Departamento de Estadística e Investigación Operativa

Compartir:

Programación del Seminario: Año 2003

 

MODELOS DE REGRESIÓN MULTINIVEL

CONVIDAT: Jesús Rosel. Dept. de Psicologia Evolutiva, E, S, y Metodología. Universitat Jaume I.

IDIOMA: Castellà

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 7 de febrer de 2003, 12h30m

RESUM: Es frecuente en la investigación empírica (psicológica, social, biológica, etc.) que el proceso de recogida de datos no siga los principios del muestreo aleatorio simple, que es supuesto por la regresión clásica por mínimos cuadrados. Esta situación ocurre cuando se ha hecho implícita o explícitamente un muestreo por agrupamientos, donde no sólo se ha seleccionado una serie de sujetos, sino también una serie de unidades contextuales a los que éstos pertenecen, tales como hospitales, clases, escuelas, municipios, empresas u otras instituciones. Los modelos multinivel son un caso particular de los Modelos Lineales generalizados, y se pueden aplicar a otros sis temas de análisis, además de la regresión lineal, como por ejemplo: al análisis de medidas repetidas, de datos categóricos, a los modelos de ecuaciones estructurales, a los procesos de series temporales, a la regresión no-lineal, etc.

RESUM EXTENS: clica aquí

Inici

COST ANALYSIS AND OPTIMAL SAMPLING DESIGN IN HYPOTHESIS TEST

CONVIDAT: Manuel Núñez. Department of Operations and Information Management School of Business. University of Connecticut.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Dilluns 17 de març de 2003, 12h30m
RESUM: The sample size determination problem for parameter estimation is typically posed with a statement such as: what sample size is needed if a statistician wants to be (1-\alpha) 100% confident of its point estimate being correct to within a given margin of error? In case of population mean estimation, the problem is straightforwardly handled by any standard applied statistics textbook. The sample size will be calculated using either the normal or t-student distribution. The particular distribution variable representing the quantity under study, and on whether \sigma, the population standard deviation of X, is known. In this case the solution is given by choosing any n such that
n >= z**2 \sigma**2 / e**2, (1) where z is the standardized normal variable associated with the confidence level \sigma and e is the acceptable interval error. The previous approach does not take into consideration the possible cost of collecting the sample. In addition, it seems reasonable that if the statistician is planning to sell the results, he can charge more for more precision e and more confidence 1 - \alpha in the report. Furthermore, even if the statistician’s fee structure is independent of e and \alpha, his reputation should be positively affected if he is known to give precise statements that have proven to be correct over time. Of course, the latter will be more likely to be true with \alpha small. Then a reasonable model for the determination of the sample size in this context is to minimize a cost function F(n, e, \alpha) subject to constraint (1).
RESUM EXTENS: clica aquí
ARTICLE COMPLET: clica aquí

Inici

THE K-CENTRUM SHORTEST PATH PROBLEM

CONVIDAT: Robert Garfinkel. Department of Operations and Information Management. School of Business. University of Connecticut.
IDIOMA: anglès
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres 21 de març de 2003, 12h30m
ARTICLE COMPLET: clica aquí

Inici

CLASIFICACIÓN NO PARAMÉTRICA MEDIANTE TÉCNICAS TIPO PROJECTION PURSUIT

CONVIDAT: Adolfo Hernández. University of Exeter. UK.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 28 de març de 2003, 12h30m
RESUM: Las reglas de clasificación no paramétricas tipo "kernel" (nucleo) presentan el inconveniente de la llamada maldición de la dimensionalidad, que provoca su mal funcionamiento en la práctica al exigir grandes tamaños muestrales cuando la dimensión del problema aumenta. Este trabajo propone dos métodos de reducción de la dimensión basados en técnicas tipo Projection Pursuit. Mediante la elección adecuada de funcionales sensibles al objetivo final de clasificación puede construirse una nueva regla kernel en un espacio de dimensión reducida. El buen comportamiento de esta regla se justifica tanto teóricamente como a través de su aplicación a conjuntos de datos reales y simulados, donde se establecen comparaciones con otros métodos propuestos en la literatura.
ARTICLE COMPLET: clica aquí

Inici

ESQUEMAS ALGORÍTMICOS GENERADORES DE HEURÍSTICAS

CONVIDAT: Julián Araoz.Department d'Estadística i Investigació Operativa. Universitat Politècnica de Catalunya / Universidad Simón Bolivar.

