Actions Panel
Maurice Bloch Seminar: Professor Noemi Kreif
Date and time
Location
Yudowitz Seminar room
Wolfson Medical Building University Avenue G12 8QQ United KingdomDescription
We are pleased to invite you to:
The Institute of Health and Wellbeing Maurice Bloch Annual Lecture Series 2018/19
Title: Machine Learning in Policy Evaluation: New Tools for Casual Inference
Presenter: Dr Noemi Kreif
Date: Tuesday 14 May 2019
Time: 1pm-2pm, a light lunch will be served 30 minutes beforehand
Venue: Yudowitz Seminar room, Wolfson Medical School
Chair: Dr Claudia Geue
Abstract
While machine learning (ML) methods have received a lot of attention in recentyears, these methods are primarily for prediction. Empirical researchers conductingpolicy evaluations are, on the other hand, pre-occupied with causal problems, tryingto answer counterfactual questions: what would have happened in the absence of apolicy? Because these counterfactuals can never be directly observed (described asthe “fundamental problem of causal inference”) prediction tools from the ML literaturecannot be readily used for causal inference. In the last decade, major innovations havetaken place incorporating supervised ML tools into estimators for causal parameterssuch as the average treatment effect (ATE). This holds the promise of attenuating model
misspecification issues, and increasing of transparency in model selection. This talkaims to review and illustrate some of these recent developments incorporating machinelearning in the estimation of the ATE of a binary treatment, under the unconfoundednessand positivity assumptions.
I will first briefly review supervised machine learning, including trees-based methods,the lasso, and ensembling approaches, in particular the Super Learner. I then reviewand illustrate the following uses of machine learning: 1) to create balance among treatedand control groups, 2) to estimate so-called nuisance models (e.g. the propensity score,or conditional expectations of the outcome) in semi-parametric estimators that targetcausal parameters (e.g. targeted maximum likelihood estimation or the double MLestimator) 3) the use of machine learning for variable selection in situations with a highnumber of covariates. Throughout I use an illustrative case study of the evaluation ofa health insurance programme in Indonesia, highlighting their potential benefits as wellas the challenges, compared to more traditional approaches.
About the speaker
Noemi Kreif joined the Centre for Health Economics in 2016 as a Research Fellow in Global Health Economics. She holds a PhD (2013) from the London School of Hygiene and Tropical Medicine. Her PhD and post-doctoral (Medical Research Council Early Career Fellowship) work focussed on advancing statistical methods for economic evaluation that uses observational data, resulting in publications in leading health economics and statistics journals, such as Health Economics, Statistical Methods in Medical Research and American Journal of Epidemiology. Her current work is centred on econometric evaluations of health policies in low and middle-income countries, with a continued interest in applying advanced causal inference and machine learning tools.