CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1353
Title: Structural equation models with mixed continuous and partially ordered data Authors:  Xiaoqing Wang - The Chinese University of Hong Kong (Hong Kong) [presenting]
Xiangnan Feng - The Chinese University of Hong Kong (Hong Kong)
Xinyuan Song - Chinese University of Hong Kong (Hong Kong)
Abstract: Structural Equation modeling (SEM) is well recognized in many disciplines as the most important multivariate technique for assessing the interrelationships among latent variables that are grouped based on correlated observable variables. Owing to the questionnaire design and problem nature, discrete categorical data, such as ordered data, unordered data, and partially ordered data, are routinely collected in social, medical and behavioural research. Among literature, substantial efforts have been devoted on investigating SEM with ordered and unordered categorical variables, whereas litter attention has been focused on the analysis of correlated partially ordered data under the SEM framework. To fill the gap, we propose a general class of SEM that is capable of operationalizing latent variables through mixed continuous and partially ordered variables. Partially ordered set theory and the specification of the model are discussed. Bayesian procedures implemented through Markov Chain Monte Carlo algorithms are developed for statistical inference. Extensive simulation studies demonstrate that the developed methodology enjoys satisfactory performance.