers' and sons' lifetime income is linear, i.e., that the intergenera-

tional income elasticity is constant along the entire income distribution. However,

Becker and Tomes (1986) point out that the intergenerational income elasticity can

62
 Transmission Channels of Intergenerational Income Persistence

assume a concave run when poor families experience credit market constraints that

do not apply for rich families. Consequently, rich families will invest in the human

capital of their children until the marginal costs equal the marginal rate of return,

while credit-constrained families might be forced to invest less than the optimal

amount in their children's education. Thus, a small increase in a poor father's in-

come will have a stronger impact on his son's income than a small increase in a

rich father's income would have. In this case, the intergenerational income per-

sistence will be more pronounced for poor families than for rich families, resulting

in a concave run of intergenerational income elasticity. However, a concave run of

the intergenerational income elasticity neither needs to follow from credit market

constraints nor is market failure implied by concavity. If the income of a father cor-

relates with the unobservable talent of his son, poor fathersregardless of whether

credit market constraints existwill reduce investments in the human capital of

their sons as a result of a lower expected rate of return. Likewise, a concave run is

not a clear indication for credit market constraints. This relationship might be trig-

gered by institutional, social, or unobservable circumstances which inuence poor

and rich families in dierent ways (Grawe, 2004).

On the other hand, a convex run of the intergenerational income elasticity can

be observed if educational policy is designed in such a way as to ensure a basic

level of human capital for all sons, regardless of their fathers' income. Beyond this

socially guaranteed level, all families experience credit market constraints, such that

the total amount of human capital investment in the son is dependent on paternal

income (Bratsberg et al., 2007). Assuming that the unobservable talent of children is

not independent from the socio-economic status of their family, the intergenerational

income persistence among poor families will consequently be lower than among rich

families, resulting in a convex run of the intergenerational income elasticity (Han

and Mulligan, 2001, Grawe and Mulligan, 2002).

Since the descriptive decomposition is to be performed along the income distribu-

tion of the sons, Equations (3.1) and (3.4) are estimated by applying unconditional

63
 Transmission Channels of Intergenerational Income Persistence

21
quantile or RIF regressions at dierent income quantiles (Firpo et al., 2009). For

this purpose, the values of the dependent variable log(yis ) are transformed into their
corresponding RIF values using the estimation formula

d (yi , ŷq , F̂) q − 1[yi ≤ ŷq ]

RIF = ŷq + , (3.6)
f̂(ŷq )

where F̂ is the estimated cumulative income distribution of the sons, q is the un-

conditional income quantile, f̂(ŷq ) gives the kernel density estimate at the income

value ŷq , and 1[yi ≤ ŷq ] is an indicator function, which takes on a value of one if a
son has an income less than or equal to ŷq at the particular quantile and a value of

zero otherwise. Equations (3.1) and (3.4) can then be estimated via ordinary least

squares (OLS) utilizing the transformed RIF values.

3.2.3 Structural decomposition

The descriptive decomposition method described in Section 3.2.2 is likely to overes-

timate the impact of education in the intergenerational transmission process if the

residuals of the respective equations are mutually correlated via, e.g., unobservable

talents or abilities (Hirvonen, 2010). To overcome this problem, a structural ap-

proach to decompose the intergenerational income elasticity into the causal eect of

nancial resources, the mechanistic transmission of human capital, and the impact

of human capital in the determination of fathers' permanent incomes is presented

in Lefgren et al. (2012). In contrast to Blanden (2013), Lefgren et al. (2012) di-

rectly model fathers' investment in the human capital of their sons by extending

and reformulating Equation (3.2) to

HCis = ψ + π1 log(yif ) + π2 HCif + εsi , (3.7)

21 The estimation method of Equations (3.2) and (3.3) remains unchanged.

64
 Transmission Channels of Intergenerational Income Persistence

where HCis = δ s Edsi and HCif = δ f Edfi . Thus, sons' and fathers' human capital

are measured in Euro or US dollar, depending on their country of residence. Ac-

cording to Equation (3.7), a father may inuence the human capital of his son via

nancial investments as well as through the direct transfer of human capital. The

rst parameter π1 represents the share of a father's income which he invests in the

human capital of his son, multiplied by the ecacy of this investment. The second

parameter π2 can be interpreted as the share of a father's human capital which is

directly passed on to his son independent of nancial investments (Lefgren et al.,

2012). Substituting Equation (3.7) into Equation (3.4), the lifetime income of a son

as a function of his father's lifetime income and human capital is expressed by

log(yis ) = π0 + π1 log(yif ) + π2 HCif + ηis , (3.8)

where π0 = ψ + θ s and ηis = εsi + νis . Finally, substituting Equation (3.3) into

Equation (3.8) yields

log(yis ) = π0 + π1 θf + (π1 + π2 )HCif + π1 νif + ηis . (3.9)

Equation (3.9) precisely depicts the notion that an increase in the lifetime income of

the father can inuence the lifetime income of his son via two dierent transmission

channels. If the father's income increase can be ascribed to the father's higher human

capital, this raises the nancial investments in the human capital and, in turn, the

adult income of his son (π1 ). Meanwhile, the higher human capital of the father

directly inuences the human capital of the son, which in turn leads to an increase

in his adult income (π2 ). In contrast, an increase in a father's lifetime income which

is due solely to his good fortune in the labor market inuences the child only via

higher nancial investments (π1 ).

65
 Transmission Channels of Intergenerational Income Persistence

Given the model of Lefgren et al. (2012), the OLS estimator β̂ OLS obtained from

Equation (3.1) converges in probability to

f
OLS Var(HC )
plim(β̂ ) = π1 + π2 . (3.10)
Var(HC f ) + Var(ν f )

The estimated intergenerational income elasticity thus depends on three dierent

factors. First, π1 is the inuence of the father's income when his human capital

remains constant. Second, π2 describes the impact of the father's human capital

f
when his income remains unchanged. Finally, Var(HC )/(Var(HC f ) + Var(ν
f
))
represents the share of the variance in the fathers' income that can be explained by

2
the variance in their human capital, which equals REd f in Equation (3.5). Thus, the

rst part of the sum can be interpreted as the investment eect, while the second

part represents the endowment eect.

Hereinafter, it will be assumed that there exists an instrument Zif for the in-

come of the father which can be used in an instrument variables (IV) estimation of

Equation (3.1). The estimated parameter β̂ IV then converges in probability to

f
IV Cov(HC , Zf )
plim(β̂ ) = π 1 + π2 . (3.11)
Cov(HC f , Z f ) + Cov(ν f , Z f )

As in Equation (3.10), π1 and π2 are the ceteris paribus inuences of the father's

f
income and human capital, respectively, while Cov(HC , Z f )/(Cov(HC f , Z f ) +
f
Cov(ν , Z f )) represents the share of the covariance between paternal income and

the instrument that can be ascribed to human capital. From Equation (3.10) and

(3.11), it follows that β̂ OLS = β̂ IV if and only if π2 = 0 or

f f
Var(HC ) Cov(HC )
= . (3.12)
Var(HC f ) + Var(ν f ) Cov(HC f , Z f ) + Cov(ν f , Z f )

Since Equation (3.12) does not generally hold, a signicant dierence between the

OLS and the IV estimator implies π2 6= 0. If a Hausman test for endogeneity is

rejected, it may thus be assumed that in addition to the pure investment eect, the

66
 Transmission Channels of Intergenerational Income Persistence

intergenerational transfer of income is carried out via the direct transfer of human

capital. In this case, dierent instruments Zif should yield dierent estimates of

β̂ IV , depending upon their covariance with the human capital and luck component

of the father's lifetime income.

This circumstance can be used to determine the magnitude of the investment

eect and the endowment eect. Consider rst the cases where the chosen instru-

ment is correlated solely with the human capital component of the father's income

f
and thus Cov(HC , Z f )/(Cov(HC f , Z f ) + Cov(ν f , Z f )) = 1. In this case, β̂ IV con-

verges in probability to π1 + π 2 . In contrast, if Zif is exclusively correlated with the

f
luck component of the father's income and thus Cov(HC , Z f )/(Cov(HC f , Z f ) +
f
Cov(ν , Z f )) = 0, β̂ IV converges in probability to π1 . A direct comparison of the

two IV estimators in combination with the OLS estimator β̂ OLS then allows for the

identication of the investment and the endowment eects.

Unfortunately, one will generally not be able to nd perfect instruments for the

father's human capital and luck income components. However, on the monotonicity

f
condition that Cov(HC , Zf ) and Cov(ν
f
, Zf ) have the same sign, each estimate

for β IV should lie in the range between π1 and π1 + π2 . Thus, if one chooses an

instrument that is highly correlated with the luck component of the father's income,

β̂ IV can be interpreted as an upper bound for π1 . In contrast, an instrument which

is primarily correlated with the human capital component of the father's income

yields a lower bound for π1 + π2 . Finally, the dierence between the two estimators

provides a lower bound for π2 . A complementary bounding procedure is possible

using only instruments for the human capital of the father. In this case, the IV

estimator β̂ IV again captures a lower bound for π 1 + π2 . A direct estimation of

2
REd f

f
via Equation (3.3) yields a lower bound for Var(HC )/(Var(HC f )+ Var(ν f )). These

results in conjunction with the OLS estimator β̂ OLS in turn allow for the estimation

of an upper bound of π1 and a lower bound of π2 .

67
 Transmission Channels of Intergenerational Income Persistence

3.3 Data and measurement issues

To examine intergenerational income mobility empirically, long-term panel data of

households that capture information on children while they are still living with

their parents and follow them into adulthood are required (Corak, 2006). For a

valid country comparison, data also need to be highly comparable. We therefore

use the Socio-economic Panel (SOEP) for Germany and the Panel Study of Income

Dynamics (PSID) for the United States. Both studies collect information on all

adult persons of a household and survey them repeatedly in the subsequent years.

Further, the SOEP and the PSID are part of the Cross-National Equivalent File

(CNEF) project, which oers a harmonized panel data set of the underlying national

household surveys (Frick et al., 2007).

3.3.1 Measurement errors and life-cycle bias

In order to measure lifetime income exactly, all of a respondent's income state-

ments over their entire working life would be required. Thus, in the case of an

academic, income observations over the course of 35 to 40 years would need to be

available (Schnitzlein, 2009). However, within very long-lasting surveys, the number

of people who continue to participate is often considerably reduced. This so-called

panel mortality can correlate with certain characteristics of a person (e.g., income

or education), resulting in a relatively homogeneous longitudinal sample (Fitzgerald

et al., 1998). This circumstance can lead to substantial distortions of the estimation

parameters (panel attrition bias ) (Solon, 1989, 1992).

For this reason, lifetime incomes are usually approximated by means of annual

income observations, which consist of a permanent component and a uctuating

component (Solon, 1989, 1992, Zimmerman, 1992). If parental income is approxi-

mated by income data from only one particular point in time, the classical errors-in-

variables problem occurs and leads to a systematic downward bias of the estimated

68
 Transmission Channels of Intergenerational Income Persistence

intergenerational income elasticity (attenuation bias ) (Wooldridge, 2010). There-

fore, Solon (1992) proposes to form an average of ve annual income observations

for the parental generation in order to reduce the variance of the uctuating compo-

nent. This procedure does not completely eliminate the bias, but it can signicantly

reduce it. The estimator for the intergenerational income elasticity can then be

22
interpreted as a lower bound for the true estimation parameter.

Haider and Solon (2006) additionally point out that the approximation of chil-

dren's lifetime income depends on the chosen stage of life. Since individual income

during a person's working life assumes a hump-shaped run, income observations at

young ages are lower and thus the lifetime income of a person is underestimated.

Meanwhile, income dierences between high- and low-skilled workers are smaller at

the beginning of their working lives and only increase over time. If incomes are thus

observed at the beginning of the son's working life, this in turn leads to a downward

bias of intergenerational income elasticity (life-cycle bias ). This circumstance is

veried by Böhlmark and Lindquist (2006) for Sweden and Brenner (2010) for Ger-

many. Haider and Solon (2006) show that the age range between the mid-30s and

mid-40s produces a good approximation of the sons' lifetime income. Schnitzlein

(2016) uses the income of sons between 35 and 42 years of age.

