Innovation und Wissenstransfer in der empirischen Sozial- und Verhaltensforschung E-Book

• Herausgeber: Campus Verlag
• Kategorie: Geisteswissenschaft
• Sprache: Deutsch
• Veröffentlichungsjahr: 2018

Seitenzahl: 492

Beschreibung

Gert G. Wagner ist als Sozial- und Wirtschaftsforscher über disziplinäre Grenzen gegangen. Als engagierter Politikberater und innovativer Wissenschaftsmanager hat er immer wieder sozial- und wissenschaftspolitische Debatten angestoßen und begleitet. Kollegen und Weggefährten liefern mit ihren Beiträgen zu diesem Band eine Bestandsaufnahme der empirisch- quantitativen Forschung, die Gert G. Wagner seit Anfang der 1980er-Jahre wesentlich geprägt hat.

Leseprobe

Marcel Erlinghagen, Karsten Hank, Michaela Kreyenfeld (Hg.)

Innovation und Wissenstransfer in der empirischen Sozial- und Verhaltensforschung

Festschrift für Gert G. Wagner

Campus Verlag

Frankfurt/New York

Über das Buch

Die EU befindet sich in einer Krise, die durch Umbrüche in den internationalen Beziehungen verschärft wird. Wie die Konflikte in der Ukraine und in Syrien zeigen, läuft sie Gefahr, zum Spielball geopolitischer Auseinandersetzungen zwischen den USA und Russland zu werden. Die EU muss zu einer ausgewogenen und selbstbewussten Politik finden, will sie ein wichtiger Akteur sein. Dabei bietet eine Verbindung der Konzeptionen von Kerneuropa und Greater Eurasia Chancen für den Abbau politischer Spannungen sowie zu einer Rückbesinnung auf eine interessengeleitete Russlandpolitik der EU und ihrer Mitgliedsstaaten. Diesem Ziel würde ein gesamteuropäischer Wirtschaftsraum dienen, der auf den Säulen von Energie- und Transportinfrastruktur beruhen könnte.

Vita

Marcel Erlinghagen ist Professor für Soziologie an der Universität Duisburg-Essen und Research Fellow am Deutschen Institut für Wirtschaftsforschung (DIW) Berlin.

Karsten Hank ist Professor für Soziologie an der Universität zu Köln und Research Fellow am Deutschen Institut für Wirtschaftsforschung (DIW) Berlin.

Michaela Kreyenfeld ist Professorin für Soziologie an der Hertie School of Governance in Berlin.

Inhalt

Marcel Erlinghagen, Karsten Hank & Michaela Kreyenfeld: Disziplinäre Grenzüberschreitungen – Gert G. Wagner zum 65. Geburtstag

Stephen P. Jenkins & Timothy M. Smeeding: In Praise of Panel Surveys, a Sonder-Panel, and a Sonder-Panel-Papa

Introduction

Why Praise Household Panel Surveys?

The Value of Longitudinal Data

Household Panel Surveys: Key Features, and Examples from Around the World

Household Panel Surveys Compared to Retrospective Designs

Household Panel Surveys Compared to Cohort Surveys

Household Panel Surveys Compared to Linked Administrative Data

Household Panel Surveys: Conclusions

Why is the SOEP Special?

Sample Design

Survey Content

Documentation and Data Access

Resources and Infrastructure

Conclusions: Gert Wagner is a Sonder-Panel-Papa

Acknowledgements

References

Bruce Headey & Ruud Muffels: Mrs. Aristotele’s Teeth: How SOEP Transformed Life Satisfaction Research

Introduction

Methods

Balanced Panel of SOEP Respondents With Complete LS Records for 1990–2014

Dependent Variables: Single and Multi-year Measures of LS

Explanatory Variables: Accounting for Change and Volatility in LS

Socio-economic Characteristics and Personality Traits

Personal Values/Life Priorities

Behavioural Choices and Health

Domain Satisfactions

Panel Effects

Imputation for Years in Which Variables Were Not Measured

Results

Accounting for Changes in LS: Longitudinal Fixed Effects Regressions

Individual Differences in LS Volatility

Discussion

References

Sandra Gerstorf, Nilam Ram & Denis Gerstorf: Glücklich und zufrieden leben bis ans Ende? Forschungsergebnisse zum terminal decline

Einleitung

Theoretischer Hintergrund: Grundprinzipien längsschnittlicher Forschung nach Baltes und Nesselroade (1979)

Grundprinzip I: Das Phänomen terminal decline

Performanz-basierte Indikatoren

Selbstbericht

Das Zwei-Phasen Modell

Offene Forschungsfragen

Grundprinzip II: Individuelle Unterschiede im terminal decline

Grundprinzip III: Wechselbeziehungen zwischen Funktionsbereichen

Grundprinzip IV: Bedingungen und Determinanten von terminal decline

Grundprinzip V: Bedingungen und Determinanten von individuellen Unterschieden im terminal decline

Bewertung und Ausblick

Dank

Literatur

Michaela Riediger: Ambulatory Assessment in Survey Research: The Multi-Method Ambulatory Assessment Project

Introduction

What is Ambulatory Assessment?

Overview of the Multi-Method Ambulatory Assessment (MMAA) Project

Selected Findings from the MMAA Project

Seeking Pleasure and Seeking Pain: Age Differences in Everyday Affect-Regulation Motivation

Feeling Good When Sleeping in? Age Differences in the Association between Sleep and Affective Well-Being

Keeping One’s Cool: Affective and Physiological Reactivity to Daily Hassles

Daily Coupling of Testosterone and Cortisol from Youth to Old Age

When Bad Moods May Not be so Bad: Valuing Negative Affect is Associated with Weakened Affect-Health Links

Personality-Situation Transactions from Adolescence to Old Age

Conclusion

Acknowledgement

References

Peter Krause: Wiedervereinigung und die Annäherung der Lebensbedingungen in Ost- und Westdeutschland

Das SOEP als Datenquelle zur Beschreibung der Lebensbedingungen in Ost- und West-Deutschland

Wiedervereinigung als (wissenschaftliche und sozialpolitische) Felderfahrung

Regionale Identität Ost-West

Preisindizes und Kaufkraftparitäten

Anpassung der Bewertungsmaßstäbe zur Evaluierung von Lebensbedingungen

Referenzrahmen zur Ermittlung von Armutsindizes

Entwicklung der Lebensbedingungen in Ost und West

Gesamtwirtschaftliche Entwicklung und Finanzkrise

Demographie und Haushaltsstrukturen

Arbeitsmarkt und Beschäftigung

Einkommen und Verteilung

Einkommensarmut und sozio-demographische Differenzierungen

Subjektive Bewertung der Angleichung

Ausblick

Literatur

Jan Goebel & Markus M. Grabka: Sensitivität von Armuts- und Ungleichheitsmessungen bei gerundeten Einkommensangaben

Einleitung

Definition und Ursachen

Ausmaß von Rundungseffekten

Auswirkungen von Rundungseffekten auf Analysen zur Einkommensungleichheit

Fazit

Literatur

Richard Hauser, Richard V. Burkhauser, Kenneth A. Couch & Gulgun Bayaz-Ozturk: Wife or Frau, Women Still Do Worse: A Comparison of Men and Women in the United States and Germany after Union Dissolutions in the 1990s and 2000s

Introduction

Data and Methods

Results

Comparing Short- and Long-Term Changes in Economic Wellbeing Following a Union Dissolution

Comparing Short- and Long-Term Changes in Sources of Income Following a Union Dissolution

Discussion

Concluding Remarks

Acknowledgements

References

Appendix: Construction of Study Variables

Hans-Jürgen Krupp: Zur wissenschaftlichen Basis von Politikberatung

Probleme unabhängiger Politikberatung

Der Wegfall der wissenschaftlichen Basis von Politikberatung am Beispiel Sozialpolitik

Fehlende Pluralität in den Wirtschaftswissenschaften als begrenzender Faktor von Politikberatung

Zur Rolle der Autonomie der Universitäten

Alternativen zur universitären Forschung als Basis der Politikberatung

Zur Zukunft der Politikberatung

Literatur

Stephan Leibfried: Sozialpolitik: Verschwindet ein zentrales wissenschaftliches Querschnittsthema aus den deutschen Universitäten?

