Bayesian conditional autoregressive model for estimating the socio-demographic effects of household on poverty in Kelantan, Malaysia

Over the past decades, there has been marked progress in reducing poverty. Although the statistics have shown a steady decrease in poverty from one year to another, the inequality gap between the regions, states and rural-urban areas remain wide. This study aims to estimate the poverty risk across 66 districts in Kelantan. The poverty data are counts of the numbers of poor household head in each areal unit. Thus, Poisson log-linear model is typically used for the analysis. However, the poverty data typically contain residual spatial autocorrelation after the covariate effects have been accounted for. This autocorrelation is commonly modelled by a set of random effects. These random effects are most often modelled by a Conditional AutoRegressive (CAR) prior as part of a hierarchical Bayesian model. Three commonly used Poisson log-linear CAR models are applied to the simulated poverty data and assessed their appropriateness via a simulation study with four different scenarios. Then, this study improved the best CAR model by providing better neighbourhood weight matricesW and investigate the performance of the best Poisson log-linear CAR model with eight different specifications of the neighbourhood weight matrix via a simulation study with sixteen different scenarios. Lastly, the best models are used to estimate the spatial pattern of poverty risk for 66 districts in Kelantan, based on data collected in the e- Kasih database. We show that, overall, a Poisson log-linear Leroux CAR model with Delaunay triangulation, contiguity, and m = 5 neighbourhood matrices performed the best. Increased number of no formal education and female of poor heads of household appear to inflate the risk of an area of poverty while increasing the age of household head appears to decrease the risk. Pengkalan Kubor, followed by Gong Datok appeared to have a higher risk of poverty than other districts in Kelantan. The findings of this study has widening the body of knowledge in solving spatial autocorrelation problem in spatial poverty data using CAR models. The findings for the estimated poverty risk for each district can help other researchers to conduct more in-depth research, particularly in areas with high poverty risks. This study will help better understand the nature of poverty in Kelantan and direct the targeted interventions to eliminate poverty. Strategic policy of poverty can be developed by using the spatial household poverty mapping information.