2014
16 Dicembre 2014
Giornata di Studio in collaborazione con ISTAT, a cura di Giulia RIVELLINI, Dipartimento di Scienze statistiche Università Cattolica di Milano
Il consumo di sostanze psicoattive nella popolazione giovanile. Comparazioni tra fonti statistiche e realtà territoriali
L’obiettivo dell’iniziativa è soffermarsi su una problematica che investe sempre più la popolazione giovanile. Sono molti gli enti che si occupano di rilevare il consumo di sostanze psicoattive, ponendo l’accento su aspetti diversi del problema. Gli interventi programmati intendono illustrare la dimensione, le specificità e le linee di prevenzione del fenomeno, a partire dalle fonti statistiche a disposizione sul tema.
11 Dicembre 2014
Enrico FABRIZI, DISES - Università Cattolica, Piacenza
Maria Rosaria FERRANTE, Carlo TRIVISANO - Dipartimento di Scienze Statistiche 'P. Fortunati',
Università di Bologna
Hierarchical Beta regression models for the estimation of poverty and inequality parameters in small areas
Many parameters that describe poverty, social exclusion and inequality can take values in the (0; 1) interval. We assume that estimates of this type of parameters are needed for small subpopulations for which only small or no samples are available. In this paper, we discuss area level models for the estimation of this parameters. The idea of area models is that of complementing often imprecise direct estimators with auxiliary information available at the area level obtained from external sources. Given the nature of the target parameters we consider Beta regression models, as the Beta distribution is very exible over the (0; 1) range and it allows for asymmetric sampling distribution.
In this paper we adopt a Bayesian approach with approximate inference for relevant posterior distributions relying on MCMC algorithms. We focus on few speci_c problems that we think that may be particularly relevant for small area applied researchers and discuss them with reference to a speci_c data set. The problems we consider are: i ) the estimation of the at-risk-of-poverty rate; ii ) the joint estimation of the material deprivation and severe material deprivation rates (i.e. two rates based on increasing thresholds); iii ) the joint estimation of two correlated parameters; speci_cally, for illustrative purposes we consider at-risk-of-poverty and Gini inequality index for the equivalized disposable income. When estimating the at-risk-of-poverty rate we face the problem of areas with no poors in the sample that leads us to consider zero-mixture Beta regressions, a class of models that will be considered also in the estimation of other parameters; to reach the goals ii and iii we introduce multivariate extensions of the Beta regression model: in the _rst case we discuss a multivariate logistic-normal model for the expected values of the Beta distributions, while in the second setting, we exploit the correlation between direct estimators using the theory of copula functions. We illustrate the models using an empirical application based on real data from the EU-SILC survey data.
18 Giugno 2014
Prof. Giovanni PETRIS
A Framework for functional Time Series Analysis
With the increasing availability of functional data collected over time, there is a need to develop flexible statistical tools for the analysis and forecasting of time series of functional data.
In the talk we propose a framework for Bayesian functional time series analysis, built on the extension of the dynamic linear model to functional-valued observations and states. The resulting functional dynamic linear model is mathematically rigorous and, at the same time, very easy to use in practical applications. While the model in its abstract formulation is very general, we will show how specific subclasses of it, inspired by the classical Structural Time Series models, can be used to analyse data exhibiting a particular structure, such as for example a seasonal pattern.
Examples and applications to economic and financial data will be provided.
27 Maggio 2014
Prof. Valentino DARDANONI - Università di Palermo,
Regression Models with Missing Covariates
21 marzo 2014
Prof.ssa Alessandra GUGLIELMI - Politecnico di Milano, Dipartimento di Matematica
Semiparametric Bayesian models for clustering and prediction of Acute Myocardial Infarction patients.
In this talk, the interest is in modeling, under a Bayesian nonparametric perspective, mixed-type multiple responses of patients hospitalized with an ST-segment elevation myocardial infarction (STEMI) diagnosis and treated with angioplasty. The responses considered here are three variables, which strategically evaluate the efficiency of the hospitals and the effectiveness of the treatment: the time between admission to the hospital and angioplasty (continuous response), in-hospital survival and survival after 60 days from admission (binary).
We consider two Bayesian semiparametric models: for both cases, the key ingredient is a nonparametric prior on the partition of the patients. Hence, we can provide model-based clustering through the posterior. Moreover, though the survival rates are strongly unbalanced, the models are able to give sufficiently reliable predictions on the survival outcomes at the patient's level.
The study is within a project, named ‘Strategic program of Regione Lombardia’; data include
clinical and process indicators, outcomes and personal information on patients admitted to all hospitals of Regione Lombardia with STEMI diagnosis.
