PREVIOUS SEMINARS | 2019/20 ACADEMIC YEAR
Wednesday 25 September: The Teaching Statistics Trust Lecture 2019 - The purpose of statistics is insight not numbers
Abstract: In recent years, statistics teaching has seen a welcome move away from formulae and calculation. Especially with the rise of ‘big data’, numerical processing is increasingly being done with software, and it is becoming much more important for students to learn the art and science of interpretation. This development requires teachers to change focus too, shifting their emphasis from numbers to language.
As with many academic disciplines, statistics overlays everyday language with specialist meaning: one familiar example is the word ‘significant’ which means very different things in everyday use and in statistics. Research shows that parallel meanings such as this make it harder for students to understand technical concepts. Research also shows that teaching with a richer vocabulary can help to overcome this problem of understanding.
But statistics is more than just an academic discipline, it is a vital element of citizenship: we all need statistical understanding to make sense of the world around us. Yet statistical data are routinely misunderstood and misinterpreted in the media. In most cases the errors arise, not from the numbers themselves, but from the confused and inaccurate language used to comment on them. Clear language is essential to clear thought.
This lecture, drawing on numerous practical examples, explored the ways in which careful use of language can help everyone – teachers, students and citizens – to understand statistics better, whether in formulating enquiries, interpreting data, or reaching trustworthy conclusions and communicating them effectively.
Neil Sheldon was a teacher for more than 40 years. He is a Chartered Statistician and a former Vice-President of the Royal Statistical Society. He was the RSS Guy Lecturer in 2007-8 and he is currently Chair of the Teaching Statistics Trust. Neil’s other academic interests include philosophy and linguistics.
The Teaching Statistics Trust Lecture is given annually at multiple locations. It is aimed at teachers of statistics, whether specialist or non-specialist, in secondary schools, colleges and early years of university.
Wednesday 2 October: Dark energy in the lab - searching for chameleons and symmetrons with atom interferometry
- Speaker: Benjamin Elder (Nottingham)
Abstract: Theories of dark energy generically introduce new degrees of freedom to the gravitational sector. These degrees of freedom mediate a fifth force between matter particles, which can in principle be an O(1) correction to General Relativity. Compatibility with traditional tests of gravity requires that such forces be dynamically suppressed, or screened, in certain environments like the Solar System. Benjamin discussed how a new generation of laboratory-based measurements of gravity, designed to be sensitive to screened forces, are playing a vital role in the search for dark energy. In particular, he focused on two popular models of screened dark energy, the chameleon and the symmetron, and the strong new bounds on those theories coming from atom interferometers.
Wednesday 16 October: Loss-based prior for variable selection in linear regression methods
- Speaker: Cristiano Villa (University of Kent)
Abstract: In this work we propose a novel model prior for variable selection in linear regression. The idea is to determine the prior mass by considering the worth of each of the regression models, given the number of possible covariates under consideration. The worth of a model consists of the information loss and the loss due to model complexity. While the information loss is determined objectively, the loss expression due to model complexity is flexible and, the penalty on model size can be even customized to include some prior knowledge. Some versions of the loss-based prior are proposed and compared empirically. Through simulation studies and real data analyses, we compare the proposed prior to the Scott and Berger prior, for non-informative scenarios, and with the Beta-Binomial prior, for informative scenarios.