13 Structural Equation Modeling

摘要: Abstract Structural equation modeling (SEM) is a multivariate statistical technique for testing hypotheses about the influences of sets variables on other variables. Hypotheses can involve correlational and regression-like relations among observed as well latent The adequacy such evaluated by mean covariance structures After an introduction, we present model. Then discuss estimation methods hypothesis tests with emphasis maximum likelihood method based assumption normal data, including issues model (parameter) identification regularity conditions. We also non-normal data misspecified models, power analysis. To supplement testing, fit indices have been developed to measure degree SEM describe major ones. When initial does not well, Lagrange Multiplier (score) Wald be used identify how might modified. In addition these standard topics, extensions multiple groups, repeated observations (growth curve SEM), hierarchical structure (multi-level nonlinear relationships between more practical topics treatment missing categorical dependent variables, software information.