Prof. Dr. Daniel Scharfstein, Professor of Biostatistics, Department of Population Health Sciences, University of Utah School of Medicine
Title: “Semiparametric Sensitivity Analysis: Unmeasured Confounding in Observational Studies”
Establishing cause-effect relationships from observational data often relies on untestable assumptions. It is crucial to know whether, and to what extent, the conclusions drawn from non-experimental studies are robust to potential unmeasured confounding. In this paper, we focus on the average causal effect (ACE) as our target of inference. We build on the work of Franks et al. (2019) and Robins et al. (2000) by specifying non-identified sensitivity parameters that govern a contrast between the conditional (on measured covariates) distributions of the outcome under treatment (control) between treated and untreated individuals. We use semi-parametric theory to derive the non-parametric efficient influence function of the ACE, for fixed sensitivity parameters. We utilize this influence function to construct a one-step, split-sample bias-corrected estimator of the ACE. Our estimator depends on semi-parametric models for the distribution of the observed data; importantly, these models do not impose any restrictions on the values of sensitivity analysis parameters. We establish that our estimator has $\sqrt{n}$ asymptotics. We utilize our methodology to evaluate the causal effect of smoking during pregnancy on birth weight. We also evaluate the performance of estimation procedure in a simulation study. This is joint work with Razieh Nabi, Edward Kennedy, Ming-Yueh Huang, Matteo Bonvini and Marcela Smid.
Prof. Dr. Daniel Scharfstein is a full professor in the Department of Biostatistics at the Johns Hopkins Bloomberg School of Public Health since 2008. While remaining on the faculty for now, he recently took a position as Professor of Biostatistics in the Department of Population Health Sciences at the University of Utah School of Medicine. His internationally acclaimed research focuses on statistical methods of causal inference, especially in the context of selection bias. He has made important contributions to the question of how to deal with informative missing values or censorship in RCTs and to the statistical analysis of observational studies with non-randomized treatment. His more than 100 mostly methodological publications have appeared in high-ranking journals such as JRSS, JASA, Biometrika, Statistics in Medicine. He was a member of the National Academy Committee that wrote the recommendation “The Prevention and Treatment of Missing Data in Clinical Trials”. He has also been and continues to be the lead statistician in a number of evaluation studies, e.g. the National Study of the Costs and Outcomes of Trauma (NSCOT), Guided Care for Chronically Ill Older Adults and Healthy Steps for Young Children. He has made important contributions to the debate on the ICH E9 Addendum on “Estimands”, e.g. in the journal “Clinical Trials”.
Topic: “Machine Learning in Biometry”
Prof. Chris Holmes, Department of Statistics, University of Oxford.
Please see here for a CV.
Title: “Machine Learning in Biometry”
Machine learning (ML) and artificial intelligence (AI) have had a major impact across many disciplines including biometrics. In the first half of this talk we will review some of the characteristics of ML that make for successful applications and also those features that present challenges, in particular around robustness and reproducibility. Relatively speaking, ML is mainly concerned with prediction while the majority of biometric analyses are focussed on inference. In the second half of the talk we will review the prediction-inference dichotomy and explore, from a Bayesian perspective, the theoretical foundations on how modern ML predictive models can be utilised for inference.