Title: La distribución Shifted-scaled Dirichlet, una generalización natural de la distribución de Dirichlet Description: Presentamos una generalización de la distribución de Dirichlet clásica sobre el símplex. En particular estudiamos la distribución resultado de aplicar las operaciones perturbación y potencia a una composición aleatoria con distribución de Dirichlet. Estas dos operaciones dotan al símplex de estructura de espacio vectorial, y juegan el mismo papel que la suma y el producto por escalares en el espacio real. Se estudia la distribución resultante desde un punto de vista probabilístico, se presenta la función de densidad, se obtienen diferentes medidas características y se discute la pertenencia a una familia exponencial. Se compara la expresión de la función de densidad con respecto a la medida de Lebesgue habitual con la densidad con respecto a la medida de Aitchison en el símplex. Finalmente se revisan las distribuciones de Dirichlet y Dirichlet escalada que resultan ser casos particulares. La distribución de Dirichlet escalada se obtiene por perturbación de una densidad de Dirichlet. Speaker: Glòria Mateu Figueras Site: Perason Seminar (2n pis, Edifici P-4) Date: 2012-05-10 Hour: 15 Link: http://ima.udg.edu/Recerca/EIO/inici_cat.html File:

Title: Linear regression with compositional explanatory variables Description: Compositional explanatory variables should not be directly used in a linear regression model because any inference statistic can become misleading. While various approaches for this problem were proposed, here an approach based on the isometric logratio (ilr) transformation is used. It turns out that the resulting model is easy to handle, and that parameter estimation can be done in like in usual linear regression. Moreover, it is possible to use the ilr variables for inference statistics in order to obtain an appropriate interpretation of the model. Speaker: Karel Hron, Eva Fiserová Site: Seminari Pearson - Edifici P4 - Campus M Date: 2012-03-28 Hour: 15:00 Link: http://ima.udg.edu/Recerca/EIO/inici_eng.html File:

Title: Predictors and Outcomes of Social Network Compositions. A Compositional Multitrait-Multimethod Approach. Description: Network compositions are expressed as proportions of a total, whose sum can only be 1. Compositional data are highly non-normal, as they range within the 0-1 interval. One component can only increase if some other(s) decrease, which results in spurious negative correlations. Statistical analysis of compositions in general, and particularly estimation of structural equation models are challenging tasks.Structural equation models are particularly valuable for network compositions, which are usually measured with survey instruments and therefore prone to measurement errors. These models make it possible to estimate unbiased effects from background variables on network compositions and from network compositions to theoretically related variables such as satisfaction with social relationships.We draw from Coenders et al. (2011) who fitted the correlated uniqueness model to repeated measurements of network composition following a multitrait-multimethod design with the aim of estimating reliability of network composition measurements. In this article we consider a new type of transformation of compositional data called isometric logratio transformation and we extend the correlated uniqueness model to include predictor and outcome variables.The data contain social network compositions expressed in percentages of partner, family, friends and other members. Background variables are found to predict network composition. For instance, women tend to have comparatively less partner support, college educated comparatively more friend support and the elderly comparatively more support from people who are neither friends nor relatives. Extroversion is found to be positively related to partner and friend support. Speaker: Tina Kogovsek, Germà Coenders, Valentina Hlebec Site: Seminari Pearson, Campus Montilivi, Edif Date: 2012-03-14 Hour: 14:30 Link: http://ima.udg.edu/Recerca/EIO/inici_cat.html File:

Title: Local regression for compositional data Description: Among nonparametric methods for regression, we focus on a class of kernel regression estimators, called "local polynomial kernel estimators" obtained by fitting "locally" a polynomial to the data. We adapt the standard results to the case of compositional data, making use of "the principle of working on coordinates". We introduce a class of kernels on the simplex in a similar manner to the normal kernel. In the regression, we consider the case when the covariates are compositions and the response variable is real or the case in which both covariates and response variable are compositions. We derive the estimator and its asymptotic properties. Future research will concentrate on the application to real data. Speaker: Catia Venieri Site: Seminari Pearson - Edifici P4 - Campus M Date: 2012-02-23 Hour: 15:30h Link: http://ima.udg.edu/Recerca/EIO/inici_cat.html File:

Title: Orthogonal regression for three-part compositional data using the linear models with type-II constraints Description: Compositional data as data carrying only relative information in their parts, demand different treatment than the absolute data, when one wants to perform statistical analysis. In the contribution we will mainly focus on calibration line problem solved using linear models with type-II constraints for three-part compositional data. We will also point out to the others less advantage approaches like for e.g. the standard approach to estimation in orthogonal regression, based on the orthogonal least squares or the maximum likelihood method. First thing that we need to do is to move isometrically the three-part compositional data from their sample space the simplex S^3, to the ordinary Euclidean space R^2. For this purpose we used the well known isometric log-ratio (ilr) transformation. Here on the ordinary Euclidean space, we will handle with our goal, we will progress orthogonal regression using linear models with type-II constraints. Naturally, under further assumption of normality of the ilr transformed data, which is equivalent to the normal or log-normal distribution in the simplex, we can provide further statistical inference e.g., construct confidence regions and bounds and test hypotheses. Speaker: Sandra Donevska Site: Pearson Seminar, Campus Montilivi, P4, Date: 2012-02-15 Hour: 15:30 Link: http://ima.udg.edu/Recerca/EIO/inici_cat.html File:

Title: Things that make us different: Analysis of variance with time-use data Description: The constrained, non-Normal nature of time-use data poses a challenge to ordinary analysis of variance. This paper investigates a computationally simple variance decomposition technique suitable for these data. As a by-product of the analysis, a measure of fit for systems of timedemand equations is proposed that possesses several useful properties. Speaker: Jorge González Chapela; Department of Economics (U. of Girona); Email: Jorge.gonzalez@udg.edu Speaker: Javier González Chapela Site: Pearson Seminar, Campus Montilivi, P4, Date: 2011-11-24 Hour: 15:30 Link: http://ima.udg.edu/Recerca/EIO/inici_eng.html File: