NaΓ―ve regression requires weaker assumptions than factor models to adjust for multiple cause confounding

07/24/2020 βˆ™ by Justin Grimmer, et al. βˆ™ 0 βˆ™

The empirical practice of using factor models to adjust for shared, unobserved confounders, 𝐙, in observational settings with multiple treatments, 𝐀, is widespread in fields including genetics, networks, medicine, and politics. Wang and Blei (2019, WB) formalizes these procedures and develops the "deconfounder," a causal inference method using factor models of 𝐀 to estimate "substitute confounders," 𝐙̂, then estimating treatment effects by regressing the outcome, 𝐘, on part of 𝐀 while adjusting for 𝐙̂. WB claim the deconfounder is unbiased when there are no single-cause confounders and 𝐙̂ is "pinpointed." We clarify pinpointing requires each confounder to affect infinitely many treatments. We prove under these assumptions, a naΓ―ve semiparametric regression of 𝐘 on 𝐀 is asymptotically unbiased. Deconfounder variants nesting this regression are therefore also asymptotically unbiased, but variants using 𝐙̂ and subsets of causes require further untestable assumptions. We replicate every deconfounder analysis with available data and find it fails to consistently outperform naΓ―ve regression. In practice, the deconfounder produces implausible estimates in WB's case study to movie earnings: estimates suggest comic author Stan Lee's cameo appearances causally contributed $15.5 billion, most of Marvel movie revenue. We conclude neither approach is a viable substitute for careful research design in real-world applications.

READ FULL TEXT
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.