Department Seminars 1999
Smoothing Conference 1999 Website Archive| DATE/PLACE: | Thursday, December 2, 1999, 16:00 CANCELLED |
| DATE/PLACE: | Thursday, November 25, 1999, 16:00 Leonard S. Klinck 301, (formerly Computer Science 301), 6356 Agricultural Road |
| TYPE: | Research Seminar |
| TITLE: | Modelling the progression of Parkinson's disease (PD) |
| SPEAKER: | Michael Schulzer Department of Statistics UBC |
| The time course of evolution of clinical deficits has been a traditional guide to the understanding of the aetiopathogenesis (ie causes and development) of neurological disease. We analyzed the influence of aging and of duration of disease on the natural history of PD from cross-sectional data on 238 patients. We concluded that age and disease duration exert independent, additive effects on the progression of the disease, and that the rate of neuronal death is more rapid in the earlier stages of evolution of the pathology, and subsequently slows down to approach the rate of attrition produced by normal aging. We used our observations to develop a mathematical model of the temporal profile of neurodegeneration in PD. Our model indicates that the hypothesis of an event that kills some neurons and damages others in such a way that their life expectation is reduced is consistent with our data. It also enables us to extrapolate back to estimate when the causal event occurred, and explains why PD proceeds more rapidly in older patients.
|
|
| DATE/PLACE: | Tuesday, November 30, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Online Prediction with Experts, Coding Theory, Model Selection Principles Under a Log Scoring Rule |
| SPEAKER: | Bertand Clarke Department of Statistics UBC |
| Here we identify a class of problems based on the log score function. This admits an information theoretic motivation and we propose solutions in a Bayesian context. We show that the Bayesian mixture density, the marginal density for the data, is an asymptotically optimal solution valid for members of this class. As part of this, we identify three members of the class of objective functions which we argue are close and conceptually related. One member of this class, an empirical version, is canonical in an invariance sense. More generally, we show senses in which the various objective functions in the class, their optimal values, and the resulting solutions are close. In addition, we characterize the difference between these Shtarkov solutions and the Bayesian mixture. | |
| PACIFIC NORTHWEST STATISTICS CONFERENCE | Friday, November 19, 1999 University of Victoria See the conference web site for more information. |
| DATE/PLACE: | Tuesday, November 23, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Bayesian Cross-Validation Choice and Assessment of Statistical Models |
| SPEAKER: | Fatemah Alqallaf Department of Statistics UBC |
| This talk will be concerned with application of a cross-validation criterion to the choice and assessment of statistical models, in which observed data are partitioned, with one part of the data compared to predictions conditional on the model and the rest of the data.
We develop three methods, gold, silver, and bronze based on the idea of splitting data in the context of measuring prediction error; however, they can also be adapted for model checking. The gold method uses analytic calculations for the posterior predictive distribution; however, the silver method avoids this mathematical intensity, instead simulating many posterior samples, and the bronze method reduces the amount of sampling to speed up computation. We also consider the Bayesian p-value in which the posterior distribution can be used to check model adequacy, in the context of cross-validation with repeated data splitting. Application to examples is detailed, using the discussed methodologies of estimation and prediction. |
|
| DATE/PLACE: | Tuesday, November 16, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Efficient Estimation in Two-phase Sampling Designs |
| SPEAKER: | Brad McNeney Department of Statistics Simon Fraser University |
| Two-phase or response-selective sampling designs can maximize the precision of inference in an epidemiologic study, while substantially reducing costs. The case-control design, in which cases are oversampled relative to their overall population frequency, may be viewed as an example, and is known to be far more efficient than a cohort study when the disease is rare. More generally, inexpensive covariates can be measured on all subjects. Then, in a second phase, those with rare or interesting exposures can be oversampled for a more informative and costly covariate. The natural models associated with these designs are semiparametric. There are a finite number of regression parameters which relate the response to the covariates, but the sampling distribution of covariates is typically left unspecified. A variety of methods have been proposed for estimating the regression parameters, most of which are inefficient. Recent attention has focused on maximum likelihood estimation, which is efficient. In this talk I will discuss large sample properties and efficiency of maximum likelihood, and the implications for study design. | |
| DATE/PLACE: | Tuesday, November 9, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Disequilibrium Likelihoods for Fine-Scale Mapping of a Rare Allele |
| SPEAKER: | Jinko Graham Department of Statistics Simon Fraser University |
