Revisiting the relationship between targeting and program performance

An important public policy question is whether improved targeting of public spending will necessarily result in more or more cost-effective poverty reduction. In an important and influential study, Ravallion (2009) shows that targeting measures perform poorly as indicators of the poverty impact and cost- effectiveness of an urban cash transfer program across China’s provinces, and warns on this basis against reliance on assessments of targeting performance to inform policy choices in relation to poverty alleviation programs. The lack of a correlation between targeting and cost-effectiveness in reducing poverty is surprising, as it is unsupported by the theoretical literature and contradicts other empirical work. We suggest that this result arises because the comparison is across programs and income distributions. Through simulations, we confirm that the targeting and poverty performance of different programs, or even the same program, implemented across different distributions might be only weakly or even negatively correlated. But a more relevant, policy-oriented concern is to compare the performance of two or more programs in relation to the same distribution. In this case, we find, again through simulations, that the link between targeting and poverty impact depends on the changes made to the underlying targeting errors, but that there is a strong and positive correlation between targeting performance and cost-effectiveness. When the distribution of interest is different to the distributions in which the targeting performance is observed – that is, when assumptions of external validity are made – the correlation is weaker, but some targeting measures still perform well. Overall, targeting results need to be interpreted with care, but do provide useful information on the cost-effectiveness of transfer programs.