CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Vine copula based structural equation models Authors:  Claudia Czado - Technische Universitaet Muenchen (Germany) [presenting]
Abstract: While there is considerable effort to construct Bayesian networks from data, there is less emphasis on understanding and quantifying conditional distributions and associated quantities of nodes given their parents from the identified Bayesian network. Often Gaussian structural equation models are utilized, which might be too restrictive. A copula-based and thus non-Gaussian non-linear structural equation model for continuous data is proposed. It utilizes a previous approach based on D-vine copulas. It allows for easy fitting and estimation of conditional quantiles. It includes a forward selection algorithm to select the most important covariates. This will be used to identify edges which can be removed from a given network. This approach will be illustrated for an experimental setting.