binary_mediation how to calculate indirect effect of specific mediators rather than all of them?

[Emma Gearon]

Hi all,


I'm using the binary_mediation command to perform a product of coefficients mediation analysis with six mediating variables, two dichotomous and four continuous.
The command I'm using is...
bootstrap r(indir_1) r(indir_2) r(indir_3) r(indir_4) r(indir_5) r(indir_6) r(tot_ind) r(dir_eff) r(tot_eff), reps(5000): binary_mediation, dv(Y) iv(X) mv(m1 m2 m3 m4 m5 m6) cv(C1)
estat bootstrap, percentile

I'm very happy with interpreting the output, which shows me the observed coefficient and bootstrapped 95% confidence interval.

However, the total indirect effect (tot_ind) gives me the observed coefficient and bootstrapped 95% CI for the sum of all six mediating variables (m1+m2+m3+m4+m5+m6). Only three mediators had a significant a and b coefficient, and while I can manually calculate the sum of the coefficients, I would like to calculate the bootstrapped percentile 95% CI for the sum of these variables (m1+m2+m4) only.

Could anyone please advise if this is possible?

I attempted running...
bootstrap r(indir_1) r(indir_2) r(indir_3) r(indir_4) r(indir_5) r(indir_6) r(indir_1 + indir_2 + indir_4) r(tot_ind) r(dir_eff) r(tot_eff), reps(5000): binary_mediation,, dv(Y) iv(X) mv(m1 m2 m3 m4 m5 m6) cv(C1)

and got the error message
indir_1+indir_2+indir_4 invalid name
error in expression: r(indir_1 + indir_2 + indir_4)
r(198);

I also attempted running a predict command after my original bootstrap
. predict indir_1+indir_2+indir_4
and got the error message
predict is not allowed after bootstrap:binary_mediation

I am also unable to simply include the non-significant mediators as confounders, as stata will include them as confounders of all relationships tested, rather than just the b and c'.

Your thoughts would be greatly appreciated.

Kind regards,

Emma.

[Emma Gearon]

Hi all,

For those playing at home, the solution was to include an additional term in the bootstrap line, sum=(r(indir_1) + r(indir_2) + r(indir_4)), to yield the coefficient and 95% CI for the sum of those standardised coefficients.
So the final code is...

bootstrap r(indir_1) r(indir_2) r(indir_3) r(indir_4) r(indir_5) r(indir_6) r(tot_ind) r(dir_eff) r(tot_eff) sum=(r(indir_1) + r(indir_2) + r(indir_4)), reps(5000): binary_mediation, dv(Y) iv(X) mv(m1 m2 m3 m4 m5 m6) cv(C1)
estat bootstrap, percentile


Cheers,
Emma.

[Clyde Schechter]

I'm happy to see that you found a solution to your problem, and extra happy to see that you shared it with the rest of us.