3.3.2 Sample denition and variables

The selected samples from the SOEP and the PSID are dened congruently so as to

ensure the reliable comparability of the results. The analysis is based on data from

the years from 1984 to 2013. The individual annual labor income is used, which in-

cludes wages and salaries from both paid employment and self-employment as well

as bonus payments, income from overtime, and prot sharing (Grabka, 2014, Lil-

23
lard, 2013). The SOEP sample does not include imputed income data. All income

22 In the approximation of the children's lifetime income, measurement errors only lead to higher
standard errors.
23 Missing income statements are estimated in the SOEP with the help of personal and household
characteristics as well as past income data (Frick et al., 2012). The CNEF-PSID features no
imputed income data.

69
 Transmission Channels of Intergenerational Income Persistence

24
statements are deated to 2010. In order to be able to compare the results with

the existing literature, annual real incomes of less than 1,200 Euro/US dollar are not

included in the estimates. To avoid a bias due to wage developments in East Ger-

many after reunication, the analysis for Germany is limited to persons who lived

in West Germany in 1989 (Schnitzlein, 2009). In order to estimate the intergenera-

tional educational persistence, we utilize years of schooling as an approximation for

25
fathers' and sons' level of education.

The generation of the parents is restricted to the income observations of the

26
fathers and the generation of the children to the income observations of the sons.

Fathers' incomes are drawn from the period from 1984 to 1993, from which at least

ve valid income observations must be available. The lifetime income of the fathers

is approximated by the formation of the average of the annual incomes. Only income

observations from the ages of 30 to 55 are considered. Thus, the fathers belong to

the birth cohorts of the period from 1933 to 1959. The income observations of the

sons are drawn from the years from 2003 to 2013, during which time period at least

one valid income observation must be available. Again, the lifetime income of the

sons is approximated by the formation of the average of the annual incomes. Only

incomes from the ages of 35 to 42 are taken into account. Thus, the sons belong to

the birth cohorts of the period from 1961 to 1978, which do not overlap with the

cohorts of their fathers.

Finally, a total of 353 and 602 father-son pairs are recorded in the SOEP and the

PSID, respectively (Table 3.1). On average, the sons earn more than their fathers

in both countries. In Germany the average income of the sons is 15.6 percent higher

than the average income of the fathers, while in the United States it is only 5.1

percent higher than the average income of the fathers. The average age of the

Consumer Price Index and, for the PSID, the Consumer Price Index of
24 For the SOEP, the
All Urban Consumers and All Items based on the recommendation of Grieger et al. (2009) are
utilized.
25 This approach implicitly assumes that the impact of one more year of schooling on the level
of education is linear and constant across nations and generations.
26 This limitation is due to the divergent labor market participation of women in both countries,
which can lead to a bias of dierences in intergenerational income elasticity.

70
 Transmission Channels of Intergenerational Income Persistence

fathers is mid-40s in both countries, older than that of the sons, whose average age

is late-30s. The younger age of the sons might also determine the higher variance

in incomes. German fathers on average spent 10.93 years in school, while their

sons received 12.75 years of schooling. In the United States, fathers' and sons'

educational attainment is relatively similar, with 13.20 and 13.82 years of schooling,

respectively. On the one hand, the fathers in the United States might spend more

years in school due to the longer compulsory school attendance period. While in

most German federal states, 9 years of schooling are mandatory, most U.S. states

require children to stay in school until the age of 16 or 18. On the other hand, the

aftermath of World War II might have signicantly contributed to the fathers' fewer

years in education in Germany.

Table 3.1: Descriptive statistics

Fathers Sons
Mean Std. Dev. Mean Std. Dev.

SOEP
Income 40,590.37 19,576.16 46,941.29 27,652.96

Age 46.78 4.54 38.17 1.79

Education years 10.93 2.55 12.75 2.94

Father-son pairs 353

PSID
Income 64,019.61 59,658.27 67,280.92 69,782.23

Age 43.82 5.46 37.87 1.88

Education years 13.20 2.41 13.82 2.03

Father-son pairs 602

Source: SOEP (1984-2013), PSID (1984-2013).

3.3.3 Descriptive evidence

The logarithmized incomes of the fathers and sons exhibit a positive correlation

(Figure 3.2). The slope of the line of best t from the bivariate OLS regression is

higher for the United States than for Germany. However, it is also obvious that the

71
 Transmission Channels of Intergenerational Income Persistence

income data points in both countries are heavily scattered around the regression

line. In order to examine the simple linear relationship more closely, a bivariate

Nadaraya-Watson (NW) estimation is additionally depicted. Both countries show

deviations compared to the OLS estimation. However, the 95 percent condence in-

tervals include the OLS regression line over nearly the entire distribution of paternal

income. From the bivariate evidence, therefore, it cannot be concluded that the in-

tergenerational income elasticity changes signicantly along the income distribution

of the fathers.

Figure 3.2: Intergenerational income correlation

Germany United States

13 14

12
12
Log. income of son

Log. income of son

11

10
10

9 8

6
9 10 11 12 9 10 11 12 13 14
Log. income of father Log. income of father

Income data OLS Income data OLS

NW 95% confidence interval NW 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a range based on
Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson.

Figure 3.3 shows the correlation between the education years of fathers and

their sons. Here, the slope for Germany is higher than that of the United States,

implying that sons' years of schooling depend more strongly on the education years

of their fathers in Germany than in the United States. However, while the 95

72
 Transmission Channels of Intergenerational Income Persistence

percent condence interval of the NW estimation almost completely contains the

OLS estimator in Germany, the results signicantly deviate from linearity in the

lower education percentiles in the United States. Thus, sons of low-skilled fathers

receive better education than the OLS estimation would predict. This nonlinearity

might also explain the lower OLS regression slope in the United States.

Figure 3.3: Intergenerational educational correlation

Germany United States

20 20
Years of education (son)

Years of education (son)

15 15

10 10

5 5
5 10 15 20 5 10 15 20
Years of education (father) Years of education (father)

Income data OLS Income data OLS

NW 95% confidence interval NW 95% confidence interval

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a range based on
Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson.

3.4 Empirical results

The bivariate estimations in the previous section give a rst impression of the dier-

ences in the intergenerational income and educational persistence between Germany

and the United States. However, in order to avoid distortions of the estimators due

to divergent age and cohort structures, additional control variables are considered

in accordance with Schnitzlein (2016). With the inclusion of age polynomials and

73
 Transmission Channels of Intergenerational Income Persistence

the birth years of fathers and sons as well as the number of valid observations of

the son, the obtained estimators slightly decrease (Table 3.2). Notwithstanding,

Germany still shows a lower intergenerational income persistence, with an estimate

of 33 percent, than the United States, with an obtained value of 45 percent.

Table 3.2: Intergenerational income and educational persistence

Germany
β 0.331*** 0.331*** 0.331***

(0.083) (0.081) (0.081)

γ 0.545*** 0.530*** 0.535***

(0.054) (0.054) (0.056)

Cohort controls No Yes Yes No Yes Yes

Age controls No No Yes No No Yes

2
R 0.053 0.083 0.089 0.224 0.236 0.243

Obs. 353 353 353 353 353 353

United States
β 0.486*** 0.455*** 0.452***

(0.069) (0.071) (0.071)

γ 0.449*** 0.441*** 0.435***

(0.029) (0.030) (0.030)

Cohort controls No Yes Yes No Yes Yes

Age controls No No Yes No No Yes

2
R 0.108 0.140 0.143 0.284 0.295 0.297

Obs. 602 602 602 602 602 602

Source: SOEP (1984-2013), PSID (1984-2013).

Note: Cohort controls include birth years of the fathers and the sons. Age controls include poly-
nomials of the average age of fathers and sons as well as the number of valid observations of the
sons. Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

In contrast, families in the United States experience a lower intergenerational

educational persistence than families in Germany. While one more education year

of a father in the United States provides his sons with an average of 0.44 additional

years of schooling, the education years of German sons increase by 0.54 years. These

contrasting results illustrate that intergenerational educational persistence cannot be

74
 Transmission Channels of Intergenerational Income Persistence

perfectly transformed into intergenerational income persistence. Instead, additional

determinants might strongly inuence intergenerational income persistence, such

that countries can even switch positions in the ranking of intergenerational mobility.

3.4.1 Descriptive decomposition

Table 3.3 shows the results of the linear descriptive decomposition as described in

Equation (3.5). In Germany, the return to education is lower in both generations

than in the United States. While in Germany, each year of education raises a father's

(son's) income by 8.9 percent (8.7 percent), the United States exhibits a value of 16.7

percent (11.5 percent) for the fathers (sons). Thus, the rate of return to education

has declined in both countries over time, but the drop is stronger in the United

States than in Germany.

Table 3.3: Linear descriptive decomposition

Cov(log(y s ),ν f ) Cov(ν s ,Edf )

β γ δf δs 2
REd f Var(ν f ) Var(Edf )
Germany 0.331*** 0.545*** 0.089*** 0.087*** 0.320*** 0.254** -0.005

(0.083) (0.054) (0.010) (0.008) (0.044) (0.104) (0.012)

United States 0.486*** 0.449*** 0.167*** 0.115*** 0.229*** 0.374*** 0.024

(0.069) (0.029) (0.017) (0.011) (0.033) (0.074) (0.012)

β Endowment eect Investment eect Cross-eect

Germany 0.331*** 0.179*** 0.172** -0.020

(0.083) (0.032) (0.071) (0.046)

United States 0.486*** 0.149*** 0.289*** 0.048**

(0.069) (0.023) (0.057) (0.024)

Dierence -0.155 0.030 -0.117 -0.068

Source: SOEP (1984-2013), PSID (1984-2013).
Note: Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

In contrast, the variation in years of schooling explains a greater portion of

the variation in parental income in Germany than in the United States. While in

Germany, 32 percent of income dierences are attributable to fathers' education,

75
 Transmission Channels of Intergenerational Income Persistence

in the United States only 23 percent of the income variance can be traced back to

educational dierences. However, the relationship between sons' income and the

predicted luck component of their fathers' income is stronger in the United States,

with a value of 0.37, than in Germany, with a value of 0.25. The estimated coecient

from a regression of the luck component of a son's income on the education years

of his father does not signicantly deviate from zero either in Germany or in the

United States. Linking these results to the estimates of intergenerational income

and educational persistence, the linear descriptive decomposition reports a higher

endowment eect for Germany, with a value of 54 percent of the intergenerational

income elasticity, than for the United States, with a value of 31 percent. In contrast,

the United States exhibits a higher investment eect, with a value of 59 percent,

than Germany, with a value of 52 percent. The cross-eect of the father's education

on the son's residual income amounts to 10 percent in the United States and -6

percent in Germany. However, the value for Germany is statistically insignicant.

Regarding the 47 percent dierence in intergenerational income elasticity between

Germany and the United States, more than three quarters is due to the dierence in

the investment eect. Thus, if the investment eect were the same in both countries,

the gap between Germany and the United States would merely be 11 percent.

The linear descriptive decomposition estimates the average endowment and in-

vestment eects in both countries. However, there might be considerable nonlineari-

ties in the relative importance of the two components across the income distribution.

Thus, in a further step we apply unconditional quantile regressions to Equations (3.1)

and (3.4) (Table 3.4). In Germany, intergenerational income persistence increases

with the son's income until the 40th income quantile and remains relatively con-

stant hereafter at about 0.4. In contrast, the United States exhibits a u-shaped run

of the intergenerational income elasticity, indicating that parental income is more

important at the edges of the sons' income distribution than in the middle.

