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.