obtaining p values after KHB disentangle

[Joanna Davies]

Hello,

I'm using Stata MP/16.1 and the user written command khb

I have the following multiple mediation model

khb logit y i.x || i.m1 i.m2 m3, concomitant(age sex) summary disentangle verbose

Question 1: I want to obtain the p value for each of the indirect effects - but i cant work out how to do it...

From reading other posts on here, i think i need to extract the coef and se from the matrix for disentangle and apply the correct formula to get the p values - but im stuck and i cant work out how to do this. Would be very grateful if someone could point me in the right direction.

So far i have obtained the correct matrix using

matrix list e(disentangle)

But when i run the following i get nothing, it doesnt display the p value (not even an error msg), so not sure where im going wrong

local t =b[1,1/b[1,2]
di 2*ttail(e(df_r),abs(`t'))

Question 2: is it necessary to present the p value for each indirect effect or is it enough to show that the direct effect of the mediators on y is significant (or not)?

Thank you,
Joanna

[Joanna Davies]

Slept on this and realised the basic mistake i was making...i need z not t!

So, the solution to get the p values for anyone else struggling with this...

matrix list e(disentangle)
mat b = e(disentangle)
/*get z score*/
di b[1,1]/b[1,2] /*to divide coef by se, corresponds to position on matrix*/
/*get p value*/
di 2*normal(-abs(z)) /*NB i manually added z but could probably automate*/
/*get ci*/
di b[1,1] - invnormal(0.975)*b[1,2]
di b[1,1] + invnormal(0.975)*b[1,2]

all helpfully explained here: https://www.stata-journal.com/sjpdf....iclenum=st0137

[daniel klein]

The code in #2 does not actually define z. Here is a revised version that does

/*get z score*/
scalar z = e(disentangle)[1,1] / e(disentangle)[1,2]
/*get p value*/
di 2*normal(-abs(z))
/*get ci*/
di e(disentangle)[1,1] - invnormal(0.975)*e(disentangle)[1,2]
di e(disentangle)[1,1] + invnormal(0.975)*e(disentangle)[1,2]