Literatur

Reinhold Thiede & Tatjana Mika: Forschungsdatenzentren

Das Forschungsdatenzentrum der Rentenversicherung als Beispiel

Kommission zur Verbesserung der informationellen Infrastruktur

Die Entwicklungshistorie der Forschungsdatenzentren

Forschungsdatenzentren öffentlicher Datenhalter

Wissenschaftsgetragene Forschungsdatenzentren

Finanzierung der Forschungsdatenzentren

Veränderung der Forschungskultur

Ausblick

Literatur

Axel Börsch-Supan: Kann die Rente generationengerecht sein?

Konzeptionelles

Rentenalter und Lebenserwartung

Altersquotient und Nachhaltigkeitsfaktor

Die Rolle der Kapitaldeckung bei der Generationengerechtigkeit

Haltelinien und Generationengerechtigkeit

Fazit: Langfristige Regeln wider die Zeitinkonsistenz

Reimund Schwarze: Die Versicherung von Naturgefahren: Das »Schweizer Modell« ein Vorbild für Deutschland und Europa?

Einleitung

Institutionelle ökonomische Grundprobleme der Versicherung von Naturgefahren

Idealtypische Modelle des Risikotransfers

Modell 1: Pflichtversicherung

Modell 2: Versicherungspflicht

Modell 3: Obligatorischer Deckungseinschluss (»Obligatorium«)

Modell 4: »Freier Markt« mit staatlicher Ad hoc-Hilfe

Modell 5: Katastrophenfonds

Das duale Versicherungssystem der Schweiz

Kantonale Gebäudeversicherungen (KGV) in 19 Kantonen

Privatversicherung in sieben Kantonen (GUSTAVO)

Institutionenökonomischer Vergleich der Risikotransfersysteme in Europa

Fazit

Literatur

C.Katharina Spieß: Die Ökonomie frühkindlicher Bildung und Betreuung: Ist sie in Deutschland angekommen?

Einleitung

Die Nachfrageseite

Betreuungsaspekte: Die Vereinbarkeit von Familie und Beruf

Bildungsaspekte: Die kindliche Entwicklung

Weitere Effekte auf der Nachfrageseite

(a) Heterogenitäten in der Nutzung von Kindertageseinrichtungen

(b) Effekte auf das mütterliche Wohlbefinden

(c) Effekte auf familiale Strukturen

Die Angebotsseite

Exkurs: Steuerungs- und Finanzierungsfragen

Schlussbemerkungen

Literatur

Nicolas R. Ziebarth: Die Symbiose von gesundheitsförderndem und gesundheitsgefährdendem Verhalten: Sport und Doping

Einleitung

Zur Geschichte des Doping und Anti-Doping-Kampfes

Zum Staatsdoping in der DDR

Zu den Erkenntnissen des Dopingmotivs und dessen empirischen Determinanten

Zur Dopingproblematik aus theoretischer ökonomischer Sicht

Schlusswort

Literatur

Jürgen Gerhards & Michael Mutz: Die empirische Vermessung der schönsten Nebensache der Welt: Fußball und sozialwissenschaftliche Forschung

Einleitung

Die Gnade der frühen Geburt

Die Entstehung der Langeweile: Ökonomisierung des Fußballs und die Berechenbarkeit von Erfolg

»Finale Dahoam« oder wie der Heimvorteil manchmal zum Nachteil gereichen kann

Ausblick

Literatur

Autorinnen und Autoren

Disziplinäre Grenzüberschreitungen – Gert G. Wagner zum 65. Geburtstag

Marcel Erlinghagen, Karsten Hank & Michaela Kreyenfeld

Zur Erklärung gesellschaftlicher Phänomene und individueller Verhaltensweisen werden seit einiger Zeit (wieder) zunehmend multi- und interdisziplinäre Perspektiven herangezogen. Auch wenn die Wissenschaftslandschaft nach wie vor stark durch disziplinäre Grenzziehungen geprägt ist, kann die komplexe Wirklichkeit globalisierter Gesellschaften und der in diesen agierenden Individuen nur durch disziplinäre Grenzüberschreitungen theoretisch und empirisch angemessen erfasst und analysiert werden. Ein beharrlicher Grenzgänger dieser Art, der in der Tradition der klassischen Nationalökonomie, die die Wirtschaftswissenschaft als Gesellschaftswissenschaft versteht und dabei neben den wirtschaftlichen und sozialen Rahmenbedingungen individuellen Handelns auch die Psychologie der Akteure mit in den Blick nimmt, ist Gert G. Wagner, der am 5. Januar 2018 seinen 65. Geburtstag feiert.

Dieses Datum nehmen wir zum Anlass, durch die im vorliegenden Band versammelten sozial- und verhaltenswissenschaftlichen Beiträge, erstens, eine aktuelle Momentaufnahme jener von Gert Wagner seit Anfang der 1980er Jahre wesentlich mitgeprägten empirisch-quantitativen Forschung in Deutschland zu geben, die sich heute nicht mehr allein auf die Wirtschafts- und Sozialwissenschaften beschränkt, sondern auch wichtige Teile der Psychologie mit einbezieht. Damit soll, zweitens, einer der einflussreichsten und produktivsten Sozial- und Wirtschaftsforscher Deutschlands gewürdigt werden, der zudem in vielfältigen Funktionen als innovativer Wissenschaftsmanager (insbesondere als langjähriger Leiter der Längsschnittstudie »Sozio-oekonomisches Panel«) sowie als engagierter Politikberater immer wieder zentrale sozial- und wissenschaftspolitische Debatten mit angestoßen und beharrlich begleitet hat.

