«DNB Occasional Studies Vol.10/No.4 (2012) The CentERpanel and the DNB Household Survey: Methodological Aspects DNB Occasional Studies Federica Teppa ...»
DNB Occasional Studies
The CentERpanel and
the DNB Household Survey:
Federica Teppa (DNB)
Corrie Vis (CentERdata)
Central bank and prudential supervisor of financial institutions
©2012 De Nederlandsche Bank NV
Authors: Federica Teppa (DNB) and Corrie Vis (CentERdata)
Aim of the Occasional Studies is to disseminate thinking on policy and analytical issues in areas
relevant to the Bank.
Views expressed are those of the individual authors and do not necessarily reflect official positions of De Nederlandsche Bank.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form by any means, electronic, mechanical, photocopy, recording or otherwise, without the prior written permission of the Nederlandsche Bank.
Subscription orders for DNB Occasional Studies and requests for specimen copies should be sent to:
De Nederlandsche Bank NV Communications P.O. Box 98 1000 AB Amsterdam The Netherlands Internet: www.dnb.nl Occ asiona l St ud ies Vol.10/No.4 (2012) Federica Teppa (DNB) Corrie Vis (CentERdata) The CentERpanel and the
DNB Household Survey:
Methodological Aspects The CentERpanel and the DNB Household Survey: Methodological Aspects Table of contents
7 1 Introduction 9 2 A brief history of the CentERpanel 11 3 Methodological aspects of the CentERpanel 15
3.1 The interview mode 15
3.2 Recruitment and selection 17
3.3 Criteria for participation and drop outs 17
3.4 Modules, response monitoring, feedback and assistance 18
3.5 Representativeness of the panel and data quality 18
3.6 Response rate 21
3.7 Incentives for participation 27 4 The projects using the CentERpanel 29 5 The DNB Household Survey 31
5.1 The questionnaires 31
5.2 Data dissemination and use 32 6 Conclusions 41 References 43 The CentERpanel and the DNB Household Survey: Methodological Aspects Abstract This study presents an overview of the main characteristics of the CentERpanel as well as of the DNB Household Survey, a longitudinal database of economic and psychological aspects of ﬁnancial behaviour of Dutch households run at CentERdata, Tilburg University and sponsored by De Nederlandsche Bank. Started in 1993, the survey is a very rich and valuable database for both academic research and policy oriented studies. Its ﬂexibility in collecting the data on a wide range of topics together with limited associated costs makes the DHS a very powerful instrument for empirical analysis.
Jel-Classiﬁcation: C5; C9; D12; G11 Keywords Survey data; Self-administered questionnaires; Survey representativeness;
Response rates Acknowlegments We thank Jakob de Haan, Marcel Das and Maarten van Rooij for their very useful comments and René Kurpershoek for proofreading. We are also grateful to the Household Finance and Consumption Network for the stimulating interactions on several sections of this study. The views expressed in this paper are those of the authors and do not necessarily reﬂect those of the institutions they belong to. Any remaining errors are our own responsibility.
1 Introduction In standard economic models, agents are assumed to act in an environment with perfect information (and the ability to process it properly), full rationality and complete markets. The substantial increase in the collection and availability of survey and administrative micro data has allowed predictions stemming from theoretical models to be tested, which revealed that in many cases empirical ﬁndings do not fully support the theory. Consequently, a huge literature developed exploiting the information conveyed by survey/administrative data, which turned out to be extremely valuable not only for the academic community, but also for policymakers.
While administrative data are a very valuable source of objective information on potentially the entire recorded population, they usually provide only limited individual characteristics and do not allow information matching between household members. Household surveys improve upon and complement administrative records as they allow for the collection of additional information from a representative sample of respondents by means of questionnaires.
In an increasing number of countries, large-scale household surveys are conducted regularly to obtain information on household income and wealth, household expenditure, job and occupational status, living conditions, access to and utilization of services, subjective expectations, health, and other critical areas of study, including psychological aspects of individual behaviour. The National Sample Survey Organization in India has been collecting household data on a regular basis since the 1940s. The Survey on Household Income and Wealth (SHIW) began in the 1960s with the aim of gathering data on the incomes and savings of Italian households. Of more recent date, the General Household Survey (GHS) is a survey conducted on an annual basis by the Ofﬁce for National Statistics (ONS) and has collected data about private households in Great Britain since 1971. The British Household Panel Survey (BHPS) is an annual survey covering a nationally representative sample of about 5,500 households recruited in 1991, with a total of approximately 10,000 interviewed individuals. The University of Michigan Health and Retirement Study (HRS) surveys more than 22,000 Americans over the age of 50 every two years since 1992, collecting information about income, work, assets, pension plans, health insurance, disability, physical health and functioning, cognitive functioning, and health care expenditures.
In the Netherlands the collection of data about economic and psychological aspects of household ﬁnancial behaviour started in in the early 1990s, with particular focus on savings. Since then a great amount of data has been collected annually through the so-called ’CentERpanel’ at CentERdata, Tilburg University. Over the past two decades, the panel has become a very rich and valuable dataset for both academic research and policy oriented studies. Its ﬂexibility in collecting the data on a wide range of topics together with limited associated costs makes the CentERpanel a very powerful and efﬁcient instrument for data collection that serves as the basis for several projects. One of the most important projects using the CentERpanel is the DNB Household Survey (DHS from now on), a longitudinal study of Dutch households sponsored by De Nederlandsche Bank.
