Can anyone give some reference material that explains how to do a sensitivity analysis of unmeasured confounding between mediator-outcome and exposure-mediator relationships when the exposure is multicategorical(with 4 categories in my case)?
I am using the natural effect model from the medflex package to estimate the natural direct effect and natural indirect effect. I've tried reading some articles by Vanderweele explaining methods to do sensitivity analysis but I believe there's no (explicit) mention of exposure being multicategorical.
