This talk presents combinatorial conditions for identifiability of linear structural equation models. These models relate random variables of interest via a linear equation system and can be represented by a graph with two types of edges that correspond to non-zero coefficients in the linear equations and correlations among noise terms, respectively. Identifiability holds if the coefficients and correlations associated with the edges of the graph can be uniquely recovered from the covariance matrix they define.
(Joint work with Jan Draisma and Rina Foygel)
