Pravděpodobnostní grafické modely pro analýzu sociologických dat

This thesis focuses on the application of advanced probabilistic graphical models for an in-depth analysis of complex sociological data. The analysis is implemented in the R statistical environment using modern methods from network psychometrics and machine learning. The input dataset originates from the Society of Distrust research project and contains categorized responses from nearly four thousand respondents to 121 questionnaire items. Using Pairwise Markov Random Fields (PMRF) and Bayesian Networks, hidden structural patterns can be identified in the data, separating direct associations from spurious correlations and exploring probabilistic flows of beliefs. The modeling process includes network regularization, robustness verification using the case-dropping bootstrap method, and predictive performance evaluation through cross-validation. The applied models proved to be highly effective, consistently identifying the individual's political and geopolitical anchoring and media habits—rather than isolated conspiracy narratives—as the core of the societal distrust ecosystem.