| Genetic linkage studies based on pedigree data have limited resolution, due to the relatively small number of segregations. Disequilibrium mapping, which uses population associations to infer the location of a disease mutation, provides one possible strategy for narrowing the candidate region. Recombination events between a disease locus and marker locus may be placed on the ancestry of a nonrecurrent disease mutation. These events define the recombinant classes, the sets of sampled disease copies descending from the meiosis at which a given recombination occurred. We show how Monte Carlo generation of the recombinant classes leads to a linkage likelihood for fine-scale mapping from disease haplotypes. We compare single-marker disequilibrium mapping with interval disequilibrium mapping. The method and its properties are illustrated with an example of simulated data, constructed to be typical of fine-scale mapping of a rare disease in the Japanese population. | |
| DATE/PLACE: | Tuesday, November 2, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Smoothing Data as a Time Series Problem |
| SPEAKER: | Piet de Jong Faculty of Commerce, UBC |
| Much of statistics can be viewed as "smoothing", that is purging the observed data of transient or uninteresting features so as to leave what is more or less permanent and interesting. The more permanent features can be extrapolated beyond the observed data.
This paper shows how smoothing is essentially a "time series" problem. Four examples taken from non time series areas are used to illustrate the methods. It is argued that time series methods have relevance to all smoothing problems. |
|
| DATE/PLACE: | Thursday, October 28, 1999, 16:00 Leonard S. Klinck 301, (formerly Computer Science 301), 6356 Agricultural Road |
| TYPE: | Research Seminar |
| TITLE: | Improving Generalized Estimating Equations Using Quadratic Inference Functions |
| SPEAKER: | Annie Qu Department of Statistics Oregon State University |
| The generalized estimating equations (GEE) method (Liang & Zeger, 1986) is widely used in longitudinal data analysis. It enables one to estimate regression parameters consistently even when the correlation structure is misspecified. However, under such misspecification, the estimator of the regression parameter can be inefficient. We introduce a new approach based on quadratic inference functions (QIF) derived from the generalized method of moments (Hansen, 1982). This approach does not involve the direct estimation of the correlation parameter, and remains optimal even if the working correlation structure is misspecified. The idea in the construction of the quadratic inference function is to represent the inverse of the working correlation matrix by the linear combination of basis matrices, which is possible for the working correlations most commonly used. Both asymptotic theory and simulation results show that under misspecified working assumptions these estimators are more efficient than those of the GEE. Another advantage of this approach is that it can be used to construct a chi-square decomposition for testing of nested models. It also provides a regression model misspecification test as a bonus. It is particularly important that the test statistics follow a chi-squared distribution asymptotically whether or not the working correlation structure is correctly specified. We use published biomedical data from children's respiratory illness and mothers' smoking habits (Laird et al., 1984) as an application for binary outcomes, and studies on the safety of the drug chenodiol for the treatment of cholesterol gallstones (Wei, 1988) as an example for continuous outcomes.
* This is joint work with Bruce Lindsay and Bing Li.
|
|
| DATE/PLACE: | Tuesday, October 26, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Stochastic Cointegration I: Estimation and Inference |
| SPEAKER: | Brendan McCabe Department of Commerce UBC |
| This paper provides an alternative framework for analysing relationships between nonstationary time series to that suggested by Engle and Granger (1987). We define a new class of nonstationary time series, called stochastically integrated processes, which are more volatile than conventionally integrated processes. We then introduce the concept of stochastic cointegration to capture interrelationships between such processes or, more generally, between stochastically and conventionally integrated processes. The requirements for stochastic cointegration are rather weaker than for conventional cointegration, as it based only on mean reversion rather than stationarity, thus allowing for more volatile interrelationships. Within this framework, we consider the appropriate analog of the cointegrating vector of conventional analysis, called the stochastically cointegrating vector, and show that this vector cannot be estimated consistently using standard OLS techniques. We propose an IV estimator that is consistent and establish asymptotic normality of statistics for testing hypotheses about the stochastically cointegrating vector. The finite sample properties of the test statistics are examined via Monte Carlo simulation and we provide an empirical application of our techniques based on the purchasing power parity hypothesis (PPP). | |
| DATE/PLACE: | Tuesday, October 12, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | A Bayesian Approach to Fuzzy Hypotheses Testing |
| SPEAKER: | S. Mahmoud Taheri Department of Statistics Shiraz University, Iran |
| In statistical decisions we may come across with imprecise (fuzzy) concepts. One important case is a situation where we are interested in testing the hypotheses that are fuzzy rather than crisp; such as: the mean is approximatly 10, the mean is very large, and so on.