Vor dem Hintergrund der interdisziplinären, kreativen und brückenbauenden Lebensleistung Gert Wagners sind »Innovation« und »Wissenstransfer« die Leitmotive des vorliegenden Bandes, die sich in den einzelnen Beiträgen in sehr unterschiedlichen Formen und Formaten – vom empirischen Fachaufsatz über Review-Artikel bis hin zum Essay – widerspiegeln. Die enorme Vielfalt des Wirkens von Gert Wagner kann in einem Band mit gerade einmal 15 Beiträgen natürlich nur im Ansatz reflektiert werden. Wir hoffen allerdings, dass es uns gelungen ist, eine zumindest halbwegs repräsentative Stichprobe gezogen zu haben:

Das Sozio-oekonomische Panel (SOEP) ist – natürlich – das zentrale, von ihm in den Jahren 1989 bis 2011 am DIW Berlin geleitete, Projekt Gert Wagners. Während Stephen P. Jenkins und Timothy M. Smeeding die Besonderheiten des SOEP im Allgemeinen würdigen, arbeiten Bruce Headey und Ruud Muffels in ihrem Beitrag heraus, wie Gert Wagner mit dem SOEP u. a. die Erforschung von Lebenszufriedenheit befruchtet und weiter vorangetrieben hat. Exemplarische Analysen hierzu finden sich bei Sandra Gerstorf, Nilam Ram und Denis Gerstorf, die über ihre – gemeinsam mit Gert Wagner durchgeführte – Forschung zu Wohlbefinden und Lebenszufriedenheit am Lebensende berichten. Gert Wagner hat jedoch nicht nur die klassischen Fragebogeninhalte einer ursprünglich fast ausschließlich auf die Untersuchung sozio-ökonomischer Fragestellungen angelegten Panelstudie sukzessive erweitert. Er hat auch surveymethodisch die Grenzen des früher im Rahmen einer repräsentativen Bevölkerungsumfrage machbar erscheinenden verschoben, wie zum Beispiel das von Michaela Riediger ausführlich beschriebene Multi-Method Ambulatory Assessment Project eindrucksvoll belegt.

Das SOEP hat die Welt (sozialwissenschaftlicher Längsschnittstudien) verändert – und ist selbst durch eine sich wandelnde Welt mitgeprägt worden. Eine besondere Rolle spielt hier natürlich die deutsche Wiedervereinigung. Wie deren Folgen für die Lebensbedingungen in Ost und West mit dem SOEP genau beschrieben und analysiert werden können, demonstriert Peter Krause in seiner Untersuchung. Dass das SOEP auch heute noch bestens dazu geeignet ist, klassische Themen der Ungleichheitsforschung – wie sie Gert Wagner immer beschäftigt haben – auf höchstem wissenschaftlichen Niveau zu untersuchen, zeigen insbesondere zwei der hier gesammelten Beiträge. Während Jan Goebel und Markus M. Grabka die Sensitivität von Armuts- und Ungleichheitsmessungen bei gerundeten Einkommensangaben betrachten, analysieren Richard Hauser, Richard V. Burkhauser, Kenneth A. Couch und Gulgun Bayaz-Ozturk auf Basis der Cross-National Equivalent Files des PSID und SOEP die wirtschaftlichen Folgen der Auflösung von Partnerschaften im deutsch-amerikanischen Vergleich.

Wissenschaftliche Forschung (ob mit oder ohne SOEP) war und ist für Gert Wagner nie – und bei aller Neugierde (zum Beispiel ob alle Wähler rechtsextremer Parteien einen Schäferhund besitzen) – akademischer Selbstzweck, sondern im Mittelpunkt steht für ihn immer deren gesellschaftliche Relevanz. Nicht zufällig unterstreicht daher Hans-Jürgen Krupp in seinem Beitrag die Bedeutung einer wissenschaftlichen Fundierung von Politikberatung. Diese hat eine wichtige Basis unter anderem in der universitären Forschung und Lehre zur Sozialpolitik, mit deren (möglichem) Verschwinden sich Stephan Leibfried kritisch auseinandersetzt (an dieser Stelle sei daran erinnert, dass die Denomination von Gert Wagners erstem Lehrstuhl – an der Ruhr-Universität Bochum – »Sozialpolitik und öffentliche Wirtschaft« lautete). Insbesondere ist eine wissenschaftlich fundierte Politikberatung jedoch auf qualitativ hochwertige Daten angewiesen, die nicht nur erhoben sondern einer breiten wissenschaftlichen Öffentlichkeit auch zugänglich gemacht werden müssen. Hierzu leisten, wie Reinhold Thiede und Tatjana Mika zeigen, Forschungsdatenzentren – für deren Einrichtung sich Gert Wagner als Vorsitzender des Rates für Sozial- und Wirtschaftsdaten stark gemacht hat – einen wichtigen Beitrag. »Gute« Politikberatung darf, last but not least, aber auch den normativen Diskurs nicht scheuen (wie Gert Wagner u. a. als Vorsitzender der Kammer für soziale Ordnung der Evangelischen Kirche in Deutschland gezeigt hat). In diesem Sinne fragt etwa Axel Börsch-Supan in seinem Essay, ob die Rente gerecht sein kann.

Die Liste der Themen, mit denen Gert Wagner sich bereits beschäftigt hat (und die neugierig auf seine Zukunftsthemen macht), lässt sich noch vielfältig fortsetzen, wie unter anderem die Beiträge von Reimund Schwarze zur Versicherung von Naturgefahren oder von C. Katharina Spieß zur Ökonomie frühkindlicher Bildung und Betreuung zeigen. Und schließlich wäre Gert Wagner nicht Gert Wagner, wenn er seiner privaten Sportbegeisterung nicht auch wissenschaftlich nachgegangen wäre. Neben dem Verhältnis von Sport und Doping, mit dem sich Nicolas Ziebarth in seinem Beitrag auseinandersetzt, geht es hier natürlich unvermeidlich auch um die – wie Jürgen Gerhards und Michael Mutz zeigen – schönste Nebensache der Welt: den Fußball (dem der Jubilar ein Semester seines Studiums widmete, um bei der Fußballweltmeisterschaft 1974 als freiwilliger Helfer mitzuwirken).

Sucht man nach den Gemeinsamkeiten im vielfältigen Wirken Gert Wagners, dann findet sich neben wissenschaftlicher Exzellenz (»Innovation«) und hoher gesellschaftlicher Relevanz (»Wissenstransfer«) vor allem ein roter Faden: die frühe Förderung junger Wissenschaftlerinnen und Wissenschaftler (oft beginnend zu einem Zeitpunkt, an dem diese sich selbst noch nicht darüber im Klaren waren, dass sie diese Richtung einmal einschlagen würden). Einige der Autorinnen und Autoren, die zum vorliegenden Band beigetragen haben (und viele, die keinen Beitrag zu dieser Festschrift leisten konnten), hätten den langen und oft steinigen Weg zu akademischen Meriten mit Sicherheit weniger erfolgreich zurück gelegt, als es ihnen durch die verlässliche, stets konstruktive Begleitung des Grenzgängers Gert Wagner tatsächlich möglich gewesen ist. Dies gilt auch und insbesondere für die Herausgeberin und die Herausgeber dieses Bandes, die allesamt an der Ruhr-Universität Bochum bei Gert Wagner studieren und arbeiten durften und sich beim Jubilar mit dieser Festschrift und einem herzlichen »Glück auf!« bedanken möchten.

In Praise of Panel Surveys, a Sonder-Panel, and a Sonder-Panel-Papa

Stephen P. Jenkins & Timothy M. Smeeding

Introduction

In this paper, we salute Gert Wagner and his work, focusing on his association with the Socio-Economic Panel (SOEP). To place Gert’s contributions in context, we argue first that household panel surveys deserve to be praised for what they contribute to science and to public policy, forming a crucial component in a portfolio of different types of longitudinal data. Second, we show that the SOEP is a very successful example of a household panel survey, comparing its characteristics and innovations with those of its counterparts from other countries. Our case is that the SOEP is truly special (it is a Sonder-Panel) – and Gert Wagner has been responsible for much of this success. He is truly a Sonder-Panel-Papa.