76
 Transmission Channels of Intergenerational Income Persistence

Table 3.4: Nonlinear descriptive decomposition

Germany β Endowment eect Investment eect Cross-eect

20th percentile 0.262** 0.148*** 0.089 0.024

(0.103) (0.032) (0.084) (0.055)

40th percentile 0.424*** 0.184*** 0.237*** 0.003

(0.076) (0.031) (0.065) (0.040)

50th percentile 0.411*** 0.188*** 0.228*** -0.005

(0.076) (0.032) (0.067) (0.036)

60th percentile 0.399*** 0.189*** 0.191*** 0.019

(0.080) (0.034) (0.069) (0.037)

80th percentile 0.418*** 0.190*** 0.191** 0.038

(0.106) (0.040) (0.086) (0.051)

United States β Endowment eect Investment eect Cross-eect

20th percentile 0.442*** 0.167*** 0.236*** 0.039

(0.086) (0.030) (0.075) (0.034)

40th percentile 0.406*** 0.129*** 0.224*** 0.053**

(0.054) (0.019) (0.049) (0.025)

50th percentile 0.407*** 0.122*** 0.223*** 0.063***

(0.051) (0.019) (0.045) (0.022)

60th percentile 0.392*** 0.132*** 0.202*** 0.058***

(0.052) (0.020) (0.045) (0.022)

80th percentile 0.467*** 0.133*** 0.256*** 0.078**

(0.079) (0.023) (0.062) (0.030)

Source: SOEP (1984-2013), PSID (1984-2013).

Note: Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

In conclusion, two suggestions can be drawn from these results. Firstly, the

Bratsberg et al. (2007) conjecture of a convex run of intergenerational income elas-

ticity applies to Germany rather than to the United States. This seems reasonable,

as the education system in Germany is largely funded by the public sector, while

privatization in the United States is strong. Thus, there might be a compensating

eect at the lower end of the income distribution in Germany. Applying uncondi-

tional quantile regressions to the intergenerational educational persistence conrms

77
 Transmission Channels of Intergenerational Income Persistence

27
this suggestion. Secondly, the pattern of the intergenerational income elasticity in

the United States at least partially reects the Becker and Tomes (1986) conjecture,

as the estimated values decrease at the upper end of the sons' income distribution.

Remarkably, the intergenerational educational persistence exhibits an exactly in-

verse shape across the ascending percentiles. The highest estimates are obtained in

the middle of the educational distribution of the sons, while the values at the two

ends are notably smaller.

Figure 3.4: Nonlinear descriptive decomposition

Endowment effect Investment and cross−effect

80 80

70 70

60 60
Percent of IIE

Percent of IIE

50 50

40 40

30 30

20 20
20 30 40 50 60 70 80 20 30 40 50 60 70 80
Income quantile (son) Income quantile (son)

Germany United States Germany United States

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Hollow circles are insignicant estimates at p = 0.05. IIE: Intergenerational income elas-
ticity.

Figure 3.4 shows the development of the endowment and investment eects across

the ascending income deciles of the sons. In the United States, the endowment eect

is signicantly smaller over the entire income distribution, with a value of around 30

percent, than the investment eect, with a value of approximatly 70 percent. At the

27 See Figure 3.5 in the Appendix.

78
 Transmission Channels of Intergenerational Income Persistence

lower end of the income distribution, the endowment eect is slightly higher and,

consequently, the investment eect is somewhat lower. In Germany, the endowment

and investment eects are relatively constant at approximately 50 percent. Overall,

the endowment eect is stronger across all income percentiles in Germany, while the

investment eect is continuously higher in the United States. Since the parameters

of Equations (3.2) and (3.3) take on the same values across the entire income distri-

bution, deviations in the relative importance of the endowment eect can be traced

back to nonlinearities in the intergenerational income elasticity or in the rate of re-

turn to sons' educational attainment. The latter is almost constant across the sons'

income distribution in Germany and only slightly deviates downwards at the 20th

28
percentile. In contrast, in the United States the sons' rate of return to education

takes on a slightly u-shaped curve over the ascending income deciles.

3.4.2 Structural decomposition

Although descriptive decomposition methods are a good starting point for rst in-

sights into the transmission channels of intergenerational income persistence, they

neglect the transfer of unobserved determinants within the family. In order to over-

come this weakness using the structural decomposition method of Lefgren et al.

(2012), valid instrument variables have to be constructed.

For the human capital component of fathers' incomes, we employ years of edu-

cation, level of education, and educational attainment. Fathers' years of education

measure total years of schooling, vocational training, and university education. As

this variable exhibits peaks at certain values due to the organization of the national

education system, the level of education is dened as an ordered variable with ve

29
levels based on a father's years of education. In contrast, educational attainment

30
indicates the level of fathers' education with respect to high school education.

28 See Table 3.8 in the Appendix.

29 (1) less than 9.5 years, (2) 9.5 to 11.4 years, (3) 11.5 to 13.4 years, (4) 13.5 to 17.4 years, (5)
more than 17.4 years.
30 (1) less than high school education, (2) high school education, (3) more than high school
education.

79
 Transmission Channels of Intergenerational Income Persistence

These instruments should be highly correlated with fathers' human capital, but can

be considered more or less independent of fathers' fortune on the labor market.

In order to measure the luck component of fathers' incomes, two instruments are

constructed based on fathers' months of unemployment in the observation period.

This variable is likely to be highly correlated with fathers' human capital, though

unemployment might also occur by chance due to exogenous shocks such as mass

layos or changing family commitments. Therefore, in the rst stage, months of

unemployment are regressed on fathers' past income, level of education, and further

human capital variables such as occupation and industry. Ultimately, the residuals

from this regression are used as an instrument for fathers' fortune on the labor

market. In addition, a second instrumental variable indicating the probability that

a fatherdepending on his income, level of education, occupation, and industry in

the rst three years of the observation periodwas unemployed at least once in

31
the subsequent two or more observed years is constructed. The obtained residuals

should thus be orthogonal to the father's human capital by construction.

Table 3.5 shows the results of an IV estimation of Equation (3.1) using the above

instrument variables. Instrumenting paternal income with years of education, level

of education, and educational attainment yields quite similar results for both coun-

tries. Regardless of the chosen instrument, the obtained values are always higher

than the corresponding OLS estimates. However, the bootstrapped Durbin-Wu-

Hausman test indicates a signicant dierence between β̂ OLS and β̂ IV only for the

United States. As the high standard errors in Germany are likely to occur due to

the relatively small number of observations, we nevertheless consider it reasonable

to reject a one-factor model of intergenerational income transmission. Using the

residuals of unemployment and employment status as instrument variables results

in lower estimates for the intergenerational income elasticity compared to the cor-

responding OLS values. However, the IV estimates do not signicantly dier from

zero in both countries and the results of the Durbin-Wu-Hausman test still do not

31 Since the dependent variable is binary, generalized residuals from a probit regressions are
calculated following the approach of Gourieroux et al. (1987).

80
 Transmission Channels of Intergenerational Income Persistence

support a rejection of the null hypothesis of equal values for Germany. Nevertheless,

Lefgren et al. (2012) show that an imperfect instrument for luck which is a valid

measure for an upper bound of π1 is sucient.

Table 3.5: IV estimation of intergenerational income elasticity

Germany Years of Level of Educational Unemployment Employment

education education attainment residuals status residuals

β 0.495*** 0.527*** 0.554*** 0.188 0.390

(0.177) (0.170) (0.192) (0.472) (1.113)

p-value Durbin- 0.280 0.150 0.203 0.280 0.280

Wu-Hausman
test

First-stage 172.750*** 45.646*** 64.575*** 20.134*** 8.141***

F-statistic

United States Years of Level of Educational Unemployment Employment

education education attainment residuals status residuals

β 0.863*** 0.899*** 0.869*** 0.341 0.401

(0.146) (0.160) (0.160) (0.288) (0.289)

p-value Durbin- 0.006 0.006 0.014 0.006 0.006

Wu-Hausman
test

First-stage 177.933*** 44.121*** 60.834*** 38.364*** 31.282***

F-statistic

Source: SOEP (1984-2013), PSID (1984-2013).

Note: Other control variables include the father's and the son's year of birth as well as the num-
ber of valid observations of the son. Standard errors are clustered at the family level and were
calculated using paired bootstrap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05,
* p < 0.1.

Since the results for the various human capital variables are of equal size, we use

fathers' years of education as an instrument to establish a lower bound for π1 + π2 .

In order to obtain an upper bound for π1 , we use fathers' unemployment residuals

because the respective IV estimates for both Germany and the United States are

smaller than those obtained with fathers' employment status residuals. Once these

f
IV estimations are combined with the OLS estimation, Var(HC )/(Var(HC f ) +
f
Var(ν )) can be calculated via Equation (3.10). The results of this rst bounding

procedure are presented in Table 3.6.

81
 Transmission Channels of Intergenerational Income Persistence

Table 3.6: Structural decomposition I

2
π1 + π2 π1 π2 REdf Investment Endowment
eect eect

Germany 0.495*** 0.188 0.307 0.466 0.568 0.432

(0.177) (0.472) (0.520) (30.265) (1.712) (1.712)

United States 0.863*** 0.341 0.523* 0.278 0.701 0.299

(0.146) (0.288) (0.317) (4.456) (0.543) (0.543)

Source: SOEP (1984-2013), PSID (1984-2013).

Note: Other control variables include the father's and the son's year of birth as well as the num-
ber of valid observations of the son. Standard errors are clustered at family level and were cal-
culated using paired bootstrap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05,
* p < 0.1.

The structural decomposition suggests an upper bound for the mechanistic eect

of fathers' nancial resources (π1 ) of 0.19 in Germany and 0.34 in the United States.

In contrast, the estimated lower bound for the mechanistic eect of fathers' human

capital (π2 ) is substantially higher, with a value of 0.31 in Germany and a value

of 0.52 in the United States. Combining these estimates with the respective OLS

estimator, an upper bound for the investment eect in Germany of 57 percent is

obtained, while the value of 70 percent for the United States is markedly higher.

Consequently, a lower bound for the endowment eect is estimated to be 43 percent

in Germany and 30 percent in the United States. However, the investment and en-

dowment eects are insignicant for both countries. Overall, the direct estimation

of π1 suers from two problems. On the one hand, constructing a valid instrument

for the luck component of a father's income within the given data set has turned

out to be somewhat problematic. On the other hand, the relatively small number

of observations produces high standard errors of the estimation parameters. Nev-

ertheless, the values are in line with the results of the descriptive decomposition in

Section 3.4.1.

The results of the alternative bounding procedure avoiding the direct estimation

of π1 are presented in Table 3.7. While a lower bound for π1 + π2 is again estimated

using fathers' years of education as an instrument for human capital, a lower bound

2
for REd f is now directly drawn from a Mincer regression of the father's income on

82
 Transmission Channels of Intergenerational Income Persistence

several human capital variables such as level of education, occupation, and industry.

Thus, in combination with the OLS results, an upper bound for π1 and a lower

bound for π2 can be calculated using Equation (3.10). This approach produces

more precise coecients for the United States. The same is true for the variation

of fathers' income due to human capital in Germany, though the results for the

investment and endowment eects are still not signicant at the 10 percent level.

Table 3.7: Structural decomposition II

2
π1 + π2 π1 π2 REdf Investment Endowment
eect eect

Germany 0.495*** 0.196 0.299 0.452*** 0.591 0.409

(0.177) (0.159) (0.285) (0.050) (0.497) (0.497)

United States 0.863*** 0.344*** 0.519*** 0.273*** 0.708*** 0.292**

(0.146) (0.094) (0.184) (0.039) (0.127) (0.127)

Source: SOEP (1984-2013), PSID (1984-2013).

Note: Other control variables include the father's and the son's year of birth as well as the num-
ber of valid observations of the son. Standard errors are clustered at the family level were cal-
culated using paired bootstrap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05,
* p < 0.1.

The second decomposition yields an upper bound for π1 of 0.20 in Germany

and 0.34 in the United States. The estimation of the extended version of Equation

2
(3.3) yields a lower bound for REd f of 45 percent in Germany and 27 percent in

the United States. Consequently, a lower bound for π2 is estimated to be 0.30 in

Germany and 0.52 in the United States. In both countries, the upper bound for the

investment eect increases slightly, to 59 percent in Germany and 71 percent in the

United States. Hence, the obtained lower bounds for the endowment eect decrease

slightly to 41 percent and 29 percent, respectively. Overall, the obtained values are

very similar to those in Table 3.6.

