Graphical Statistical Methods for Studying Causal Effects. Bayesian Networks

Revival of interest to statistical causality theory from the beginning of the 1990s was brought about, in our opinion, by two reasons. First, realization of the fact that Markovian properties allow, though partial, statistical testing of causal relations a priori determined by a researcher. Second, introduction to statistics of operator "do" by Judea Pearl (Pearl, 1995). The latter contributed to a great extent to basing ausality theory upon formal probability theory and understanding causal effects as external interference in data generating process. The paper has three major objectives: 1) to present, on a sufficiently rigorous mathematical level, basic concepts and ideas of modern statistical causality theory based on graphical representation of Bayesian networks; 2) to demonstrate how graphical methods of statistical causality theory can be applied to economics and economic policy by means of "do" operator; 3) to show how those methods may be effectively used for determining indirect causes of economic factors. For this purpose we developed a new method of representing the Bayesian graph as a sequence of layers about the factor under consideration.