Invited commentary: variable selection versus shrinkage in the control of multiple confounders

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Variable selection methods are tempting for reducing numerous potential confounders in regression analyses.
• Conventional selection methods lack consensus and can lead to inaccurate statistical inference (narrow confidence intervals, small p-values).

Purpose of the Study:

• To evaluate the utility and potential pitfalls of statistical confounder selection in regression.
• To compare confounder selection methods against adjusting for all measured confounders.

Main Methods:

• Review of theoretical and simulation evidence regarding confounder selection.
• Discussion of challenges in controlling all measured confounders with conventional methods.
• Introduction to modern techniques like shrinkage estimation and exposure modeling.

Main Results:

• No statistical selection method is consistently superior to adjusting for all well-measured confounders.
• Conventional selection methods often produce biased confidence intervals and p-values.
• Controlling all measured confounders can pose issues for standard model-fitting approaches.

Conclusions:

• Statistical confounder selection may be an unnecessary complication in most regression analyses.
• Modern techniques offer alternatives when controlling all confounders presents challenges.
• Focusing on accurate measurement and appropriate modeling is key, rather than complex selection procedures.