DAGs as communication tools

Using directed acyclic graphs to align stakeholders on assumptions before modeling.

Core Idea

DAGs are useful because they force a causal story into the open. They help teams separate observed covariates, confounders, mediators, colliders, and outcomes before estimation begins.

Discussion Prompts

• Which variables determine treatment assignment?
• Which variables determine the outcome?
• Which variables are measured before treatment?
• Which variables are downstream of treatment?
• Which paths should be blocked?
• Which variables should not be adjusted for?

Applied Framing

In industry settings, DAGs can prevent teams from adjusting for post-treatment variables such as engagement, usage, or retention when those variables are themselves affected by the intervention.