statistical network model
Definition
Statistical network model refers to a class of probabilistic graphical models used to analyze multivariate psychological data by representing variables as nodes and their conditional statistical relationships as edges. The most common instantiation is the Gaussian Graphical Model, a parameterization of the Pairwise Markov Random Field in which edges reflect partial correlations between variables after conditioning on all others in the network. In empirical practice, researchers frequently interpret these estimated conditional dependencies as direct causal relationships between symptoms of mental disorder or other psychological variables, an approach that amounts to a form of causal discovery. Because network edges are strictly undirected, however, the models are likely to perform poorly in that role, since observed associations can be produced by conditioning on common effects and other inferential artifacts rather than by genuine direct causal links.
Sources: Park et al. (2024)
Related Terms
- constraint-based (1 shared article)
- cyclic causal discovery (1 shared article)
- directed cyclic graph (DCG) (1 shared article)
- partial ancestral graph (PAG) (1 shared article)
Applications
Statistical Network Model and Cyclic Causal Discovery
Statistical network models and cyclic causal discovery methods address overlapping questions about the structure of psychological phenomena, yet they do so through fundamentally different formalisms. Network models estimate undirected conditional associations, whereas cyclic causal discovery methods attempt to recover directed causal structures that explicitly permit feedback loops, such as the reciprocal symptom relationships theorized to sustain psychopathology. Applying constraint-based cyclic discovery algorithms to the same empirical data as a network analysis allows researchers to assess how well the undirected network edges correspond to the directed causal relations the models are often assumed to reflect.
Sources: Park et al. (2024)
Statistical Network Model and Causal Interpretation
A persistent issue in psychological network research is that parameters estimated from models such as the Gaussian Graphical Model are routinely given causal interpretations despite the models carrying no causal content by construction. Undirected edges in a Pairwise Markov Random Field indicate conditional statistical dependence, not intervention-based causation, and associations visible in the network may arise from conditioning on colliders or other structural features of the underlying causal graph rather than from direct causal links. This gap between statistical and causal graphical models motivates the development and evaluation of purpose-built causal discovery procedures for psychological data.
Sources: Park et al. (2024)
