>>11662901man, if anything, you sound too ambitious
if I was you I would have done
logit occ_outcome i.parent_edu##i.homework_knowledge i.parent_edu##i.uni_desire
i.parent_edu##i.nurture_knowledge i.parent_occ##i.homework_knowledge i.parent_occ##i.uni_desire
i.parent_occ##i.nurture_knowledge
as my final model, with adding interactions one after the other and testing via lrtest to see if the model fits better. so you start with
logit occ_status i.parent_edu i.parent_occ
est store A
then you go
logit occ_status i.parent_edu##homework_knowledge i.parent_occ
est store B
lrtest A B
if you have a better fit (so p<0.05), then you add the next interaction and so on. Now, this is not the best way by far, but it is quite simple and might give you some results. why don't you run it in this way just for the education interactions (mediation) and see if you have better fit with each subsequent interaction? it will take you 5 mins, let me know what happened
