Events
Thu 17 Jan, '08- |
CRiSM SeminarDr Elena Kulinskaya, Statistical Advisory Service, Imperial College |
|
Thu 24 Jan, '08- |
CRiSM SeminarA1.01Dr Richard Samworth, Statistical Laboratory, Cambridge We show that if
$X_1,...,X_n$ are a random sample from a log-concave density $f$ in
$\mathbb{R}^d$, then with probability one there exists a unique maximum
likelihood estimator $\hat{f}_n$ of $f$. The use of this estimator is
attractive because, unlike kernel density estimation, the estimator is fully
automatic, with no smoothing parameters to choose. The existence proof is
non-constructive, however, and in practice we require an iterative algorithm
that converges to the estimator. By reformulating the problem as one of
non-differentiable convex optimisation, we are able to exhibit such an
algorithm. We will also show how the method can be combined with the EM
algorithm to fit finite mixtures of log-concave densities. The talk will be
illustrated with pictures from the R package LogConcDEAD. |
|
Thu 31 Jan, '08- |
CRiSM SeminarA1.01Dr Robert Gramacy, Statistical Laboratory Cambridge |
|
Thu 14 Feb, '08- |
CRiSM SeminarA1.01Professor Simon Wood, University of Bath |
|
Mon 18 Feb, '08- |
CRiSM SeminarA1.01Terry Speed, University of California, Berkeley |
|
Thu 28 Feb, '08- |
CRiSM SeminarA1.01Alexey Koloydenko & Juri Lember (Joint Talk), University of Nottingham |
|
Thu 6 Mar, '08- |
CRiSM SeminarA1.01Dr Cliona Golden, UCD, Dublin
On the validity of ICA for fMRI data Functional Magnetic Resonance Imaging (fMRI) is a brain-imaging technique which, over time, records changes in blood oxygenation level that can be associated with underlying neural activity. However, fMRI images are very noisy and extracting useful information from them calls for a variety of methods of analysis. I will discuss the validity of the use of two popular Independent Component Analysis (ICA) algorithms, InfoMax and FastICA, which are commonly used for fMRI data analysis. Tests of the two algorithms on simulated, as well as real, fMRI data, suggest that their successes are related to their ability to detect "sparsity" rather than the independence which ICA is designed to seek. |
|
Thu 13 Mar, '08- |
CRiSM SeminarA1.01Prof Antony Pettitt, Lancaster University
|
|
Thu 3 Apr, '08- |
CRiSM SeminarA1.01Professor Jay Kadane, Carnegie Mellon University
Driving While Black: Statisticians Measure Discriminatory Law Enforcement (joint work with John Lamberth) The US Constitution guarantees "equal protection under the law" regardless of race, but sometimes law enforcement practices have failed to adhere to this standard.In the 1990's, a suit was brought alleging that the New Jersey State Police were stopping Blacks at disproportionately high rates in the southern end of the New Jersey Turnpike. In this talk I * review the evidence in that case, the decision, and its immediate aftermath * discuss criticisms of that decision * examine new evidence that rebuts those criticisms * comment on the extent to which the Constitutional standard is now being met. |
|
Thu 1 May, '08- |
CRiSM SeminarA1.01Alastair Young, Imperial College London |
|
Thu 8 May, '08- |
CRiSM SeminarA1.01Cees Diks, University of Amsterdam
Linear and Nonlinear Causal Relations in Exchange Rates and Oil Spot and Futures Prices Various tests have been proposed recently in the literature for detecting causal relationships between time series. I will briefly review the traditional linear methods and some more recent contributions on testing for nonlinear Granger causality. The relative benefits and limitations of these methods are then compared in two different case studies with real data. In the first case study causal relations between six main currency exchange rates are considered. After correcting for linear causal dependence using VAR models there is still evidence presence for nonlinear causal relations between these currencies. ARCH and GARCH effects are insufficient to fully account for the nonlinear causality found. The second case study focuses on nonlinear causal linkages between daily spot and futures prices at different maturities of West Texas Intermediate crude oil. The results indicate that after correcting for possible cointegration, linear dependence and multivariate GARCH effects, some causal relations are still statistically significant. In both case studies the conclusion is that non-standard models need to be developed to fully capture the higher-order nonlinear dependence in the data. |
|
Thu 22 May, '08- |
