Simulation and big data: in search of causality in big data-related managial decision making
Other
Overview
Overview
abstract
The unprecedented availability of digitized human behavioral data offers new research opportunities for discovering hidden patterns in Big Data that may not be apparent in smaller samples. At the same time, there are potential pitfalls associated with Big Data analytics in the absence of also working to identify causal relationships among the constructs thought to be involved. Indeed, despite the seemingly advanced modeling techniques applied to the analysis of Big Data, they are not well suited to addressing issues of causality. We illustrate the potential issues involved, using the context of human resources selection, in which the relationship between résumé typos and future job performance is of interest. Specifically, using computer simulation methodology, we demonstrate that including résumé typos along with the personality trait of conscientiousness to predict performance is likely to result in adverse impact on job applicants based on their country of birth, without significantly improving prediction. This outcome would leave the employer open to equal employment opportunity lawsuits and raise ethical concerns. In all, we suggest guidelines in which the analytical approaches typically used in the analysis of Big Data be supplemented with experimental and/or statistical approaches better suited to identification of causal relationships. Valuation Insight: This paper illustrates the potential pitfalls of Big Data Analytics via a simulation example of using the incidence of typos in applicant résumés as a criterion in hiring decisions. The answer is that, whereas conscientiousness adds positive value to the corporation, there is no evidence that fewer résumé errors add positive value through better employee performance. In addition, there is an ethical issue in employing the résumé error criterion directly, making it essential to qualify the approach so that transparent models are employed that identify causality.
