When does social desirability become a problem? Detection and reduction of social desirability bias in information systems research

Abstract

Social desirability (SD) bias occurs in self-report surveys when subjects give socially desirable responses by over- or underreporting their behavior. Despite knowledge of SD as a potential threat to the validity of information systems (IS) research, little has been done to systematically assess its extent. Furthermore, we are uncertain of how to recover reliable estimates of the relationships between research variables contaminated by SD bias. We sought in this study to assess the extent of SD bias in causal inferences when independent and/or dependent variables are contaminated. We also evaluated whether an SD scale in conjunction with partial correlation could effectively and efficiently correct SD bias when it is found. To achieve these purposes, we designed a survey study and collected data from Amazon's Mechanical Turk in the context of mobile loafing, which refers to employees’ personal use of the mobile Internet during business hours. Using various detection methods, we found that SD bias existed in the context of mobile loafing. From the results of the variance reduction rate and a covariate technique, we found that SD bias becomes problematic when both the independent and dependent variables are susceptible to SD bias. Overall, our study contributes significantly to the IS literature by revealing the extent of SD bias and the magnitude of the possible correction for it in IS research.