Table 3.

Examples of identification assumptions

Assumption	Basic explanation of meaning
Exchangeability †	This assumption is generally true if there are no unmeasured common causes of variables that are part of the treatment strategies (Table 2: e.g., baseline or postbaseline treatment(s), censoring) and the outcome (informally, if there is no unmeasured confounding) and we have not conditioned on a variable that is affected by the treatment variable(s) [32,46].
Positivity	This assumption is true if, for every possible combination of measured confounding variables, individuals with those characteristics have a positive probability of following any of the treatment strategies of interest.

^† Full exchangeability is generally not required if weaker conditions (e.g., mean exchangeability, sequential conditional exchangeability, or others) hold [32].