David, thank you for the very apt comments about the Heifer study. In doing this research, propensity score matching was literally our third or fourth choice for impact evaluation. Our original plan was to work with the organization to carry out an RCT, but logistics in the field prevented any kind of controlled approach. We then searched for some kind of IV or a rule where we could use a regression discontinuity, but obviously one needs (reasonably) firm allocation rules, and even after considerable effort, there just wasn't a clear rule we could find to get identification. This is one type of program that simply yearns to be evaluated with a nice, moderately sized RCT, and someone should take up the task because a lot of money is given to animal donation by everyday people who would be very interested in its impact. The main issue, however, is that the program isn't about just giving people animals; it's about the support staff that households need to manage a new animal, ag extension, insemination services, etc. So one can't just randomly allocate animals to households alone to evaluate animal donation programs. It's necessary to have the whole array of services that come with them, and for this you need to work closely with an NGO that is in this business. But we found it difficult to carry out randomization working with Heifer simply because it is hard in this particular case to build randomization into something resembling their normal operations, and too far a departure from their normal approach would render the study invalid as an evaluation of Heifer's program per se.