hierarchical linear models

Educational research has advanced theoretical approaches by defining and designing integrated models oriented up to the study of relationships between constructs not directly observable embedded in complex contexts. In parallel, data analysis software has allowed testing these complex models in reasonable computing environments in terms of time and effort. This techno-scientific revolution in the developed software for quantitative data analysis has transformed scientific practice in educational research. In this paper is assumed the growing trend in educational research on design and fit of comprehensive models of educational phenomena. More specifically we will focus on the statistical decision trees, in structural equation models and multilevel models, which come to dominate the landscape of educational research since the mid 70s and have recently been generalized thanks to technological development. The impact on educational research has been a widespread use of sophisticated statistical models than ever so far. It is necessary to think about what theoretically grounded models have to be tested with statistical tools as powerful as those being described in this paper.

This article deals with the following issue: When reliability indices of teaching evaluation at university level are broken down into students personal components and contextual components, reduction of reliability is expected.

Results suggest that it is necessary to consider a methodological alternative to reliability calculated with traditional methods, and to interpret cautiously results obtained from questionnaires, especially if they are going to be used for summative assesment by members of the academic authorities.

 

Key words: Hierarchical linear models, reliability indices, teaching effectiveness