Global regression assumes that the relationships being measured are stationary over space or the model is applied equally over the whole region. If there is spatial heterogeneity on the data, then the global model is not suitable to the reality. To overcome multivariate spatial over dispersed negative binomial data, we evaluate geographically weighted multivariate negative binomial (local method) and compare it to the global method (multivariate negative binomial). The results show that the geographically weighted negative binomial performs better than the global method. The log likelihood of the local method is higher than the global method. The deviance and mean square prediction error of the local method are smaller than the global method. Moreover, the prediction of dependent variables of the local method are closer to the observed data than the global method. The estimated coefficients of the local method vary, depending on where the data are observed.
|Journal of Physics: Conference Series
|Published - 19 Dec 2019
|6th International Conference on Research, Implementation, and Education of Mathematics and Science, ICRIEMS 2019 - Yogyakarta, Indonesia
Duration: 12 Jul 2019 → 13 Jul 2019