Ian McKeague is Professor Emeritus of Biostatistics at Columbia University in the City of New York, and Distinguished Visiting Professor at the City University of Hong Kong. He has served as a co-editor of the Journal of the American Statistical Association, and will soon start a term as a co-editor of Statistica Sinica. He is a Fellow of the IMS and the ASA. His research focuses on inference for stochastic processes, post-selection inference, and functional data analysis, with diverse applications in biomedical settings. In particular, he has developed inferential methods for selecting high-dimensional predictors and causal mediators of survival outcomes, for the analysis of growth trajectories, and for the analysis of wearable device data. His work has also included studies of Bayesian/Markov chain Monte Carlo methods with applications in ocean circulation inverse problems, signature verification, and statistical problems inspired by relativity and quantum physics. This lecture will be given at the 2026 IMS Annual Meeting.

What we can know from complex functional data
The first part of the talk will consist of an informal discussion of the nature of novelty in statistics. Recent headlines ask: Does A.I. think? Can A.I. become conscious? Fortunately, so far at least, consciousness is necessary for a statistician. The famous remark of John Tukey that “the best thing about being a statistician is that you get to play in everyone’s backyard” essentially made the same point, and that being a statistician gives the freedom to pursue an interest in almost any field, given you have a slight corner on something. This point has a parallel in the thinking of the British writer Geoff Dyer. For instance, in discussing his memoir The Missing of the Somme at a recent event at Columbia, Dyer remarked that his initial idea for the book was “a great insight of such originality that nobody had had it before … what I realized about the First World War is that it took place in the past.” I will argue for what I call the Tukey–Dyer method: the inspired choice of “a slight corner on something” (in Dyer’s words) as the starting point for creative work, even when you are not an expert. I will discuss the surprising role that this method can play in searching for novel ideas in statistics, and how it has occasionally led to some projects I have been involved with.
The second part of the talk will discuss some specific examples of the Tukey–Dyer method, and, in particular, how a paper concerning inference for non-smooth functional data having sample paths of bounded variation [1] was inspired in this way. This paper originated in the need to understand complex wearable device data collected in a Columbia-based study of an experimental therapy for mitochondrial disease, a group of disorders that affect the body’s ability to produce energy. The resulting clinical paper [2] provided a bias-­adjusted outcome measure of acceleration across a range of subjects’ activities and has played a role in assessing the efficacy of a nucleoside therapy (recently approved by the FDA) for thymidine kinase 2 deficiency, an ultra-­rare autosomal recessive mitochondrial disease. In addition, I will briefly touch on the potential, in terms of what we can know in a statistical sense, of having access to complex functional data collected along causal paths in a possibly non-smooth metric spacetime. This relaxes the smooth Lorentzian framework of Einstein’s field equations to allow rougher settings, and is related to similar problems for standard metric space-valued random objects. In this connection, I will also discuss how a statistical approach might be useful in addressing the question (recently raised in the physics literature) of whether the discreteness of spacetime has an observable signature.

References
[1] Chang, H-W, and McKeague, IW (2022). Empirical Likelihood-based Inference for Functional Means with Application to Wearable Device Data. Journal of the Royal Statistical Society, Series B, 84(5), pp.1947–1968
[2] McKeague, IW, et al. (2025). Assessing a Mitochondrial Disease Treatment via a Novel Statistical Technique for Accelerometer Data. Annals of Clinical and Translational Neurology, 12: 2505–2513