Research Seminar
Clusterwise SCA-ECP with different numbers of components for the clusters: Model estimation and model selectionKim De RooverKU Leuven | |
| Abstract: | When researchers measure the same set of variables for a number of subjects, who are nested in groups (e.g., students of different classes; inhabitants of different countries), the obtained multivariate data have a hierarchical structure. Therefore, such data can be called ‘multivariate multiblock’ data as there is a ‘block’ of data for each group. Given such data, an interesting question is what the underlying structure is of each data block, and to what extent this structure differs across the data blocks. Recently, a component method called Clusterwise Simultaneous Component Analysis (Clusterwise SCA; De Roover et al., in press; De Roover, Ceulemans, Timmerman, & Onghena, 2011) was proposed that captures the most important structural differences and similarities between a set of multivariate data blocks by assigning the data blocks to a number of mutually exclusive clusters. Data blocks that belong to the same cluster are modeled using the same loadings, but the loadings may differ across the clusters. However, the Clusterwise SCA as proposed has an important drawback in that the number of components is restricted to be equal across the clusters, which is often not very realistic. Therefore, in this presentation, we present a more general version of Clusterwise SCA-ECP (where ‘ECP’ refers to Equal Cross-Product constraints on the data blocks within the same cluster; see De Roover et al., in press) in which the number of components may vary over the clusters. This generalisation is not as straightforward as it may seem, as serious challenges arise on the level of model estimation and model selection. These challenges are discussed and resolved. |
| Date: | Tue Oct 4, 12:00 pm - 1:00 pm |
| Place: | room 01.07 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |