Bayesian Networks to Evaluate and Test the Raven's Colored Progressive Matrices

This study explores the conditional dependencies among items in Raven’s Colored Progressive Matrices (CPMs), a non-verbal measure of cognitive ability and fluid intelligence, within a sample of school-aged children. To this aim, we applied Bayesian networks (BNs) to data obtained from 255 participants, aged 4 to 11 years. We evaluated a variety of models, starting from the transitive independence model and progressing to sequential, data-informed structures. The results highlighted that inter-item dependencies were particularly significant among kindergarten-aged children, suggesting that their problem-solving approaches may be influenced by their previous responses. The Sequential Data-Driven model consistently outperformed traditional theory-driven hypothesis models, indicating the presence of complex interrelationships that could challenge the assumption of local independence in psychometric assessments. Cross-validation analyses revealed developmental differences in model fit, with optimal performance in kindergarten and increased variability as children transition to primary school. These findings advocate for the development of refined scoring methodologies that incorporate item interdependencies and prompt further investigation into the cognitive processes underlying these dependencies. Our study has implications for the psychometric validation of cognitive assessments, underscoring the importance of integrating complex item relationships in test design and analysis to enhance measurement accuracy and validity.