Identifying Likely Duplicates by Record Linkage in a Survey of Prostitutes

• Activity Number: 478
• Type: Topic Contributed
• Date/Time: Thursday, August 11, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Thomas R. Belin*+ and Hemant Ishwaran and Naihua Duan and Sandra H. Berry and David E. Kanouse
• Companies: University of California, Los Angeles and Cleveland Clinic Foundation and University of California, Los Angeles and RAND Corporation and RAND Corporation
• Keywords: calibration ; mixture model ; HIV ; distance metric ; survey incentives

Abstract

Duplication is a concern in surveys offering an incentive. The Los Angeles Women's Health Risk Study was a survey of female prostitutes aiming to provide insight into the evolution of the AIDS epidemic, where fieldworkers assigned to a probability sample of areas sought interviews in a systematic fashion. The study produced 998 completed interviews, but on the basis of a code that distinguished individuals (including elements such as the first letter of the mother's maiden name), there were 55 individuals "known" to be interviewed more than once, raising concerns about further duplication in the sample. This talk describes a strategy for estimating the probability of duplication using a record-linkage calibration strategy (Belin and Rubin 1995). The method was applied to a distance metric for agreement among other items in the interview record and made use of differences in the distributions of distances between known duplicates and known nonduplicates. Duplicates were estimated to represent roughly 10-15% of the sample, enough to be a concern worth addressing but not so great to undermine the findings about condom use and HIV seroprevalence.