Differentiation is an important inference method in Bayesian networks and intervention is a basic notion in causal Bayesian networks. In this paper, we reveal the connection between differentiation and intervention in Bayesian networks. We first encode an intervention as changing a conditional probabilistic table into a partial intervention table. We next introduce a jointree algorithm to compute the full atomic interventions of all nodes with respect to evidence in a Bayesian network. We further discover that an intervention has differential semantics if the intervention variables can reach the evidence in Bayesian networks and the output of the state-of-the-art algorithm is not the differentiation but the intervention of a Bayesian network if the differential nodes cannot reach any one of the evidence nodes. Finally, we present experimental results to demonstrate the efficiency of our algorithm to infer the causal effect in Bayesian networks.
Schooling rewards people with labor market returns and nonpecuniary benefits in other realms of life. However, there is no experimental evidence showing that education interventions improve individual economic rationality. We examine this hypothesis by studying a randomized 1-year financial support program for education in Malawi that reduced absence and dropout rates and increased scores on a qualification exam of female secondary school students. We measure economic rationality 4 years after the intervention by using lab-in-the-field experiments to create scores of consistency with utility maximization that are derived from revealed preference theory. We find that students assigned to the intervention had higher scores of rationality.