Centre for Mathematical
Sciences
Wilberforce Road, Cambridge, CB3 0WB
Tel: (01223) 337958
Fax: (01223) 337956
Email: secretary@statslab.cam.ac.uk
All interested are welcome
This list is subject to revision
Select a date to view the relevant seminar abstracts:
A sketch will be given of work in progress on the measurement of inter-generational mobility between social classes using statistical models. Measurement based on the "UNIDIFF" or "log multiplicative" model, commonly used in sociology, is shown to be non-robust to failure of the modelling assumptions; and more general methods based on odds ratios lack a clear interpretation. An alternative approach is developed based on transition rates, in which fluidity corresponds to the speed of an underlying mobility process.
Lagrange multiplier tests against nonstationary unobserved components such as stochastic trends and seasonals are based on statistics which, under the null hypothesis, have asymptotic distributions belonging to the class of generalised Cramér-von Mises distributions. Converseley, unit root tests can be formulated, again using the Lagrange multiplier principle, so as to yield test statistics which also have Cramér-von Mises distributions usnder the null hypothesis. These ideas may be extended to multivariate models and to models with structural breaks thereby providing a simple unified approach to testing in nonstationary time series.
A ``mortal'' faced with a sequential decision situation, may choose the best stopping rule to receive an expected value V . A ``prophet'' on the other hand simply observes the whole sequence and picks the largest one to receive an expected value of M is always larger than V, and the question is by now much larger. For non-negative variables the question may naturally be asked in terms of the ratio M/V. Perhaps surprisingly, it turns out that this ratio is uniformly bounded. The talk will begin with some well known basics on optimal stopping and prophet inequalities, continuing with recent results when the mortal has several stopping options rather than one.
Prompted by Spirtes and Pearl working in AI, there is now considerable interest from the statistical community in defining and using operationally various concepts of causation. Models are called causal if certain structural relationships are retained when a system is manipulated rather than simply observed. In this talk I will discuss how comparable ideas can be developed in a dynamic probabilistic forecasting setting. Links are made with stochastic control theory. My running example investigates the effects of external manipulations in forecasts associated with an oil supply chain.
Spatial models have been at the heart of the development of Bayesian hierarchical models and computations. In epidemiology, spatial analyses abound and have been used to raise hypotheses concerning disease patterns and their potential relationship with environmental variables. In this talk, we focus on classes of spatial models (parametric and semi-parametric) that have been proposed for analysing epidemiological applications, review some of their properties and illustrate their use for disease mapping or for detecting areas of high risk.In many instances, covariate effects are the central question in spatial analyses and the spatial structure in the residuals can be thought as a proxy for many unidentified covariates. We discuss some current issues concerning the interplay between the regression modelling of explanatory variables and that of residual spatial variability.