Every causal analysis starts by naming three roles: treatment (the intervention), outcome (what you measure), and confounders (variables that influence both treatment and outcome).
Treatment can be binary (received email yes/no), continuous (dose level), or categorical (program type). Outcome can be continuous (revenue), binary (converted), or time-to-event (survival).
Confounders in practice
In medicine, age and baseline severity often confound drug assignment and health outcomes. In marketing, prior engagement confounds who receives a promotion and who purchases. In policy, regional wealth confounds program rollout and employment.
Including the right confounders in your analysis is not optional — it is how you justify a causal claim in observational data.
What to list before you analyze
- Treatment: exactly what changed and for whom?
- Outcome: measured when and how?
- Confounders: what influenced both treatment assignment and outcome?
- Instruments or time structure: is DiD or IV plausible?