Assessing Latent Structure in Questionnaire Data: an integrated approach between Consensus Bayesian Networks and Community detection

Assessing the dimensionality of questionnaire data is a well-established research task in the literature, essential to ensure the validity and reliability of measurement instruments. This study integrates Bayesian networks (BNs) with community detection techniques to achieve two primary objectives: first, to use the graphical structure arising from a consensus BN as a starting point for community detection, inferring the number of latent dimensions (the number of communities) and the association of items with the underlying constructs; second, to leverage BNs as a probabilistic tool for exploring what-if scenarios via their inference engine. Through simulation studies, we test the effectiveness of this proposal and demonstrate its superior performance relative to a benchmark method.