In this seminar, we consider the problem of fuzzy hypotheses testing on the basis of a Bayesian approach. |
|
| DATE/PLACE: | Tuesday, October 5, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Fuzzy Sets (An Introduction) |
| SPEAKER: | S. Mahmoud Taheri Department of Statistics Shiraz University, Iran |
| A Fuzzy set is a generalization of idea of ordinary (crisp) set. The concept of fuzzy set was first introduced by, an Iranian scientist, L.A. Zadeh in 1965. It is a mathematical tool for representation, modeling, and analyzing of imprecise Knowledges.
Fuzzy set theory provides a strict mathematical framework (there is nothing fuzzy about fuzzy set theory!) in which vague concepts can be precisely and rigorously studied. In this seminar, some elementary concepts will be intoduced. Also we will explain how the fuzzy concepts can enter the world of statistics and probability. |
|
| DATE/PLACE: | Thursday, September 30, 1999, 16:00 Leonard S. Klinck 301, (formerly Computer Science 301), 6356 Agricultural Road |
| TYPE: | Research Seminar |
| TITLE: | Neutralizing Antibodies and the Efficacy of Interferon Beta-1b in Relapsing-Remitting Multiple Sclerosis |
| SPEAKER: | John Petkau & Rick White Department of Statistics UBC |
| Several large multi-center clinical trials completed over the last five years or so have established that Type I interferons favorably influence clinical and MRI outcomes in both relapsing-remitting and secondary-progressive multiple sclerosis (MS). Some patients develop neutralizing antibodies (NABs) to these treatments, reflecting an immune system response. The clinical significance of these NABs has been uncertain as titers vary widely and elevated NAB titers often diminish to undetectable levels. The question of whether NABs diminish the efficacy of these treatments is an unresolved scientific issue directly related to the question of how MS patients should be treated. It has also become a part of the marketing strategy of pharmaceutical companies with different Type I interferons approved for the treatment of MS: hundreds of millions of dollars per year are at stake.
In this presentation, we will discuss our involvement with this issue for the pivotal trial of interferon beta-1b in relapsing-remitting MS. We will describe the original cross-sectional analyses which initially raised the concern, and the subsequent analyses we have carried out attempting to resolve this issue. A fascinating part of our involvement has been attempting to persuade a neurological audience of the need for a more detailed statistical analysis than is customary in the field. We are interested in reactions to the general approach we have taken and suggestions of other approaches that might be utilized.
|
|
| DATE/PLACE: | Tuesday, September 28, 1999, 16:00 Leonard S. Klinck 301 (formerly Computer Science 301), 6356 Agricultural Road |
| TITLE: | Log Hazard Regression |
| SPEAKER: | Huiying Sun Department of Statistics UBC |
| We propose using regression splines to estimate the two log marginal hazard functions of bivariate survival times, where each time could be censored. The method is a modified version of Kooperberg, Stone and Truong's (JASA, 1995) hazard regression for estimating a univariate survival time. We derive an approach to find standard errors for estimates of the difference of the log hazard functions. The approach is inspired by Wei, Lin, and Weissfeld (JASA, 1989).