In total, the results of the structural decomposition support the ndings of the

descriptive decomposition with an investment eect and an endowment eect of

approximately equal size in Germany and a signicantly higher investment eect

in the United States. Thus, we conclude that sons in the United States are more

83
 Transmission Channels of Intergenerational Income Persistence

reliant on the nancial resources of their fathers, whereas the transmission of human

capital within the family is more substantial in Germany.

3.5 Conclusion

This chapter analyzes the transmission channels of intergenerational income persis-

tence in Germany and the United States. Using a descriptive decomposition method,

we nd that the mechanistic eects of fathers' nancial resources and human capital

are about equally high in Germany, while the investment eect in the United States

accounts for approximately 70 percent of the intergenerational income elasticity.

We estimate stronger nonlinearities only in the lower income quantiles in the United

States, where the endowment eect is somewhat more pronounced. The results of

the structural decomposition method using dierent bounding procedures support

this supposition. However, the values should be interpreted with caution since they

are insignicant for the most part.

The overall result of a stronger impact of the investment eect in the United

States seems reasonable in light of a signicantly higher level of privatization in

the education system than in Germany. The large cross-country dierences in the

relative contribution of the two transmission channels emphasize that policy makers

should not only focus on the level of intergenerational income mobility alone, but

also on the underlying transmission mechanisms. If the endowment eectas in the

case of Germanyis very pronounced, equality of opportunity for children born to

poor parents cannot be reached by the supply of nancial means alone. Conversely,

an ecient policy must additionally substitute for the missing direct transmission

of human capital within socio-economically weak families. Appropriate means to

improve intergenerational income mobility in this case might be the expansion and

improvement of early childcare facilities, kindergartens, and (full-time) schools with

a good sta-to-student ratio and well-trained educators.

84
 Transmission Channels of Intergenerational Income Persistence

References

Becker, G. S. and Tomes, N. (1979). An equilibrium theory of the distribution

of income and intergenerational mobility. Journal of Political Economy, 87 (6),

11531189.

Becker, G. S. and Tomes, N. (1986). Human capital and the rise and fall of families.

Journal of Labor Economics, 4 (3), S1S39.

Behrman, J. R. and Rosenzweig, M. R. (2002). Does increasing women's schooling

raise the schooling of the next generation? American Economic Review, 92 (1),

323334.

Björklund, A. and Jäntti, M. (1997). Intergenerational income mobility in Sweden

compared to the United States. American Economic Review, 87 (5), 10091018.

Björklund, A. and Jäntti, M. (2009). Intergenerational income mobility and the role

of family background. In Salverda, W., Nolan, B., and Smeeding, T. M., editors,

Oxford Handbook of Economic Inequality, 491521.

Black, S. E. and Devereux, P. J. (2011). Recent developments in intergenerational

mobility. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics,

4B, 14871541.

Blanden, J. (2013). Cross-country rankings in intergenerational mobility: A compar-

ison of approaches from economics and sociology. Journal of Economic Surveys,

27 (1), 3873.

Blanden, J. and Gregg, P. (2004). Family income and educational attainment: A

review of approaches and evidence for Britain. Oxford Review of Economic Policy,

20 (2), 245263.

Blau, D. M. (1999). The eect of income on child development. Review of Economics

and Statistics, 81 (2), 261276.

85
 Transmission Channels of Intergenerational Income Persistence

Böhlmark, A. and Lindquist, M. J. (2006). Life-cycle variations in the associa-

tion between current and lifetime income: Replication and extension for Sweden.

Journal of Labor Economics, 24 (4), 879896.

Bratberg, E., Davis, J., Mazumder, B., Nybom, M., Schnitzlein, D. D., and Vaage,

K. (2017). A comparison of intergenerational mobility curves in Germany, Norway,

Sweden, and the US. Scandinavian Journal of Economics, 119 (1), 72101.

Bratsberg, B., Røed, K., Raaum, O., Naylor, R., Jännti, M., Eriksson, T., and

Österbacka, E. (2007). Nonlinearities in intergenerational earnings mobility: Con-

sequences for cross-country comparisons. Economic Journal, 117 (519), C72C92.

Brenner, J. (2010). Life-cycle variations in the association between current and

lifetime earnings: Evidence for German natives and guest workers. Labour Eco-

nomics, 17 (2), 392406.

Cardak, B. A., Johnston, D. W., and Martin, V. L. (2013). Intergenerational earn-

ings mobility: A new decomposition of investment and endowment eects. Labour

Economics, 24, 3947.

Chetty, R., Hendren, N., Kline, P., Saez, E., and Turner, N. (2014). Is the United

States still a land of opportunity? Recent trends in intergenerational mobility.

American Economic Review: Papers & Proceedings, 104 (5), 141147.

Corak, M. (2006). Do poor children become poor adults? Lessons from a cross-

country comparison of generational earnings mobility. In Creedy, J. and Kalb,

G., editors, Dynamics of Inequality and Poverty, 143188.

Corak, M., Lindquist, M. J., and Mazumder, B. (2014). A comparison of upward and

downward intergenerational mobility in Canada, Sweden and the United States.

Labour Economics, 30, 185200.

Dahl, G. B. and Lochner, L. (2012). The impact of family income on child achieve-

ment: Evidence from the Earned Income Tax Credit. American Economic Review,

102 (5), 19271956.

86
 Transmission Channels of Intergenerational Income Persistence

Fertig, A. R. (2003). Trends in intergenerational earnings mobility in the U.S.

Journal of Income Distribution, 12 (34), 108130.

Firpo, S., Fortin, N. M., and Lemieux, T. (2009). Unconditional quantile regressions.

Econometrica, 77 (3), 953973.

Fitzgerald, J., Gottschalk, P., and Mott, R. (1998). An analysis of sample attrition

in panel data: The Michigan Panel Study of Income Dynamics. Journal of Human

Resources, 33 (2), 251299.

Frick, J. R., Grabka, M. M., and Groh-Samberg, O. (2012). Dealing with incomplete

household panel data in inequality research. Sociological Methods & Research, 41

(1), 89123.

Frick, J. R., Jenkins, S. P., Lillard, D. R., Lipps, O., and Wooden, M. (2007). The

Cross-National Equivalent File (CNEF) and its member country household panel

studies. Journal of Contextual Economics, 127 (4), 627654.

Gourieroux, C., Monfort, A., Renault, E., and Trognon, A. (1987). Generalised

residuals. Journal of Econometrics, 34 (12), 532.

Grabka, M. M. (2014). Codebook for the $PEQUIV File 1984-2013: CNEF vari-

ables with extended income information for the SOEP. Data Documentation 74,

German Institute for Economic Research, Berlin, Germany.

Grawe, N. D. (2004). Reconsidering the use of nonlinearities in intergenerational

earnings mobility as a test for credit constraints. Journal of Human Resources,

39 (3), 813827.

Grawe, N. D. and Mulligan, C. B. (2002). Economic interpretations of intergenera-

tional correlations. Journal of Economic Perspectives, 16 (3), 4558.

Grieger, L. D., Danziger, S., and Schoeni, R. F. (2009). Accurately measuring the

trend in poverty in the United States using the Panel Study of Income Dynamics.

Journal of Economic and Social Measurement, 34 (23), 105117.

87
 Transmission Channels of Intergenerational Income Persistence

Haider, S. and Solon, G. (2006). Life-cycle variation in the association between

current and lifetime earnings. American Economic Review, 96 (4), 13081320.

Han, S. and Mulligan, C. B. (2001). Human capital, heterogeneity and estimated

degrees of intergenerational mobility. Economic Journal, 111 (470), 207243.

Hauser, R. M. and Logan, J. A. (1992). How not to measure intergenerational

occupational persistence. American Journal of Sociology, 97 (6), 16891711.

Hirvonen, L. (2010). Accounting for intergenerational earnings persistence: Can

we distinguish between education, skills, and health? Working Paper 2/2010,

Swedish Institute for Social Research, Stockholm, Sweden.

Holmlund, H., Lindahl, M., and Plug, E. (2011). The causal eect of parents'

schooling on children's schooling: A comparison of estimation methods. Journal

of Economic Literature, 49 (3), 615651.

Lee, C. and Solon, G. (2009). Trends in intergenerational income mobility. Review

of Economics and Statistics, 91 (4), 766772.

Lefgren, L., Lindquist, M. J., and Sims, D. (2012). Rich dad, smart dad: Decompos-

ing the intergenerational transmission of income. Journal of Political Economy,

120 (2), 268303.

Lillard, D. R. (2013). Codebook for the Cross-National Equivalent File 1970-2013:

BHPS  HILDA  KLIPS  PSID  RLMS-HSE  SHP  SLID  SOEP. Colum-

bus, United States.

Oreopoulos, P., Page, M. E., and Stevens, A. H. (2006). The intergenerational eects

of compulsory schooling. Journal of Labor Economics, 24 (4), 729760.

Schnitzlein, D. D. (2009). Structure and extent of intergenerational income mobility

in Germany. Journal of Economics and Statistics, 229 (4), 450466.

Schnitzlein, D. D. (2016). A new look at intergenerational income mobility in Ger-

many compared to the U.S. Review of Income and Wealth, 62 (4), 650667.

88
 Transmission Channels of Intergenerational Income Persistence

Solon, G. (1989). Biases in the estimation of intergenerational earnings correlations.

Review of Economics and Statistics, 71 (1), 172174.

Solon, G. (1992). Intergenerational income mobility in the United States. American

Economic Review, 82 (3), 393408.

Solon, G. (1999). Intergenerational mobility in the labor market. In Ashenfelter, O.

and Card, D., editors, Handbook of Labor Economics, 3A, 17611800.

Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data.

MIT Press, Cambridge, United States.

Zimmerman, D. J. (1992). Regression toward mediocrity in economic stature. Amer-

ican Economic Review, 82 (3), 409492.

89
 Transmission Channels of Intergenerational Income Persistence

Appendix

Figure 3.5: Intergenerational persistence along the income distribution

Income Years of education

1.0
1.0

0.8
Intergenerational persistence

Intergenerational persistence
0.8

0.6 0.6

0.4
0.4

0.2
0.2
20 30 40 50 60 70 80 20 30 40 50 60 70 80
Income quantile (son) Income quantile (son)

Germany United States Germany United States

Source: SOEP (1984-2013), PSID (1984-2013).

90
 Transmission Channels of Intergenerational Income Persistence

Table 3.8: Nonlinear descriptive decomposition (detailed estimates)

Germany β γ δf δs 2
REd f
Cov(log(y s ),ν f ) Cov(ν s ,Edf )
Var(ν f ) Var(Edf )
20th percentile 0.262** 0.545*** 0.087*** 0.074*** 0.320*** 0.131 0.007

(0.103) (0.054) (0.008) (0.012) (0.044) (0.124) (0.015)

40th percentile 0.424*** 0.545*** 0.087*** 0.092*** 0.320*** 0.349*** 0.001

(0.076) (0.054) (0.008) (0.008) (0.044) (0.094) (0.011)

50th percentile 0.411*** 0.545*** 0.087*** 0.094*** 0.320*** 0.336*** -0.001

(0.076) (0.054) (0.008) (0.008) (0.044) (0.098) (0.010)

60th percentile 0.399*** 0.545*** 0.087*** 0.094*** 0.320*** 0.280*** 0.005

(0.080) (0.054) (0.008) (0.009) (0.044) (0.100) (0.010)

80th percentile 0.418*** 0.545*** 0.087*** 0.095*** 0.320*** 0.280** 0.010

(0.106) (0.054) (0.008) (0.013) (0.044) (0.125) (0.014)

United States β γ δf δs 2
REd f
Cov(log(y s ),ν f ) Cov(ν s ,Edf )
Var(ν f ) Var(Edf )
20th percentile 0.442*** 0.449*** 0.115*** 0.186*** 0.229*** 0.306*** 0.019

(0.086) (0.029) (0.011) (0.024) (0.033) (0.097) (0.017)

40th percentile 0.406*** 0.449*** 0.115*** 0.144*** 0.229*** 0.291*** 0.027**

(0.054) (0.029) (0.011) (0.015) (0.033) (0.063) (0.012)

50th percentile 0.407*** 0.449*** 0.115*** 0.136*** 0.229*** 0.289*** 0.031***

(0.051) (0.029) (0.011) (0.015) (0.033) (0.059) (0.010)

60th percentile 0.392*** 0.449*** 0.115*** 0.148*** 0.229*** 0.262*** 0.029***

(0.052) (0.029) (0.011) (0.015) (0.033) (0.058) (0.010)

80th percentile 0.467*** 0.449*** 0.115*** 0.149*** 0.229*** 0.331*** 0.039***

(0.079) (0.029) (0.011) (0.020) (0.033) (0.082) (0.014)

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Standard errors are clustered at the family level and were calculated using paired boot-
strap resampling with 1,000 replications. *** p < 0.01, ** p < 0.05, * p < 0.1.