Why Praise Household Panel Surveys?

Praise household surveys because they are a valuable source of longitudinal data, and longitudinal data are an important type of collection mechanism for addressing many social science issues relevant to policy. Longitudinal datasets are those in which the same set of individual units is tracked over time; we have movies on the same units rather than a series of snapshots on different samples of units as one does with repeated cross-section data. In principle, the movies may be created using surveys with retrospective recall questions, or using prospective data collection based on temporally-linked administrative registers, cohort studies, or – the focus of this paper – household panel surveys.

The Value of Longitudinal Data

Longitudinal data are valuable for three main reasons. They describe phenomena and relationships that are intrinsically longitudinal (and their correlates); they provide a better understanding of socioeconomic processes over the life course and behaviour and, thereby, they better inform policy.

First, considering better description, longitudinal data enable us to distinguish gross change from net change. We can relate a fall in the poverty rate to increases in flows out of poverty or reductions in flows into poverty (Bane/Ellwood 1986). In addition, some phenomena of scientific and policy interest are inherently longitudinal. Examples include how long people remain poor (poverty persistence) or sick, the extent to which exits from unemployment are sustained or represent a »low pay – no pay« cycle, the prevalence of residential mobility, and household formation and dissolution (marriage and divorce, births and deaths, and children leaving home or boomeranging back). We can look at not only events per se, but take spell-based perspectives, and assess how long spells last, how the chances of spell endings vary with elapsed duration, and with characteristics that change during the spell. Longitudinal data also provide information about the associations between current events and outcome experienced by individuals and their past history. We can study questions such as the relationships between current unemployment chances and past unemployment, children’s development and life chances and their family background, income in old age and work-life history, current earnings and job tenure, labour market experience, and we can measure differences between current outcomes and past expectations (»surprises«).

Second, concerning greater understanding, longitudinal data, by contrast with cross-sectional snapshot data, allow us to better align our models with underlying constituent processes. Rather than modelling changes in the unemployment rate directly, we model the chances of leaving work among people who have a job, and the chances of finding work for the people who do not currently have a job. The drivers of each process (and the people at risk of the events) differ, and should not be thought of as the same. Going further, one can understand not only transitions per se but also – with a spell-based perspective – how the chances of getting a job vary with how long the spell of unemployment has been, and how the chances vary with circumstances that change during the spell (e.g. the amount of unemployment benefits the person is eligible for).

Empirical modelling and hence understanding is further enhanced by longitudinal data because they allow for the possibility of controlling for the effects on outcomes of not only observed characteristics such as age, sex, educational qualifications, but also the persistent characteristics of individuals that are unobserved (or intrinsically unobservable). Having repeated observations on individuals allows one to difference out time-constant factors of all kinds, observed and unobserved. Or one can exploit the fact that past histories of outcomes incorporate information about the realised effects of unobservables, and summarise their distribution.

More generally, we can make better causal inferences from our empirical models using longitudinal data because there is a temporal ordering in the data of outcomes (later) and hypothesized drivers (earlier). Indeed, longitudinal data are the essential ingredient of the social experiment revolution in evidence-based policy analysis and impact evaluations, using methods such as randomised control trials as well as several types of quasi-experimental designs (including differences-in-differences based on before-after comparisons for the same individuals).

Third, and a consequence of the two features just described, longitudinal data enable us to better inform policy. They enable better focus on the underlying processes, rather than on »problem groups« at a point in time (such as »the poor« or »single parent families«) that may be subject to a high degree of turnover in any case rather than being a fixed and unchanging population. The importance of this orientation is illustrated by David Ellwood, one-time advisor on welfare policy to President Clinton, who stated that:

»[D]ynamic analysis gets us closer to treating causes, where static analysis often leads us towards treating symptoms. … The obvious static solution to poverty is to give the poor more money. If instead, we ask what leads people into poverty, we are drawn to events and structures, and our focus shifts to looking for ways to ensure people escape poverty.« (Ellwood 1998: 49)

The same point was picked up on by the UK’s reform-minded New Labour government:

»In the past, analysis … has focused on static, snapshot pictures of where people are at a particular point in time. Snapshot data can lead people to focus on the symptoms of the problem rather than addressing the underlying processes which lead people to have or be denied opportunities.« (HM Treasury 1999: 5)

Longitudinal data contribute to policy design because generally they provide policy-relevant contextual information about key risks and potential intervention points relevant to policy focus and policy design and, specifically, they can be employed to evaluate the impacts of specific programmes. In short, they help us to understand not only the »Whats« of social indicators (such as poverty rates) but also the much more difficult causal »Whys« that help in successful policy design.

Household Panel Surveys: Key Features, and Examples from Around the World

The discussion above is about longitudinal data speaking generically, and there are multiple ways of collecting these. What are the particular features of household panel surveys such as the SOEP compared to other sources?

Household panel surveys are prospective longitudinal designs. Data collection is undertaken in an initial year (call it t) with repeated follow-up data collection points typically at (approximately) annual intervals thereafter (years t+1, t+2, …). The number of these has increased significantly over the last few decades. The pioneer and longest-running is the US Panel Study of Income Dynamics, which began in 1968 and celebrates its 50th anniversary in 2018. Major household panel surveys began in 1984 in Germany (Socio-Economic Panel; on-going), the Netherlands (Dutch Socio-Economic Panel, 1984–1997), and Sweden (Panel Study of Market and Nonmarket Activities, HUS, 1984–1998, and the Level of Living Surveys, from 1968 onwards). Over the following two decades, household panels began in Australia, Belgium, Canada, Korea, Luxembourg, the Lorraine region of France, Hungary, New Zealand, Switzerland, and Britain (the BHPS, 1991–2008). The BHPS has been superseded by Understanding Society – the UK Household Longitudinal Study, which not only incorporates the BHPS sample, but adds a new large sample of respondents (from 2009). There is the multi-country European Community Household Panel (ECHP) survey which used a cross-nationally harmonized instrument. In 1994, the first waves of surveys were fielded in twelfe member states, with some member states joining later. There were eight waves of fieldwork, with the final one in 2001. There are also a growing number of household panel surveys in other countries, including developing ones. Examples include the Korean Labour and Income Panel Survey (KLIPS), the KwaZulu-Natal Income Dynamics Study (KIDS), the Russia Longitudinal Monitoring Survey (RLMS), and the Indonesia Family Life Survey.

A distinction can also be made between perpetual (or indefinite) life panels such as the SOEP for which data collection is intended to carry on indefinitely, with no final collection date set at the outset, and rotating panel surveys for which the number of data collection points is fixed at the outset by design, and there are typically new panels starting each year (e.g. panel I starts in year t, panel II starts in year t+1, etc.), so that for any given calendar year, there are data from multiple panels. Leading examples of rotating panels are the panel surveys used to contribute longitudinal data for EU-SILC (there are four annual data collection points per panel), European labour force surveys (five quarterly data collection points per panel), and the US Surveys of Income and Program Participation (interviews every four months over periods of 2½ to four years depending on the panel).