We also propose procedures for testing the four hypotheses that the marginals follow an exponential or Weibull distribution and that the two failure times have the same distribution or have proportional hazards. A simulation study is conducted to assess the performance of our estimates and test procedures. We study the effects of the censoring rates, correlation levels, and number of knots. The regression is applied to the data set of the Diabetic Retinopathy Study (Diabetic Retinopathy Study Research Group, 1981). Our analysis for the data set matches study results of Huster, Brookmeyer , and Self (1989). |
|
| DATE/PLACE: | Wednesday, September 22, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TITLE: | Improved Estimation of Location Parameters: Part 3 of 3 |
| SPEAKER: | STATISTICAN IN RESIDENCE Professor William E. Strawderman Department of Statistics Rutgers University Professor Stawderman's visit is supported by the Constance van Eeden Fund |
| We consider estimation of a p-dimensional location vector where performance is measured via the expected sum of squared errors. For much of the talk the distribution will be assumed to be spherically symmetric and it will be assumed that a residual vector is present so that scale can be estimated. We will give a number of heuristic arguments that so called shrinkage (Stein) estimators can be expected to out-perform the usual (least squares) estimator under minimal assumptions. We will review known results for the case of a normal distribution and demonstrate that most of these results have an extension to the general spherically case.
In a series of 3 follow-up lectures I will go into more detail on general results for spherically symmetric distributions and discuss related problems. Professor Strawderman's visit is supported by the Constance van Eeden Statistician in Residence program. |
|
| DATE/PLACE: | Tuesday, September 21, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TITLE: | Improved Estimation of Location Parameters: Part 2 of 3 |
| SPEAKER: | STATISTICAN IN RESIDENCE Professor William E. Strawderman Department of Statistics Rutgers University Professor Stawderman's visit is supported by the Constance van Eeden Fund |
| We consider estimation of a p-dimensional location vector where performance is measured via the expected sum of squared errors. For much of the talk the distribution will be assumed to be spherically symmetric and it will be assumed that a residual vector is present so that scale can be estimated. We will give a number of heuristic arguments that so called shrinkage (Stein) estimators can be expected to out-perform the usual (least squares) estimator under minimal assumptions. We will review known results for the case of a normal distribution and demonstrate that most of these results have an extension to the general spherically case.
In a series of 3 follow-up lectures I will go into more detail on general results for spherically symmetric distributions and discuss related problems. Professor Strawderman's visit is supported by the Constance van Eeden Statistician in Residence program. |
|
| DATE/PLACE: | Thursday, September 16, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TITLE: | Improved Estimation of Location Parameters: Part 1 of 3 |
| SPEAKER: | STATISTICAN IN RESIDENCE Professor William E. Strawderman Department of Statistics Rutgers University Professor Stawderman's visit is supported by the Constance van Eeden Fund |
| We consider estimation of a p-dimensional location vector where performance is measured via the expected sum of squared errors. For much of the talk the distribution will be assumed to be spherically symmetric and it will be assumed that a residual vector is present so that scale can be estimated. We will give a number of heuristic arguments that so called shrinkage (Stein) estimators can be expected to out-perform the usual (least squares) estimator under minimal assumptions. We will review known results for the case of a normal distribution and demonstrate that most of these results have an extension to the general spherically case.
In a series of 3 follow-up lectures I will go into more detail on general results for spherically symmetric distributions and discuss related problems. Professor Strawderman's visit is supported by the Constance van Eeden Statistician in Residence program. |
|
| DATE/PLACE: | Tuesday, September 14, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TITLE: | Improved Estimation of Location Parameters: Introduction |
| SPEAKER: | STATISTICAN IN RESIDENCE Professor William E. Strawderman Department of Statistics Rutgers University Professor Stawderman's visit is supported by the Constance van Eeden Fund |
| We consider estimation of a p-dimensional location vector where performance is measured via the expected sum of squared errors. For much of the talk the distribution will be assumed to be spherically symmetric and it will be assumed that a residual vector is present so that scale can be estimated. We will give a number of heuristic arguments that so called shrinkage (Stein) estimators can be expected to out-perform the usual (least squares) estimator under minimal assumptions. We will review known results for the case of a normal distribution and demonstrate that most of these results have an extension to the general spherically case.
