International Conference on Robust Statistics 2012

Dear colleagues, We are pleased to announce that the University of Vermont will be hosting the upcoming International Conference on Robust Statistics. The conference will be held from August 5 to August 10, 2012 on the campus of the University of Vermont in Burlington, Vermont, USA. August 5-10 is the week after the Joint Statistical Meetings in San Diego.

The International Conference on Robust Statistics (ICORS) has been an annual international conference since 2001. The aim of the conferences is to bring together researchers interested in robust statistics, data analysis and related areas. This includes theoretical and applied statisticians as well as data analysts from other fields, and leading experts as well as junior researchers and graduate students.

ICORS welcomes contributions to applied statistics as well as theoretical statistics, and in particular new problems related to robust statistics and data analysis. The following areas are expected to be well represented at the conference, but contributed talks on other related topics are also welcomed.

    * Concepts and theory of robust statistics
    * Asymptotic theory and efficiency
    * Novel applications of robust statistical methods
    * Robust and nonparametric multivariate statistics
    * Robust functional data analysis
    * Robust regression, including quantile regression
    * Linear and generalized linear models;  mixed models
    * Biostatistics
    * Statistical methods in bioinformatics/genetics
    * Statistical computing and graphics and data mining
    * Data mining and machine learning

The website for the ICORS 2012 conference can be found at:

http://www.rci.rutgers.edu/~dtyler/ICORS2012/

For any questions please email to:

icors12@gmail.com

We wish to acknowledge the generous support of the Minerva Research Foundation, and pending support of the National Science Foundation.

Announce of Seminar

Dr. Luigi Spezia

Biomathematics and Statistics Scotland, Aberdeen, UK

MODELLING THE STATES OF SCOTLAND'S RIVERS

January 10th, 2012 at 14.00

Aula B, Santa Marta, Venezia

Anyone interested is invited to participate

Analyses of Scotland's rivers are presented both on a local and on a global scale, under the environmental and the ecological point of view. First, the multivariate time series of nitrogen concentrations are considered by modelling rivers in isolation, then modelling all data simultaneously. Next, the dynamics of the signatures of two stream isotopes is investigated. Finally, the mapping of the species distribution of freshwater pearl mussels in a river is shown. The analyses are performed by means of Bayesian hierarchical models, driven by a hidden Markov chain. The states of the Markov process allows gathering of the observations in a small set of homogeneous groups. The models shown belong to the class of hidden Markov models and Markov switching autoregressive models.

The talk is based on joint work with Christian Birkel (University of Aberdeen), Mark Brewer (Biomathematics& Statistics Scotland, Aberdeen), Susan Cooksley (The James Hutton Institute, Aberdeen), Martyn Futter (Swedish University of Agricultural Sciences, Uppsala), Roberta Paroli (Catholic University, Milan).

The announcement is available here.

Il giorno Martedì 15 Novembre 2011 alle ore 11.15 presso l'Aula Consiglio, Ala 2C secondo piano (ex Dipartimento di Statistica) si terrà un seminario tenuto dal J.F. Rosco Nieves (Università dell'Estremadura) dal titolo

SKEWNESS - INVARIANT MEASURES

central moment is a characteristic used to describe the shape of a probability distribution. However, since its introduction more than a century ago, numerous interpretations of it have been suggested within the literature. A historical review of these interpretations is made and the measurement of kurtosis in the presence of asymmetry is considered. Blest’s kurtosis measure adjusted for skewness is studied and an alternative coefficient is proposed, both measures also being based on moments. Since the existence of moments of a distribution cannot always be assured, a quantile-based approach is considered. Two forms of kurtosis measures based on quantiles are identified which are invariant to the presence of skewness for certain families of distributions obtained via the transformation of a base symmetric random variable. A very general condition is established which can be used to determine when the two types of kurtosis measures will be skewness invariant, and two wide families of distributions are identified for which the measures are skewness invariant. These two family of distributions are the well-known Johnson’s unbounded family and the more recently proposed sinh-arcsinh family. Also we identify a family of distributions not arising via transformation for which the measures are skewness invariant, namely the Tukey lambda family.

La locandina è disponibile qui.

Vi segnalo questo seminario organizzato dal Prof. Andrea Collevecchio. Il seminario si terrà il 13 Ottobre 2011 alle ore 14.30 presso aula Saraceno a San Giobbe. Relatore del seminario con titolo

Large deviations for cluster size distributions in a classical many-body system

è il Prof. Wolfgang Koenig (Weierstrass Institute and TU Berlin).

La locandina è disponibile qui.

Il giorno Venerdì 28 Ottobre 2011 alle ore 14.30 presso l'Aula Consiglio, Ala 2C secondo piano (ex Dipartimento di Statistica) si terrà un seminario tenuto dal Prof. C. Hennig (UCL) dal titolo

How to find the best cluster analysis method (for social stratification based on mixed data)?

This presentation will treat two issues. 1. A general approach for the decision about the ``best'' cluster analysis method will be sketched. This is based on the idea that a subjective decision is needed about the cluster concept required in a particular application, and that there is no such thing as an objectively true clustering determined by the data alone. For example, even if data can be perfectly fitted by a Gaussian mixture, this does not necessarily mean that the mixture components correspond to the "true" clusters, because in a given application unimodal mixtures of more than one Gaussian may be considered as a single cluster (see Hennig 2010). On the other hand, large within-cluster dissimilarities may not be admissible, in which case certain Gaussian components need to be split up. 2. The general philosophy is applied to the problem of defining social strata from data with mixed continuous, ordinal and categorical variables. Clusterings as obtained from a mixture/latent class approach (as for example fitted by the LatentGOLD? software, Vermunt and Magidson, 2005) are compared with clusterings obtained from applying k-medoids (Kaufman and Rousseeuw, 1990) to dissimilarities balancing the different types of variables in a sensible user-specified way.

References HENNIG, C. (2010): Methods for merging Gaussian mixture components. Advances in Data Analysis and Classification, 4, 3-34. KAUFMAN, L. and ROUSEEUW, P. J. (1990): Finding Groups in Data. Wiley, New York. VERMUNT, J. K. and MAGIDSON, J. (2005): Technical Guide for Latent GOLD 4.0: Basic and Advanced. Statistical Innovations Inc., Belmont Massachussetts.

La locandina è disponibile qui.