Conference Programme 2015
Tuesday 14th July Wednesday 15th July Thursday 16th July Friday 17th July
Thursday 16th July, 16:00 - 17:30 Room: HT-103
Methodological issues of using administrative data to improve the quality of survey data 2
|Convenor||Dr Emanuela Sala (Dipartimento di Sociologia e Ricerca Sociale, Università di MIlano Bicocca )|
|Coordinator 1||Dr Jonathan Burton (ISER, University of Essex)|
|Coordinator 2||Dr Gundi Knies (ISER, University of Essex)|
Session DetailsThere is a body of research on eliciting respondents' consent to link their survey data to their administrative records. However, the process of linking survey and administrative data is complex and the issues that survey practitioners need to tackle go beyond the asking for respondents' consent. Furthermore, using that linked data also provides potential for further methodological research. The aim of this session is to foster discussion on i) methodological issues that concern the data linkage process, ii) the research potential of the linked data,
iii) use of administrative data to improve the quality of survey data
We welcome papers on the following topics:
1. Analysis of mis-reporting
2. Measurement 'bias' using linked data
3. Impact on research findings compared to using survey data only
4. Validation studies
5. Methodological lessons from linkage process
6. Implications of consent bias
Relevant papers on other aspects of data linkage will also be considered.
Paper Details1. Using Register Data in Income Statistics in the Austrian EU-SILC: (Why) Do People Get Poorer?
Mr Richard Heuberger (Statistics Austria)
Ms Nadja Lamei (Statistics Austria)
Mr Stefan Angel (WU Wien)
We take advantage of the fact that for EU-SILC 2011 in Austria comparable income data from register and from questionnaires are available. Our analysis shows that 1) the collection of income is more complete with register data and 2) the distribution for the majority of income types becomes more unequal as compared to questionnaire data. After a description of households whose poverty status changes if register data is used, explanatory factors will be analysed. A focus on explanations for the increase is provided. Moreover, the consequences of using register data for the calculation of income and weights are examined.
2. Determining recall errors in retrospective life course data – an approach using linked survey and administrative data
Ms Stefanie Unger (Institute for Employment Research)
Dr Britta Matthes (Institute for Employment Research)
Because of the imperfections in autobiographical recall, there is often skepticism about the usefulness of retrospective data. One way of analyzing retrospective recall errors is using linked survey and administrative data. However, the reliability of administrative data also has been distrusted. By using the linked data-set ALWA-ADIAB (which combines interview data and administrative data from the same individuals) we analyze recall errors of retrospective life course data considering the “notification error” – the error which can be ascribed to the administrative notification process. We show that the extent of the “notification error” is large and cannot remain unconsidered.
3. Integration of Administrative Data in the Survey of Income and Program Participation
Dr Jason Fields (US Census Bureau)
Dr Martha Stinson (US Census Bureau)
Dr Gary Benedetto (US Census Bureau)
This presentation highlights the uses and challenges of integrating administrative data in the Survey of Income and Program Participation (SIPP). We discuss four areas where administrative records are being used with household survey processes and products. These include imputation, data synthesis, data validation, and evaluation of bias. This work represents some of the important areas of innovation and are key areas of the SIPP program’s future. We discuss several of the challenges associated with integrating administrative and survey data systems, producing linked data resources, providing public access opportunities, and protecting respondent confidentiality.
4. Use of Medicare linkage data to study nonresponse bias in the distribution of physical measure and biomarker data: Examination of the Health and Retirement Study
Dr Jessica Faul (Survey Research Center, University of Michigan)
Dr Mary Beth Ofstedal (Survey Research Center, University of Michigan)
Increasingly studies are incorporating biological data to help elucidate mechanisms linking social factors and health. However, differences in obtaining consent to collect these data can affect representativeness. Administrative data sources can be used to identify characteristics associated with nonresponse that are otherwise unobserved in a panel survey. We use administrative health records linked to 20,000+ Health and Retirement Study respondents to investigate potential nonresponse bias in measured biological data. We examine whether the current nonresponse propensity adjustment method used to create subsample weights sufficiently adjusts for potential bias once characteristics beyond those collected in the panel study are considered.
5. Consent to Data Linkage in Business Surveys: Correlates and linked dataset representativeness.
Dr Jamie Moore (Administrative Data Research Centre for England and Department of Social Statistics and Demography)
Dr Gabriele Durrant (Administrative Data Research Centre for England and Department of Social Statistics and Demography)
Professor Peter Smith (Administrative Data Research Centre for England and Department of Social Statistics and Demography)
The utility of research utilising data linkage depends on linked dataset representativeness: inferences may be biased if population subgroup relative frequencies differ from those in the parent population. Non-representativeness can occur, for example, because respondent propensities to consent to linkage vary. We study such questions in four business surveys. To quantify representativeness and estimate impacts of consent propensity variation associated with different business attribute variables, we utilise for the first time methods developed to investigate survey non-response bias (R indicators, Coefficients of Variation, Dissimilarity indices). We describe our findings considering data collection strategies, and also evaluate differences observed