A stochastic dominance approach to program evaluation with an application to child nutritional status in Kenya

Most existing program evaluation methods examine the average impact of a program. This necessarily overlooks the potential for different program impacts over different parts of the distribution of the variable of interest. To overcome this limitation, we develop a novel methodology for program evaluation which combines stochastic dominance with difference‐in‐differences methods. We use this new method to evaluate the impact of a large decentralization program in Kenya on changes in child nutritional status, where one's primary concern is about sharp adverse (i.e., negative) changes. Using standard difference‐in‐differences regression as a baseline, we find , no statistically or practically significant mean impact. In contrast, our stochastic dominance estimations reveal that project expenditures have had different impacts on different parts of the distribution. In particular, they are associated with less deterioration in children's nutritional status among the worst‐off children, indicating that the program effectively functions as a nutritional safety net.