Household Panel Surveys Compared to Retrospective Designs

Both types of prospective panel survey can be contrasted with retrospective designs in which there is a single data collection point, with the data for previous periods collected by retrospective recall of respondents about their circumstances and characteristics now and in the past. Because it is difficult to reliably collect information about income amounts and some other detailed aspects of people’s lives, retrospective designs have focused on topics for which this is less of an issue, e.g. less detailed information about a respondent’s parents such as job type or occupation at the time the respondent was a teenager (for studies of social class mobility), or fertility histories for mothers of young children (as in many Demographic and Health Surveys). Otherwise, the most common form of retrospective data collection is within household panel surveys, to collect information about the period between the annual data collection points, e.g. monthly job histories, with recall reliability issues mitigated by the shorter recall period.

Household Panel Surveys Compared to Cohort Surveys

Household panel surveys can also be contrasted with cohort surveys which are also perpetual panel surveys. The key distinctions relate to features such as the population of interest, frequency of data collection, and the nature of the data collected. Household panel surveys are surveys of the private household population (individuals and their households), and are designed to maintain representativeness of the sampled population over an extended period. Representativeness is achieved by implementing particular »following« rules for data collections after the initial one. Original panel members are followed even if the household splits (e.g. a husband and wife divorce and move to form two separate households) or is geographically mobile. Children who are members of respondent households become respondents in their own right when they reach a particular age (in the SOEP it is the year the child turns 17). The survey design mimics the way in which the population reproduces itself over time.

By contrast, cohort surveys are more narrowly focused on individuals with a particular set of defining characteristics, and hence are designed to maintain representativeness of the sampled cohort (they are individual- rather than household-focused). The leading examples are birth cohort surveys, in which there is sampling of many (or all) individuals born round a particular date. For example, the UK has had birth cohort studies following individuals born in 1946, 1958, 1970, 1980, and 2000/1.

In a cohort survey, each cohort member is followed over time and, although there may be some data collection about co-resident individuals on each occasion, the co-residents are not always followed. Interviews are typically less regular than for household panel surveys (often several years apart but not always thus) and cover much longer periods of individuals’ lives. (The UK’s 1958 birth cohort study recently interviewed individuals aged 55.) Other types of cohort surveys cover transitions from school to work and thereafter (e.g. the National Longitudinal Studies of Youth in the USA), or from work to retirement (such as the US Health and Retirement Study, and the English Longitudinal Study of Ageing, each focusing on individuals aged 50+).

Data collection in cohort studies is relatively frequent initially when development is relatively rapid (early childhood in birth cohort surveys) and less frequent thereafter through the life course. (The UK’s Millennium Cohort Survey which started in 2000/1 has collected data so far at ages nine months, three, five, seven, eleven, and 14.) The long-running nature of birth cohort surveys means that they focus on developmental and life course and intergenerational issues, and the topic focus varies between sweeps. By contrast, household panels with their annual data collection focus on topics for which short-term changes are more relevant, notably subjects such as labour market activity, incomes and other factors related to living standards, housing conditions, demographic change, and so on. High priority is given to repeated measurement of the same phenomena: the same topics are covered at each wave rather than changing from wave to wave as with cohort surveys. In addition, data collection refers to all individuals within the household by design (all of whom are followed over time), rather than one particular person and a varying degree of information about their household context.

Household Panel Surveys Compared to Linked Administrative Data

All the discussion so far has tacitly assumed that data collection is undertaken using a survey of the targeted respondents, whether the survey is done face to face or by other modes, such as telephone or web. Longitudinal datasets can also be compiled by temporal linkage of administrative register data.

Administrative data have distinct advantages. They are typically based on very much larger samples and more comprehensive coverage than possible in surveys, participation is not a choice of the targeted individuals (reducing problems of unit non-response and loss to follow-up), and data are often viewed as being more accurate than respondent recall (e.g. income data included in the registers may come directly from employer payroll records, and penalties against tax avoidance may reduce incentives to under-report income). Also, the data are cheap to collect by comparison with surveys – they already exist as a by-product of the administrative process.

However, the by-product nature of the data collection process also signals the main disadvantages of administrative register data. The scope of data collection is limited to the sponsoring agency’s purposes, not the goals of researchers. The outcome variables in the longitudinal data may be rather limited in number and definition, and there may be few of the additional covariates that are routinely wanted for empirical modelling, e.g. income tax return data do not include information about a tax-payer’s educational status because this is not relevant to assessing tax due. Similarly, no information about household composition may be collected. Precise details about pay may not exist for individuals earning below the social insurance liability threshold or above the maximum amount (as in the Integrated Employment Biographies from the German Institute for Employment Research (IAB)). Furthermore, payroll tax records do not capture non-covered earnings, which are reported on surveys, thus underestimating the variable of interest (Hoyakem et al. 2016). Major changes in a tax system may introduce non-comparabilities over time in coverage or variables collected. In countries with individual-based tax systems, it is usually impossible to link individuals with other household members. For this reason, longitudinal administrative data are most useful for individual-focused analyses and less useful for studies in which household context is important (which is of course the forte of household panel surveys). Finally, because of the very nature of much administrative data, there are concerns about privacy and confidentiality, so that researchers’ access may be only under restrictive (or inconvenient) conditions, or the variables made available in the public use data may be censored or in banded form to reduce disclosure risks.

This is not to say that administrative data are not valuable to panel surveys. Indeed the ability to link records for panel survey respondents to administrative data can help us understand the topic of panel attrition and its possible biases in much greater detail than by any other method (US National Academy of Sciences 2016).

Household Panel Surveys: Conclusions

On balance, it is impossible to say generally whether household survey data or linked administrative register data are best: it depends a lot on the national context and also the research question. At one extreme lie the Nordic countries with widespread use of administrative register data, characterized also by extensively linking across different types of registers. This means that it is possible to look at household context as well as individual circumstances per se, and a wide range of both outcomes and covariates. Also facilitating use are national cultures in which using a national identity (or social security) number for many purposes is widely accepted, and there are fewer concerns with personal privacy issues related to income and taxation than in most other countries. In other countries, the use of longitudinal administrative data is growing but not as developed. A notable example is the work of Chetty and colleagues linking US Internal Revenue Service records to derive income histories and to link individuals and their parents, and also exploiting detailed information about geographical location and correlates of intra-area mobility (see e.g. Chetty et al. 2014; 2016).

The upshot is that there remains a substantial role for household panel surveys as a source of longitudinal data, particularly for research questions that require information about household context – including interactions among household members, whether concerning their living standards or demographic behaviour – and across multiple life domains (e.g. work, family, attitudes and beliefs, etc.). In addition, most countries use measures of household- or family-level income and resources when monitoring levels and trends in individual economic well-being and for assessing eligibility for social assistance and other income support programmes. Even where longitudinal administrative data are not available, administrative data may be used to supplement and enhance survey data collection. In some cases, one may be able to link administrative register data with survey respondents. (An example is the linking of test scores and other information in the English National Pupil Database with members of the UK birth cohort surveys.) This raises issues of informed consent to data linkage, and other linkage biases arising when statistical matching across registers is required. Another form of panel survey data supplementation is the matching of geocoded data about the areas in which respondents live rather than linking at the individual-level data. We return to this below with reference to the SOEP.

