Latente variabelen gebruiken om leren te ondersteunen

This dissertation leverages latent variables in order to explain, understand and optimise learning. Latent variables are variables that cannot be measured directly, but instead are approximated with models that explain observable outcomes. Psychometric latent modelling allows the analysis of the learning process and its underlying factors. The most common Artificial Intelligence techniques prioritise prediction neglecting generally the interpretability of the learning process. This dissertation intends to explore both sides in order to propose novel combinations within four studies that describe and explain different learning settings. Chapter 2 is centred around instructional support, which is one of the latent variables of interest. This chapter explores how instructional support can affect the difficulty or the time items demands in a learning environment. The chapter explores the modelling of instructional support effect on response time. Several models are proposed and compared with their Bayes factors. This study not only serves for explaining how support affects response time, but also lays the ground for adaptivity, so that learning environments could automatically change the level of support depending on the learner characteristics. Chapters 3, 4 and 5 are focused on soft skill proficiency. Soft skills are non-technical, personal and social skills such as Stress Management or Leadership. The effect of having taken a course on soft skill proficiency is one of the main latent variables in these chapters. The chapters explain the soft skill proficiency of postgraduate students based on the students' course history and individuals' characteristics. That is, by the courses they took and how the students differ across each other. The models estimate the effect of having taken a course on the students' soft skill proficiency. Based on this, we can describe which course better promotes students to develop a higher proficiency in soft skill terms. Moreover, this dissertation implements innovative ways to combine psychometric latent modelling with Artificial Intelligence techniques. These novel combinations maintain interpretability, which is essential for the continued improvement of the learning factors involved in the process. These proposed combinations intent to tackle problems which would be otherwise difficult to face with either technique on their own. For instance, in chapter 4, Genetic Algorithms are combined with a psychometric model in order to propose recommendations of courses to students given their soft skills.