[Darina Peycheva]

Thank you so much! I thought of this yesterday but considered that [1,1] should be right. Thanks for the alternative coding of this. The -bootstrap- worked smoothly and I now have the coeffects.

An only thing that does not work is the following (but as long as the bootstrap estimations are displayed this is less important, I guess):
. scalar list b_total b_direct b_indirect
scalar b_indirect not found
r(111);

Why would _bs_1: r(b_indirect) will be displayed but not scalar b_indirect? Thanks once again for the enormous help!

[Clyde Schechter]

Why would _bs_1: r(b_indirect) will be displayed but not scalar b_indirect? Thanks once again for the enormous help!

Because the code never defines a scalar named b_indirect.

Review the code and you will see that there are explicit commands defining scalars b_total and b_direct:

scalar b_total = bb_total[1,1]
return scalar b_total = bb_total[1,1]

*matrix list r(table)

*display b_total
*scalar list

*Estimate NDE
glm early_menopause i.smokingm [pw=wt_iow], fam(binomial) link(log) vce(robust)  nolog base

matrix bb_direct=e(b)
*matrix list e(b)

scalar b_direct = bb_direct[1,1]

There is no similar command defining scalar b_indirect. What you do have for b_indirect is:

return scalar b_indirect = b_total-b_direct

But this command does not define a scalar. Rather it causes a number to be placed in r(b_indirect). But it does not create any scalar. The mention of the word -scalar- in that command simply tells Stata that what is to be placed in r() is a number, not a matrix, nor a macro.

[Darina Peycheva]

Yes, that's right. Thanks again, this was very helpful.