In a series of 3 follow-up lectures I will go into more detail on general results for spherically symmetric distributions and discuss related problems. Professor Strawderman's visit is supported by the Constance van Eeden Statistician in Residence program. |
|
| DATE/PLACE: | Tuesday, August 31, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TYPE: | Research Seminar |
| TITLE: | New Design Strategies for Repeated MRI Scanning in MS Clinical Trials |
| SPEAKER: | Alex Smith Department of Statistics UBC |
| A rising trend in multiple sclerosis clinical trials is the inclusion of a cohort study in which patients undergo repeated MRI scanning, most commonly in monthly intervals. Outcomes of these cohorts are based on newly active lesions which are detected on the brain stem, and which are now commonly believed to be highly associated with MS exacerbations. These cohorts have the benefit of allowing researchers to see a clear treatment effect in a relatively shorter amount of time than with the more common clinical outcomes. Employing two large data sets, we attempt to address two distinct research topics relating to designs of such trials. The first topic is related to a published algorithm that uses placebo data to repeatedly simulate trials for a given efficacy as a means of producing sample size calculations. We present a validation of this algorithm, and propose a theoretical alternative to the recommended simulation. The second topic involves the question of optimal designs for repeated MRI trials in MS using summary statistics. Typically, designs have included one baseline scan, followed by several scans while under treatment. We propose a model and means of examining optimality to recommend improvements to this common design, such as the use of an ANCOVA model for the response, and the use of multiple baselines as opposed to abundant treatment period scans.
|
|
| DATE/PLACE: | Thursday, August 19, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TITLE: | Dynamic Modelling for Foreign Exchange Rates. |
| SPEAKER: | Kelly Kwok Department of Statistics UBC |
| Since the advent of generalized floating exchange rates in 1973, the behavior of exchange rate movements has become an extremely challenging research area. Recent research indicates that standard linear macroeconomic models generally fail to improve upon the simple random walk model in out-of-sample forecasting. We present an empirical study (based on over 20 years of monthly data) of several models with dynamic state structure to forecast the foreign exchange rates of seven major currencies. As part of our study, we also employ various measures and visualization techniques to evaluate the performance of our candidate models in terms of expectation and risk forecastability. In addition to the commonly used statistical measures, we compare the models in seemingly practical situations. The performance measurement methodology for risk management is based on the concept of Value-at-Risk.
|
|
| DATE/PLACE: | Thursday, August 12, 1999, 16:00 Computer Science 301, 6356 Agricultural Road |
| TYPE: | Research Seminar |
| TITLE: | A Model for Markers and Latent Disease Progression |
| SPEAKER: | Dr. Mei-Ling Ting Lee Channing Lab., Medical School Harvard University |
| In HIV/AIDS research, statistical marker models help in investigating surrogacy of laboratory measures in studies of antiretroviral drugs. Markers contribute in controlling costs and completion times for clinical trials. In this paper (joint with Victor DeGruttola) we consider a bivariate Wiener process to model the joint distribution of markers and the latent disease progression. The process of disease progression is assumed to be latent or unobservable. The time to reach the primary endpoint of failure (death, disease onset, etc) is the time when the latent disease process first crosses a failure threshold level. Inference for the model is based on two kinds of data: censored survival data and marker measurements. To demonstrate the usefulness of the model, we apply the methods in analyzing data from the protocol 116a of the AIDS Clinical Trials Group (ACTG). | |
| WORKSHOP ON SMOOTHING APPLICATIONS | June 23-25, 1999 University of British Columbia See the workshop web site for more information. |
| DATE/PLACE: | Thursday, May 27, 1999, 16:00 NOTE DIFFERENT DAY Computer Science 301 |