IDIOMA: Castellà

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 11 d'abril de 2003, 12h30m

RESUM: Se presentara A*, un esquema de algoritmo para caminos mínimos en grafos, que permite, mediante funciones estimativas, dirigir el comportamiento del algoritmo, así, si por ejemplo usamos como estimado 0 tendremos el conocido algoritmo de Dijtra, con otras funciones se obtienen los algoritmos de Camino Crítico, Bellman, Greedy, Simulation Anealing, etc. En grafos de grandes dimensiones ha sido utilizado como generador de métodos Heurísticos exitosos en problemas como Mantenimiento de Centrales Eléctricas, Colocación de Torres de Alta Tensión, Distribución de Horas Hombre en Pert/CPM , Asignación de Tripulaciones a Vuelos, Agente Viajero, etc.

Inici

POLÍTICA SANITARIA Y SOCIAL EN EUROPA Y EN FRANCIA: EL PAPEL DE LOS OBSERVATORIOS REGIONALES DE LA SALUD DE FRANCIA

CONVIDAT: André Ochoa. Observatoire régional de la santé d'Aquitaine.
IDIOMA:Castellà

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 25 d'abril de 2003, 12h30m

RESUM: Se presentarán: los “Observatorios regionales de Salud” franceses y su federacion, sus misiones y trabajos; un proyecto europeo sobre los indicadores de salud en las regiones de la Unión Europea y, finalmente, un caso concreto de planificación : las personas mayores, metodología de estudio y resultados.

Inici

OPTIMUM FUNDAMENTAL CYCLE BASIS PROBLEMS

CONVIDAT: Francesco Maffioli. DEI, Politecnico di Milano, Italy.
IDIOMA: Anglès
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 9 de maig de 2003, 12h30m
RESUM: Let G=(V,E) be a bi-connected graph with non-negative weights on its m edges. Let n be the number of vertices of G, i.e. the cardinality of set V. With respect to any spanning tree T of G we have a fundamental cycle basis of G formed by the m-n+1 fundamental cycles, each one corresponding to one of the co-tree edges. The characterisation of fundamental cycle basis is due to Syslo and the problem of determining one such basis minimising the sum of the length of its cycles has been proved NP-hard by Deo et alii. Applications exist in organic chemistry, biology and electronics. More recently Galbiati has proved that also the problem of determining a tree with respect to which the longest fundamental cycle is minimum is NP-hard. Another recent result proves that no PTAS (polynomial time approximation scheme) can exist for the sum problem unless P=NP. The talk is reviewing these and other complexity results for both problems. Then some formulations as 0-1 linear programming problems are presented and their suitability for a branch-and-bound approach is analysed. (Joint work with E.Amaldi, G.Galbiati and N.Maculan.)

Inici

STOCHASTIC INTEGER PROGRAMMING: WHY AND HOW?

CONVIDAT: Maarten van der Vlerk. Department of Econometrics and Operations Research. University of Groningen. Netherlands.
IDIOMA: Anglès
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Dimecres, 4 de JUNY de 2003, 12h30m
RESUM: Stochastic programming models arise as reformulations of optimization problems with uncertain data. We motivate and review examples of so-called recourse models, in particular for the case that integer decision variables are involved. Since such models are non-convex in general, we discuss algorithms based on convex approximations and other specific solution approaches.

Inici

REGRESIÓN LOGÍSTICA Y ANÁLISIS DISCRIMINANTE ROBUSTOS: UN ENFOQUE BAYESIANO

CONVIDAT: Javier Girón. Dept. de Estadística e Investigación Operativa. Universitat de Málaga.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 13 de juny de 2003, 12h30m
RESUM: La regresión binaria, mal llamada -a veces- regresión logística, habitualmente utiliza cuatro funciones de nexo, a saber: la "logística" (que da origen al modelo logístico, el más usual), la"normal" que da origen al modelo probit, habitualmente usado en bioensayo y, con menor frecuencia se utilizan los nexos "Loglog" y "Loglog complementario". Los modelos de regresión logística basados en estos nexos presentan el defecto de ser muy sensibles ante la presencia de observaciones atípicas o anómalas, es decir son "poco robustos". La idea de función de nexo robusta se justifica por la relación existente entre la regresión ordinaria y la binaria. En esta charla proponemos dos familias de funciones de nexo, que dan lugar a modelos de regresi&oa cute;n binaria más robustos, y aplicamos estos modelos a problemas de discriminación de dos o más poblaciones en pres encia de observaciones anómalas. El análisis bayesiano de estos modelos se realiza a través de métodos MCMC. Lo anterior se ilustra con varios ejemplos.
TRANSPARÈNCIES DE LA SESSIÓ: clica aquí

