Nick,
Thanks for this interesting example. When you mention marital status and relationships as possible mediators, do you mean mediators or modifiers? (or both?)
A mediator is something that is altered by the exposure and in turn alters the outcome, so marital status would be a mediator if receiving the lone influenced a woman's marital status and her marital status in turn influenced her health. A modifier is something that alters the relationship between the exposure and the outcome, so for example the loans might be more important for unmarried women than married women.
What are the most rigorous evaluations of microfinance that you have seen? Could you use a randomized wait list design, which can sometimes be a very rigorous model?
Maria