| TITLE: | Extensions of the VaR Approach to Portfolio Selection With Non-normal Returns |
| SPEAKER: | Ryan W. Tse Department of Statistics, UBC |
|
The well-known mean-variance approach to portfolio selection problem proposed by Markowitz (1952) is often criticized for its use of variance as a measure of risk exposure. Recently, Value at Risk (VaR) has become a popular alternative for measuring risk in many firms. Using the idea of VaR, we formulated a chance constrained programming problem for portfolio selection. Until recently, most real life applications rely on the normality distributional assumption of the asset returns, which seems to be inconsistent with the empirical distributions. To relax this assumption, our study focused on the extensions of the VaR approach to portfolio selection to the class of Elliptically Contoured Distributed returns, and time-varying distributed returns. For the latter case, we proposed a new solution via empirical distributions. Moreover, a profile map of returns versus risks was proposed so that the optimal portfolio could be identified for various time-window sizes. The performance of various portfolios over different time periods were evaluated by means of off-sample cumulative returns and a new reward-to-risk measure. |
|
| PACIFIC NORTHWEST STATISTICS CONFERENCE | Friday, April 23, 1999 Washington State University See the conference web site for more information. |
| DATE/PLACE: | Thursday, April 15, 1999, 16:00 Computer Science 301 |
| TYPE: | Research Seminar |
| TITLE: | Recent Work on Statistical Methods for Mapping Incidence or Mortality Risks |
| SPEAKER: | Ying MacNab, Department of Mathematics and Statistics, Simon Fraser University |
| DATE/PLACE: | Thursday, April 1, 1999, 16:00 Computer Science 301 |
| TYPE: | Journal Club Session |
| TITLE: | Extending Zero-Inflated Poisson Models to Longitudinal Count Data |
| SPEAKER: | Alex Smith, Department of Statistics, University of British Columbia, Vancouver, B.C. |
|
I will introduce a paper which addresses the problem of an inordinate number of zeroes in count data sets. The paper proposes a zero-inflated Poisson (ZIP) regression model and discusses circumstances under which this model proves most superior to a standard Poisson regression model. Next I will introduce a data set which contains repeated (longitudinal) counts of active brain lesions in MS patients. I will then extend the theories of the paper from a cross-sectional to a longitudinal setting, in order to implement the model on this data set. I will conclude with the results of this implementation, in comparison with standard Poisson regression. |
|
| DATE/PLACE: | Monday, March 15, 1999, 16:00 NOTE DIFFERENT DAY Computer Science 301 |
| TITLE: | Symbolic algorithms for statistical inference |
| SPEAKER: | Jamie Stafford, Department of Statistics and Acturial Science, University of Western Ontario |
| The use of computer algebra in statistics will be demonstrated through a series of examples - some simple and some complex. The simple examples will reflect fundamental strategies as the familiar is viewed from a different perspective. This deepens our understanding of material we already ``understand''. The complex examples derive from seemingly distinct areas of statistical methodology including likelihood inference, the jackknife and bootstrap, M-estimation, Survey methodology, Experimental design, Regression, Edgeworth, Cornish-Fisher and Saddlepoint approximations. These examples share a common structure due to the tools used. These tools, like the cumulant or the delta method, can be automated. Algorithms developed, capitalize on the common structure and these areas of methodology begin to look very similar.
In the process we will negate the original motivation for the bootstrap, eliminate tedious algebra from a first course in sampling, develop an important complement to any course in asymptotics, apply the methods to research problems in sampling and experimental design and address fundamental research issues in computer algebra including Symbolic-Numerics. An example of the latter concerns the evaluation for survey data of algebraic expressions derived in the sampling context. Giving this capacity to the product SymSS2.1, which has been developed for Statistics Canada, would make it commercially viable.