Inici

A CONVERGING STRATEGY TO GLOBAL MINIMIZATION

CONVIDAT: Ubaldo García Palomares. Professor visitant del Dept. Ingeniería de Telecomunicaciones. Universitat de Vigo.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 27 de juny de 2003, 12h30m
RESUM: Global minimization has been a topic that has attracted a lot of attention from both practitioners and researches in the optimization field. There is a vast bibliography available in specialized and applied journals. This talk presents an algorithm that hopefully finds a global minimum of an objective scalar function f(.):R^n->R subjected to bounds on the variables. A set of heuristics successfully used in evolutionary algorithms are embodied into locally converging derivative-free algorithm for box constrained minimization. In our approach a population P of individuals is characterized by its worst species value F, which is sufficiently improved in the next generation. This talk shows that the proposed algorithm inherits the convergence features of the derivative-free algorithm and discusses some appropiate heuristics to improve its performance when it searches for a global minimum. In short, the proposed algorithm: 1) Is easy to implement. 2) possesses guaranteed convergence to local minima of functions with directional derivatives f'(x,d) defined for x feasible and feasible direction d and 3) Can be conveniently adapted to a multi-processing environment.

 

Inici

THE EARLY DETECTION OF DISEASE AND STOCHASTIC MODELS

CONVIDAT: Marvin Zelen. Department of Biostatistics. Harvard University, USA.
IDIOMA: Anglès
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 19 de setembre de 2003, 12h30m
RESUM: Early detection of disease presents opportunities for using existing technologies to significantly improve patient benefit. The possibility of diagnosing a chronic disease early, while it is asymptomatic, may result in treating the disease in an earlier stage leading to better prognosis. Many cancers, diabetes, tuberculosis, cardiovascular disease, HIV related diseases, etc. may have better prognosis when combined with an effective treatment. However gathering scientific evidence to demonstrate benefit has proved to be difficult. Clinical trials have been arduous to carry out, because of the need to have large numbers of subjects, long follow-up periods and problems of non-compliance. Implementing public health early detection programs have proved to be costly and not based on analytic considerations. Many of these difficulties are a result of not understanding the early disease detection process and the disease natural histories. One way to approach these problems is to model the early detection process. This talk will discuss stochastic models for the early detection of disease. The talk will include : the role of length biased sampling, early prediction of clinical trials, the scheduling of early detection programs and the over diagnosis of disease. Breast cancer will be used to illustrate some of the ideas.

Inici

REMARKS ON LOCAL LIKELIHOOD DENSITY ESTIMATION

CONVIDAT: Pedro Delicado. Departament d'Estadística i Investigació Operativa. Univ. Politècnica de Catalunya.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 7 de novembre de 2003, 12h30m
RESUM: Local likelihood is recognized as a very successful and intuitively appealing method for nonparametric regression, but on the other hand the density estimation version of local likelihood is rarely used, in spite of the fact that theoretical properties are comparable to those in regression. It is our belief that the main reason for that is the lack of straightforward motivation for the formulas recently put forward by the literature in the topic of local likelihood density estimation. An alternative approach to achieve these formulas is considered in this paper. It is based on truncation arguments. Moreover, it is established that apparently different approaches to local likelihood density estimation are equivalent when the parametric model they are based on is closed for product by constants.
ARTICLE: clica aquí

Inici

SMALL-SAMPLE COMPARISONS FOR GOODNESS-OF-FIT STATISTICS IN ONE-WAY MULTINOMIALS WITH COMPOSITE HYPOTHESES

CONVIDAT: Vicente Núñez-Antón. Departamento de Econometría y Estadística. Universidad del País Vasco.
IDIOMA: Castellà
LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.
DATA: Divendres, 5 de desembre de 2003, 12h30m
RESUM: (Joint work with Miguel A. García-Pérez, Universidad Complutense). In this work, we evaluate the small-sample behavior of power-divergence goodness-of-fit statistics with composite hypotheses in multinomial models of up to five cells and up to three parameters. In addition, their performance was assessed by comparing asymptotic test sizes with exact test sizes obtained by enumeration in the near right tail, where (1-\alpha) \in (0.90, 0.95], and in the far right tail, where (1-\alpha) \in (0.95, 0.99]. Our study addresses all combinations of power-divergence estimates of indices \nu \in \{-1/2, 0, 1/3, 1/2, 2/3, 1, 3/2\} and power-divergence statistics of indices \lambda \in \{-1/2, 0, 1/3, 1/2, 2/3, 1, 3/2\}. The results we have obtained indicate that the asymptotic approximation is sufficiently accurate (by the criterion that the average exact size in no larger than +/-10% of the nominal asymptotic test size) in the near right tail when the power-divergence index estimate \nu= 0 and the power-divergence index statistics \lambda= 1/2 and, in the far right tail, when the power-divergence index estimate \nu= 0 and the power-divergence index statistics equals \lambda= 1/3, in both cases providing that the smallest expectation in the composite hypothesis exceeds five. The only exception to this rule is the case of models that render a near-equiprobable composite hypothesis on the boundaries of the parameter space, where average exact sizes are usually quite different from nominal sizes despite the fact that the smallest expectation in these conditions is usually well above five.
ARTICLE: clica aquí
Inici