Generalized Structural Equations Approach in the of Elderly Self-rated Health

Abstract

Abstract To model classified and ordinal data, the Generalized Structural Equation Model (GSEM), which is based on the integration of two generalized linear model (GLM) and Structural Equation Modeling (SEM) algorithms, is applied. Unlike the SEM, this model does not require the normality assumption. The main purpose of this paper is to introduce and compare weighted (WLSMV) and unweighted (ULSMV) least squares mean and variance adjusted methods, two of the most applicable estimators of GSEM, for studying factors affecting the elderly self-rated health in 2015 in Tehran, Iran. 600 elderly people aged 60 years and above from 22 regions of Tehran were selected using multi-stage sampling. Self-rated health of the elderly variable (a 5-point Likert scale) was analyzed as an ordinal variable and was modeled considering the variables of social support, financial and environment security, spirituality, mental and physical health, functional health and health-related behaviors by Mplus software. The results showed that WLSMV outperformed ULSMV according to the smaller values of RMSEA and larger values for CFI and TLI indexes (RMSEA WLSMV = 0.04, CFI WLSMV = 0.965, and TLI WLSMV = 0.936). To prevent concluding invalid results in studying ordinal data due to considering them as a continues variable, it is important selecting correct statistical method according to the type of variables.