cyclic causal discovery
Definition
Cyclic causal discovery refers to a class of methods for inferring causal structures that include feedback loops and reciprocal relationships, where variable X may cause variable Y and Y may simultaneously cause X. Standard causal discovery approaches assume acyclicity, making them unsuitable for psychological phenomena such as psychopathologies, which are theoretically characterized by cycles among symptoms, as in the feedback loop perceived stress, negative affect, rumination, perceived stress proposed for depression. Constraint-based cyclic causal discovery methods work by using patterns of statistical dependence estimated from observational data to infer these cyclic structures, and their output can be interpreted as reflecting causal relations between equilibrium states of a dynamic system rather than moment-to-moment temporal sequences. Simulation results show that an autoregressive-based method outperforms FCI-variants and latent-variable-based methods in recovering cyclic causal structures from psychologically plausible data, particularly under varying sample sizes, network densities, and the presence of unobserved confounders.
Sources: Park et al. (2024)
Related Terms
- constraint-based (1 shared article)
- directed cyclic graph (DCG) (1 shared article)
- partial ancestral graph (PAG) (1 shared article)
- statistical network model (1 shared article)
Applications
Cyclic Causal Discovery and Statistical Network Models
Statistical network models such as the Gaussian Graphical Model are commonly used in psychology to analyze multivariate observational data, and researchers frequently interpret their parameters as reflecting direct causal relationships between symptoms. This practice amounts to an informal form of causal discovery, yet undirected network models are poorly suited to that task, since edges in a Pairwise Markov Random Field encode conditional statistical associations rather than directed causal effects and cannot represent the feedback loops that cyclic causal models are designed to capture.
Sources: Park et al. (2024)
Cyclic Causal Discovery and Psychopathology
Network theory in psychopathology holds that mental disorder arises from direct causal interactions among symptoms, and several theoretical accounts of conditions such as depression explicitly posit cyclic feedback among symptom processes. Because acyclic causal models cannot represent these feedback structures, cyclic causal discovery methods are necessary for empirically evaluating such theories from observational data.
Sources: Park et al. (2024)
Cyclic Causal Discovery and Structural Equation Modeling
In structural equation modeling, all acyclic path models with independent error terms and no latent variables are statistically identified, but path models containing cycles are not identified in general. This identification problem is one reason cyclic causal models have been less studied than their acyclic counterparts, and it adds practical difficulty to both fitting and interpreting the output of cyclic causal discovery algorithms.
Sources: Park et al. (2024)
