Speaker：Gang Chen, Ph.D. (NIMH/NIH)

Time： 14:00-15:30, Sep 25, 2025

Venue：王克桢楼1206

Host：杨炯炯

Abstract

Statistical modeling is central to how we analyze and interpret data. Yet too often, the statistical “tail” wags the scientific “dog.” This talk highlights common pitfalls where unquestioned reliance on conventional statistical practices—however standard they may appear—risks obscuring rather than advancing scientific understanding.

We will tackle thorny issues such as variable selection, multiple comparisons, result reporting, and experimental design. I will argue that, to improve reproducibility and to make scientifically meaningful inferences, statistical methods should be guided by scientific knowledge (and, where possible, common sense). I will also share some of my own misadventures at the intersection of statistical theory and the messy realities of real-world data.

Biography

Gang Chen’s background is in applied mathematics and statistics. Since 2003, he has been part of the NIH team developing and maintaining AFNI, a widely used software package for neuroimaging data analysis. He is currently a senior associate scientist.

His present focus is on grappling with hierarchical structures in neuroimaging data: the brain’s structural heterogeneity, the challenges of experimental design, and the elusive profile of the hemodynamic response. He approaches these not merely as statistical nuisances, but as interconnected pieces of a larger ontological puzzle. His fascination with causal inference is a recurring theme, and sees statistical modeling as a unique blend of science, philosophy, and performance art.

2025-09-07