Sensitivity analysis in causal mediation using the mediate command

[Annet Nassolo]

I am currently conducting causal mediation analysis using the new mediate command in Stata 18. My outcome model is linear regression and the mediator model is logistic regression. I have been able to obtain the total, direct and indirect effects of my exposure on the outcome. I would now like to check assumptions under which my estimates would have a causal interpretation. As a way of conducting sensitivity analysis for unmeasured mediator-outcome confounding, I have come across mediation E values explained by Louisa H. Smith and Tyler J. VanderWeele. I am wondering how to obtain the mediation E values in Stata. My mediation command is below.

mediate (outcome age i.sex) ///
(binary_mediator age i.sex, logit) ///
(study), ///
con(2)

[Erik Ruzek]

I am not familiar with Stata's new mediate command. You may want to look at Hicks' and Tingly's medeff and medsens commands, detailed in the Stata Journal. The medsens command in particular conducts sensitivity analyses for mediation, but not necessarily E values. Linden et al's (2020) evalue command does this, but it does so for single equation models, I believe. It was also detailed in a Stata Journal article.