We've been reading a new summary of the literature on microcredit impacts. The paper reviews technical issues using technical language, so it's not the paper I'd read first as an introduction unless you're doing a PhD in Economics or something similar.
The paper covers terrain familar from The Economics of Microfinance, 2nd edition, but offers an independent review (with a couple of helpful summary tables at the end). The paper comes to similar conclusions as Armendariz-Morduch, so there will be no surprises if you've read the book. If you haven't read the book, the paper offers a smart synthesis.
Here's my summary of the state of play: After 30 years of microcredit and the rise of randomized trials, we still don't have an impact evaluation that is ideal yet, but we're getting closer. One big lesson on which we can all agree (or should all agree) is that flawed evaluations can be seriously misleading (due to self-selection, attrition, and non-comparable control groups). So getting the details right matters.
In the end, the big question is not whether access to microfinance will likely yield miraculous transformations for poor families. It won't likely do that. The more interesting question is whether it will yield benefits that exceed the costs of donors. Since those costs can be relative modest, even modest benefits can lead to a winning formula. A second question is who will benefit most? The average impact for the population is a starting point, but in practice it's a less interesting result than the impact for particular sub-samples. A third question is whether adding (or subtracting) services can improve benefit-cost ratios. Will, say, a bit of training help? Simple health interventions?
What we need is more research. Of course.