Psychometric network analysis draws a Gaussian graphical model whose edges are significance-thresholded partial correlations, laid out as a circular network with centrality indices.
A partial-correlation network shows which variable pairs remain associated after conditioning on every other variable — marginal correlations that vanish under conditioning draw no edge. Two estimators: EBICglasso (default) runs the graphical lasso over a λ path and picks the model by EBIC (γ tunes the sparsity preference), shrinking small edges to exactly zero; the pcor alternative computes unregularized partial correlations from the inverse correlation matrix, t-tests each (df = n − p), and keeps edges surviving Benjamini-Hochberg FDR at the chosen α.
Green edges are positive partial correlations, red negative; width tracks magnitude.
Centrality: strength (Σ|weights| at a node) and expected influence (signed sum — better when negative edges exist). Closeness/betweenness are omitted deliberately; the field increasingly discourages them for psychological networks.
≥ 3 numeric columns + estimator (EBICglasso with γ, or pcor with FDR α).
Circular network plot, edge count / sparsity, strength + expected-influence centrality table.
High-strength nodes are the most connected; an edge is a conditional association, not causation. Networks are unstable in small samples — replicate before interpreting fine structure.