Fraud by interviewers is a ubiquitous threat to data quality in survey practice, whenever face-to-face surveys are conducted. Particularly if interviewers use stereotypes about respondents to fill in questionnaires, falsifications can limit the variety of possible answers, lead erroneously to significant correlations and distort survey results. In addition to external control mechanisms to detect fraud (such as postcards or time stamps) more recent research has started to also consider internal indicators (such as the number of missing values or open answers) as a monitoring strategy. This latter approach relies on ex-post statistical analyses and implicitly assumes that falsifiers apply rational behavioral strategies which result in detectable response patterns. This study examines to what extent fieldwork monitoring can benefit from such approaches, by empirically assessing how effective different indicators are at detecting known cases of fabrication. In contrast to most previous research, which often relies on laboratory fabrications, this study uses authentic cases of detected interviewer fraud from a survey on the fairness of earnings conducted in Germany. The main goal of this study is to examine to what extent the falsifiers’ attempts to produce unsuspicious data led to recognizable response patterns. For this purpose, we test a wide range of indicators that could potentially identify falsifications: avoidance of extreme categories and open text-based answers, low rates of item-nonresponse, strategic use of filter questions to shorten the questionnaire and non-compliance of responses to numeric questions with Benford‘s Law. Furthermore, we compare authentic and fabricated interviews according to their values on a social desirability scale and report results from an innovative trick question that was especially designed to detect falsifiers.

methods, data, analyses, 15(2), 125-160