MULTILEVEL STRUCTURAL EQUATION MODELING: ANALISIS METODE WEIGHTED LEAST SQUARE PADA VARIABEL OBSERVASI NON-NORMAL (Multilevel Structural Equation Modeling: Analysis of Weighted Least Square On Non-Normal Observed Variables)

SEPTIANAWATI, ERNI, Abdurakhman

2016 | Disertasi | FMIPA

Structural Equation Modeling is used to show how the relationship between latent variable and observed variable, however, it tends to be less focused on structural and hierarchical data. On the other hand, multilevel modeling is used when the structure of the data are in the hierarchical form, however, it has a limitation, that it often fails to have direct modulation result under complex causal process. Some studies synthesized both models and enhance them as Multilevel Structural Equation Modeling (ML-SEM). Multilevel Structural Equation Modeling (ML-SEM) is employed using non-normal observed variable and will be estimated under Weighted Least Square (WLS), because it does not depend on form of distributed data. It means Weighted Least Square (WLS) method accepting non-normality. In this study, Multilevel Structural Equation Modeling is applied on Maternal Health Care (MHC) sample (n=8933) using Survey Demografi Kesehatan Indonesia (SDKI) 2012 data. The objective is to know the variance of SDKI in the both of individual and province level using dichotomous categorical indicators.The result shows variance in the both of individual and province level under its indicators from SDKI and Delivery Care (DC) factors. On the other hand, invariance is generated in the both of individual and province level from Post Natal Care (PNC) factors.

Kata Kunci : Structural Equation Modeling (SEM), Multilevel Model, Weighted Least Square, Limited Information Estimation, dichotomous, SDKI 2012.