A further important characteristic of household panel survey designs is that they have been implemented in very similar ways in a number of countries, and there is a core set of variables that is common to each of the surveys. Both features mean that production of cross-national harmonized data is relatively straightforward, at least in principle, though also dependent on securing the resources to make it happen. The notable success in this area is the Cross-National Equivalent File (CNEF), to which almost all national household panel surveys contribute data. Comparable cross-national panel data are available in the CNEF from eight countries, the contributing surveys being the PSID, SOEP, SLID, BHPS, HILDA, Swiss HPS, and KLIPS. See Frick et al. (2007) for a description of the CNEF.

This picture of richness of cross-nationally comparative data is a marked contrast with that for longitudinal administrative register data, because countries differ so much in their social policy institutions and the systems used to administer them. Cross-nationally comparable data are also rare for birth cohorts because designs have differed, but exist for cohort surveys of elder people – precisely because comparability and harmonization were built in at the start. We are referring to the Survey of Health, Ageing, and Retirement in Europe (SHARE), modelled on the US Health and Retirement Study, which began with twelve participant countries and since expanded to include many more.

We summarize the principal features of household panel survey designs for the collection of longitudinal data in Table 1, comparing their advantages and disadvantages relative to other data collection designs. The main advantages of household panels lie in their focus on individuals within their household context, the coverage of multiple life domains, and the relatively high frequency of data collection, enabling coverage of relatively frequent life events and exploitation of repeated measures modelling techniques.

In the next section, we continue the story, but elaborating some details not covered so far. Focusing on the case of the SOEP, we demonstrate how it stands out as an exemplar of a good household panel survey.

Design feature

Advantage or disadvantage relative to other longitudinal designs

Sample size

Small relative to longitudinally-linked registers and rotating panels

Panel length

For perpetual life panels, depends on maturity (but longer than rotating panels)

Data collection frequency

Annual (mostly by personal interview); more frequent than most cohort surveys

Coverage and representativeness

Focus on national populations of individuals living in private households (cf. individual focus in cohort surveys and most linked registers)

Topics covered

Intentionally broad, covering all life domains (broader than cohort or linked registers), typically with topic-specific modules on a multi-year rotating cycle

Attrition

Potentially a greater problem than for linked registers or rotating panels

Measurement error

Greater than for longitudinally-linked registers

Availability and access

Much greater than for longitudinally-linked registers

Cross-national comparable data

Good by comparison with longitudinally-linked registers and most cohort surveys

Table 1: Household panel surveys: design features, and their advantages and disadvantages

Note: Adapted from Jenkins (2011: Table 3.1).

Why is the SOEP Special?

The SOEP is an example of a household panel survey, as we described in the previous section. But what makes it such a good example? In this section, we provide answers to this question. We write as researchers based outside Germany, and emphasize a number of features that strike us personally; we are not aiming to be comprehensive (for the SOEP team’s own view of the situation a decade ago, see Wagner et al. 2007.) The gist of our story is that, as the SOEP has evolved, it has incorporated changes and innovations that address a number of the disadvantages or vulnerabilities that are often associated with household panel surveys. We focus on developments in sample design, content, user support and access, and resources.

The SOEP is now the longest-running household panel that has not experienced major changes in design and content, 1984 to present day (33 years). To be sure, the PSID started in 1968 and is still going, but it has had a major design change (the switch to data collection every second year in 1999, the change from face-to-face to computer-assisted telephone interviewing in 1993), and the PSID’s content coverage of life domains is not as comprehensive as the SOEP’s. The long-running nature of the SOEP means that one can look at not only short-run change (as with all household panels), but also increasingly able to address intergenerational issues by having data collected for parents and their offspring, and there is greater potential for following individuals over their course from cradle to grave and across multiple generations – increasingly the PSID’s focus and comparative advantage.

Sample Design

Among the major household panel surveys, the SOEP was the first to move away from the PSID model of using a single respondent to provide information about all household members and the household itself. Instead, all adult members of SOEP households receive an individual questionnaire (and there is also a household questionnaire completed by one person). Clearly, collecting data in this way is more expensive, but has great advantages in terms of reliability (adults report about their own circumstances rather than relying on the reports of a proxy), and also makes it possible to address new research questions relating to within-household bargaining and other matters.

Representativeness and sample composition are important issues for all household panels as they mature. Respondent drop-out (attrition) is the problem most commonly flagged in this respect, but there are also more fundamental questions concerning the on-going representativeness of the target population (individuals in the private household population).

One consequence of attrition is a fall in sample size over time, leading to less precise estimates. This is a particular problem when looking at small-sized population subgroups (e.g. lone parents; some minority groups). More than any other household panel survey, the SOEP has systematically and repeatedly introduced new »refreshment« samples of the German population purely for this reason: sample E in 1998, sample F in 2000, sample H in 2006, sample J in 2011, and sample K in 2012. In each case, more than 1,000 households were added and often more (3,136 in sample J).

By construction, the first wave of a household panel survey aims to be representative of the private household population in that initial year, and representativeness of contemporary society in subsequent years is maintained by the survey’s following rules – as long as the society does not change its fundamental character (or there is differential attrition – see below). But, as is well known, what constitutes »Germany« has changed in very fundamental ways, because of reunification and immigration, and the SOEP has responded to this challenge.

Perhaps the most far-sighted innovation of the SOEP was to introduce a new sample of more than 2,000 households in Eastern Germany in 1990, just months after the fall of the Wall and around the time of formal reunification. Thus, the reconstitution of German society has been tracked from the very start by the SOEP, and it has provided invaluable information about how differences between East and West have dissolved or persisted. Interestingly, a recent SOEP-based commentary, two decades on, states that »differences between East and West still exist in many areas. But they depend much more on the concrete living conditions in a specific place than on whether people or their parents lived on one or the other side of the inner-German border« (Krause, cited in DIW Berlin 2013). This is an incredible achievement in such a relatively short period of time, and a tribute to the underlying strength of the German economy and related social and political institutions.

Immigration has long been a major feature of German post-WWII society, including the arrival of »guest-workers« decades ago, the migration of ethnic Germans after German reunification, including the arrival of refugees more recently. German society today is much more diverse than German society in 1984. If the SOEP had continued to be based on the original 1984 sample, its contemporary samples would provide a biased picture of society today. But this has not been the case.

The SOEP has accounted for immigration all along. In order to be representative of Germany in 1990, the SOEP already included a special sample of households headed by someone from Turkey, Italy, Spain, Greece, and the former Yugoslavia. In 1995, a sample of around 500 households of immigrants who had moved to Germany after 1984 was added. And a substantially larger sample, of around 2,700 migrant households, was added in 2013 using register data held by the Federal Employment Agency (IAB) to develop the sampling frame. In 2016 the SOEP added a special random sub-sample of refugees coming to Germany between 2013 to 2015 (Brücker et al. 2016). Societal change is also reflected in the way these samples and their target populations are referred to by the SOEP. At the outset, there were many references to »guest-workers« or »foreigners« rather than the more generic »immigrants« used now. Other high immigration countries such as the USA and the UK have introduced special samples of immigrants and ethnic minority groups (mostly immigrants) in their household panel surveys, but in neither case has it been as thorough or as successful as in the SOEP.

