The Interplay Between Clusters, Covariates, and Spatial Priors in Spatial Modelling of Covid-19 in South Sulawesi Province, Indonesia

Abstract

A number of previous studies on Covid-19 have used Bayesian spatial Conditional Autoregressive (CAR) models. However, basic CAR models are at risk of over-smoothing if adjacent areas genuinely differ in risk. More complex forms, such as localised CAR models, allow for sudden disparities, but have rarely been applied to modelling Covid-19, and never with covariates. This study aims to evaluate the most suitable Bayesian spatial CAR localised models in modelling the number of Covid-19 cases with and without covariates, examine the impact of covariates and spatial priors on the identified clusters and which factors affect the Covid-19 risk in South Sulawesi Province. Data on the number of confirmed cases of Covid-19 (19 March 2020 -25 February 2022) were analyzed using the Bayesian spatial CAR localised model with a different number of clusters and priors. The results show that the Bayesian spatial CAR localised model with population density included fits the data better than a corresponding model without covariates. There was a positive correlation between the Covid-19 risk and population density. The interplay between covariates, spatial priors, and clustering structure influenced the performance of models. Makassar city and Bone have the highest and the lowest relative risk (RR) of Covid-19 respectively.

Modeling Factors Influencing Covid-19 Cases in South Sulawesi Using Bayesian Conditional Autoregressive Localised

“…Pemodelan untuk melihat faktor yang mempengaruhi meningkatnya kasus Covid-19 dan analisis risiko relatif di Sulawesi Selatan perlu dilakukan, untuk mengetahui daerah yang berisiko tinggi terinfeksi virus serta faktor yang mempengaruhinya. Pendekatan yang dilakukan dalam menaksir angka risiko relatif Covid-19 di Sulawesi Selatan yang mampu mengatasi permasalahan overdispersi, salah satunya adalah Bayesian Conditional Autoregressive (CAR), melalui pemilihan parameter prior yang tepat (Aswi, 2020;Aswi et al, 2024;Moraga, 2020).…”