Causal machine learning masterclass

Learn more Applications are now closed Add to Calendar 03/03/2020 10:00 AM 03/03/2020 05:30 PM Europe/London Causal machine learning masterclass Location of the event
Tuesday 03 Mar 2020
Time: 10:00 - 17:30

Audience type



This is an apply to attend workshop. Please await written confirmation of your acceptance before planning your travel. Applications are now closed.

About the event


Most scientific questions, such as those asked when evaluating policies or exposures, henceforth referred as treatments, are causal in nature, even if they are not specifically framed as such. Causal inference reasoning helps clarify the scientific question, and the assumptions necessary to express it in terms of the observed data. Once this is achieved, the focus shifts to estimation and inference. Estimating causal effects typically requires adjustment for confounding. This is the result of a lack of comparability between subjects due to possibly many factors that are related simultaneously to the outcome and the variable whose effect we aim to estimate. These adjustments can be achieved via parametric modelling. However, such traditional statistical tools are not entirely satisfactory as high-dimensional confounding is difficult to handle and model misspecification is likely. As even minor misspecifications can induce large bias in the treatment effect estimate, the task of learning functional relationships between variables in order to adjust for confounding is critical. Unsurprisingly, machine learning methods are increasingly being used to assist in this task. This is challenging because, while the prediction performance of a given machine learning algorithm can be measured by contrasting observed and predicted outcomes, performance evaluation becomes impossible for treatment effect estimation since the `ground truth’, i.e. the true treatment effect, is unknown.


The aim of this masterclass is to introduce machine learning-based methods for the evaluation of (causal) treatment effects. We will highlight that bias can be introduced if using standard machine learning methods that are tuned for prediction performance, as opposed to estimation of treatment effects. We will then introduce the framework of “Targeted Learning” and other causal machine learning approaches, as a principled solution with optimal statistical properties for the estimation of causal treatment effects. The masterclass will include a number of hands-on sessions in R where participants can experience the problems with naive machine learning and understand how Targeted Learning works by implementing it in real-world settings.


We have some funds set aside to support attendees who need to travel to attend. Successful applicants will have the opportunity to request this support after their application has been accepted. 





The Alan Turing Institute

1st floor of the British Library, 96 Euston Road, London, NW1 2DB

51.5297753, -0.12665390000006