Some contributions to efficient statistics in structural models: Specification and estimation of moment structures.

摘要: Current practice in structural modeling of observed continuous random variables is limited to representation systems for first and second moments (e.g., means covariances), distribution theory based on multivariate normality. In psychometrics the multinormality assumption often incorrect, so that statistical tests parameters, or model goodness fit, will frequently be incorrect as well. It shown higher order product yield important information when arbitrary. Structural representations are developed generalizations Bentler-Weeks, Joreskog-Keesling-Wiley, factor analytic models. Some asymptotically distribution-free efficient estimators such arbitrary models developed. Limited obtained The special case elliptical distributions allow nonzero but equal kurtoses discussed some detail. argument made normal covariance structure should abandoned favor theory, which only slightly more difficult apply specializes traditional normality holds. Many open research areas described.