This article lacked the power to tell if the effect (ie, different language in reviews) was correlated to their gender or because of a selection bias in the peo

This article lacked the power to tell if the effect (ie, different language in reviews) was correlated to their gender or because of a selection bias in the people studied, eg because of a possible difference in the distribution of personality types in the field from each gender. It's possible that there's a hiring bias, and reviews are being done fairly, but show biased results when naively studied because of the underlying bias. This is what makes social statistics hard, and there are many faults in studies involving gender that fail to account for possible confounding effects in the data.