Font Size: 

The classical data representation model is too restrictive when the data to be analysed are not real numbers but comprise variability. In this talk, we are interested in numerical distributional data, where units are described by histogram or interval-valued variables. We consider parametric probabilistic models, which are based on the representation of each distribution by a location measure and interquantile ranges. A multivariate outlier detection method  is proposed that makes use of restricted configurations for the covariance matrix, and is based on a sparse robust estimator of its inverse. The computations rely on an efficient adaptation of the graphical lasso algorithm. A simulation study puts in evidence the usefulness of the robust estimates for outlier detection.