|
|
| DATE/PLACE: | Tuesday, March 9, 1999, 16:00 Computer Science 301 |
| TITLE: | The Effective Use of Complete Auxiliary Information From Surveys Through Model Calibration and Empirical Likelihood |
| SPEAKER: | Changbao Wu, Department of Mathematics and Statistics, Simon Fraser University |
| Complete auxiliary information is often available from survey data. By complete we mean the values of auxiliary variable(s) are known for the entire finite population, not just in the sample. One of the fundamental questions is how to effectively use the complete auxiliary information at the estimation stage. In this work, a unified framework has been attempted under a general modeling process. The proposed model calibration estimators of the population mean and total can effectively handle any linear or non-linear models and reduce to the generalized regression estimators under linear models. The newly proposed pseudo empirical likelihood estimator (Chen and Sitter, 1999), when used in this setting, gives an estimator that is asymptotically equivalent to the model calibration estimator but with positive weights, and therefore is prefered. Some existing estimators using auxiliary information are re-examined and some related issues, in particular, the estimation of the finite population distribution function and quantiles, are discussed under the framework. Results of a limited simulation study are reported. | |
| DATE/PLACE: | Tuesday, March 9, 1999, 16:00 Computer Science 301 CANCELLED |
| TITLE: | A few state-space approaches to the modeling of longitudinal data |
| SPEAKER: | Soeren Lundbye-Christensen, Aalborg University, Denmark |
| DATE/PLACE: | Thursday, March 4, 1999, 16:00 Computer Science 301 |
| TYPE: | Journal Club Session |
| TITLE: | Modeling Two-state Disease Processes with Random Effects |
| SPEAKER: | Jochen Brumm, Department of Statistics, University of British Columbia, Vancouver, B.C. |
|
The article deals with disease processes in which subjects alternate between symptom free and symptomatic states. The authors develop a mixed-effect two-state model in which covariate effects are modeled multiplicatively on transition intensities. Bivariate log-normal random effects are used to accomodate heterogeneity in disease activity between subjects and to admit correlation between the transition intensities. We will present the methodology from this article and apply some of it to a data set from a clinical trial with relapsing-remitting Multiple Sclerosis patients. |
|
| DATE/PLACE: | Tuesday, March 2, 1999, 16:00 Computer Science 301 |
| TITLE: | Speed of convergence of the hit-and-run sampler |
| SPEAKER: | Claude Belisle, Laval University |
| The hit-and-run sampler is a Monte Carlo sampling method related to the Gibbs sampler. I will discuss its convergence properties and I will show that it is possible to construct simple examples where the convergence is as slow as we wish. I will also show that in cases where there is geometric convergence, a simple epsilon-transformation of the target distribution can make the convergence as slow as we wish. | |
| DATE/PLACE: | Tuesday, February 23, 1999, 16:00 Computer Science 301 |
| TITLE: | Applications of State-space Models |
| SPEAKERS: | Claus Dethlefsen and Bjarke Klein, Department of Mathematics, Aalborg University, Denmark |
| First, we will introduce you to being a mathematics student at Aalborg University. This is different from traditional universities because the teaching and learning environments are concentrated around project work in groups of 4-6 students. Each group has a professional supervisor. The written projects (100-130 pages) are assessed during oral examinations of approx. 2-3 hours. The students present their results and a discussion between the external examiner, the supervisor and the students take place.
Second, our Master's thesis project in Statistics, "State Space Models and Kalman Filtering in the GPS system", is presented. State space models for Longitudinal data/time series are introduced in a Bayesian framework. The Kalman filter, used to assess the state of the modelled system, will be described and an application of state space models on data obtained from the global positioning system is presented. Finally, as a part of a Ph.D. study, a state space model is proposed for a sequence of images recorded from an ultrasound examination of coronary arteries. By identifying objects in the image and describing their shape, orientation and motion by a low-dimensional vector of image characteristics, a substantial reduction in dimension is achieved. An algorithm is discussed for processing the images based on the extended Kalman filter. |
|
| DATE/PLACE: | Thursday, February 18, 1999, 16:00 Computer Science 301 |
| TYPE: | Research Seminar |
| TITLE: | Ecological Fallacy: Can ecological data measure our health? |
| SPEAKER: | K.C. Carriere, Department of Mathematical Sciences, University of Alberta, Edmonton, AB |
|
This talk deals with dilemmas faced by researchers who confront data from different levels of aggregation (spatial or longitudinal). Efficient, reliable and robust analytic methods to determine an appropriate and equitable distribution of health care funding and resources for health regions are in great demand. Substantial small area variation in practice patterns, care seeking behavior and political influence creates differences in utilization and health care resources that do not necessarily reflect need. Diverse, and often contradictory, estimates of needs are possible depending on the chosen methodological approaches and also on the level of aggregation of the data being used. We derive empirical and theoretical conditions under which ecological level analysis can be an efficient alternative to individual level analysis without unduly worrying about creating ecological fallacy. It will be demonstrated that, with careful consideration in aggregation of the data to satisfy the thoretical conditions, analyses based on aggregate data can be free of ecological fallacy. |
|
| DATE/PLACE: | Tuesday, February 9, 1999, 16:00 Computer Science 301 |
| TITLE: | GIW and its application to BC pollution |
| SPEAKER: | Li Sun, Department of Computer Science, University of British Columbia, Vancouver, B.C. and Cancer Control Research, BC Cancer Agency |
| Generalized Inverted Wishart (GIW) model will be introduced with a good reason - most BEAUTIFUL BRITISH COLUMBIANS are interested in the dirty pollutants in the air.
