Speeding problem detection in business surveys: benefits of statistical outlier detection methods

Abstract

Speeding describes the unusually fast responses provided to survey questions. A characteristic of speeders is that answers by-pass cognitive process. Consequently, this low respondent engagement results in the poor quality and validity of data.

The issue at hand is how to detect speeders in a survey. The presumption is the use of different statistical outlier detection methods. This paper presents graphical methods for outlier detection, such as: dot-plot diagrams, scatter diagrams, histograms and box-plot diagrams. Furthermore, the quantitative methods for outlier detection in this paper are the z-score, modified z-score, Dixons’ test, Grubbs’ test, Tietjen-Moore test, Rosners’ or the generalized extreme studentized deviate (ESD) test. The performance of these outlier detection methods was observed on completion times of 217 surveys from enterprises which participated in a web survey on the use of statistical methods, and which use them in their business processes.

The analysis has shown that none of the observed outlier detection methods were able to detect speeders in an appropriate and satisfactory way as shown by the threshold. The main reasons for this can be found in slowers, the violations of assumptions on normal distribution and in masking. Hence, existing outlier detection methods should be improved and adjusted in future research in order to detect speeders. The introduction of novel speeders detection methods would be a good choice for future research.