New data on COVID-19 is undermined by old statistical problems

Event Information

Share this event

Date and Time



Online Event

Event description
A talk by Gibran Hemani, Senior Research Fellow, University of Bristol

About this Event

This event is part of the 2020 Seminar Series on behalf of the Health Data Science Research Strand with the Elizabeth Blackwell Institute at the University of Bristol.

About the speaker

Gibran Hemani is a Senior Research Fellow with a background in quantitative genetics. He holds a Sir Henry Dale Fellowship, which focuses on leveraging large scale genetic data to improve causal inference in epidemiological models.

Talk description

Who is more at risk of COVID-19 infection? What causes some people to die from the disease? Answers to these questions would be valuable, and data are being generated rapidly to try to answer them.

In this talk I will describe the types of data being generated. My main aim is convince you that a problem known as ‘collider bias’, which arises when your samples are selected non-randomly, is a serious threat to the validity of inferences being made from these data. I will use some examples e.g. claims about ACE-inhibitors and smoking on COVID-19 to illustrate the potential impact of collider bias on causal inference. I’ll also talk about some potential solutions to the problem. Have a look at this twitter thread for a gentle introduction, and links to a blog post and recent paper on the subject.

Data Week Online 2020

This event is part of Data Week 2020, organised by the Jean Golding Institute. Running from 15 - 19 June 2020, this will be our 3rd annual week long series of workshops, talks and events in the world of data science and data-intensive research. This year, all events are online.

Join our mailing list

To keep up to date with events like this, as well as funding, news, blogs and more, please join our mailing list.

Date and Time


Online Event

Save This Event

Event Saved