The seminar is based on joint work with Francesca Ieva, Anna Maria Paganoni, Elena Prandoni, Fernando Quintana, Valerio Valvo.
14 Marzo 2014
Dott. Giacomo ZANELLA - University of Warwick
Bayesian complementary clustering, MCMC for data association and Anglo-Saxon placenames
Common cluster models for multi-type point processes look for unusual aggregations (clusters) of points of the same type. Motivated by the study of Anglo-Saxon settlements locations, where administrative clusters involving a variety of complementary names tend to appear, we develop a Bayesian Random Partition Model that looks for clusters formed by points of different types.
We obtain a multimodal, intractable posterior distribution on the space of matchings contained in a k-partite hypergraph. We develop an efficient Metropolis-Hastings algorithm to sample from such posterior distribution. We consider the problem of choosing an optimal proposal distribution in such framework. Simulated Tempering and multiple proposal techniques are used.
This allows us to study the Anglo-Saxon placenames locations dataset. Without strong prior knowledge required the model allows for explicit estimation of the number of clusters, the average intra-cluster dispersion, the ratio of noise points and the distribution of clusters sizes. First results seems to support the hypothesis of the settlements being organized into administrative clusters.
13 Febbraio 2014
Presentazione Volume a cura di Filomena RACIOPPI e Giulia RIVELLINI
La Demografia per le aziende e la governance locale
E’ realistico sostenere che tra gli interessi professionali di chi opera per le imprese e per il mercato, e per la governance in genere, possa esserci spazio – o sia opportuno che ci sia – anche per conoscenze e competenze della Demografia? Il volume propone una risposta affermativa.
6 Febbraio 2014
Dott.ssa Laura SANGALLI - Politecnico di Milano
Modeling demographic data and medical imaging data via spatial regression models with differential regularization
I will present a novel class of models for spatial data analysis, that merges advanced statistical methodology with numerical analysis techniques. Thanks to the combination of potentialities from these two scientific areas, the proposed class of models has important advantages with respect to classical techniques used to analyze spatially distributed data. The models are able to efficiently deal with data distributed over irregularly shaped domains, including non-planar domains, only few methods existing in literature for this type of data structures. Moreover, they can incorporate problem-specific priori information about the spatial structure of the phenomenon under study, with a very flexible modeling of space variation, allowing naturally for anisotropy and non-stationarity. The models have a generalized additive framework with a regularizing term involving a differential quantity of the spatial field. The estimators have good inferential properties; moreover, thanks to the use of numerical analysis techniques, they are computationally highly efficient. The method is illustrated in various applied contexts, including demographic data and medical imaging data.
The seminar is based on joint work with Laura Azzimonti, Bree Ettinger, Fabio Nobile, Simona Perotto, Jim Ramsay, Piercesare Secchi. This research is developed within FIRB2008 Futuro in Ricerca research project SNAPLE (http://mox.polimi.it/users/sangalli/firbSNAPLE.html).
29 Gennaio 2014
Dott.ssa Silvia FACCHINETTI - Università Cattolica del Sacro Cuore
Portfolio selection with Lasso algorithm
Abstract In this talk we describe the application of the Lasso algorithm to financial time series. The aim is to estimate the inverse covariance matrix of the portfolio and detect and remove elements that presents a spurious correlations
17 Gennaio 2014
Prof. Dimitris FOUSKAKIS, Department of Mathematics, School of Applied Mathematical and Physical Sciences National Technical University of Athens
Power-Conditional-Expected Priors: Using g-priors with Random Imaginary Data for Variable Selection
The talk consists of two parts. In Part 1 a description of how the Bayesian community deals with the variable selection problem will be presented. Several popular approaches to Bayesian variable selection, including computational methods for posterior evaluation and exploration, will be briefly reviewed. Additionally, different approaches on specifying the prior distribution (a) on the parameter space of each candidate model and (b) on the model space, will be shown. Emphasis will be given on “Objective” Bayesian variable selection techniques.
In Part 2 we borrow ideas from the power-expected-posterior (PEP) priors in order to introduce, under the g-prior approach, an extra hierarchical level that accounts for the imaginary data uncertainty. For normal regression variable selection problems, the resulting power-conditional-expected-posterior (PCEP) prior is a conjugate normalinverse gamma prior which provides a consistent variable selection procedure and gives support to more parsimonious models than the ones supported using the g-prior and the hyper-g prior for finite samples. Detailed illustrations and comparisons of the variable selection procedures using the proposed method, the g-prior and the hyper-g prior are provided using both simulated and real data examples.