Wayfair DS Explains It All: Foundational Assumptions of Experimentation and Linear Regression
April 8, 2019
Wayfair has a strong emphasis on causal inference when evaluating the impact of business strategies. The assumptions of experimentation and regression are critical to know your models are capturing true effects and not confounding variables. In this video, Wayfair data scientist Dan VanLunen outlines some of these key assumptions along with some tips around using regression.
The following charts show the key assumptions of linear regression and their implications in the finite and asymptotic cases.
Fig 1: Finite Sample OLS
Fig 2: Asymptotic OLS
At Wayfair, Dan’s team focuses on evaluating the effects of product labels (e.g. “Sale”) on demand through experiments to get causal estimates, predictive models to project estimates where data is sparse, and ongoing learning to keep estimates fresh. Dan enjoys hiking in nature around the world with his wife (including on multiple glaciers), breakdancing, hanging out with his 12lb poodle Teddy, and writing data science raps about his co-workers.