Table 1.
Variables used for the analysis.
Table 2.
Operationalisation of social trust, political trust, external and internal political efficacy.
Fig 1.
The step-wise description of the procedure in analysing the data.
Table 3.
The results of cross-validation using different BN structures.
Fig 2.
Results of the over-fit testing on the synthetic dataset ALARM.
Source: [63]. N = 20 000 observations. Notes: The two-step approach was used to balance the structures. The over-fitting was tested on the original dataset and on the same dataset with an increased percentage of noise (i.e., 0.5%, 1%, 2%, 5%, 10% and 15% of noise). Eight rows on the top show the accuracy of the models in identifying the correct arcs. Thus, longer bars display higher accuracy. On the bottom, the rows show the number of falsely identified arcs. The shorter the bar the better as it signifies lower over-fitting.
Fig 3.
Directed acyclic graphs of the relations between factors associated with participation in online activism.
Source: [35]. N = 27 379 individuals in 19 countries. Notes: Within Bayesian network analysis, score-based Tabu and hybrid H2PC algorithms were applied to analyse the data and learn the structure of the causal relations between the variables. Dashed blue lines represent false positives, i.e., edges that are not present in the structure learned by the Tabu algorithm but present in the structure learned by H2PC. Orange lines represent false negatives, i.e., edges that are present in the structure learned by the Tabu algorithm but absent in the structure learned by H2PC. The direction of each arc represents the orientation of the causality between two variables: e.g., in a relationship A → B, A is a parent and B is its child. All the edges from the other nodes to “Age”, “Gender” and “Born in the country” were blacklisted prior to learning the structure. No other edges were blacklisted. In the figure, those nodes that can only be parents have a darker blue color. The node “Country” (i.e., the country of the respondent’s residency) is present in the structure but not depicted by the figure to facilitate the apprehension of the relations between the nodes of interest. All variables are individual-level variables.
Fig 4.
Directed acyclic graph of the relations between factors associated with participation in online activism.
Source: ESS 2018 [35]. N = 27 379 individuals in 19 countries. Notes: Structural equation modeling was applied to analyse the data. Entities depicted in association with the edges are parameter estimates of the structural equation modeling. Sign.: *p < 0.05; **p < 0.01; ***p < 0.001. All variables are individual level variables.
Fig 5.
Probability distribution table of participation in online activism.
Source: ESS 2018 [35]. N = 27 379 individuals in 19 countries. Notes: Bayesian parameter estimation, conditional on the acquired structure of the network, was applied to analyse the data. Entities are the probability of participation in online activism in percentage.
Fig 6.
Probability distribution of all factors associated with participation in online activism.
Source: ESS 2018 [35]. N = 27 379 individuals in 19 countries. Notes: Bayesian parameter estimation, conditional on the acquired structure of the network, was applied to analyse the data. Entities are the probabilities of events in percentage.