Tuesday, January 27, 3.15 pm, 2015. Seminar in Mathematical Statistics.

Runs in coin tossing: a general approach for deriving distributions for functionals
Takis Konstantopoulos, Mathematical Statistics, Uppsala University.
Abstract: We take a fresh look at the classical problem of runs in a sequence of i.i.d. coin tosses and derive a general identity/recursion which can beused to compute (joint)distributions of functionals of run types. This generalizes and unifies already existing approaches. We give several examples, derive asymptotics, and pose some further questions.
Location: Hopningspunkten.

Tuesday, February 10, 3.15 pm, 2015. Seminar in Mathematical Statistics.

Extended GMANOVA model with a linearly structured covariance matrix
Joseph Nzabanita, Mathematical Statistics, Linköping University.
Abstract:We consider the problem of estimating a linearly structured covariance matrix in the extended GMANOVA (Generalized Multivariate Analysis of Variance) model. In the talk we will show how a decomposition of the residual space, the orthogonal complement to the design space, into m + 1 orthogonal subspaces and a study of residuals obtained from projections of observations on these subspaces yields explicit consistent estimators of the covariance matrix. An explicit consistent estimator of the mean will be also given.
Location: Hopningspunkten.

Tuesday, March 10, 3.15 pm, 2015. Seminar in Statistics.

Statistical Interpretation of DNA Mixtures
Therese Graversen, Dept. of Mathematical Sciences, University of Copenhagen.
Abstract: In this talk, I will discuss novel statistical and computational methodology for analysing mixed samples of DNA, where typical questions of interest evolve around determining the DNA profiles of the donors to a sample. In a recent discussion paper co-authored with R. Cowell, S. Lauritzen, and J. Mortera in Journal of the Royal Statistical Society, I propose a statistical framework, where the focus lies on establishing a joint statistical model for the unobservable DNA profiles and the observable set of peaks in the associated electropherogram. An important advantage of this approach is that a wide range of questions can be addressed entirely within this highly flexible framework, leading to a very consistent way of reasoning. Importantly, the approach enables a systematic assessment of the model. I will give some examples that illustrate the use of model checking methods in DNA mixture analysis. Concurrent development of methodology for efficient exact computations based on standard machinery for inference in Bayesian Networks has ensured that the approach is also practically feasible. A full implementation is available through the R package DNAmixtures.
Location: Alan Turing

Tuesday, March 24, 3.15 pm, 2015. Seminar in Mathematical Statistics.

The inverse first-passage problem and optimal stopping
Erik Ekström, Mathematical Statistics, Uppsala University.
Abstract: Given a survival distribution on the positive half-axis and a Brownian motion, a solution of the inverse first-passage problem consists of a boundary so that the first passage time over the boundary has the given distribution. We show that the solution of the inverse first-passage problem coincides with the solution of a related optimal stopping problem. Classical methods in optimal stopping theory thus may be applied in the study of the inverse first-passage problem.
Location: Hopningspunkten.

Tuesday, April 7, 3.15 pm, 2015. Seminar in Statistics.

Geostatistical modeling using Matérn fields driven by Levy noise
Jonas Wallin, Mathematical Sciences, Chalmers University.
Abstract:During the last ten years, "big data" has become the center of attention for the research in spatial statistics and geostatistics, as it has in many other scientific disciplines. Latent Gaussian models are predominant for that research, mainly because they are easy to work with. Still there are many geostatistical applications where Gaussian models are not suitable, and where this is a need for easy-to-use latent non-Gaussian models. I will present a type of latent non-Gaussian model generated through Stochastic partial differential equation (SPDE). Using Finite element methods approximation on the models, computationally efficient methods for parameter estimation and Kriging is obtained.
I will start by reviewing the SPDE approach from a Gaussian setting, and explain why it generates computationally efficient models. Then, I will show how to extend the methods to a class of Levy fields. And finally go over surprising results obtained when applying the models to a precipitation data set.
Location: Alan Turing

Tuesday, April 21, 3.15 pm, 2015. Seminar in Mathematical Statistics.