Immigrants are an example of a group of particular interest. Another such group is »the rich«. The interest stems from both substantive reasons – growing concerns about inequality and gaps between the rich and the poor – but also methodological reasons. Household surveys of all kinds (not only panel surveys) are often cited as under-representing the richest households in a society by comparison with benchmarks derived from personal income tax register data. The SOEP has been a pioneer among household panel surveys in its introduction of a »high income« sample, starting to track around 1,200 households from 2002 onwards. Interestingly, around 100,000 households had to be screened in order to generate this number of respondents (Wagner et al. 2007: 13) – establishing contact and securing a successful interview is difficult – and yet, once interviewed, retention rates are as high as for other types of respondents (Kroh et al. 2015, Table 3.1).

As a result of its specially targeted and refreshment samples, there are now twelve SOEP samples. This is another distinctive feature of the SOEP – no other household panel has so many – and it gives rise to complexities for the data producers and users that are perhaps under-appreciated. The issues stem from the fact that the respondents to each of the different samples had a different probability of selection into the survey, and this has to be taken account of in any analysis. In principle, doing so is straightforward: to derive population estimates, data should be weighted by the »design weight« appropriate to each sample (the inverse of the probability of selection), and these are known. In practice, things are rather messier, because weights also need to take account of cross-sectional non-response (as all samples do) and non-response over time (attrition) in order to enable derivation of population-representative estimates – non-respondents may differ in significant ways from respondents, as hinted above. In addition, weights may be further adjusted so that sample estimates correspond to population totals (grossing-up or post-stratification).

Practical complications increase the larger number of survey waves there are and, related, because there are a very large number of potential samples that analysts might use in longitudinal analyses – ranging from the pooling of year-on-year transitions over a number of years to spell-based analysis based on samples defined by response over a large number of waves, and users may wish to base analysis on some of the various SOEP samples but not others. This is an example of the more general issue that the weights supplied by survey data producers are often general-purpose weights, not tailored to a particular analysis or »population«.

There is no easy remedy for addressing the complexities associated with weights (especially since many of them are conceptual). However, the SOEP has been exemplary in its development of different types of weights (cross-sectional and longitudinal) for users and been responsive to their needs. From early on, the SOEP has provided information about survey design (clustering and stratification) and separate design weights as well as weights accounting for non-response, attrition, and grossing-up. Compare this with the PSID (which supplies only one type of weight) and the BHPS (which supplies only one type of longitudinal weight, relevant to individuals with continuous panel participation from the start of the survey), whereas the SOEP has separate longitudinal weights referring to wave-on-wave non-response for each wave. This richness of detail has come at the price of complexity: the SOEP staff member responsible for weighting has remarked that »our impression from users« feedback was that after 25 waves, the growing number of weighting variables for each wave but also for different combinations of sub-samples … made the SOEP less accessible to new users. One aim of the revision was thus to concentrate on the »standard« variables in the data distribution. (Kroh 2009: 1–2) At the same time, there is detailed documentation and training in the meaning and use of the different sets of SOEP weights available: see e.g. Kroh (2010).

We cannot leave discussion of SOEP samples without also mentioning the Innovation Sample (SOEP-IS), introduced in 2012 as an enhanced replacement of the existing pre-test sample, combining a new subsample with subsamples of respondents from existing samples E and I. (SOEP-IS also replaces the previous SOEP pre-test sample.) The primary goal is »to test innovative survey methodologies and apply them to a representative longitudinal sample of the German population. A further objective is to test innovative procedures that go beyond the classic survey components (after pretesting if required) with an adequate sample size for high-quality data analysis« (Richter/Schupp 2012: 4). This is a major methodological enhancement. SOEP-IS has many features in common with the Innovation Panel component of the UK Understanding Society household panel survey (introduced in 2008), for example in its aims and the use of annual competitions among users for questionnaire space, but is distinctive nonetheless. There is the combination of old and new samples, and the sample size in total is more than twice as large, which substantially expands the scope for methodological experimentation.

Survey Content

Innovations in survey content are the second main area in which SOEP is special. By content, we mean the areas of people’s lives about which data are collected – the questions that are asked of respondents and the range of variables available to users. Compared to other household panel surveys, the SOEP has been notable for its emphasis on psychological and attitudinal measures. From the very outset, the SOEP was innovative in collecting data about self-reported life satisfaction (on an eleven-point Likert scale) as part of a more general goal to assess social progress using a battery of social indicators in addition to income alone. (For a detailed discussion of the genesis of this dimension of the SOEP, see Schupp 2015.) As a result, the SOEP has become an essential core component of survey-based studies of life satisfaction. The most widely cited research paper in the history of the SOEP, according to the SOEP’s online Newsletter (February 2017) is Winkelmann and Winkelmann’s (1998) study showing that job loss makes people unhappy and this unhappiness arises from more than simply the loss of earned income.

Even if one restricts oneself to more traditional money-based measures of well-being, the SOEP has been innovative. From early on, the SOEP has routinely provided, as part of the user database, a measure of household disposable income that is consistent with international standards such as set down by the Canberra Group, and constructed from respondents’ reports about the income received from each element of an extensive menu of potential sources over the previous calendar year. The PSID produced these household disposable income measures only in its early years, and the BHPS never did it. In both cases, however, these household disposable income variables have been produced separately from the main survey releases and without the same level of institutional investment, quality control and timeliness of data release. In addition to the »Canberra« household disposable income variable referring to the previous year’s income, the SOEP also has data about the current household net income derived from household heads’ responses to a single question.

To return to the psychological aspects of people’s lives, the SOEP has been a pioneer among household panel surveys in its use of occasional supplementary modules to collect data about personal psychological traits. By these, we refer in particular to the »Big Five« personality traits (openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism), as well as measures of risk aversion, trust, fairness, and reciprocity. Data collection on two measures of cognitive ability, based on simple tests, has also been introduced. Hence both cognitive and socio-behavioural traits can be tracked over time for the same individuals and related to current and past events, including educational experiences, as they move through the life course.

Although the SOEP was conceived as a »socioeconomic« panel, it has been an earlier responder to scientific interest in the overlaps between socioeconomic and health domains of people’s lives, for example, considering issues such as the relationship between obesity and income or education. The SOEP has long collected self-reported measures of height and weight, but it was also the first household panel to collect physical health measures (after specialist panels focusing on older people), notably a measure of grip strength from 2006 onwards.

As the SOEP has matured, its potential for looking at the whole life course has been matched by a greater focus on specific population groups and life events using specially-tailored survey instruments. The principal examples of this are the Mother and Child questionnaire modules. Since 2003, data have been collected about new-born children; from 2005, data on infants; from 2008, data on children entering primary school; and from 2013, children entering secondary school. Since 2001, adolescence-specific data have also been collected from first-time respondents (aged 16 or 17). Event-triggered instruments are being planned around partnership dissolution, and transitions from work to retirement. The extra information about early childhood is reminiscent of the PSID’s innovative Child Development Supplement – in 1997, additional data were collected about children aged 0 to twelfe and their parents, with follow-ups of these families in 2002/3 and 2007/8. A new cohort began in 2014. The BHPS has a self-completion Youth Questionnaire directed at children aged eleven to 15 (i. e. not yet full non-respondents) that started in 1994. The SOEP’s Mother and Child modules borrow the same basic ideas from these predecessors, but have been more systematic in their coverage of different childhood stages.

