Fabbricatore, Rosa (2023) Latent Class Analysis for proficiency assessment in Higher Education: Integrating multidimensional latent traits and learning topics. [Tesi di dottorato]

Preview

Text fabbricatore_rosa_35.pdf Download (2MB) | Preview

Collection description

The present work is devoted to the study and the methodological development of statistical approaches in the framework of Learning Analytics, with a particular focus on the issues of students’ assessment, profiling and tutoring. An accurate assessment of students’ ability undoubtedly represents a considerable part of personalized learning activity: conceived as a complex process, it accounts for different learning topics, dimensions of students’ ability (specific skills), and individual characteristics affecting students’ achievements and performance (e.g., emotional, psychological, and motivational aspects). The derived vast amount of data requires advanced statistical modeling to bring the whole latent underlying structure into the light, transforming data in knowledge as support to address the evaluation’s final aim. In this vein, this contribution proposed two novel statistical approaches in the framework of non-parametric latent variable models, which allow to detect homogeneous groups of students according to their ability characteristics, concurrently accounting for the effect of individual factors on achievements. As such, they offer helpful statistical models for handling students’ learning activities data or as a knowledge base in a self recommendation learning environment. Specifically, the first proposal exploits a non-standard implementation of multilevel latent class analysis defining a multidimensional latent structure at the low level of the hierarchy to account for the multidimensional nature of students’ ability. The second proposal presents a bias-adjusted three step rectangular latent Markov modeling, with an IRT parameterization for the measurement part of the model, either exploiting BCH or ML-based correction methods at the third step. Issues related to changes leading to different nature and number of latent classes and students’ dropouts during learning are also discussed. The results from the empirical application of the proposed approaches in the context of learning Statistics in non-STEM degree courses complement the theoretical aspects.

Downloads

Actions (login required)

View Item