Hypothesis Testing of Patterned Covariance Matrices: Two Situations.
Yuli Liang, Statistics, Stockholm University.
Abstract: Testing in multivariate normal models with two types of patterned covariance matrices are considered.In the models with a covariance structure based on the Kronecker product of compound symmetry matrices, we explore possible methods to combine testing procedures based on a certain number of independent F-tests. To compare the performance of the new combined test procedures with the so-called higher-order accurate testing procedures, we compute the attained significance level and the empirical power. In the models with a block circular covariance structure, we consider various hypotheses concerning testing mean and covariance matrices. The corresponding likelihood ratio test statistics are derived and their distributions are studied.
Location: Hopningspunkten.

Tuesday, May 5, 3.15 pm, 2015. Seminar in Statistics.

Study of the Multivariate Structure of the Estonian Alchemilla L. (Rosaceae) Microspecies: an Example of the Structural Indices Approach
Tatjana von Rosen, Statistics, Stockholm University.
Abstract: Among the characteristics of a plant population, a particular value is vested in those remaining invariant in different ecological conditions and random environmental fluctuations. These features can be interpreted within the metapopulation concept (here, the metapopulation means a set of relatively isolated local populations). The genus Alchemilla L. consists of more than 1000 taxa, about 300 of which have been described in Europe. Because of the large variation, the genus has been an object of widespread scientific interest since the last century. The structural indices approach was used in the taxonomy of 23 microspecies of Alchemilla L. (occurring in the Estonian flora) to overcome the problem of statistical incorrectness caused by testing the objectivity of taxa applying the same variables as those used to define them. The questions should be answered are the following. How distinct are microspecies according to the metric and count variables? How do the structural indices distinguish microspecies? What are the most stable proportions between characteristic variables? Which characteristics are most informative in microspecies distinction?
Location: Alan Turing

Tuesday, May 19, 3.15 pm, 2015. Seminar in Statistics.

Hierarchical generalized linear models - a Lego approach to mixed models
Lars Rönnegård, Dalarna University and SLU Uppsala.
Abstract: The method of hierarchical generalized linear models (HGLM) fits generalized linear models with random effects and was introduced by Lee & Nelder (1996). It is based on the h-ikelihood and is a complete statistical framework including inference and model selection tools. The h-likelihood theory will be explained and in the presentation I also give a couple of examples from genetics where HGLM has been applied. The HGLM approach allows extended modelling in a building-block type of structure, like Lego. Together with my colleagues, I have implemented the HGLM method in the R package hglm (available on CRAN) and I will show some of its functionality. The examples and the presented theory are included in a book that I am writing on currently together with Youngjo Lee (Seoul National University) and Emmanuel Lesaffre (KU Leuven).
Location: Alan Turing

Thursday, June 4, 10.15 am, 2015. Seminar in Mathematical Statistics.

An Economic Evaluation of SPREAD on Rwanda's Rural Population
Alexandre Lyambaje, University of Rwanda.
Abstract: Sustainable Partnerships to Enhance Rural Enterprises and Agricultural Development (SPREAD) was a program to enhance the value chain for commodities in Rwanda including coffee. The implicit concept was that improving the value chain would increase the incomes for smallholders and, hence, reduce the poverty rate. The results indicate that Rwanda coffee prices increased relative to an index price for traded coffee with the implementation of SPREAD. In addition, the results indicate that participation in the coffee market at this time was associated with higher household income and lower rates of poverty.
Location: Hopningspunkten.

Wednesday, June 10, 10.15 am, 2015. Seminar in Mathematical Statistics.

Optimal Estimation for Doubly Multivariate Data in Blocked Compound Symmetric Covariance Structure
Miguel Fonseca, Center of Mathematics and Applications, NOVA University of Lisbon.
Abstract: The paper deals with the best unbiased estimators of the blocked compound symmetric covariance structure for m-variate observations over u sites under the assumption of multivariate normality. The free-coordinate approach is used prove thay the quadratic estimation of covariance parameters is equivalent to linear estimation with a properly defined inner product in the space of symmetric matrices. Complete statistics are then derived to prove that the estimators are best unbiased. Finally, strong consistency is proven.
Location: Hopningspunkten.