3 thoughts on “Links I liked

  1. Is it possible for any normal person to understand number 2?

    Is it trying to say how to design and implement an experiment to determine whether giving girls some cash will actually cause their friends (not them) to not get married or pregnant?

    Does that have enough plausibility for such an experiment over other things that might need testing? For example, how to protect the long term psychosocial well being of children enduring crises with clowns. I ask genuinely. I know nothing about the efficacy of clowns nor about the effect of girl’s income on their friends’ behaviour. If you do it would be nice to hear; you’re pretty good at explaining things to the non-economist.

  2. Understanding spillovers and having a sophisticated design where they are present is hugely important for policy.

    For instance, in principle the main impact of a vaccine or an employment program is to have effects on people who are not directly treated. If a vaccine interrupts a disease vector this is crucial to know because it affects the cost-benefit analysis compared to other interventions. An economic development program such as cash transfer could stimulate employment among others (e.g. if businesses grow and hire others, or if there is a positive demand shock in the economy) or it could have negative effects (such as crowding other people out of the same activity). This is hugely important to know from a policy perspective.

    On the other hand, testing the effect of clowns is a relatively pointless exercise. Randomized evaluations are useful for testing our understanding of the world–for re-evaluating theory. Testing the efficacy of a particular program may or may not accomplish this goal. Many program evaluations do not. If they were cheap and easy, then that would be fine. But they are so expensive and disruptive we probably need to use them selectively, to the questions with the most uncertainty, in ways that provide the most generalizabilty.