Spatial Dependence and Heterogeneity in Bayesian Factor Analysis: A Cross-National Investigation of Schwartz Values

Area of Science:

• Statistics
• Spatial Analysis
• Psychology

Background:

• Traditional factor analysis models often overlook spatial relationships in data.
• Geographically distributed data may exhibit spatial autocorrelation and heterogeneity, violating model assumptions.
• Existing methods may not adequately capture complex spatial dependencies in latent variable structures.

Purpose of the Study:

• To present a novel Bayesian spatial factor analysis model.
• To extend confirmatory factor analysis by incorporating geographically distributed latent variables.
• To account for heterogeneity and spatial autocorrelation in statistical modeling.

Main Methods:

• Developed a Bayesian spatial factor analysis model.
• Incorporated geographically distributed latent variables.
• Accounted for heterogeneity and spatial autocorrelation.
• Conducted simulation studies to assess parameter recovery and the impact of ignoring spatial dependence.
• Applied the model to Schwartz value priority data from 5 European countries.

Main Results:

• Simulation studies demonstrated excellent recovery of model parameters.
• Ignoring spatial dependence resulted in biased and inefficient estimates of factor score means and covariance matrices.
• Schwartz value types (Conformity, Tradition, Benevolence, Hedonism) exhibited significant spatial autocorrelation.
• Identified distinct spatial patterns: country-specific structures for Conformity and Hedonism, a North-South gradient for Tradition, and a South-North gradient for Benevolence.

Conclusions:

• The proposed Bayesian spatial factor analysis model effectively captures spatial dependencies in latent variable structures.
• Ignoring spatial autocorrelation can lead to substantial biases and inefficiencies in statistical inference.
• The model reveals significant spatial patterns in Schwartz values across European countries.
• Conventional factor analysis may obscure valuable insights present in spatially structured data.