Multilevel latent class modelling to advise students in self-learning platforms: An application in the context of learning Statistics

The growing interest in using technology in education has brought new challenges for data analysis, especially when the main aim is providing students with an appropriate feedback on their latent ability level. Any recommender system should be able to properly account for the multidimensional nature of such latent trait, while correctly modelling the relationships between variables and individuals at possibly different levels. In this perspective, we focus on a non-standard application of multilevel latent class analysis in the context of learning Statistics. The goal is to detect homogeneous groups of students according to their level of ability, concurrently accounting for the hierarchical structure of the data.