91
Chapter 4

Intergenerational Income Mobility

among Daughters

4.1 Introduction

Intergenerational income mobility as measured by the extent to which children's

adult income is predetermined by their parents' income has been discussed in the

economic literature for several decades (Solon, 1999, Björklund and Jäntti, 2009,

Black and Devereux, 2011). However, the majority of empirical studies focus on

the association between fathers and their sons. This restriction is commonly made

due to the lower labor market participation of women as compared to men. While

most men work full-time, married women in particular still tend to work only part-

time or not at all. Thus, the individual labor income of a daughter might be an

unreliable indicator for her actual economic status. This is especially true under

the assumption of assortative mating, i.e., if daughters from well-o families are

likely to marry rich men and decide to reduce their labor supply as a result of their

32
husbands' higher income (Chadwick and Solon, 2002).

Studies that consider daughters are relatively rare and the results vary substan-

tially. A comprehensive literature review is, for example, presented by Raaum et al.

(2007). Early analyses that consider individual labor income of daughters, such as

32 The expression assortative mating refers to any nonrandomness in the process of who marries
whom (Chadwick and Solon, 2002). The reasons for systematic mate selection are discussed in the
theoretical analyses of Lam (1988) and Becker (1991). The mostly noneconomic empirical literature
documents positive correlations between spouses with respect to age, physical size, intelligence test
scores, religion, ethnicity, and other personality traits (Epstein and Guttman, 1984). Empirical
research by economists has focused mainly on educational attainment and earnings (Kremer, 1997).

92
 Intergenerational Income Mobility among Daughters

Altonji and Dunn (1991), Peters (1992), Couch and Dunn (1997), and Mazumder

(2005), estimate about equally high intergenerational elasticities for sons and daugh-

ters in the United States. In contrast, Österberg (2000), Österbacka (2001), Brat-

berg et al. (2005, 2007), Holmlund (2006), Jäntti et al. (2006), and Hirvonen (2008)

estimate lower elasticities for women than for men in the Scandinavian countries.

Dearden et al. (1997) and Blanden et al. (2004) report higher elasticities for daugh-

ters than for sons using data for the United Kingdom. For Germany, Couch and

Dunn (1997) nd a very low and at times even negative elasticity of women's indi-

vidual earnings with respect to their parents' income. However, the cross-country

pattern of daughters is similar to that of sons, with smaller estimated elasticities in

the Nordic countries and larger values in the United States and the United Kingdom

(Raaum et al., 2007).

Chadwick and Solon (2002) show that in the United States, the elasticity of

daughters' household labor income with respect to her parents' income is of the

same magnitude as the elasticity typically found for individual earnings of sons

and their fathers. Raaum et al. (2007) nd similar elasticities of family earnings

with respect to parental earnings as the elasticities of individual earnings. They

argue that these somewhat surprising results can be explained by strong assortative

mating, which ensures that the earnings of the spouse are as closely correlated with

parents' income as the children's own earnings. Atkinson et al. (1983) estimate the

elasticity of the daughters' husbands' individual income with respect to their fathers'

earnings to be just as great as the elasticity of sons' income with respect to their

own fathers' earnings. Altonji and Dunn (1991) and Chadwick and Solon (2002)

support these ndings.

This chapter contributes to the literature on intergenerational income mobility

among daughters in Germany and the United States by presenting new results based

on data from the Socio-economic Panel (SOEP) and the Panel Study of Income Dy-

namics (PSID). The baseline regression analysis shows a higher intergenerational

income elasticity in Germany and a lower intergenerational income elasticity in the

93
 Intergenerational Income Mobility among Daughters

United States for women as compared to men. However, a separation by marital

status reveals that in both countries, unmarried women exhibit a higher intergener-

ational income elasticity than unmarried men, while married women feature a lower

intergenerational income elasticity than married men. The reason for the lower mo-

bility of unmarried women turns out to be a stronger human capital transmission

from fathers to daughters than to sons. The higher mobility of married women is

driven by a weaker human capital transmission and a higher labor supply elasticity

with respect to the spouse's income of women as compared to men. In order to

further study the eects of assortative mating, the subsample of married children is

analyzed by dierent types of income. It shows that the estimated intergenerational

income elasticity of children's household incomes is even higher than that of their in-

dividual incomes. This can be seen as an indication for strong assortative mating. If

household income is interpreted as a measure of children's actual economic welfare,

there are barely any dierences between sons and daughters. The intergenerational

income elasticity of spouses' income with respect to fathers' income is again rela-

tively high, which in turn supports the hypothesis of strong assortative mating. The

elasticity of the sons-in-law with respect to their fathers-in-law in Germany is even

higher than that of the sons with respect to their own fathers. In the following,

Section 4.2 presents a theoretical model for the interpretation of the dierences in

the intergenerational income elasticity. Section 4.3 discusses potential measurement

errors and reports some descriptive statistics of the data used. Section 4.4 presents

the empirical results before Section 4.5 concludes.

4.2 Theoretical framework

Raaum et al. (2007) provide a framework to understand why intergenerational in-

come elasticities may dier between women and men, wherein the primary mech-

anisms are assortative mating and labor supply responses both with respect to a

person's own hourly wage and with respect to the spouse's wage. In a two-adult

94
 Intergenerational Income Mobility among Daughters

household, family earnings zi consist of a person i's own earnings yi and those of

the spouse yis , i.e., zi = yi + yis . If earnings are written as the product of an average

hourly wage wi and hours worked li , then log(yi ) = log(wi )+log(li ). To represent the

intergenerational association between parents and their children, it is assumed that

the logarithmized wage of the child is a function of the logarithmized income of their

parents log(yip ) and a term εi capturing the combined eect of factors orthogonal to

parental earnings:

log(wi ) = α + λ log(yip ) + εi , (4.1)

where 0 < λ < 1. The positive correlation between the economic status of parents

and their children might be due to the genetic transmission of certain traits and

talents, the endowments a child receives at home, and nancial investments in the

human capital of the child (Becker and Tomes, 1979, 1986).

Raaum et al. (2007) emphasize that the matching of spouses into marriage is not

a completely random process, as it takes place in numerous private and professional

environments. Educational institutions are, for example, important meeting places

where the density of potential partners is high and search costs are low (Blossfeld

and Timms, 2003). Evidence from dierent countries indicates that about 20 percent

have met their spouse in school, college, or university (Lewis and Oppenheimer, 2001,

Skyt Nielsen and Svarer, 2006). Another reason might be that marriage is motivated

by the economic resources and the risk insurance it provides (Hess, 2004). Finally,

assortative mating may also arise if individual traits and skills are complements in

household production (Becker, 1973, 1974). In this model, the degree of assortative

mating is captured by means of wage resemblance within partnerships:

log(wis ) = π log(wi ) + (1 − π) log(w̄s ) + ξi , (4.2)

where 0 < π < 1. Equation (4.2) expresses how the logarithmized wage of the

spouse is composed of a weighted average of a person's own logarithmized wage

and the average logarithmized wage in the pool of potential matches log(w̄s ) plus

95
 Intergenerational Income Mobility among Daughters

a residual term ξi representing factors orthogonal to wages, where the parameter

33
π captures the extent of assortative mating. Assume further that logarithmized

hours worked are a linear function of a person's own and their spouse's logarithmized

wages represented by

log(li ) = η log(wi ) − η s log(wis ) + κi , (4.3)

where η >0 denotes the elasticity of labor supply with respect to a person's own

wage, ηs > 0 is the cross-elasticity with respect to the wage of the spouse, and κi
34
includes any individual labor supply components orthogonal to wages.

Combining Equations (4.1) to (4.3) and allowing η and ηs to dier between men

and women, the logarithmized income of a daughter log(yif ) can be expressed by

log(yif ) = β f log(yip ) + Ki , (4.4)

where β f = ((1 + η f ) − πη sf )λ denotes the intergenerational earnings elasticity

of the daughter with respect to her parents' income and Ki = κi + ((1 + η f ) −

πη sf )(α + εi ) − η sf ((1 − π) log(w̄s ) + ξi ) is the combined residual term. Thus, the

elasticity of daughters' earnings with respect to parents' income is increased by the

intergenerational transmission of human capital λ as well as by their own wage labor

supply elasticity ηf , and is lowered by the strength of assortative mating π as well

as by the cross-elasticity η sf . Assortative mating and family labor supply decisions

are therefore important components of intergenerational income mobility, even if the

analysis is restricted to individual earnings. In the same manner, Equations (4.1) to

(4.3) can be utilized to derive the association between the income of the daughter's

33 Note that assortative mating is dened in terms of potential incomes rather than realized
incomes. When spouses make joint decisions with respect to labor supply, even a close matching
on potential earnings will not necessarily imply that actual earnings are highly correlated as a
higher spousal wage induces a negative own labor supply response (Raaum et al., 2007).
34 In principle, wage elasticities can be positive or negative as long as leisure is a normal good.
The assumption of positive elasticities is justied by empirical results indicating that a person's
own wage elasticity is positive but close to zero for men and strictly positive for married women
(Killingsworth, 1983, Blundell and MaCurdy, 1999), whereas the elasticity with respect to the
partner's wage is close to zero for men and signicantly negative for women (Lundberg, 1988, Juhn
and Murphy, 1997, Eckstein and Wolpin, 1989, Devereux, 2004, Blau and Kahn, 2007).

96
 Intergenerational Income Mobility among Daughters

husband and that of her parents:

log(yisf ) = β sf log(yip ) + Kis , (4.5)

where β sf = ((1 + η m )π − η sm )λ denotes the elasticity of the husband's earnings with

respect to the income of his parents-in-law and Kis = κi +((1+η m )π −η sm )(α +εi )−
(1 + η sm )((1 − π) log(w̄s ) + ξi ) is the combined error term. Thus, the in-law elasticity

depends positively on the intergenerational transmission of human capital between

a daughter and her parents λ, the husband's own wage labor supply elasticity ηm,
and the degree of marital sorting π, while it is aected negatively by the cross-

elasticity of husbands' labor supply with respect to their wives' hourly wages η sm .
The elasticity of the husband's earnings may exceed that of the daughter's earnings

if (1 + η m + η sf )π > 1 + η f + η sm , i.e., if the degree of marital sorting is strong,

if the wife's labor supply is highly responsive to her husband's wage while being

less inuenced by her own wage, and if the husband's labor supply responds more

strongly to his own wage than to the wage of his wife. The association between

parental earnings and combined earnings of partners µf is nally represented by a

weighted average of daughters' and husbands' elasticities:

µf = (1 − θ)β f + θβ sf , (4.6)

where 0 ≤ θ = ȳ s /(ȳ + ȳ s ) ≤ 1 is the husbands' average share of household earnings.

As men on average work longer market hours and frequently receive higher wages

than women, θ will typically exceed one half.

4.3 Data and measurement issues

In order to analyze intergenerational income mobility, individual data for at least

two generations of parents and their children are required. For a valid country com-

parison, it is also necessary that the data used are highly comparable. Therefore, the

97
 Intergenerational Income Mobility among Daughters

Socio-economic Panel (SOEP) for Germany and the Panel Study of Income Dynam-

ics (PSID) for the United States are utilized in this study. Both data sets represent

long-term household surveys that capture information on children while they are

still living with their parents and follow them into adulthood. Thus, children who

leave their homes and establish their own households can continue to be covered over

time. In addition, both household surveys are part of the Cross-National Equivalent

File (CNEF) project, which oers a harmonized individual data set of the underly-

ing national household surveys. In particular, it provides a reliable data basis for

international comparisons of incomes, taxes, and transfers (Frick et al., 2007).

