Discovering Cyclic Causal Models in Psychological Research

Statistical network models have become popular tools for analyzing multivariatepsychological data. In empirical practice, network parameters are often interpretedas reflecting causal relationships – an approach that can be characterizedas a form of causal discovery. Recent research has shown that undirected networkmodels are likely to perform poorly as causal discovery tools in the contextof discovering acyclic causal structures, a task for which many alternative methodsare available. However, acyclic causal models are likely unsuitable for manypsychological phenomena, such as psychopathologies, which are often characterizedby cycles or feedback loop relationships between symptoms. A numberof cyclic causal discovery methods have been developed, largely in the computerscience literature, but they are not as well studied or widely applied inempirical practice. In this paper, we provide an accessible introduction to thebasics of cyclic causal discovery for empirical researchers. We examine threedifferent cyclic causal discovery methods and investigate their performance intypical psychological research contexts by means of a simulation study. We alsodemonstrate the practical applicability of these methods using an empirical exampleand conclude the paper with a discussion of how the insights we gainfrom cyclic causal discovery relate to statistical network analysis.