Abstract:

The concern for the quality of university education has stimulated the formulation of indicators that aim to summarize and synthesize the behavior, performance and results of the academic activity of students and teachers. Academic performance, as a result of a series of factors acting on and from the learner, is a quality indicator that is developed in a context where large amounts of data are used to guide teaching - learning in order to create a "map of learning opportunities". When the scale of data manipulation, exploration and inference, grows beyond human abilities, one must resort to new technologies. Moodle is one of them and following its tracks allows us to give usefulness to the data generated in the university system (internal data) and to face the study of academic performance from a different perspective to the traditional one. Our starting hypothesis states that the behavior of students is influenced by the context in which they develop; That is, the interaction between the individual characteristics of the students and the variables that describe the different contexts (group, course, degree, etc.) are nested and form a hierarchical data structure.

Due to the hierarchical association of data generated internally -by universities- is not accidental, it should not be ignored, disregarding the effects of the context may invalidate the use of traditional statistical models, since they incur errors in assigning the same value for variables of the macro units (context) to the micro units, or when deductions are inferred about the individual dynamics based on aggregated data or context level.

In summary, the purpose of this paper is to predict the academic performance of university students, based on their educational characteristics generated by Moodle, applying multilevel models. The multilevel models solve the problems derived from the atomistic and ecological fallacy, when we are working with the different levels at the same time. With these models it is possible to differentiate the variance explained by each predictor in the different levels of aggregation selected. In addition, it is possible to make inferences with variables that act at different levels.

Keywords:

Nested data, Academic performance, interaction, context or level.