Interesting paper, but I have issues with some of the statistics--not helped by the large number of seemingly exploratory analyses that give the impression of a

Interesting paper, but I have issues with some of the statistics--not helped by the large number of seemingly exploratory analyses that give the impression of a garden of forking paths. 1) The authors go through the work of doing several simpler analyses which are obviated by the final multiple regression. The analyses (in Wilkinson notation) P&L[0] ~ HBD[1] and HBD ~ covariates only provide a subset of the information in the final P&L ~ HBD + covariates. The paper would improve if the preliminary analyses were replaced by a more thorough treatment of the multiple regression. 2) From what I gather, the maximal model containing HBD and the covariates was fit and then variables were subtracted successively based on whether doing so reduced the AIC. This does not seem like the right choice at all, and could possibly have influenced the final result. Selecting an optimal model is not the goal here, yet that is what this process is designed to do. The research question is about the influence of HBD on P&L, not predicting P&L from some optimal combination of variables. So, the theoretically important covariates should have remained when testing the significance of P&L. The point of the covariates is to "control" for confounding factors or "stress test" the importance of HBD in the presence of confounds, not to increase the predictive power of the model. 3) Trader and session length are actually random variables but their error is not modeled. Instead, the results from all sessions are averaged for each trader. Taking the average has the result of undercounting the variance and misspecifying the variance structure. Something like a multilevel model with crossed random effects should have been used instead. [0] "Average daily profit and loss" - measure of trading performance. [1] "Heartbeat detection score" - how well the trader self-assesses their heartbeat.