Abstract

Establishing cause and effect relationships outside of experimental settings is well known to be a challenging task with far reaching implications for the scientific enterprise across research disciplines. Nonetheless, due to ethical concerns or practical considerations, experimental evaluation of causal effects is often not feasible. For example, learning about the potential harmful effect of air pollution on asthma attacks cannot readily be studied in a randomized experiment for the obvious ethical reasons; likewise evaluating the real-world effectiveness and safety of COVID -19 vaccines necessarily requires working with observational data where vaccine uptake and exposure to COVID -19 reflect their natural population dynamics. The main challenge of causal attribution is bias due to hidden confounding. We will present recent advances in Statistical Science that demonstrate how credible causal attribution from observational studies remains possible despite hidden confounding, provided certain proxies (even if noisy) of hidden confounders can be measured.  As we argue, such proxies abound in many practical settings, examples of which we will describe from infectious disease, environmental and cancer epidemiology, providing a revealing window into the beautiful and challenging world of causal attribution in epidemiological inquiry. What is a Kirk Distinguished Visiting Fellow? The fellowship allows field-leading mathematicians from under-represented groups to join the programme as influential role models, helping to address historical imbalances and promote diversity within mathematical research.