Fixed effect geographically weighted panel regression: A comparison of Kernel Gaussian and Bi-Square for modeling sugarcane yield in East Java

Geographically weighted panel regression (GWPR) combines geographically weighted regression models with panel regression. The Weighted Least Squares (WLS) method is used for parameter estimation in the GWPR. Each location’s result model will be distinct from the others. Similar to GWR, the GWPR used weighting functions, such as Gaussian and Bi-Square kernel functions. There has never been any study done on comparing two kernels for GWPR modeling. This study aims to examine two commonly used kernel functions Gaussian and Bi-Square to determine their effectiveness in capturing the spatial variations in sugarcane yield. The Akaike Information Criterion (AIC) value is used to compare the Gaussian and Bi-Square kernel functions in GWPR modeling. The application of GWPR to modeling sugarcane yield at district and city of East Java shows that the estimated model give different result between one location and another. GWPR with Gaussian kernel function is the optimal model for analyzing sugarcane yield in East Java, as determined by the AIC and R2. The results show that the choice of kernel function significantly affects the models performance, highlighting the importance of selecting an appropriate kernel function in GWPR modeling.