Minimizing Statistical Non-Reproducibility

Date: Wednesday, 18 January 2023.
Time: 12:30 p.m. to 2:30 p.m.
Location: hybrid (in person at Burnside Hall room 1104 and online via Zoom).
Instructor: Prof. James Hanley. Dept of Epidemiology, Biostatistics and Occupational Health, Faculty of Medicine, McGill.

Overview: This workshop will combine lecture and hands-on with R to show, via concrete examples drawn from the instructor’s research and teaching experience, how faulty statistical reasoning regarding probabilities, and misuse of statistical tests, lead to misleading and non-reproducible inferences.

The workshop will begin with some probability calculations, all easily calculated, but also easily miscalculated. Then, by first analyzing a real data set generated by a known physical process, participants will try out ways to avoid finding false signals. They will then apply these and other principles to a second dataset generated by human behaviour.

At the end of this workshop, participants will be able to understand some of the statistical principles and practices that make inferences more reproducible.

Prerequisites:
· Prior exposure to multiple regression models, and the ability to fit them in R would be hepful.
· Introductory knowledge of R, e.g. from our workshop Introduction to programming in R, or from McGill's R summer camp.

Resources: Some of the examples to be used early in the workshop will be drawn from the links on the right-hand side of the 2022 SLIDES on Prof. Hanley's website and from this article. For a quite thorough (and forceful) description of how and how not to use regression models, see Frank Harrel’s book Regression Modeling Strategies e-available from the McGill library.