Validating Sensitive Questions: A Comparison of Survey and Register Data
Journal of Official Statistics, 31 (1):
31--59 (2015)DOI: 10.1515/jos-2015-0002
Abstract
This article explores the randomized response technique (RRT) - to be specific, a symmetric forced-choice implementation - as a means of improving the quality of survey data collected on receipt of basic income support. Because the sampled persons in this study were selected from administrative records, the proportion of respondents who have received transfer payments for basic income support, and thus the proportion of respondents who should have reported receipt is known.
The article addresses two research questions: First, it assesses whether the proportion of socially undesirable responses (indication of receipt of transfer payments) can be increased by applying the RRT. Estimates obtained in the RRT condition are compared to those from direct questioning, as well as to the known true prevalence. Such administrative record data are rare in the literature on sensitive questions and provide a unique opportunity to evaluate the ‘more-is-better’ assumption. Second, using multivariate analyses, mechanisms contributing to response accuracy are analyzed for one of the subsamples.
The main results can be summarized as follows: reporting accuracy of welfare benefit receipt cannot be increased using this particular variant of the RRT. Further, there is only weak evidence that the RRT elicits more accurate information compared to direct questioning in specific subpopulations.