Outlier Detection and Handling in Spatial Autoregressive Models with Variance Shift Outlier Models (VSOM): (Case Study: GDRP Data of Agriculture Sector in Java Island)

Agricultural GRDP data is spatial data that has inter-regional dependency. Therefore, the Autoregressive Spatial SAR model is used to analyze the data. The SAR model is often used to analyze spatial data that has inter-regional dependence in Java. However, the presence of outliers can affect the results of SAR analysis and lead to inaccurate conclusions. By testing spatial dependency using Moran's I, there is an effect of spatial dependency on GDP data for the agricultural sector in Java Island, thus this study aims to detect and handle outliers in GDP data for the agricultural sector in Indonesia. This study uses a method to detect outliers, namely the Variance Shift Outlier Model (VSOM) method which is a defined random effects model that focuses on detecting and handling outliers with a variance approach. This method allows the identification and handling of outliers that arise due to changes in variance between observations in the SAR model, and the variables used include agricultural sector GRDP (Y), total agricultural sector labor (X1), agricultural sector real wages (X2), agricultural sector domestic investment (X3), and agricultural sector FDI (X4). The results showed that there were 6 observations of boostrap results that indicated the presence of outliers and then resolved with VSOM which resulted in a smaller SSR and MSE value of VSOM in detecting and accommodating outliers in the GDP data of the Agricultural Sector in Java the Adj R2 value of VSOM is greater at 0.59655 compared to the SAR model.