On the Statistical Properties of the Dankel-Loenneke Method

Abstract

Dankel & Loenneke (2019) recently presented a new approach to identifying subgroups in parallel group study designs. Here, we briefly discuss our statistical concerns with proposed approach. We reveal that the error rates of the Danke-Loenneke approach are much higher than the claimed 5%, and that these error rates are dependent on numerous factors, including sample size, effect variance, and random error. The Dankel-Loenneke method has poor statistical properties; as such, we suggest that the method not be used and the manuscript constitutes an "honest error" per the Committee on Publication Ethics (COPE) guidelines.