How do you handle intervening variables in causal diagrams to estimate causal-effect?

• A. Do not include them in statistical model; include only confounders
• B. Intervening variables are the primary effect
• C. Intervening variables are used to adjust for measurement error
• D. Include them as primary variables in the model

Answer: (A)

Intervening variables, or mediators, lie on the causal path from the exposure to the outcome. When you want the total causal effect of the exposure, you should not adjust for these mediators because conditioning on them blocks part of the pathway through which the exposure influences the outcome. This would bias the estimate of the total effect by removing the indirect effect that runs through the mediator.

Instead, you adjust for confounders—variables that influence both the exposure and the outcome. By controlling for these, you remove spurious associations and obtain an unbiased estimate of the overall causal effect. If you did include the intervening variables, you’d be estimating the direct effect (the effect not mediated), not the total effect. So the recommended approach is to include confounders in the model and exclude intervening variables when estimating the total causal effect.