This study provides an overview of the main characteristics of both the CentERpanel and the DNB Household Survey. The paper is organized as follows. Section 2 reports a brief overview of the history of the CentERpanel. Section 3 elaborates on the methodological aspects of the survey, with particular emphasis on recruitment and selection, criteria for participation and dropping out, economic incentives, response monitoring, feedback from respondents and assistance, representativeness of the panel, and response rate. Section 4 reviews the projects using the CentERpanel.
Section 5 describes the DHS questionnaires, the data dissemination and use. Section 6 concludes.
2 A brief history of the CentERpanel The origins of the panel In 1986 Willem Saris and Marius De Pijper of the University of Amsterdam developed a new procedure for computer-assisted data collection speciﬁcally designed for panel research. The main idea was to offer a good and reasonably priced instrument for data collection to be used for research in social sciences. A random sample of the Dutch-speaking population was provided with a computer, modem and software to communicate by telephone with a central computer located at the University of Amsterdam. Using this system, interviews could be sent to the respondents, who could then ﬁll in the questionnaires (off line), and return the responses to the central computer. In late 1990, after a few years of testing, a panel of about 2,000 households in the Netherlands was set up at the University of Amsterdam with grants from Eurostat and the University of Amsterdam.
The sample was drawn in collaboration with Statistics Netherlands (CBS), the national statistics institute of the Netherlands. The recruitment was done as follows.
At that time the ﬁxed phone net covered virtually the whole population, with a 99 percent penetration rate. In order to include unlisted phone numbers, the last two digits of the phone numbers in the sample were randomized (Random Digit Dialing, RDD). An Equal Probability Selection Method (EPSEM) was used to select the telephone numbers (see Nyhus, 1996 for further details). Twenty contact persons were sent to visit the households at home. The recruitment interview contained questions about both demographic and non-demographic topics. The data collected this way were stored in a database. At the end of the interview the respondents were asked whether they were willing to participate in a longitudinal panel. Nobody had a private computer at that time. However, the households that had expressed their interest to participate were provided with a (Sinclair) computer and a modem to connect to the central computer. The contact person helped the respondent ﬁll in the ﬁrst questionnaire about the composition of the household and explained how to use the computer and the modem. If a household refused to participate, the contact person would visit the household next door and repeat the same procedure (Random Walk Method).
Over the years, in order to deal with panel attrition, new households with the same characteristics as the households that left the panel (in terms of region, age, household composition, degree of urbanization and monthly household income) were drawn from the database. The database was updated regularly. Every now and then new, faster modems were purchased in order to facilitate the panel members and to keep pace with the ongoing technical development.
The development of the panel In 1993 the panel included the overrepresentation of the wealthiest population segment, which is usually the most difﬁcult segment to be inteÐrviewed. A subsample of about 1,000 households belonging to the 10 percent highest income group (deﬁned as monthly net household income of at least 5,000 guilders) was created. The households belonging to this High Income Panel also received a small (Quaderno) laptop computer in use and a modem.
In 1997 the panel moved to the University of Tilburg. The High Income Panel component was stopped due to high associated costs and the households belonging to that group were added to the rest of the panel. The RDD-method was no longer used due to the increasing costs associated with two phenomena: the diffusion of many telephone numbers belonging to companies rather than to households on the one hand, and an increasing number of disconnected telephone numbers on the other. Instead telephone ﬁles from GeoMarktproﬁel, the ﬁrst postal code segmentation system that was developed and exploited in the Netherlands, were adopted. It was then possible to stratify the samples (e.g. with respect to age).
In addition, in 1998 the recruitment of the panel members, done until then by CentERdata, was taken over by Marktonderzoeksbureau M4.
Moreover, the Netherlands Organisation for Scientiﬁc Research (NWO) sponsored the full replacement of the Sinclairs and Quadernos in the panel. In the second half of 1999, CentERdata launched a pilot study for the purchase and distribution of PCs for panel members which eventually led to the adoption of the Net.Box, a so called ’set top box’ allowing Internet access through a connection to the television and the telephone line. There followed a period of transition: from DOS to Windows, from ofﬂine to online interviewing, to new interviewing software and a new panel management system. By end-2000 the move to online communication and new software had been completed.
The following years saw the updating and improvement of technical details and the emergence of new phenomena. The coverage of ﬁxed phones decreased from 99 percent to 70 percent and the willingness of people in the Netherlands to participate in survey research dropped after 1990 (De Leeuw and Hox, 1998; De Heer and De Leeuw, 2002). In a study about non-response in the Dutch Labour Force Survey (LSF) Cobben and Schouten (2007) found that the respondents of the LFS and the non-respondents in the follow-up study differed with respect to the possession of a ﬁxed phone and to their country of origin.
The CentERpanel and the DNB Household Survey: Methodological Aspects The current situation Since 2008 TNS-NIPO has been in charge of panel member recruitment. TNS-NIPO is part of the TNS Group, one of the world’s largest agencies in the ﬁeld of marketing information and business insight. The TNS network provides access to research experience and expertise in more than eighty countries. TNS is an international leader in customer research, ﬁnance and public opinion. The recruitment procedure in place nowadays is described in detail in the next paragraph. For the households without a computer or Internet access, the Net.Box has been substituted by the SimPC, a simple computer speciﬁcally developed for older people and people with limited computer skills.
3 Methodological aspects of the CentERpanel