A0276
Title: Multivariate scan statistics for spatial data
Authors: Lionel Cucala - Universite de Montpellier (France) [presenting]
Abstract: A parametric and a nonparametric scan method for multivariate data indexed in space are introduced. The parametric one relies on a generalized likelihood ratio associated with the multivariate Gaussian distribution whereas the nonparametric one is completely distribution-free as it is based on so-called multivariate ranks. In contrast to existing scan methods, they both take into account the covariance structure between all observed variables. These methods are compared through a simulation study and then applied to a dataset recording the levels of metallic pollutants for two areas in the North of France.