CRiSM SeminarA1.01Thomas Richardson, University of Washington I will first review well-known differences between odds ratios, relative risks and risk differences. These results motivate the development of methods, analogous to logistic regression, for estimating the latter two quantities. I will then describe simple parametrizations that facilitate maximum-likelihood estimation of the relative risk and risk-difference. Further, these parametrizations allow for doubly-robust g-estimation of the relative risk and risk difference. (Joint work with James Robins, Harvard School of Public Health). |
|
Thu 29 May, '08- |
CRiSM SeminarA1.01Geoff McLachlan, University of Queensland An important problem in microarray experiments is the detection of genes that are differentially expresse in a given number of classes. As there are usually thousands of genes to be considered simultaneously, one encounters high-dimensional testing problems. We provide a straightforward and easily implemented method for estimating the posterior probability that an individual gene is null (not differentially expressed). The problem can be expressed in a two-component mixture framework. Current methods of implementing this approach either have some limitations due to the minimal assumptions made or with the computationally intensive nature of more specific assumptions. By converting to a z-score the value of the test statistic used to test the significance of each gene, we propose a simple two-component normal mixture that models adequately the distribution of this score. The approach provides an estimate of the local false discovery rate (FDR) for each gene, which is taken to be the posterior probability that the gene is null. Genes with the local FDR less than a specified threshold C are taken to be differentially expressed. For a given C, this approach also provides estimates of the implied overall errors such as the (global) FDR and the false negative/positive rates. |
|
Thu 5 Jun, '08- |
CRiSM SeminarA1.01Thomas Nichols, GlaxoSmithKline Clinical Imaging Centre |
|
Thu 12 Jun, '08- |
CRiSM SeminarA1.01Jonathan Dark, University of Melbourne
Dynamic hedging with futures that are subject to price limits The standard approaches to estimating minimum variance hedge ratios (MVHRs) are mis-specified when futures prices are subject to price limits. This paper proposes a bivariate tobit-FIGARCH model with maturity effects to estimate dynamic MVHRs using single and multiple period approaches. Simulations and an application to a commodity futures hedge support the proposed approach and highlight the importance of allowing for price limits when hedging. |
|
Thu 19 Jun, '08- |
CRiSM SeminarA1.01Ian Dryden, University of Nottingham |
|
Thu 26 Jun, '08- |
CRiSM SeminarA1.01Gersende Fort, ENST (Ecole Nationale Superieure Des Telecommunications, France
Stability of Markov Chains based on fluid limit techniques. Applications to MCMC We propose a transformation of some Markov chains which will allow us to define its fluid limit: by renormalization in time, space, and initial value of the chain, we exhibit a time-continuous process which governs the dynamic of the initial chain. The goal is to identify the quantities that govern the ergodic behavior of the Markov chain, by showing their impact on the dynamics of the associated fluid process which, by definition, gives information on the transient steps of the chain. We will consider applications of these techniques to the choice of the design parameters of some MCMC samplers. |
|
Mon 14 Jul, '08- |
CRiSM SeminarA1.01Prof Donald Martin, North Carolina State University
Markov chain pattern distributions We give a method for predicting statistics of hidden state sequences, where the conditional distribution of states given observations is modeled by a factor graph with factors that depend on past states but not future ones. Model structure is exploited to develop a deterministic finite automaton and an associated Markov chain that facilitates efficient computation of the distributions. Examples of applications of the methodology are the computation of distributions of patterns and statistics in a discrete hidden state sequence perturbed by noise and/or missing values, and patterns in a state sequence that serves to classify the observations. Two detailed examples are given to illustrate the computational procedure. |
|
Mon 25 Aug, '08- |
Economics/Stats SeminarS2.79Donald Rubin (Harvard)
For Objective
Causal Inference, Design Trumps Analysis
For obtaining causal inference that are
objective, and therefore have the best chance of revealing scientific truths,
carefully designed and executed randomized experiments are generally
considered to be the gold standard. Observational studies, in contrast,
are generally fraught with problems that compromise any claim for objectivity
of the resulting causal inferences. The thesis here is that
observational studies have to be carefully designed to approximate randomized
experiments, in particular, without examining any final outcome data.
