Hierarchical Linear Mixed Model for Poverty Analysis in Indonesia

Abstract

Hierarchical Linear Mixed Model (HLMM) is an extension of Linear Mixed Model with hierarchical levels of observation. HLMM allows researchers to model different types of covariance structure to describe data properly while in classical linear model the covariance structure defines in constant variance and correlation that hardly applicable for longitudinal data. This paper describes two levels HLMM which represents a single growth curves model. Level-1 presents growth shape to capture within-subject effect and level-2 presents growth parameters that characterized between-subject differences. We model the covariance structure of level-1 random effect to excavate individual growth performance and applied to longitudinal data from poverty data of 34 provinces in Indonesia. Different types of covariance structures are modeled using PROC MIXED in SAS system, produce that AR(1) is the alternative of constant covariance structure and ARH(1) as an alternative for non-constant variance structure based on -2RLL, AIC and BIC criteria.