Structural Equation Modelling-Partial Least Square to Determine the Correlation of Factors Affecting Poverty in Indonesian Provinces

Abstract

Structural Equation Modelling (SEM) is a multivariate statistical technique to analyze the pattern of relationships between latent variables (unobserved) with indicator variables (observed). One aspect that can be solved by SEM is poverty. Poverty is a major problem in economic development in developing countries like Indonesia. The problem of poverty can also be seen from the dimensions of education and health. This research uses SEM model with Partial Least Square (PLS) approach on poverty data in all provinces in Indonesia based on 2015-2017 Badan Pusat Statistik (BPS) data with latent variables of poverty, education, economy, and health. The results show that there is an indicator of latent educational variables that must be excluded from the model because it has a loading factor value > 0.5. The results of the outer model evaluation with Convergent Validity, Composite Reliability (CR), and Average Variance Extracted (AVE) show significant and reliable values that mean the indicators used can explain latent variables well. The SEM-PLS inner model evaluation results can be known from the value of R-Square (R2) through the bootstrapping stage with 500 sample cases. R-square value for Economics is 0.751 means the model can explain variations of Economy in poverty cases in the Provinces of Indonesia at 75.1%, Education at 0.583 or 58.3% and the poverty model at 0.645 or 64.5%. The structural model for poverty cases in Indonesia in 2015-2017 is obtained Poverty = -0.548 Health - 0.085 Education + 0.128 Economy. The relationship of the poverty model shows that if poverty in Indonesia increases, the health of poor households and Education in Indonesia will decrease by 0.548 and 0.085 units respectively, whereas if poverty in Indonesia increases, the economy of poor households will increase by 0.128 assuming looking at the indicators that form latent economic variables.