Executive summary: This exploratory critique argues that “controlling for a variable” in observational studies often fails to clarify causality and can be deeply misleading, because the statistical technique—typically just adding variables to regressions—relies on untestable assumptions about causal direction, ignores feedback loops and confounding, and is frequently misunderstood or misrepresented in scientific communication.
Key points:
“Controlling” usually means adding a variable to a regression, which doesn’t resolve deeper issues like reverse causality, nonlinear relationships, or missing variables—it only creates the illusion of causal clarity.
Reverse causality and feedback loops make observational data ambiguous, as the same dataset could be explained by entirely different causal models, making causal inference impossible without experimental intervention.
Controlling for a variable can mislead if variables are interdependent, potentially obscuring real causal pathways (e.g., blocking mediation effects or misrepresenting indirect causation).
Additional problems include measurement noise, poor variable encoding, linearity assumptions, and omitted variable bias, all of which weaken the reliability of regression-based causal claims.
Many scientific communities use evasive language to imply causality from observational studies, substituting phrases like “associated with” to suggest effects while avoiding scrutiny of causal assumptions.
The author calls for intellectual honesty and humility, urging researchers to either pursue experimental designs when possible or be transparent about the limitations and assumptions behind their observational findings.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.
Executive summary: This exploratory critique argues that “controlling for a variable” in observational studies often fails to clarify causality and can be deeply misleading, because the statistical technique—typically just adding variables to regressions—relies on untestable assumptions about causal direction, ignores feedback loops and confounding, and is frequently misunderstood or misrepresented in scientific communication.
Key points:
“Controlling” usually means adding a variable to a regression, which doesn’t resolve deeper issues like reverse causality, nonlinear relationships, or missing variables—it only creates the illusion of causal clarity.
Reverse causality and feedback loops make observational data ambiguous, as the same dataset could be explained by entirely different causal models, making causal inference impossible without experimental intervention.
Controlling for a variable can mislead if variables are interdependent, potentially obscuring real causal pathways (e.g., blocking mediation effects or misrepresenting indirect causation).
Additional problems include measurement noise, poor variable encoding, linearity assumptions, and omitted variable bias, all of which weaken the reliability of regression-based causal claims.
Many scientific communities use evasive language to imply causality from observational studies, substituting phrases like “associated with” to suggest effects while avoiding scrutiny of causal assumptions.
The author calls for intellectual honesty and humility, urging researchers to either pursue experimental designs when possible or be transparent about the limitations and assumptions behind their observational findings.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.