You Can't Drive a Car With Only Three Wheels

Area of Science:

• Epidemiology
• Causal Inference
• Observational Data Analysis

Background:

• Causal inference from observational data typically relies on exchangeability, consistency, and positivity assumptions.
• Measurement bias, especially exposure misclassification, is often overlooked but critically impacts causal effect estimation.
• Existing identifiability conditions are insufficient when measurement bias is present.

Purpose of the Study:

• To highlight the critical role of measurement bias in causal inference.
• To emphasize that exposure misclassification bias can invalidate causal effect estimates.
• To advocate for quantitative strategies to assess misclassification bias.

Main Methods:

• Discussion of fundamental identifiability assumptions in causal inference.
• Review of methodological challenges in estimating causal effects with observational data.
• Examination of empirical strategies for nonexchangeability and consistency violations.
• Emphasis on the need for quantitative strategies to address misclassification bias.

Main Results:

• Standard identifiability assumptions are insufficient in the presence of measurement bias.
• Accurate exposure classification is a fundamental requirement for causal effect estimation.
• Exposure misclassification bias can prevent valid causal effect estimation, analogous to a car missing a wheel.

Conclusions:

• Authors must address exposure misclassification bias for valid causal effect estimation.
• Quantitative methods are needed to examine the impact of misclassification.
• Causal inference requires careful consideration of all potential biases, including measurement error.