Visualizing Influential Observations in Multivariate Dependent Structures

Two graphical representations are proposed: the dmax plot and the Parablot for spatial correlated data. The goal is to provide simple and easy-to-use tools for multivariate structures with spatial dependence.

These tools exploit the normal curvature technique of the local influence plot in dependent multivariate statistics. Through applications in simulated and real data, the methods demonstrated great power in identifying possible influential observations. To verify this, we analyzed the impact of removing the detected observations using Moran’s I index, as well as visualizing the empirical semivariogram.