So many ways for assessing outliers: What really works and does it matter?

Recent research in leading business journals has varied widely in how statistical outliers are identified and handled; many techniques were reported. But most articles with empirical data have not mentioned outliers; many others simply referred to their removal without details. This wide variety of methods and frequent non-disclosure of methods cannot represent the best practice. Proper outlier identification and handling are important issues for any research community that performs quantitative research based on empirical data. We document this diversity of methods by examining articles published in the UTD-24 business journals during a 12-year period, and we explain why almost all methods described in these articles are ill-advised. To achieve an effective assessment of outliers, we propose a process of outlier identification based on testing hypotheses using a controlled significance level. The implementation of our suggested method is feasible using commonly available statistical analysis software.