Often a candidate data set will have to be rejected as inadequate because of
lack of data on key covariates, or because of lack of overlap in the
distributions of key covariates between treatment and control groups, often
revealed by careful propensity score analyses. Sometimes the template
for the approximating randomized experiment will have to be altered, and the
use of principal stratification can be helpful in doing this. These
issues are discussed and illustrated using the framework of potential
outcomes to define causal effects, which greatly clarifies critical
issues. |
|
Tue 9 Sep, '08- |
CRiSM SeminarA1.01Professor P Cheng, Academia Sinica, Taipei, Republic of China
Linear Information Models and Applications Log-likelihood information identities and Venn diagrams for categorical data exhibit fundamental differences from those of continuous variables. This presentation will start with three-way contingency tables and the associated likelihood ratio tests. It will introduce linear information models that deviate from hierarchical log-linear models, beginning with three-way tables. A connection to latent class analysis with two-way tables and the geometry of the one-degree-freedom chi-square test and exact test for two-way independence is also investigated. Key Names: Pearson; Fisher; Neyman and Pearson; Kullback and Leibler; Cochran, Mantel, and Heanszel; Goodman. Co-authors: John A. D. Aston, Jiun W. Liou, and Michelle Liou. |
|
Mon 22 Sep, '08- |
CRiSM SeminarA1.01Dr Jonathan Evans, Institute of Linguistics, Academia Sinica
Statistical Modelling in Linguistics: Approaches and Challenges in pitch analysis
This talk introduces the use of Linear Mixed Effects (LME) analysis to model f0 (pitch) production in a language with two tones, and demonstrates the advantages of using such a method of analysis. LME can be used to weigh the impact of a large number of effects, it can demonstrate the interaction among those effects, and can also show how both fixed and random effects contribute to the model. Unlike previous analytical methods for modeling f0 in tone languages, LME analysis allows researchers to have more freedom in designing experiments, and to have sufficient variety in the dataset without having to rely on nonsense words and phrases to fill out a data matrix. LME makes it is possible to put a multitude of effects and interactions into a single comprehensive model of f0. The ensuing model is easy to interpret and straightforward to compare crosslinguistically. LME analysis makes possible a quantitative typology that shows clearly how linguistic and nonlinguistic factors combine in the production of f0 for each language thus analyzed. The talk will also veer into discussion of how to model f0 based on the pitch curve of each syllable. Although each curve contains an infinite number of points, there is striking similarity between the curve-based model and the point-based model. |
|
Thu 2 Oct, '08- |
CRiSM Seminar: Martin BaxterA1.01Dr Martin Baxter, Nomura International
Levy Modelling of Credit This talk will start with some simple models of credit dynamics, and embed them in a general Levy process framework. A particular instance, the Gamma process, will then be studied with reference to both its theoretical and practical properties. A brief analysis of the ongoing credit crisis in terms of Levy modelling. Time permitting, we will also look at some other applications.
|
|
Thu 23 Oct, '08- |
Oxford-Warwick Joint SeminarOxford
3rd JOINT WARWICK-OXFORD STATISTICS SEMINAR 2:30 – 5:00 pm at The Mary Ogilvie Lecture Theatre, St. Anne’s College, University of Oxford 2:30 p.m. Speaker 1: Julian Besag (University of Bath, University of Washington, Seattle) Title:
Continuum limits of Gaussian Markov random fields: resolving the
conflict with geostatistics MRFs refer to fixed regular or irregular discrete lattices or arrays and questions arise regarding inconsistencies between MRFs specified at differing scales, especially for regional data. Ideally, one would often prefer an underlying continuum formulation, as in geostatistics, which can then be integrated to the regions of interest. However, limiting continuum versions of MRFs, as lattice spacing decreases, proved elusive until recently. This talk briefly presents some motivating examples and shows that limiting processes indeed exist but are defined on arbitrary regions of the plane rather than pointwise. Especially common is the conformally invariant de Wijs process, which coincidentally was used originally by mining engineers but which became unfashionable as geostatistics developed. Convergence is generally very fast. The de Wijs process is also shown to be a natural extension of Brownian motion to the plane. Other processes, including the thin-plate spline, can be derived as limits of MRFs. The talk closes by briefly discussing data analysis.
