CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Area-level Poisson mixed models for estimating forest fire occurrences Authors:  Miguel Boubeta - Universidade da Coruna (Spain) [presenting]
Maria Jose Lombardia - Universidade da Coruna (Spain)
Manuel Marey-Perez - Universidade de Santiago de Compostela (Spain)
Domingo Morales - University Miguel Hernandez of Elche (Spain)
Abstract: Area-level Poisson mixed models are good tools for modelling count data at area level. However, its most basic version has several limitations. It does not take into account complex spatial structures or temporal components. A spatiotemporal extension of the basic Poisson mixed model by incorporating a SAR(1) spatial structure and independent time effects is considered. Model fitting based on the method of moments is proposed. The empirical best predictor of the Poisson parameter is obtained and compared against other competitors such as the synthetic or plug-in estimators. These estimators are empirically investigated by several simulation experiments. As accuracy measure of the proposed estimator, the mean squared error is considered and a bootstrap approach is given. Finally, an application to real data of forest fires occurrences in Galicia during 2007-2008 is carried out.