Generalized structured component analysis

[Speaker] Choi, Ji Yeh:1
[Co-author] Hwang, Heungsun:1
1:McGill University (Canada)

Generalized structured component analysis (GSCA) (Hwang & Takane, 2004, 2014) is a component-based approach to structural equation modeling, which combines component analysis and path analysis models. To improve data-analytic flexibility, several extensions of GSCA have been proposed; for example, multilevel GSCA (Hwang, Takane, & Malhotra, 2007) and nonlinear GSCA (Hwang & Takane, 2010). Despite such various extensions, to date, there has been no attempt to apply Bayesian approach to GSCA. In this paper, a new extension of GSCA is proposed that estimates parameters based on a Bayesian framework. Compared to the original GSCA, the proposed approach enables to (1) infer the probability distribution for parameters that are treated as random variables, (2) provide more measures of fit indices for model assessment and comparison, (3) specify various structures of error terms in the measurement model, and (4) incorporate external information about data as a prior distribution in the modelling process.
Advanced Search