ÍÎ
3:30 to 4.00 - Tea, Coffee and biscuits in foyer outside lecture theatre
4:00 p.m. Speaker 2: Susan Lewis (University of Southampton, UK) Title:
Screening experiments |
|
Thu 30 Oct, '08- |
CRiSM Seminar: Christopher SherlockA1.01Dr Christopher Sherlock
Optimal scaling of the random walk Metropolis University of Lancaster
Abstract: The random walk Metropolis (RWM) is one of the most commonly used Metropolis-Hasting algorithms, and choosing the appropriate scaling for the proposal is an important practical problem. Previous theoretical approaches have focussed on high-dimensional algorithms and have revolved around a diffusion approximation of the trajectory. For certain specific classes of targets it has been possible to show that the algorithm is optimal when the acceptance rate is approximately
0.234. We develop a novel approach which avoids the need for diffusion limits. Focussing on spherically symmetric targets, it is possible to derive simple exact formulae for efficiency and acceptance rate for a "real" RWM algorithm, as opposed to a limit process. The limiting behaviour of these formulae can then be explored. This in some sense "simpler" approach allows important general intuitions as to when and why the 0.234 rule holds, when the rule fails, and what may happen when it does fail. By extending the theory to include elliptically symmetric targets we obtain further intuitions about the role of the proposal's shape. |
|
Thu 6 Nov, '08- |
CRiSM Seminar: Ming-Yen ChengA1.01Prof Ming-Yen Cheng UCL |
|
Wed 12 Nov, '08- |
CRiSM Seminar - John CussensA1.01, Zeeman BuildingJames Cussens, University of York Model Selection using weighted MAX-SAT Solvers This talk concerns encoding problems of statistical model selection in such a way that "weighted MAX-SAT solvers" can be used to search for the 'best' model. In this approach each model is (implicitly) encoded as a joint instantiation of n binary variables. Each of these binary variables encodes the truth/falsity of a logical proposition and weighted logical formulae are used to represent the model selection problem. Once encoded in this way we can tap into years of research and use any of the state-of-the-art solvers to conduct the search. In the talk I will show how to use this approach when the model class is that of Bayesian networks, and also for clustering. I will briefly touch on related methods which permit the calculation of marginal probabilities in discrete distributions. |
|
Wed 12 Nov, '08- |
Joint Stats/Econometrics Seminar: Sylvia Fruehwirth-SchnatterA1.01, Zeeman BuildingProf Sylvia Fruehwirth-Schnatter, Johannes Kepler University, Austria
Latent Variable models are widely used in applied statistics and econometrics to deal with data where the underlying processes change either over time or between units. Whereas estimation of these models is well understood, model selection problems are rarely studies, because such an issue usually leads to a non-regular testing problem. Bayesian statistics offers in principle a framework for model selection even for non-regular problems, as is shortly discussed in the first part of the talk. The practical application of the Bayesian approach, however, proves to be challenging and numerical technique like marginal likelihoods, RJMCMC or the variable selection approach have to be used. The main contribution of this talk is to demonstrate that the Bayesian variable selection approach is useful far beyond the common problem of selecting covariates in a classical regression model and may be extended to deal model selection problems in various latent variable models. First, it is extended to testing for the presence of unobserved heterogeneity in random effects models. Second, dynamic regression models are considered, where one has to choose between fixed and random coefficients. Finally, the variable selection approach is extended to state space models, where testing problems like discriminating between models with a stochastic trend, a deterministic trend and a model without trend arise. Case studies from marketing, economics and finance will be considered for illustration. |
|
Thu 27 Nov, '08- |
CRiSM Seminar - Anthony ReveillacA1.01Anthony Reveillac
Humboldt University - Berlin
Stein estimators and SURE shrinkage estimation for Gaussian processes using the Malliavin calculus |
|
Fri 12 Dec, '08- |
Joint CRiSM/Applied Maths/Stats SeminarA1.01Professor Jeff Rosenthal, University of Toronto
Adaptive MCMC |
|
Thu 22 Jan, '09- |
CRiSM SeminarRoom A1.01, Zeeman BuildingDr Laura Sangalli, Politecnico di Milano Title : Efficient estimation of curves in more than one dimension by free-knot regression splines, with applications to the analysis of 3D cerebral vascular geometries. Abstract : We deal with the problem of efficiently estimating a 3D curve and its derivatives, starting from a discrete and noisy observation of the curve. We develop a regression technique based on free-knot splines, ie. regression splines where the number and position of knots are not fixed in advance but chosen in a way to minimize a penalized sum of squared errors criterion. We thoroughly compare this technique to a classical regression method, local polynomial smoothing, via simulation studies and application to the analysis of inner carotid artery centerlines (AneuRisk Project dataset). We show that 3D free-knot regression splines yield more accurate and efficient estimates. Joint work with Piercesare Secchi and Simone Vantini. |