A new approach to structural equation modeling based on so-called extended redundancy analysis (ERA) is proposed. In ERA, latent variables are obtained as exact linear combinations of observed variables, and model parameters are estimated by consistently minimizing a single criterion. As a result, the method can avoid limitations of covariance structure analysis (e.g., stringent distributional assumptions, improper solutions, and factor score indeterminacy) in addition to those of partial least squares (e.g., the lack of a global optimization procedure). The method is simple yet versatile enough to fit more complex models; e.g., those with higher-order latent variables and direct effects of observed variables. It can also fit a model to more than one sample simultaneously. Other relevant topics are also discussed, including data transformations, missing data, metric matrices, robust estimation, and efficient estimation. Examples are given to illustrate the proposed method.
| Identifer | oai:union.ndltd.org:LACETR/oai:collectionscanada.gc.ca:QMM.36954 |
| Date | January 2000 |
| Creators | Hwang, Heungsun, 1969- |
| Contributors | Takane, Yoshio (advisor) |
| Publisher | McGill University |
| Source Sets | Library and Archives Canada ETDs Repository / Centre d'archives des thèses électroniques de Bibliothèque et Archives Canada |
| Language | English |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Format | application/pdf |
| Coverage | Doctor of Philosophy (Department of Psychology.) |
| Rights | All items in eScholarship@McGill are protected by copyright with all rights reserved unless otherwise indicated. |
| Relation | alephsysno: 001810504, proquestno: NQ70045, Theses scanned by UMI/ProQuest. |
Page generated in 0.0019 seconds