Psychometric Network Analysis

Description

Estimating a network in which questionnaire items are nodes and edges are partial correlations, to see how individual beliefs, feelings and behaviours reinforce each other instead of assuming they reflect one latent factor.

Survey analysis in UX usually assumes that items reflect a common latent cause, such as satisfaction or trust, and sums them. Network psychometrics treats the items themselves as interacting components. A Gaussian graphical model (or an Ising model for binary items, or a mixed graphical model for mixed data) estimates conditional associations between every pair of items after controlling for all others, usually with regularisation to remove spurious edges. The network is visualised, central items are identified with measures such as strength, betweenness or bridge centrality, communities of items are detected, and networks from different groups are compared with permutation tests or bootstrapping. The book stresses the limits: cross-sectional networks are descriptive, not causal; missing nodes distort results; self-report dominates; and replicability needs stability checks. For UX researchers the method shows which perceptions are hubs (for example whether 'feels in control' links usability and trust items), how attitude structures differ between segments, and which components might be worth targeting.

Attributes

Loss type
Siloed
Effort
Days

Steps

  1. Select items that each measure a distinct belief, feeling or behaviour and avoid near-duplicate wording that creates trivial edges.
  2. Collect a large sample and choose the estimator matching the data type (Gaussian graphical model, Ising model or mixed graphical model).
  3. Estimate a regularised network and visualise it.
  4. Compute centrality and detect communities of items.
  5. Check edge and centrality stability with bootstrapping before interpreting.
  6. Compare networks across segments or time with a network comparison test and report the findings as associations, not causes.

Sources

  • van der Maas, H., Complex Systems Research in Psychology (2024)

Also known as

  • Network psychometrics
  • Gaussian graphical model
  • Partial correlation network
  • Item network analysis

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics