Local Influence for Spatially Correlated Binomial Data: An Application to the Spodoptera frugiperda Infestation in Corn

Influence diagnostics are valuable tools for understanding the influence of data and/or model assumptions on the results of a statistical analysis. This paper proposes local influence for the analysis of spatially correlated binomial data. We consider a spatial model with a binomial marginal distribution and logit link function. Generalized estimating equations via Fisher’s scoring are used for estimating the parameters. We present an application to the spatial Spodoptera frugiperda infestation where the generalized estimating equations are used to identify potential influential observations by the local influence analysis. The spatial prediction with and without the influential points is compared. The results show that the presence of the influential observation in the data changes statistical inference, the predicted values and the respective maps. A simulation study considering different scenarios shows the performance of the local influence diagnostic method.