Tuesday 16th July
Wednesday 17th July
Thursday 18th July
Friday 19th July
Download the conference book
Linking Survey and Administrative Records: Processes and selectivities of consent 2 |
|
| Convenor | Ms Julie Korbmacher (Max Planck Institute For Social Law and Social Policy, Munich) |
| Coordinator 1 | Dr Joseph Sakshaug (Institute for Employment Research, Nuremberg and Department of Statistics, Ludwig-Maximilians-University, Munich) |
| Coordinator 2 | Dr Gundi Knies (ISER, University of Essex) |
Researchers are invited to submit presentation proposals at the session "Linking Survey and Administrative Records: Processes and selectivities of consent" at the European Survey Research Association conference, July, 15-19, 2013 in Ljubljana.
The number of studies linking survey and administrative records is still increasing in the social sciences. There are two ways how to link data, with different implications. In statistical or propensity score matching sample units from a survey are matched to "similar" (in a statistical sense) units in the administrative records. The other way is to link the two data sources directly which requires respondents' consent. The latter approach is usually thought to be more promising, but the fact the respondents' consent is needed could lead to problems. For example, if not all respondents consent, the sample size of the linked data set decreases. Additionally, consent bias might occur in the case of systematic differences between consenters and non-consenters. Compared to the number of studies asking for consent to record linkage, relatively little is known about the mechanisms behind the consent question.
Papers in this session focus on experiences and consequences when asking respondents for consent to record linkage in a survey. Topics might include issues such as determinants of consent; ways to increase consent rates; interviewer effects on consent; consent bias; experimenting with consent questions; consequences of asking for consent; etc.
Proposals should be no more than 500 words in length and should also be sent to Korbmacher@mea.mpisoc.mpg.de.
Despite its wide spread use, there is very little methodological research on data linkage in household panel surveys. Most of the survey design decisions regarding data linkage are often based on anecdotal accounts and common sense rather than sound empirical evidence. This paper is an attempt to fulfill this gap.
The aims of the paper are threefold: it aims to: (i) investigate the reasons why people consent/do not consent to data linkage, (ii) explore the effects of the position of the consent question in the questionnaire on response rates and bias, and (iii) confront and contrast the effects of different ways of re-asking consent (i. e., using dependent and independent questions) on consent rates and bias.
We report findings from a study and two experiments on data linkage carried out in wave 4 of the Innovation Panel (IP) of the UK Household Longitudinal Study. Detailed information on the Innovation Panel can be found at http://www.understandingsociety.org.uk/design/innovation/default.aspx. In the paper we also use recordings of the IP survey interviews.
The number of studies combining survey data and administrative records has grown recently, both because researchers seek to increase data quality and reduce the survey burden. Two basic approaches exist: one is the direct link, where the respondent's survey data are linked to her administrative records by identifiers in both datasets. The other is to use matching procedures, where statistically similar records are found and matched with the help of variables common to both datasets. The first approach ensures that the records linked belong to the respondent, whereas the quality of the matching depends on the available matching variables. However, the direct linkage usually requires the respondent´s consent, which reduces the coverage. Matching procedures do not require such consent, which allows matching records to every respondent.
This paper seeks to compare the two methods to give researchers the opportunity to choose between matching and direct linkage. In SHARE-RV, German respondents of the third wave of the Survey of Health, Ageing, and Retirement in Europe, SHARE, were linked to administrative records of the German Pension Fund via the direct linkage. We use matching procedures to find the respondent´s "statistical twins" from the German Pension Fund. We then compare the matched data from the statistical twins to the data obtained through direct linkage, which will provide insights to the costs and benefits of the two methods. Focus will be on the coverage error through record linkage and on the possible deviations induced by matching procedures.
Numerous studies have tried to explore sources of variation in linkage consent rates across surveys and administrative data targets. While most studies have focused on using respondent-level attributes to explain this variation, few have explored consent variation attributed to interviewers. Interviewers are usually responsible for administering the consent request and studies have shown significant variation in consent rates across interviewers. This begs the question of whether interviewer attributes play a role in respondents’ willingness to consent to link their records. In this paper, we address this question using detailed questionnaire data collected from interviewers prior to conducting interviews and asking respondents for linkage consent in the Panel Study “Labour Market and Social Security.” Specifically, we investigate whether interviewers’ personality traits, behavior during interviews, motivation for becoming an interviewer, and attitudes toward data linkage are related to respondent’s likelihood of consent.
Usually selectivity of respondent's consent to linkage of administrative with survey data is analyzed by looking at the conditions at the point of interview time. Further research has shown that consent is selective when focusing on actual status of respondent. By analyzing the consent in ALWA-ADIAB - a dataset which linked ALWA survey data to administrative data of the IAB - Antoni (2011) showed that respondent characteristics (such as age, employment status and data protection concerns), interviewer characteristics (such as gender and number of previous ALWA interviews) and interview situation (such as share of refused answers to sensitive questions) are important determinants of consent.
In my presentation I will look at consent in ALWA from a longitudinal point of view. ALWA conducted the individual educational and employment history of respondents retrospectively in every detail. So, I'm able to identify systematic differences between consenters and non-consenters by looking at the respondent's life course. Are respondents who have been unemployed frequently or for a longer time more probably non-consenters to the data linkage? What is more important for the consent: the actual situation or the life course experiences? In a multivariate analyze I will look at further interesting longitudinal variables which might answer this question.