Research Seminar
Clusterwise simultaneous component analysis for the analysis of structural differences in multivariate multilevel dataKim De RooverKU Leuven | |
| Abstract: | Numerous research questions in educational sciences and psychology concern the structure of a set of variables. In educational sciences, one is, for instance, interested in the structure of the beliefs, competencies, and expectancies of students (e.g., Vanhoof, Castro Sotos, Onghena, Verschaffel, Van Dooren & Van den Noortgate, 2006). The debate about the structure of emotions (e.g., Kuppens, Ceulemans, Timmerman, Diener & Kim-Prieto, 2006) is an example from psychology; in this debate, one often assumes that emotions can be organized in a low-dimensional space; however, considerable disagreement exists about the number and the nature of the dimensions of this space (e.g., Fontaine, Scherer, Roesch & Ellsworth, 2007). To study the structure of a set of variables, one typically relies on scores from a group of persons on those variables. One may, however, wonder whether the same structure would have been retrieved if another group of persons had been studied. As such, the structure of the beliefs, competencies, and expectations of students may differ strongly across study disciplines. Similarly, the covariation of emotions may vary strongly across cultures (e.g., Eid & Diener, 2001). It goes without saying that to trace such structural differences, one will have to gather data from students from different study disciplines, or from inhabitants from different nations. Formally, the resulting data then constitute multivariate multilevel data, with persons being nested in groups. Obviously, the crucial question is how such data have to be analyzed to find out whether and in what way the structure of the variables differs across the groups of persons. A number of principal component analysis techniques exist to study such structural differences, for instance, simultaneous component analysis (Timmerman & Kiers, 2003). However, these techniques suffer from some important limitations. Therefore, in this presentation, we propose a novel generic modelling strategy, called clusterwise SCA, which solves these limitations by combining clustering and SCA analysis. This strategy will be shown to encompass several existing techniques as special cases. |
| Date: | Tue Oct 13, 12:15 pm - 1:15 pm |
| Place: | room 02.51 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |