BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:BSU Seminar: "\;IV-learner: learning conditional average treat ment effects using instrumental variables"\; - Karla Diaz Ordaz\, Univ ersity College London DTSTART:20241119T140000Z DTEND:20241119T150000Z UID:TALK224275@talks.cam.ac.uk CONTACT:Alison Quenault DESCRIPTION:Instrumental variable methods are very popular in econometrics and biostatistics for inferring causal average effects of an exposure on an outcome where there is unmeasured confounding. However\, their applicat ion for learning heterogeneous treatment effects\, such as conditional av erage treatment effects (CATE)\, in combination with machine learning\, is somewhat limited.\n \nA generic approach that allows the use of arbitrary machine learning algorithms can be based on the popular two-stage princip le. We first "regress" the exposure on the instrumental variables (and pre -exposure covariates) and then learn the causal treatment effects by regre ssing the outcome on the predicted exposure. This is the approach of Foste r and Syrgkanis (2023)\, referred to as IV-debiased machine learning (IV-D ML). \nUnfortunately\, the slow convergence rates of the data-adaptive es timators that affect the first-stage predictions propagate into the result ing CATE estimates. \n\nIn view of this\, we propose the IV-learner\, insp ired by infinite-dimensional targeted learning procedures (Vansteelandt 20 23\, van der Laan et al 2024). It strategically targets the first-stage pr edictions so they perform well in their ultimate task: CATE estimation. Th e resulting learner is easy to construct based on arbitrary\, off-the-shel f algorithms. \n\nWe study the finite sample performance of our proposal u sing simulations\, and compare it to existing methods. We also illustrate it using a real data example. LOCATION:MRC Biostatistics Unit\, East Forvie Building\, Forvie Site Robin son Way Cambridge CB2 0SR. END:VEVENT END:VCALENDAR