Queen Mary and Westfield College School of Mathematical Sciences Autumn 1998 STATISTICS SEMINARS : DESIGN OF EXPERIMENTS All are welcome The talks are held at 1630 hours in the Mathematics Seminar Room (103) on Level 1 of the Mathematics Building, Queen Mary and Westfield College. Tea and coffee are available in the Mathematics Common Room (102) from 1500 hours. The nearest underground station is Stepney Green. Turn left at the exit and walk 400 yards. ____________________________________________________________________________ ____ Date Speaker Title University 08 Oct. 98 Karen Ayres "Measuring Genetic Correlations Reading Within and Between Loci". 29 Oct. Neil Butler "A Comparison of Optimality Reading Criteria for Quadratic Response Surface and Spline Smoothing Models". 19 Nov. Deborah Ashby "Casting a Sceptical Eye over the QMW. Data : Implementing Bayesian Data Monitoring in Cancer Clinical Trials". 17 Dec. Ben Torsney "Results on D-Optimal Designs for Glasgow Weighted Regression and Binary Regression Models". ____________________________________________________________________________ ______ For more information ask: Barbara Bogacka School of Mathematical Sciences Queen Mary and Westfield College Mile End Road, London E1 4NS Tel: 0171 975 5497 e-mail: [log in to unmask] --------------------------------------------- The seminar information is kept on: http://www.maths.qmw.ac.uk/~rab/seminars.html Please see attached Abstracts and Important Notice ABSTRACTS "Measuring Genetic Correlations Within and Between Loci" Karen Ayres Forensic DNA match probabilities often assume independence of genes within and between loci. It is therefore important to investigate these assumptions of independence, and one approach is hypothesis testing. However, more information is available via posterior probability density curves of parameters measuring dependence. Useful parameters include the inbreeding coefficient, which measures dependence within loci, and gametic disequilibrium coefficients, which measure the dependence of genes at different loci within gametes (sex cells). Adopting a Bayesian approach, Markov chain Monte Carlo methods are described for sampling from the posterior distributions of these parameters. The methods are demonstrated via application to simulated data, and also to population data from New Zealand. The implications for DNA match probabilities are briefly discussed. oooOOOooo "A Comparison of Optimality Criteria for Quadratic Response Surface and Spline Smoothing Models" Neil Butler This talk will consist of two parts. The first part introduces some optimality criteria for quadratic response surface designs based on the gradient of the fitted response. These are then compared with more established optimality criteria. The second part considers optimal designs for smoothing spline and linear variance models, which are shown to provide a link between optimal design theory and sampling methods on an interval. oooOOOooo ABSTRACTS CONTINUED "Casting a Sceptical Eye over the Data: Implementing Bayesian Data Monitoring in Cancer Clinical Trials" Deborah Ashby Many clinical trials organisations use regular interim analyses to monitor the accruing results in large clinical trials. Classical rules, such as the group-sequential procedures of Peto, O'Brien and Fleming or Pocock have traditionally been based on P-values. However, none of these rules formally assess the impact that the results of a clinical trial might have on clinical practice. Thus a trial might be terminated early because of apparent treatment benefit, but fail to influence future clinicians to modify their future treatment policy. Bayesian rules have been proposed that have the potential to overcome these difficulties (Freedman, Spiegelhalter and Parmar, Statistics in Medicine, 1994). Like many Bayesian techniques, they were originally presented as reanalyses of trials that have already been published. They are now being used `live' (Fayers, Ashby and Parmar, Statistics in Medicine, 1997). We present details of their implementation in a trial of surgery with and without pre-operative chemotherapy for oesophageal cancer. Practical issues encountered will be discussed, including the choice of prior distribution and `clinically relevant' values for use in monitoring, and presentation of ideas to clinical members of the data monitoring committee. As one clinician put it `It's just casting a sceptical eye over the data'. oooOOOooo "Results on D-Optimal Designs for Weighted Regression and Binary Regression Models". Ben Torsney We consider initially designing for weighted regression models in one design variable x (with a 'constant' term). Weight functions considered include established choices in the literature - choices for which the weight function is bounded when x is not. A 'widest' possible design interval, X, can be an unbounded one. Ford, Torsney and Wu (JRSSB 1992) revealed a further class of such weight functions, in establishing that a design problem for a binary regression model (and other generalised linear models) can be transformed to designing for a weighted regression model. These new weight functions are bounded for all real x, so that we can allow X = R. D-optimal designs for all possible interval subsets of X are derived under conditions on the weight function which are satisfied by a large class. These designs are not simply linear shifts of the optimal design on X, This conflicts with hints in the literature. The results extend to higher order models. -------------------------------------------------------------- PLEASE NOTE In future, we intend to replace most of the mailing addresses by electronic addresses. If you would like to receive notification by e-mail, please send a message to [log in to unmask] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%