Identification

Conditions for identifiability

Key Conditions for Identifiability

Consistency: The observed outcome equals the potential outcome under the observed treatment.

Positivity: Every treatment level has a positive probability for each level of covariates.

Exchangeability / Ignorability: No unmeasured confounding (in the potential outcomes framework).

Graphical Criteria: Like backdoor or front-door adjustment sets (in structural causal models).

If these hold, identifiability is possible.

Also, full observability does not imply identifiability.