Online Program

Patterns of Consent: Linking Longitudinal Health Survey and Social Security Administration Records
Mick P. Couper, University of Michigan 
Mary Beth Ofstedal, University of Michigan 
*Joseph W. Sakshaug, University of Michigan 

Keywords: consent, data linkage, confidentiality, longitudinal survey, paradata

Longitudinal survey records are increasingly being linked to federal administrative records to enhance the quality and utility of the collected survey data. A prerequisite for linking survey and administrative records is to obtain informed consent and an id number used to facilitate the linkage (e.g., social security number) from respondents. However, this necessary step has the potential to introduce both random and systematic estimation errors if not all respondents consent or provide this information. Several factors have been shown to affect the propensity of data linkage consent, including demographics, health status, income, and mode of data collection (Olson, 1999; Jenkins et al., 2006; Marmot et al., 2003; Lessof et al., 2004).

We extend these prior studies by examining two additional mechanisms that may influence whether respondents consent to having their survey and administrative records linked. First, we hypothesize that interviewers (and their attributes) play a vital role in obtaining consent from respondents and may vary in how they present the consent request and address respondent concerns. Second, we expect that the main cause of non-consent is due to confidentiality and privacy concerns among respondents who are reluctant to provide a social security number to facilitate record matching. Although confidentiality and privacy concerns are not typically collected from survey respondents, this information is being collected in the form of interviewer observations and other paradata. Longitudinal surveys have an advantage over cross-sectional surveys in that a history of respondent behavior is collected over time. Interviewer observations and paradata from prior waves could potentially be used to preidentify respondents with low consent propensities, who may be eligible for intervention strategies that attempt to address respondent concerns.

We investigated these hypotheses by analyzing data from the 2004 Health and Retirement Study (HRS). HRS is a biennial longitudinal health survey that is representative of adults over the age of 50. In 2004, a consent request was presented to all respondents asking for permission to link their HRS record with their Social Security Administration record. A consent rate of 65% was achieved. A multilevel model was used to disentangle interviewer- and respondent-level associations with consent. In addition, we utilized prior wave information to assess the predictive capability of these measures on consent.