4.3.1 Measurement errors and life-cycle bias

To exactly measure the lifetime incomes of parents and their children, all income

statements of a respondent over their entire working life would be required. Thus,

in the case of an academic, for example, income observations of 35 to 40 years would

need to be available (Schnitzlein, 2009). However, with such a long survey period,

the number of people who continue to participate is often signicantly reduced. This

so-called panel mortality might correlate with certain characteristics of a respondent

(e.g., income or education), resulting in a relatively homogeneous longitudinal sam-

ple (Fitzgerald et al., 1998). Solon (1989, 1992) shows that this homogeneity leads

to substantial downward distortions of the estimated parameters (panel attrition

bias ).

Lifetime incomes are thus typically approximated by means of annual income

observations. However, these income statements consist of a permanent as well as

a uctuating component, where the second causes lifetime income to be determined

with measurement errors (Solon, 1989, 1992, Zimmerman, 1992). Thus, if parental

income is approximated by income data from only one particular point in time, the

classical errors-in-variables problem occurs (Wooldridge, 2010). This, in turn, leads

to a systematic downward bias of the estimated intergenerational income elasticity

(attenuation bias ). Solon (1992) proposes to form an average of ve valid annual

98
 Intergenerational Income Mobility among Daughters

income statements for the parental generation in order to reduce the variance of the

uctuating component. This procedure does not completely eliminate the bias, but

can signicantly reduce it. Since the direction of the bias is known, an estimate of

the intergenerational income elasticity can be interpreted as a lower bound for the

true estimation parameter. In the approximation of the children's lifetime income,

measurement errors only lead to higher standard errors.

In addition, Haider and Solon (2006) point out that the observations of children's

lifetime incomes depend on the chosen stage of life. On the one hand, individual

income during the working life assumes a hump-shaped run, so that the income at

the beginning of the working life is lower and thus the lifetime income of a person

is underestimated. On the other hand, dierences in income between high- and

low-skilled workers are smaller at the beginning of their working lives and steadily

increase over time. If incomes are thus observed at the beginning of the children's

working life, this leads to an underestimation of intergenerational income elasticity

(life-cycle bias ). This circumstance is veried by Böhlmark and Lindquist (2006) for

Sweden and Brenner (2010) for Germany. For the United States, Haider and Solon

(2006) show that for sons the age range between mid-30s and mid-40s produces a

good approximation of the lifetime incomes. Schnitzlein (2016) uses the income of

sons between 35 and 42 years of age for Germany.

4.3.2 Sample denition

Taking the previously mentioned problems into consideration, a sample for the anal-

ysis of intergenerational income mobility has to be formed. The selected samples

from the SOEP and the PSID are dened congruently so as to ensure reliable com-

parability of the results. The analysis is based on data from the years of 1984 to

35
2013. All income statements are deated to the year 2010. In the baseline sam-

Consumer Price Index and, for the PSID, the Consumer Price Index of
35 For the SOEP, the
All Urban Consumers and All Items based on the recommendation of Grieger et al. (2009) are
utilized.

99
 Intergenerational Income Mobility among Daughters

36
ple, individual labor earnings are used. Imputed income data are excluded from

37
the SOEP sample. In order to be able to compare the results with the existing

literature, annual real individual incomes of less than 1,200 Euro/US dollar are not

included in the estimates. To avoid a bias due to wage developments in East Ger-

many after reunication, the analysis for Germany is limited to persons who lived

in West Germany in 1989 (Schnitzlein, 2009).

Fathers' incomes are drawn from the period of 1984 to 1993, from which at least

ve valid income observations must be available. The lifetime income of the father is

approximated by the formation of the average of the annual incomes. Only income

observations from the age of 30 to 55 years are considered. Thus, the fathers belong

to the birth cohorts from the period of 1933 to 1958. The incomes of the children are

drawn from the years from 2003 to 2013, during which time period at least one valid

income observation must be available. Again, the lifetime income of the children is

approximated by the formation of the average of the annual incomes. Only incomes

from the age of 35 to 42 years are taken into account. Thus, the children belong to

the birth cohorts from the period of 1961 to 1978, which do not overlap with the

cohorts of their fathers.

A total of 354 (601) father-son and 261 (623) father-daughter pairs are thus

recorded in the SOEP (PSID). In Germany (the United States), sons' incomes aver-

age to 46,868 Euro (68,599 US dollar) and are thus higher than daughters' average

incomes of 21,553 Euro (39,937 US dollar). Fathers' incomes average to 40,333 Euro

(66,418 US dollar) and are thus slightly lower than the income average of the sons.

The average age of the fathers is mid-40s in both countries, while the children's

average age is late 30s (Table 4.1).

36 This variable captures wages and salaries from employees as well as self-employed individuals,
and includes bonus payments, overtime pay, and shares in prots (Lillard, 2013, Grabka, 2014).
37 Missing income statements are estimated in the SOEP with the help of personal and household
characteristics as well as past income data (Frick et al., 2012). The CNEF-PSID features no
imputed income data.

100
 Intergenerational Income Mobility among Daughters

Table 4.1: Descriptive statistics

SOEP Mean Std. Dev. Min. Max. N

Fathers
Income 40,332.98 18,668.57 10,347.98 153,308.70 615

Age 46.55 4.68 32 53 615

Sons
Income 46,868.19 27,724.28 1,891.89 345,753.30 354

Age 38.13 1.80 35 42 354

Daughters
Income 21,552.95 17,488.95 1,798.41 129,572.60 261

Age 37.90 1.83 35 42 261

PSID Mean Std. Dev. Min. Max. N

Fathers
Income 66,418.02 68,892.81 6,026.50 981,877.40 1,224

Age 43.80 5.60 32 53 1224

Sons
Income 68,599.45 71,238.90 1,234.57 915,955.40 601

Age 37.90 1.88 35 42 601

Daughters
Income 39,936.94 39,241.34 1,212.12 532,325.60 623

Age 37.89 1.89 35 42 623

Source: SOEP (1984-2013), PSID (1984-2013).

Figure (4.1) shows the intergenerational income correlation between sons and fa-

thers and daughters and fathers, respectively. The univariate ordinary least squares

(OLS) estimation implies a higher intergenerational income elasticity for the daugh-

ters in Germany, while in the United States, the OLS estimate of the sons seems

to be stronger. The Nadaraya-Watson (NW) estimation shows slight deviations

at the lower and upper ends of the income distribution. However, they are most

likely caused by single outliers in the top and bottom income quantiles. Overall, the

income data points are heavily scattered around the respective regression lines, indi-

cating that parental earnings are not the only determinant for sons' and daughters'

incomes.

101
 Intergenerational Income Mobility among Daughters

Figure 4.1: Intergenerational income correlation by gender

Germany

13 13

12 12

Log. income of daughter

Log. income of son

11 11

10 10

9 9

8 8

7 7
9 10 11 12 9 10 11 12
Log. income of father Log. income of father

Income data OLS Income data OLS

NW 95% CI NW 95% CI

United States

14 14
Log. income of daughter

12 12
Log. income of son

10 10

8 8

6 6
9 10 11 12 13 14 9 10 11 12 13 14
Log. income of father Log. income of father

Income data OLS Income data OLS

NW 95% CI NW 95% CI

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: The Nadaraya-Watson estimation uses the Epanechnikov kernel with a range based on
Silverman's rule of thumb. OLS: Ordinary least squares, NW: Nadaraya-Watson, CI: Condence
interval.

102
 Intergenerational Income Mobility among Daughters

4.4 Empirical results

4.4.1 Baseline estimation

As a starting point, the intergenerational income elasticity of men and women is

estimated using a lower income bound of 1,200 Euro/US Dollar per year in order to

compare the results to the available literature (Table 4.2). In Germany, the intergen-

erational income elasticity appears to be lower for the sons, with a value of 0.3428,

than for the daughters, with a value of 0.4980. These results can be interpreted in

such a way that 34 percent (50 percent) of the income advantage or disadvantage of

a father is transmitted to his son (daughter). Thus, if a father's income is twice as

high as the average income in the parental generation, the expected income of his

son (daughter) will exceed the average income of the lial generation by 34 percent

(50 percent). The German sons are therefore more mobile than the German daugh-

ters. In the United States, in contrast, the sons exhibit a higher intergenerational

income elasticity, with an estimate of 0.4624, than the daughters, with an estimate

of 0.2607, and are therefore less mobile.

To further explore the eect of divergent labor market participation, the elastici-

ties are re-estimated for married and unmarried sons and daughters, respectively. In

both countries, unmarried women show a higher elasticity than unmarried men. In

Germany, the value for unmarried daughters is 0.4791, while the value for unmarried

sons does not signicantly dier from zero. In the United States, unmarried daugh-

ters exhibit a value of 0.4699, whereas the estimate for unmarried sons is 0.3213.

In contrast, married sons exhibit a higher elasticity than married daughters in both

countries. While for married sons, a value of 0.4804 is estimated, the estimate for

married daughters is lower, with a value of 0.3918, in Germany. In the United States,

married sons exhibit a value of 0.4350, while the value for married daughters does

not signicantly deviate from zero. Thus, the higher overall elasticity for daughters

in Germany is driven by unmarried children, while the higher value for sons in the

103
 Intergenerational Income Mobility among Daughters

United States is driven by married children. However, the dierences between sons

and daughters are only statistically signicant in the case of the United States for all

sons and daughters (p = 0.0341) and for married sons and daughters (p = 0.0064).38

Table 4.2: Intergenerational income elasticity by gender and marital status

Germany United States

Sons Daughters Sons Daughters

All
IIE 0.3428*** 0.4980*** 0.4624*** 0.2607***

(0.0785) (0.1484) (0.0702) (0.0662)

2
R 0.0847 0.1419 0.1412 0.0783

N 354 261 601 623

Not married
IIE 0.2024 0.4791** 0.3213*** 0.4699***

(0.1513) (0.1858) (0.1179) (0.0877)

2
R 0.1198 0.3401 0.1227 0.1624

N 134 89 203 252

Married
IIE 0.4804*** 0.3918* 0.4350*** 0.1194

(0.0929) (0.2025) (0.0780) (0.0855)

2
R 0.1347 0.1324 0.1380 0.0576

N 220 172 398 371

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Covariates include polynomials of fathers' and children's age as well as their birth year
and the number of valid observations of the children. Standard errors are clustered at the family
level and were calculated using paired bootstrap resampling with 1,000 replications. *** p <
0.01, ** p < 0.05, * p < 0.1. IIE: Intergenerational income elasticity.

The model described in Section 4.2 suggests that the intergenerational income

elasticity of unmarried children is positively inuenced by the impact of fathers' in-

come on their own hourly wages λ and by the elasticity of hours worked with respect
to their own wages η. To explore these determinants, λ and η are estimated for the

subsamples of unmarried sons and daughters, respectively (Table 4.3). In Germany,

38 Utilizing mothers' individual incomes, the obtained estimates are mostly insignicant.

104
 Intergenerational Income Mobility among Daughters

the elasticity of hourly wages with respect to fathers' income is 0.3519 for unmarried

daughters, while the estimate for unmarried sons is not statistically signicant. The

elasticity of hours worked with respect to their own hourly wages is about equally

high for unmarried sons (0.2049) and daughters (0.2031). In the United States, the

eect of fathers' income on children's hourly wages is again lower for unmarried

sons (0.3238) than for unmarried daughters (0.4642). The elasticity of hours worked

with respect to a person's own hourly wages is, however, not statistically signicant

for either gender. Thus, the higher intergenerational income elasticity of unmarried

daughters is driven by a higher estimated impact of fathers' incomes on daughters'

hourly wages in both countries.

