Proceedings of the American Statistical Association, Survey Research Methods Section, pp. 3562-3657.
ISSN/ISBN: Not available at this time. DOI: Not available at this time.
Abstract: Interviewer falsification is a constant worry in survey data collection. Traditionally, phone verification of a subsample of completed interviews is the primary method used for detection. However, it is slow and imperfect. Among the alternatives, we are exploring the use of Benford’s Law (the distribution of the first digits in a random set of variables) to detect anomalies. Theoretically, numerical data can be compared to Benford’s Law on a flow basis and suspicious interviews and interviewers can be flagged for investigation. To explore the effectiveness of using Benford’s Law, we used a data set with many financial (numerical) variables, SCF 2004. Knowing two falsifying interviewers detected through other methods, we examined SCF 2004 data near the end of the field period to see if Benford’s Law could be used to pick out these interviewers. Unfortunately, we found that even with this number-rich data, Benford’s Law could not pick out the falsified interviews. Also, they were not the top two, but they were near the top.
Bibtex:
@inproceedings{,
title={Detecting falsified cases in scf 2004 using Benford’s Law},
author={Wang, Yongyi and Pedlow, Steven},
booktitle={Proceedings of the American Statistical Association (Survey Research Methods Section)},
pages={3652--3657},
year={2005}
}
Reference Type: Conference Paper
Subject Area(s): Statistics