Usual time: Thursdays 14:00-15:00
Location: Room 102, Department of Statistical Science, 1-19 Torrington Place (1st floor). Some seminars are held at different locations and at different times. Please click on the abstract for further details.
- 04 October 2018 1400-1500: Prof. Michael Goldstein (University of Durham)
Inverting the Pareto boundary: Bayesian uncertainty quantification for decision support
We consider problems of decision support arising in the context of uncertainty quantification for complex physical systems modelled by computer simulators, where we must take into account a soft constraint on our decision choices. This leads to the problem of identifying and inverting the Pareto boundary for the decision. We show how Bayes linear methods may be used for this purpose and how the sensitivity of the decision choices may be quantified and explored. The approach is illustrated with a problem on wind farm construction. (This is joint work with Hailiang Du.
- 04 October 2018 1530-1630: Prof. Stephen Stigler (University of Chicago)
The miraculous theorem: Bayes and Price and the replication crisis (Homage to Dennis Lindley)
Thomas Bayes' article "An Essay towards solving a Problem in the Doctrine of Chances" is perhaps the most cited article in the history of statistics, even though it has only seldom been carefully read. In 2013 I made a surprising discovery about this article, which I shared with Dennis Lindley. His reply (the last letter I received from him, in March 2013) will be shown, and a sequence of further discoveries about the article will be discussed and related to current controversies.
- 11 October 2018: Dr. Chen Qu (University College London)
Modelling of mean-covariance structures in marginal structural models
Marginal structural models (MSMs) are used to more accurately model the causal effect of a time-dependent treatment, especially when confounders are present. Previous literature were mainly focused on estimating the mean structure for the MSMs, but only few studies considered the structures of variance or covariance. However, it is believed that the impacts of confounders may be more severe in relation to the mean estimation, as estimating the mean involves the estimation of the second moments.
In this talk, we propose to model the mean-covariance structures for marginal structural models within the framework of the weighted generalized estimating equations (WGEE). These models offer appropriate adjustment for time-dependent confounding on both mean and covariance estimation. We investigate the performance of the proposed approach in the simulation studies based on a real-life data set.
- 18 October 2018: Dr. Ben Powell (University of York)
Confidence intervals for least absolute deviations regression
Least-absolute-deviations (LAD) regression, which uses an L_1 penalty to punish prediction errors, provides a workable alternative to least-squares (LS) regression with superior robustness properties. I will argue, however, that wide-spread adoption of LAD regression is impeded by a relative lack of supporting theory. In this talk I describe methodology for producing conservative confidence and credible regions for regression coefficients and predictions for LAD regression problems. Accompanied by these intervals, robust LAD estimates are better placed to compete with their LS counterparts.