Table 4.3: Determinants of intergenerational income mobility

Germany United States

Sons Daughters Sons Daughters

Not married
λ 0.0510 0.3519** 0.3238*** 0.4642***

(0.0926) (0.1472) (0.0845) (0.0851)

η 0.2049*** 0.2031** -0.0284 0.0689

(0.0661) (0.0820) (0.1584) (0.0800)

Married
λ 0.3524*** 0.2731** 0.4149*** 0.2860***

(0.0876) (0.1266) (0.0661) (0.0626)

η 0.1262* 0.1735* -0.0433 0.0436

(0.0737) (0.0915) (0.0707) (0.0575)

ηs 0.0097 -0.1710*** -0.0113 -0.0958***

(0.0196) (0.0638) (0.0207) (0.0297)

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Covariates include polynomials of fathers' and children's age as well as their birth year
and the number of valid observations of the children. In the estimation of η and ηs , the number
of children age 0-14 in the household is additionally included. Standard errors are clustered at
the family level and were calculated using paired bootstrap resampling with 1,000 replications.
*** p < 0.01, ** p < 0.05, * p < 0.1.

105
 Intergenerational Income Mobility among Daughters

For married children, in addition to λ and η, the intergenerational income mo-

bility is presumably negatively inuenced by the cross-elasticity of working hours

with respect to the spouse's income η s .39 For married daughters in Germany, the

eect of fathers' income on hourly wages is smaller (0.2731) than for married sons

(0.3524). While the elasticity of hours worked with respect to one's own wages is

slightly higher for women (0.1735) than for men (0.1262), there exist stronger dier-

ences in the elasticity of hours worked with respect to the partner's income. While

sons do not react signicantly to their wives' higher wages, daughters tend to reduce

their labor supply with their husbands' increasing income (-0.1710). In Germany,

the higher intergenerational income mobility of married daughters as compared to

married sons is thus driven by a lower impact of fathers' income on hourly wages and

a stronger reduction of hours worked with respect to their partner's income. These

eects are counteracted though not oset by a higher elasticity of daughters' hours

worked with respect to their own wages. In the United States, the impact of father's

income on the hourly wage is higher for sons (0.4149) than for daughters (0.2860).

The elasticity of hours worked with respect to their own hourly wages is insigni-

cant for both genders. The elasticity of hours worked with respect to their partner's

income is not statistically signicant for the sons, however, daughters again reduce

their labor supply with their husbands' rising income (-0.0958). Interestingly, the

absolute value of this estimate is markedly lower than in Germany, implying that

German women react more strongly to the income of their husbands than American

women. Thus, the higher income elasticity of married sons as compared to mar-

ried daughters in the United States is driven by a higher impact of fathers' income

on hourly wages and a weaker reduction of hours worked with respect to partners'

income.

39 Spouses' incomes are approximated by the dierence between children's household incomes
and their individual incomes. In principle, partners in the SOEP and the PSID could also be
matched via their personal identication number. However, this procedure further reduces the size
of the already small sample. In addition, the eect of ηs is moderated by the level of assortative
mating π (see Section 4.2). However, as hourly wages of the spouse cannot be observed, the
comparison implicitly assumes that the strength of assortative mating is approximately equal in
Germany and the United States.

106
 Intergenerational Income Mobility among Daughters

4.4.2 Eects of assortative mating

The common restriction of the sample to incomes higher than 1,200 Euro/US dollar

seems reasonable as an income of less than 100 Euro/US dollar per month is not

sucient for the subsistence of an individual person in either country. However, mar-

ried individuals might indeed have an individual income of less than 1,200 Euro/US

dollar if the labor division within the family is designed in such a way that primarily

one of the spouses is active in the labor market. Therefore, the intergenerational in-

come elasticities for married sons and daughters are re-estimated without the lower

income limit of 1,200 Euro/US dollar (Table 4.4). The estimated elasticities are

somewhat lower than those presented in Table 4.2, with a value of 0.2006 (0.1531)

for married sons (daughters) in Germany and a value of 0.2860 (0.1451) for married

sons (daughters) in the United States. However, married sons still exhibit a higher

40
intergenerational income elasticity than married daughters in both countries.

To further analyze the eects of assortative mating, the elasticity of children's

household incomes with respect to fathers' incomes are additionally reported in

Table 4.4. In Germany, the estimates are about equally high for sons and daughters,

with values of 0.2269 and 0.2360, respectively. In the United States, the estimates are

also very similar for men and women, with values of 0.2744 for the sons and 0.2671

for the daughters, though they appear to be somewhat higher when compared to

Germany. Thus, although married women's individual incomes depend less strongly

on their family background than those of married men, the genders do not dier

much with regard to household incomes. If household income is interpreted as the

actual economic status of a child, sons and daughters in the respective countries are

therefore about equally mobile.

40 If an individual features an income of zero, the logarithmized income cannot be calculated

and is thus dropped from the regression analysis. To control for a sample selection bias due to this
issue, a parallel Heckman estimation with the number of children younger than 14 years of age as
the selection variable is performed (see Table 4.5 in the Appendix).

107
 Intergenerational Income Mobility among Daughters

Table 4.4: Intergenerational income elasticities by type of income

Germany United States

Sons Daughters Sons Daughters

Own individual income

IIE 0.2006*** 0.1531*** 0.2860*** 0.1451

(0.0771) (0.1187) (0.0693) (0.1056)

2
R 0.0763 0.0459 0.0973 0.0523

N 230 188 408 398

Household income
IIE 0.2269*** 0.2360*** 0.2744*** 0.2671***

(0.0684) (0.0527) (0.0485) (0.0563)

2
R 0.1101 0.1351 0.1172 0.1073

N 235 226 417 441

Spouse's individual income

IIE 0.4029*** 0.3755** 0.0474 0.2561***

(0.1531) (0.1808) (0.0913) (0.0613)

2
R 0.0742 0.1430 0.0244 0.0772

N 193 222 366 427

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Covariates include polynomials of fathers' and children's age as well as their birth year
and the number of valid observations of the children. Standard errors are clustered at the family
level and were calculated using paired bootstrap resampling with 1,000 replications. *** p <
0.01, ** p < 0.05, * p < 0.1. IIE: Intergenerational income elasticity.

Because the elasticity of household incomes with respect to parental earnings is

a weighted average of the children's and their spouses' elasticities (Section 4.2), the

observed dierences between the elasticities of individual incomes and household

incomes are driven by the correlation between fathers-in-law and their children-in-

law. This suggestion is supported by the estimated elasticities from a regression of

the logarithmized incomes of children's spouses on the logarithmized incomes of the

fathers. The elasticity of the son-in-law is estimated to be 0.3755 in Germany and is

thus even higher than the corresponding value of the sons. Surprisingly, the income

of the daughter-in-law is also strongly correlated with the income of her husband's

father with an estimated elasticity of 0.4029. In the United States, there is no

108
 Intergenerational Income Mobility among Daughters

signicant impact of a father's income on his daughter-in-law's income. However,

the elasticity of sons-in-law with respect to their fathers-in-law is again relatively

strong with an estimated value of 0.2561. This estimate is again equal in size to the

corresponding value for the sons. Overall, the results imply a considerable extent of

41
assorative mating in both countries.

4.5 Conclusion

This chapter analyzes intergenerational income mobility among daughters in Ger-

many and the United States. The baseline estimation shows a higher intergenera-

tional income elasticity in Germany and a lower intergenerational income elasticity

in the United States for women as compared to men. However, a separation by

marital status reveals that in both countries, unmarried women exhibit a higher in-

tergenerational income elasticity than unmarried men, while married women feature

a lower intergenerational income elasticity than married men. The reason for the

lower mobility of unmarried women appears to be a stronger human capital trans-

mission from fathers to daughters than to sons. The higher mobility of married

women is driven by a weaker human capital transmission and a higher labor supply

elasticity with respect to spousal income. The estimated intergenerational income

elasticity of married children's household incomes is even higher than that of their

individual incomes. This can be seen as an indication for strong assortative mating.

If household income is interpreted as a measure of children's actual economic welfare,

there are barely any dierences between sons and daughters. The intergenerational

income elasticity of spousal income with respect to parental income is again rela-

tively high, which in turn supports the hypothesis of strong assortative mating. The

elasticity of the sons-in-law with respect to their fathers-in-law in Germany is even

higher than that of the sons with respect to their own fathers.

41 In the cases of the daughters-in-law in Germany and the sons-in-law in the United States, the
Heckman estimation reports a signicant sample selection bias. However, the results are relatively
similar with values of 0.4574 and 0.2439, respectively (Table 4.5).

109
 Intergenerational Income Mobility among Daughters

References

Altonji, J. G. and Dunn, T. A. (1991). Relationships among the family incomes

and labor market outcomes of relatives. In Ehrenberg, R. G., editor, Research in

Labor Economics, 12, 269310.

Atkinson, A. B., Maynard, A. K., and Trinder, C. G. (1983). Parents and Children:

Incomes in Two Generations. Heinmann Educational Books, London, United

Kingdom.

Becker, G. S. (1973). A theory of marriage: Part I. Journal of Political Economy,

81 (4), 813846.

Becker, G. S. (1974). A theory of marriage: Part II. Journal of Political Economy,

82 (2), S11S26.

Becker, G. S. (1991). A treatise on the family. Harvard University Press, Cambridge,

United States.

Becker, G. S. and Tomes, N. (1979). An equilibrium theory of the distribution

of income and intergenerational mobility. Journal of Political Economy, 87 (6),

11531189.

Becker, G. S. and Tomes, N. (1986). Human capital and the rise and fall of families.

Journal of Labor Economics, 4 (3), S1S39.

Björklund, A. and Jäntti, M. (2009). Intergenerational income mobility and the role

of family background. In Salverda, W., Nolan, B., and Smeeding, T. M., editors,

Oxford Handbook of Economic Inequality, 491521.

Black, S. E. and Devereux, P. J. (2011). Recent developments in intergenerational

mobility. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics,

4B, 14871541.

110
 Intergenerational Income Mobility among Daughters

Blanden, J., Goodman, A., Gregg, P., and Machin, S. (2004). Changes in intergen-

erational mobility in Britain. In Corak, M., editor, Generational Income Mobility

in North America and Europe, 122146.

Blau, F. D. and Kahn, L. W. (2007). Changes in the labor supply behavior of

married women: 1980-2000. Journal of Labour Economics, 25 (3), 393438.

Blossfeld, H. and Timms, A. (2003). Educational systems as marriage markets in

modern societies: A conceptual framework. In Blossfeld, H. and Timms, A., edi-

tors, Who Marries Whom? Educational Systems as Marriage Markets in Modern

Societies, 118.

Blundell, R. and MaCurdy, T. (1999). Labor supply: A review of alternative ap-

proaches. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics,

3A, 15591695.

Böhlmark, A. and Lindquist, M. J. (2006). Life-cycle variations in the associa-

tion between current and lifetime income: Replication and extension for Sweden.

Journal of Labor Economics, 24 (4), 879896.

Bratberg, E., Nielsen, Ø. A., and Vaage, K. (2005). Intergenerational earnings

mobility in Norway: Levels and trends. Scandinavian Journal of Economics, 107

(3), 419435.

Bratberg, E., Nielsen, Ø. A., and Vaage, K. (2007). Trends in intergenerational

mobility across ospring's earnings distribution in Norway. Industrial Relations,

46 (1), 112129.

Brenner, J. (2010). Life-cycle variations in the association between current and

lifetime earnings: Evidence for German natives and guest workers. Labour Eco-

nomics, 17 (2), 392406.

Chadwick, L. and Solon, G. (2002). Intergenerational income mobility among daugh-

ters. American Economic Review, 92 (1), 335344.

111
 Intergenerational Income Mobility among Daughters

Couch, K. A. and Dunn, T. A. (1997). Intergenerational correlations in labor market

status: A comparison of the United States and Germany. Journal of Human

Resources, 32 (1), 210232.

Dearden, L., Machin, S., and Reed, H. (1997). Intergenerational mobility in Britain.

Economic Journal, 107 (1), 4766.

Devereux, P. J. (2004). Changes in relative wages and family labor supply. Journal

of Human Resources, 39 (3), 696722.

Eckstein, Z. and Wolpin, K. I. (1989). Dynamic labour force participation of married

women and endogenous work experience. Review of Economic Studies, 56 (3),

375390.