Some beautiful results, such as prediction and backcasting theory will be briefly described/discussed, followed by the application to BC ozone data. This is a joint work with Nhu Le and Jim Zidek.
|
|
| DATE/PLACE: | Thursday, February 4, 1999, 16:00 Computer Science 301 |
| TYPE: | Journal Club Session |
| TITLE: | Modeling Partly Conditional Means With Longitudinal Data |
| SPEAKER: | Huiying Sun, Department of Statistics, University of British Columbia, Vancouver, B.C. |
|
The paper proposes a general modeling approach to longitudinal data that is a hybrid of the marginal regression models and of the transition models. Rather than conditioning at time t only on covariate values, as is typical with the marginal approach, or on the entire history of the process up to t, as is typical with the transition model approach, the suggested models condition on a subset of the process history. Estimation proceeds using generalized estimating equation methodology but with the restriction that the working covariance matrix is diagonal. The proposed regression models share common features with Cox regression models for failure time data in that they are composed of a nuisance baseline function of time and a simple parametric function of the covariates. Two illustrative examples are presented. |
|
| DATE/PLACE: | Tuesday, February 2, 1999, 16:00 Computer Science 301 |
| TITLE: | Stochastic planning using decision diagrams |
| SPEAKER: | Robert St-Aubin, Department of Computer Science, University of British Columbia, Vancouver, B.C. |
| We propose a method of performing value iteration for Markov decision processes in stochastic environments using algebraic decision diagrams (ADDs), called SPUDD (Stochastic Planning using Decision Diagrams). We claim that using ADDs yield a more compact representation of the problem than decision trees do. Our preliminary experimental results tend to support this claim. | |
| DATE/PLACE: | Tuesday, January 26, 1999, 16:00 Computer Science 301 |
| TITLE: | Bias robustness for regression models |
| SPEAKER: | Jorge Adrover, FaMAF-University of Cordoba, Argentina |
| The robustness properties of an estimator are thoroughly described by its maximum bias function (maxbias) over a contamination neighborhood of the target model. Furthermore, the maximum bias function is known to be an important tool for inferential purposes. Therefore, it is rather natural to seek an estimator that minimizes the maxbias function over a class of estimators. We review some results about minimax bias estimation obtained in the context of the location and regression models. These previous results have not included the intercept in the regression model. Based on minimax bias considerations for the regression through the origin model, we analyze the behavior of the maximum bias function for GM-estimators when the intercept is also included. We next analyze a new proposal which seems to be quite promising in order to obtain minimax bias. | |
| JOINT SEMINAR WITH STATISTICS WORKSHOP SEMINAR SERIES | |
| DATE/PLACE: | Tuesday, January 19, 1999, 16:00 Computer Science 301 |
| TYPE: | Research Seminar |
| TITLE: | Biostatistical Consulting at the Mayo Clinic |
| SPEAKER: | Jeff A. Sloan, Mayo Clinic, Rochester, MN |
|
This talk will present the structure of and approach to biostatistical consulting carried out at the Mayo Clinic, particularly within the Cancer Center Statistics Unit. The framework of the co-operative group structure and its impact on statistical design and analysis issues will be highlighted. The bulk of the talk will present brief summaries of oncology clinical research studies and the statistical methods used and/or developed at Mayo to address issues encountered. Specific examples include phase III clinical trials comparing two radiation schedules for advanced non-small cell lung cancer, investigations of alternative agents for chemotherapy-related stomatitis, a crossover trial evaluating capsaicin cream for post-mastectomy pain, limited sampling models for pharmacokinetic data drawn from a phase I study of Irinotecan, and an assessment for the efficacy of coumarin for lymphedema. |
|
