Dear all,

I run a linear model to test for a mediation (outcome and mediator are both metric, the treatment is binary). This works fine but I got the comment that I should run a sensitivity check on this analysis. While I am aware of several ados that can do that in cross-sectional models (medeff), I could not figure out how to do this in a longitudinal model. Can someone give advice? My model basically looks like the following example:

Testing whether the effect of union membership on wages is mediated by working time
Union Membership --> Working time--> Wage

Code:
clear all
version 16

webuse nlswork
xtset idcode year

drop if missing(ln_wage, union, union, wks_work)
xtreg ln_wage union                //M1 without the mediator
local coef_first = _b[union]
xtreg ln_wage union wks_work    //M2 including the mediator
local coef_second = _b[union]

*Share mediated:
display 1 - (`coef_second' / `coef_first')
How can I test sensitivity here?