Statistics and Operations Research Seminars 2010--2011,
Department of Economics and Business, Pompeu Fabra University
Schedule.
Thursday, September 30, 17:00, room 20.287.
Enrique Cabaña (Universidad de la República, Uruguay)
Consistent and focused tests based on transformations of processes with application to the comparison of AR(p) vs AR(p + 1) models for stationary time series
Abstract.
Friday, October 8, 15:00, room 20.237.
Nikos Demiris
Agricultural University of Athens
Survival Extrapolation with applications in health economics
Abstract.
Survival extrapolation appears implicitly or explicitly in several applications of medical statistics and health economics. In particular, the mean survival and other functions of the complete survivor curve are the object of interest in cost-effectiveness studies. In this talk we shall give a short introduction to survival analysis and subsequently present two distinct ways to extrapolate survival. The first method synthesizes evidence from different sources such as patient registries, population statistics and meta-analyses. The second approach is based on a flexible parametric model that retains a natural competing risks interpretation. Both methods are implemented in freely available software and will be illustrated using medical examples. This is joint work with Linda Sharples (Cambridge, UK).
Double feature special seminar!!
Thursday, November 11, 15:00, room 40S14
Ioannis Kosmidis, UCL
The reduction of bias in parametric estimation with emphasis on models for categorical responses.
Abstract:
For estimation in exponential family models, Kosmidis & Firth (2009, Biometrika) show how the bias of the maximum likelihood estimator may be reduced by an appropriate adjustment of the efficient score function. In this presentation some of the results of that study are presented along with a generic algorithm of easy implementation for reducing the bias in parametric estimation in general. Recent work is presented on the beneficial side-effects that bias reduction can have in the estimation of some well-used models with particular emphasis on models for categorical responses.
Cristiano Varin, Venice
Gaussian dependence models for non-Gaussian marginal regression
Abstract: This talk identifies and develops a comprehensive class of models for marginal regression analysis of non-normal dependent observations. The proposed class provides a natural extension of traditional linear regression models with normal correlated errors. Any kind of continuous, discrete and categorical responses is allowed. Dependence is conveniently modelled in terms of multivariate normal errors. The model class stems from and generalizes previous work on Gaussian copula regression models and multivariate probit models. Inference is performed through a likelihood approach. While the likelihood function is available in closed form for continuous responses, in the non-continuous setting numerical approximations are used. Residual analysis and a specification test are suggested for validating the adequacy of the assumed multivariate model. Methodology is implemented in a R package called mr. Illustrations include simulations and real data applications regarding time series, cross-design data, longitudinal studies, survival analysis and spatial regression. Joint work with Guido Masarotto, University of Padova (Italy).
Thursday, March 17, 12:00, room 40.S03.
Cluster Analysis Through Model Selection
ElĂas Moreno
Abstract: Clustering is an important and challenging statistical
problem for which there is an extensive literature. Modelling
approaches include mixture models and product partition models. Here
we develop a product partition model and search algorithm driven by
Bayes factors from intrinsic priors. The priors we develop for the
partitions, and the number of clusters in the partition, lead to
finding partitions with a smaller number of clusters, which does not
happen if a uniform prior is used. We argue that this is desirable,
since one reason for doing a cluster analysis is to find a small
number of clusters that can help to understand underlying
structure. However, we are also able to establish that our procedure
is consistent, and hence will find the true underlying structure as
the sample size increases. We illustrate our cluster algorithm with
both simulated and real examples.
Thursday, March 24, 12:00, room 20.287.
Ab Mooijaart (University of Leiden)
Moment Method (MM) versus Maximum likelihood (ML) estimates of structural equation models
with interaction terms
Abstract.
A standard assumption in structural equation models with interaction
terms is normality of predictors and latent variables. However, in
practice these predictors are seldom normally distributed. In this paper
two estimation methods are compared for these kinds of models: the
maximum likelihood (ML) method and a moment method (MM).
In MM the means, variances/covariances and a selection of third order
moments will be fitted. These third order moments will be formulated
in matrix notation. A characteristic of MM is that less assumptions
are made than in the ML method. This means that MM may not be
sensitive to violation of the assumption that the predictors are normally
distributed, whereas such a violation may have dramatic consequences
for the ML method.
In this talk more information about both estimation methods for
structural equation models with interaction terms will be given.
To illustrate the differences between the two estimation methods, the
results of a small Monte Carlo study are given. The conclusion of our
paper is that for practical research we strongly recommend to use some
MM method in models with interaction terms.
Thursday, May 19, 15:00, room 20.287.
Michael Wolf, Zurich
Nonlinear Shrinkage Estimation of Large-Dimensional Covariance Matrices
Abstract. Many applied problems require an estimate of a covariance matrix and/or its inverse. When the matrix dimension is large compared to the sample size, which happens fre- quently, the sample covariance matrix is known to perform poorly and may suffer from ill-conditioning. There already exists an extensive literature concerning improved estimators in such situations. In the absence of further knowledge about the structure of the true covariance matrix, the most successful approach so far, arguably, has been shrinkage estimation. Shrinking the sample covariance matrix to a multiple of the identity, by taking a weighted average of the two, turns out to be equivalent to linearly shrinking the sample eigenvalues to their grand mean, while retaining the sample eigenvectors. Our paper extends this approach by considering nonlinear transformations of the sample eigenvalues. We show how to construct an estimator which is asymptotically equivalent to an oracle estimator suggested in previous work. As demonstrated in extensive Monte Carlo simulations, the resulting bona fide estimator can result in sizeable improvements over the sample covariance matrix and also over linear shrinkage.
Thursday, June 9, 15:00, room 20.137.
Charles Bordenave, Toulouse
Spectrum of large non-hermitian random matrices
Abstract.
In this talk, we will revisit recent works of T. Tao, V. Vu and others
on large non-hermitian random matrices with independent and
identically distributed entries. We will explain the general strategy
behind this class of problems. As a byproduct, we will explain why the
generalized eigenvalues distribution of two independent matrices
converges almost surely to the uniform measure on the Riemann sphere.