Speaker

Kino Zhao 

Date and Time

Wednesday, March 22, 2023
1:00 - 2:30 PM PST

Location

Robert C. Brown Hall 6152
Simon Fraser University
Burnaby, BC V5A 1S6

Abstract

Statistical models are built on various assumptions -- that the sample is i.i.d., that measurement errors are homoscedastic, that we should model a variable as latent instead of composite, etc. All of them can be challenged or relaxed, but it is unclear what criteria we should use to judge the reasonableness of these assumptions. What are they about, after all?

One popular answer is that model assumptions are about the data generating process. The sample is i.i.d. just in case it is drawn randomly. Errors are homoscedastic just in case they are trembling-hand-style measurement errors. In this talk, I argue that this view of modeling assumptions underestimates the creative power of statistical modeling as a form of knowledge generation. I propose a distinction between data and knowledge by highlighting the context dependency and perspectivity of the latter. I explore an alternative way of understanding model assumptions as about the knowledge generating process.