Fig. 1: Measurement and structural models.

a EVS model 1, b EVS model 2.1, c EVS model 2.2. Standardized parameter values estimated by the structural equation models applied on the training dataset (90% of the full sample) and averaged on the 1000 iterations set for cross-validation. Ellipses represent latent variables, rectangles represent their indicators. In the WVS model, social mistrust is modeled as a single composite variable, here represented by a rectangle. Paths between the latent variables and/or the single composite represent regressions. Paths between the indicators and the latent variables represent factor loadings. Note that all indicators and the single composite variable “Social mistrust” are regressed against the covariate “age of the respondent”. Significant paths at the 5% level are represented with bold arrows. d Cross-validation. Distributions of predictive accuracy ratios obtained from the stratified tenfolds cross-validation of the EVS model 1, 2.1 and 2.2 (10*100 rounds in total for each model).