Epstein, E. and Guttman, R. (1984). Mate selection in man: Evidence, theory, and

outcome. Social Biology, 31 (34), 243278.

Fitzgerald, J., Gottschalk, P., and Mott, R. (1998). An analysis of sample attrition

in panel data: The Michigan Panel Study of Income Dynamics. Journal of Human

Resources, 33 (2), 251299.

Frick, J. R., Grabka, M. M., and Groh-Samberg, O. (2012). Dealing with incomplete

household panel data in inequality research. Sociological Methods & Research, 41

(1), 89123.

Frick, J. R., Jenkins, S. P., Lillard, D. R., Lipps, O., and Wooden, M. (2007). The

Cross-National Equivalent File (CNEF) and its member country household panel

studies. Journal of Contextual Economics, 127 (4), 627654.

Grabka, M. M. (2014). Codebook for the $PEQUIV File 1984-2013: CNEF vari-

ables with extended income information for the SOEP. Data Documentation 74,

German Institute for Economic Research, Berlin, Germany.

Grieger, L. D., Danziger, S., and Schoeni, R. F. (2009). Accurately measuring the

112
 Intergenerational Income Mobility among Daughters

trend in poverty in the United States using the Panel Study of Income Dynamics.

Journal of Economic and Social Measurement, 34 (23), 105117.

Haider, S. and Solon, G. (2006). Life-cycle variation in the association between

current and lifetime earnings. American Economic Review, 96 (4), 13081320.

Hess, G. D. (2004). Marriage and consumption insurance: What's love got to do

with it? Journal of Political Economy, 112 (2), 290318.

Hirvonen, L. H. (2008). Intergenerational earnings mobility among daughters and

sons: Evidence from Sweden and a comparison with the United States. American

Journal of Economics and Sociology, 67 (5), 777826.

Holmlund, H. (2006). Intergenerational mobility and assortative mating: Eects

of an educational reform. Working Paper 4/2006, Swedish Institute for Social

Research, Stockholm, Sweden.

Jäntti, M., Bratsberg, B., Røed, K., Raaum, O., Naylor, R., Österbacka, E., Björk-

lund, A., and Eriksson, T. (2006). American exceptionalism in a new light: A

comparison of intergenerational earnings mobility in the Nordic countries, the

United Kingdom and the United States. IZA Discussion Paper 1938, Institute of

Labor Economics, Bonn, Germany.

Juhn, C. and Murphy, K. M. (1997). Wage inequality and family labor supply.

Journal of Labor Economics, 15 (1), 7297.

Killingsworth, M. R. (1983). Labor Supply. Cambridge University Press, Cambridge,

United Kingdom.

Kremer, M. (1997). How much does sorting increase inequality? Quarterly Journal

of Economics, 112 (1), 115139.

Lam, D. (1988). Marriage markets and assortative mating with household public

goods: Theoretical results and empirical implications. Journal of Human Re-

sources, 23 (4), 462487.

113
 Intergenerational Income Mobility among Daughters

Lewis, S. K. and Oppenheimer, V. K. (2001). Educational assortative mating across

marriage markets: Non-Hispanic whites in the US. Demography, 37 (1), 2940.

Lillard, D. R. (2013). Codebook for the Cross-National Equivalent File 1970-2013:

BHPS  HILDA  KLIPS  PSID  RLMS-HSE  SHP  SLID  SOEP. Colum-

bus, United States.

Lundberg, S. (1988). Labor supply of husbans and wives: A simultaneous equations

approach. Review of Economics and Statistics, 70 (2), 224235.

Mazumder, B. (2005). Fortunate sons: New estimates of intergenerational mobility

in the United States using social security earnings data. Review of Economics and

Statistics, 87 (2), 235255.

Österbacka, E. (2001). Family background and economic status in Finland. Scan-

dinavian Journal of Economics, 103 (3), 467484.

Österberg, T. (2000). Intergenerational income mobility in Sweden: What do tax-

data show? Review of Income and Wealth, 46 (4), 421436.

Peters, H. E. (1992). Patterns of intergenerational mobility in income and earnings.

Review of Economics and Statistics, 74 (3), 456466.

Raaum, O., Bratsberg, B., Røed, K., Österbacka, E., Eriksson, T., Jäentti, M., and

Naylor, R. A. (2007). Marital sorting, household labor supply, and intergenera-

tional earnings mobility across countries. B. E. Journal of Economic Analysis &

Policy, 7 (2), Article 7.

Schnitzlein, D. D. (2009). Structure and extent of intergenerational income mobility

in Germany. Journal of Economics and Statistics, 229 (4), 450466.

Schnitzlein, D. D. (2016). A new look at intergenerational income mobility in Ger-

many compared to the U.S. Review of Income and Wealth, 62 (4), 650667.

114
 Intergenerational Income Mobility among Daughters

Skyt Nielsen, H. and Svarer, M. (2006). Educational homogamy: Preferences or

opportunities? IZA Discussion Paper 2271, Institute of Labor Research, Bonn,

Germany.

Solon, G. (1989). Biases in the estimation of intergenerational earnings correlations.

Review of Economics and Statistics, 71 (1), 172174.

Solon, G. (1992). Intergenerational income mobility in the United States. American

Economic Review, 82 (3), 393408.

Solon, G. (1999). Intergenerational mobility in the labor market. In Ashenfelter, O.

and Card, D., editors, Handbook of Labor Economics, 3A, 17611800.

Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data.

MIT Press, Cambridge, United States.

Zimmerman, D. J. (1992). Regression toward mediocrity in economic stature. Amer-

ican Economic Review, 82 (3), 409492.

115
 Intergenerational Income Mobility among Daughters

Appendix

Table 4.5: Heckman estimation of intergenerational income elasticity

Germany United States

Sons Daughters Sons Daughters

Own individual income

IIE 0.1921** 0.2101* 0.2628*** 0.1458

(0.0777) (0.1261) (0.0860) (0.1127)

N 236 228 421 443

p(λ) 1.000 0.102 0.449 0.970

Household income
IIE 0.2264*** 0.2372*** 0.27722*** 0.2672***

(0.0705) (0.0522) (0.0514) (0.0566)

N 236 228 421 443

p(λ) 0.639 1.000 1.000 1.000

Spouse's individual income

IIE 0.4574** 0.3631* -0.1203 0.2439***

(0.1872) (0.1886) (0.2534) (0.0672)

N 236 228 421 443

p(λ) 0.006 1.000 0.391 0.069

Source: SOEP (1984-2013), PSID (1984-2013).

Notes: Covariates include polynomials of fathers' and children's age as well as their birth year
and the number of valid observations of the children. The number of children age 0-14 in the
household is used as the selection variable. Standard errors are clustered at the family level
and were calculated using paired bootstrap resampling with 1,000 replications. *** p < 0.01,
** p < 0.05, * p < 0.1. IIE: Intergenerational income elasticity.

116
Chapter 5

Conclusive remarks
The present dissertation deals with intergenerational income mobility in Germany

and the United States. The transmission process of income dierences from one

generation to the next is an important aspect of income inequality as it captures the

dynamic component of the income distribution. A high level of intergenerational

income elasticity implies that income dierences are mainly caused by divergent

talents, abilities, and preferences. In this case, income inequality is less problematic

and might even encourage investments in human capital and eorts to increase

earnings. On the contrary, if children's future level of income is predetermined

for the most part by their family background, future prospects of poor children

are literally eliminated. This means that a high level of intergenerational income

mobility and, closely related to this, equality of opportunity should be a primary

goal for economic policy in light of the rising level of interpersonal income inequality.

The analyses in the previous chapters use comparable data from Germany and

the United States in order to contrast the results for the two countries. Thus, they

contribute to the empirical literature on intergenerational income mobility in both

countries as well as to the literature on country comparisons. The rst part is moti-

vated by an unclear position of Germany in the international ranking of intergenera-

tional mobility levels. We therefore conduct a direct comparison of the structure and

extent of intergenerational income mobility in Germany as compared to the United

States. The results support the widely accepted view that the intergenerational

income elasticity is higher in the United States than in Germany. However, while

the results for the intergenerational rank mobility do not dier much between the

two countries, Germany exhibits a higher intergenerational income share persistence

117
 Conclusive remarks

than the United States. We nd no indications for a nonlinear run of the intergen-

erational income elasticity which might point to credit market constraints. A nal

decomposition of intergenerational income inequality shows both higher income mo-

bility and stronger progressive income growth for Germany compared to the United

States. Overall, we cannot identify a clear ranking between the two countries.

The second contribution examines the transmission channels of intergenerational

income persistence. To deduce concrete economic policy actions, it is important

to know why intergenerational income persistence is present in a certain country.

Firstly, income dierences could be inherited due to the actual higher income of par-

ents allowing them to directly invest in the human capital of their children. Secondly,

human capital might also be transferred from parents to children without nancial

expenditures. We perform a descriptive as well as a structural decomposition and

nd that the direct eect of a father's nancial means is much more important in the

United States, whereas the indirect eect of a father's nonmonetary human capital

transmission is predominant in Germany. These results are in line with the fact that

the share of private expenditures in the American education system is signicantly

higher than in Germany. In conclusion, equality of opportunity in Germany cannot

be reached by the supply of nancial means for poor children alone. Rather, so-

cial policy must substitute for the missing direct human capital transmission within

low-income families.

The third analysis deals with the dierences in intergenerational income mobility

between sons and daughters. Individual incomes are likely to be an unreliable mea-

sure of daughters' actual economic status in most cases because married women in

particular tend to reduce their working hours as a result of joint decisions on house-

hold labor division. The existence of assortative matingthe tendency of women

from well-o families to marry rich menaggravates this problem. The baseline

analysis shows that women in Germany exhibit higher elasticities and thus lower

mobility levels than men, while in the United States this relation is reversed. A de-

tailed analysis by marital status suggests that in both countries unmarried women

118
 Conclusive remarks

exhibit higher elasticities than unmarried men, while married women show lower

elasticities. This is obviously due to a stronger human capital transmission from fa-

thers to their daughters as compared to their sons in the case of unmarried children,

and due to a lower human capital transmission and a higher cross-elasticity with

respect to husbands' wages for daughters in the case of married children. Overall,

there appears to be a signicant extent of assortative mating as spouses' incomes

as well as overall household incomes are likewise highly correlated with parental

earnings.

Nevertheless, the statistical analyses in this dissertation suer from at least two

problems concerning the available data sets. On the one hand, the number of ob-

servation in the parent-child samples is relatively small such that the estimated

standard errors are often quite large. This becomes especially obvious in the struc-

tural decomposition analysis in Chapter 3.4.2, where hardly any signicant estimates

can be obtained. A further restriction to more specic subsamples is thus often not

possible. On the other hand, data for Germany is only available for one generation of

parents and their children. A dynamic analysis of intergenerational income mobility,

as performed in several studies available for the United States and the Scandinavian

countries, is thus not feasible and the implications derived in this dissertation might

not be transferable to past and future generations. Here, an improvement of data

availability might enable future researchers to obtained more detailed results.

119
Lebenslauf

Persönliche Daten

geboren 24. Oktober 1990 in Ansbach

Ausbildung

12/15 [Link]. in Economics an der Universität Würzburg

08/13 [Link]. in Wirtschaftswissenschaft an der Universität Würzburg

Praktische Tätigkeit

07/14-02/18 Wissenschaftliche Mitarbeiterin am Lehrstuhl für VWL, insb.

Wirtschaftsordnung und Sozialpolitik an der Universität Würzburg

Ausgewählte Veröentlichungen

2018 Potenzial des griechischen Exportsektors: Neues Wirtschaftswachstum

zur Eindämmung der Schuldenkrise?, WiSt  Wirtschaftswissenschaft-

liches Studium 47 (1), S. 30-37 (mit Julia Rumler)

2017 Do Expectations Matter? Reassessing the Eect of Government

Spending on Key Macroeconomic Variables in Germany, Applied

Economics Letters, im Erscheinen (mit Klaus Gründler)

2016 Intergenerative Einkommensmobilität in Deutschland und den USA,

ORDO Jahrbuch für die Ordnung von Wirtschaft und Gesellschaft 67,

S. 101-131 (mit Mustafa Çoban)