A further major development over the last decade of the SOEP is its improved capacity for spatial analysis. Given the sample sizes of household panels (and their clustered design), estimates for country regions below the national level are rarely possible, the major exception being the UK’s Understanding Society panel survey with its very large sample size (around 40,000 households). Sub-national estimates for the SOEP are only representative of the very largest federal states in Germany.

The enhanced capacities that we are referring to is the potential for linking geocoded data to respondents using multiple levels and definitions of geographical areas. Facilitating these improvements have been changes in perceptions about the potential net benefits of having such data combined with the design of appropriately tailored user contracts to ensure that respondent confidentiality and privacy are protected, plus technological advances that make it possible to use sensitive data securely. One can now link in geocoded information at the county level and do so remotely using SOEP-Remote (a remote access platform based on the successful LISSY platform developed by the LIS Datacenter) and having signed a special user contract. Data can now even be linked at the postcode level for users making special arrangements with DIW Berlin and using the in-house secure research data centre. From being perhaps a laggard among panel surveys in facilitating geocoded data analysis, the SOEP is now among the leaders, with innovative data access solutions.

Documentation and Data Access

Among household panel surveys, the SOEP has always stood out for its extensive documentation, and for documentation in an easily digestible form to help new users (see SOEP Group 2017). The SOEP team has developed these in English as well as German – a smart move that has ensured that the SOEP’s use among non-German users has always been remarkable. Similarly inspired was the »95 percent scientific use version« directed at international users from very early on, enabling them to use the data in a manner that would not broach Germany’s data protection laws.

The DeskTop Companion has long been the essential starting place for new users and a useful reference for experienced ones. There is a bespoke data extraction tool, John Haisken-deNew’s PanelWhiz that makes it much easier to extract complex longitudinal data from the multiple data files of different type and year. With the same goal, the SOEP has recently introduced a new data release format, SOEP-Long, in which data are ready-supplied in long (panel) form and so do not have to be combined and reformatted from wave-specific files. We already mentioned SOEP-Remote above. The SOEP team has also developed a user-friendly metadata system that is also used by other longitudinal studies (http://paneldata.org).

Resources and Infrastructure

It should be clear from our discussion so far that the SOEP is definitely special – it is a Sonder-Panel. Other household panel surveys have some of the features we have highlighted; it is the one that has them all. The explanation for this, and an additional distinctive feature in itself, is the SOEP’s resources and infrastructure. (The team and its leader are also important, a point we return to in the next section.)

Our perception is that the SOEP has received greater core funding support than other household panel surveys, if not in terms of resource levels per se, then in longer-term stability of support. The SOEP began life in 1983 as a special research area (Sonderforschungsbereich, Sfb) in the »Microanalytical basis of social politics« based at the Universities of Frankfurt am Main and Mannheim. (Sfbs are collaborative research centres with long-term funding from the German Science Foundation, and multidisciplinary research programmes.) Sfb funding lasted until 1989 and then between 1990 and 2002 the SOEP was funded directly by the DFG, with additional support from the Federal Ministry of Education and Research. Nowadays, the SOEP receives its funding through the Joint Science Conference (Gemeinsame Wissenschaftskonferenz, GMK, an organisation with oversight over research funding, science and research policy issues jointly affecting the federal and state governments), by the Federal Ministry of Education and Research (BMBF), and the State of Berlin and other federal states. In short, the SOEP has moved from project-based science foundation funding to being incorporated into the national institutional infrastructures supporting science and research. And yet, at the same time, the SOEP retains greater discretion over the scientific direction.

This contrasts with the experience of household panels in the USA and UK. The PSID was initially funded by a government agency (Office of Economic Opportunity), but for a long time has been supported by a portfolio of national science foundations (including the National Science Foundation, which is currently the largest single funder, and the National Institute on Aging) and other co-funder organisations which include government departments. Refunding rounds occur every three to five years, and have involved a tough battle on each occasion, with pressures to include special modules at the expense of core longitudinal content, and there have been some cuts to sample size. Similarly, the funding for the BHPS and for its successor, Understanding Society, come from the Economic and Social Research Council (the main national social science research funder) together with a number of co-funders, mainly government departments, and also for time-limited periods (around five years). Refunding rounds have been and remain a fraught process involving submissions in a competition with other major science funding bids every five years. Australia’s HILDA funding model is closer to the SOEP’s; it is supported by the Department for Social Services, a federal government department, and has funding guaranteed for 18 waves. The use in the UK of competitive tendering processes led to a different fieldwork agency being used for Understanding Society data collection from the one used for 18 waves of BHPS data collection. The fact that the SOEP has been able to use the same agency (TNS Infratest Sozialforschung, known as Kantar public since 2016) for decades and develop a very close relationship with it, is another source of stability that other studies surely envy.

Along with greater and more stable funding support, our impression is that the SOEP team has been able to retain greater discretion over scientific direction than other study teams. In the USA, UK, and Australia, the competitive funding environment places greater control in the hands of the funders and the other constituencies represented among referees for funding bids and boards of governance (and their membership changes over time). One of us, a two-term member of the PSID Board of Overseers, experienced this turbulence first hand. The other of us, part of the BHPS team, was a member of the SOEP Scientific Committee around a decade ago and observed first-hand how the SOEP governance structure allowed the team to make strategic choices relatively independently and quickly. This was the era in which the strategic choice was made to incorporate collection of cognitive measures and behavioural experiments and quickly actioned in a way that could never have happened with the BHPS.

There is another feature of national context that has worked in the SOEP’s favour and which contrasts with the US and UK environments. Our understanding is that when the SOEP began in 1984, it was one of the only socioeconomic data sources for Germany to which researchers had access to unit record data – for repeated cross-section data let alone longitudinal data – and thereby could better address a wide range of scientific research questions. Thus the SOEP was able to embed itself in German social science early on, and this influence has persisted. The SOEP’s reputation for high quality data has reinforced this position, and it is not only for its longitudinal data – the cross-sectional data are also highly valued.

Our arguments are illustrated by the debates about the quality of the income data in the Survey of Income and Expenditure (EVS) and the data for Germany (that initially) contributed to EU-SILC. The SOEP was widely accepted as providing a benchmark to assess the relatively poor quality of these two sources from both cross-sectional and longitudinal perspectives: see e.g. Becker et al. (2003), Hauser (2008), and Frick/Krell (2010). The preeminence of the SOEP is further illustrated by the fact that it is used as the data source in chapters about income and wealth in the annual national report on social conditions and trends and, for this purpose, the SOEP’s cross-sectional data are used more than the longitudinal data (Goebel/Krause 2016, Grabka/Westermeier 2016). The repeated cross-section data for Germany available from the LIS Cross-National Data Center have continued to be sourced from the SOEP for many years, and not from the other German sources now available.

The multiple roles that the SOEP has come to play in the data portfolio for Germany have no parallels in other countries. For example, in the USA, the Current Population Survey has (since the start of the 1960s) always been used as the principal source of cross-sectional data about the income distribution, and the PSID has never threatened this role. Similarly, in the UK, the Family Expenditure Survey and, since the mid-1990s, the Family Resources Survey, have always been the preeminent cross-sectional data sources, never the BHPS. Moreover, US and UK researchers have long had access to unit record data from the surveys mentioned. The situation is quite different from Germany.