Today’s Aid Watch post by Dennis Whittle raises a great question – shouldn’t we care about the distribution of aid impacts, and not just the average? – but then doesn’t seem to answer it at all. He talks about how aid proponents claim the distribution looks like one thing, and critics argue otherwise. But his distributions have different means! The plots he shows are just standard normal distribution-looking things with different expected values. So this is just a different way of talking about the same argument.
We can do better. Take a look at his first plot. Let’s assume this is the real distribution of impacts from some policy:
The average impact is 1. Why should we care about the low tail? After all, this is just random variation, right? No – in fact, policy impacts are often correlated with important stuff like income and education level. It’s not a stretch to assume we might be hurting the poorest and most vulnerable people in our sample.
But the problem is worse than that: even if this is random chance, we might be doing more harm than good. Suppose the effect of the policy is to change people’s incomes. It’s a standard to assume that the marginal utility of income is diminishing, and for poor people it can diminish really fast. Taking $30/month away from a rural Malawian could be devastating, and while giving them $30/month is good, it’s not as important as a loss that might mean life or death. The distribution is very important.