多效应GTWPR模型的局部线性极大似然估计OA

Geographically and temporally weighted Poisson regression(GTWPR)models extend geographically and temporally weighted regression(GTWR)by accommodating count data while account-ing for spatial and temporal heterogeneity.Leveraging the advantages of local linear fitting techniques for mitigating boundary effects,this paper proposes a local linear maximum likelihood estimation(LLMLE)method for GTWPR.Furthermore,based on the idea of coefficient averaging,we present a two-stage local linear maximum likelihood estimation(TSLLMLE)method for multi-effect geographically and temporal-ly weighted Poisson regression(MEGTWPR)models.This approach can fit nonstationarity in certain explanatory variables across spatiotemporal,spatial,and temporal changes,while also characterizing glob-ally stationary for other variables,thereby improving the estimation accuracy of coefficients.Numerical simulations show that,compared with the existing coefficient-average-based estimation(CABE)method,the TSLLMLE method yields more accurate coefficient estimates and better boundary performance.