Detecting Possibly Fraudulent or Error‐Prone Survey Data Using Benford’s Law

David Swanson, MoonJung Cho, and John L. Eltinge

Abstract

The quality of any survey's results depends on the ability to collect data that accurately represent the underlying phenomena of interest. In some interview surveys two potential problems with data collection are inaccurate reporting by the respondent, and fabrication of data by a data collector who does not contact the selected sample unit (a process called curbstoning). For each of these cases, one may be able to identify problematic interviews by evaluating the distribution of the leading digits of the responses to the questionnaire. In the aggregate, the distribution tends to follow a pattern known as Benford's Law. Consequently, it may be appropriate to re-interview the cases that display a markedly different distribution of leading digits. This paper describes a potential application of this idea to the Consumer Expenditure Interview Survey.