Detecting computer-generated random responding in online questionnaires: An extension of Dupuis, Meier & Cuneo (2019) on dichotomous data. (15th April 2020)

• Record Type: Journal Article
• Authors: Dupuis, Marc; Meier, Emanuele; Gholam-Rezaee, Mehdi; Gmel, Gerhard; Strippoli, Marie-Pierre F.; Renaud, Olivier
• Is Part Of: Personality and individual differences. Volume 157(2020)
• Publication Date: 2020-04-15
• Subjects: Botnet -- Dichotomous data -- Functional method -- Mahalanobis distance -- Person-total correlation -- Random responding -- Response coherence; Personality -- Periodicals; Individuality -- Periodicals; Individuality -- Periodicals; Personality Development -- Periodicals; Personnalité -- Périodiques; Individualité -- Périodiques; 155.205
• DOI: 10.1016/j.paid.2020.109812
• Languages: English
• ISSNs: 0191-8869

Abstract

Highlights: Computer-generated fraudulent data represents a major threat for online survey research. Several indices are relevant to detect randomly distributed sets of responses. Research so far focused on computer-generated data using Likert-type scales. The Mahalanobis distance, and the person-total correlation are the most useful indices. Abstract: Some authors recently underlined the existence of programs generating invalid responses in online surveys as an emerging threat for the different crowdsourced research fields (e.g., botnets, form fillers or survey bots). Accordingly, online data research might include computer-generated sets of responses representing invalid data at risk of largely distorting study results. Several statistical indices exist in order to detect problematic data. In line with a previous study that compared these indices in Likert-type scale questionnaire data, this study purported to extend the analyses with dichotomous-itemed questionnaires. Three samples of about more than 2, 000 participants were mixed with different proportions (i.e., 5% to 50%) of simulated data to mimic their effect. Then, seven indices were compared in terms of correct detections of non-human response sets. Consistent with former findings, three indices resulted in superior correct detection rates: response coherence, the Mahalanobis distance and the person-total correlation. Two of them can easily be computed using basic statistical software. The current study findings represent an encouragement to use them in priority as routine for data screening.