Causal sufficiency

Causal sufficiency is the assumption that all common causes of the measured variables in a causal system have themselves been measured. It is the silent partner of the Causal Markov condition: the Markov condition tells us how to read independencies from a graph, but causal sufficiency guarantees that the graph contains all the variables that matter. When causal sufficiency is violated — when unmeasured confounders lurk behind the scenes — the conditional independencies observed among measured variables become misleading, and algorithms like the PC algorithm or MMPC recover not the true graph but a distorted projection of it onto the observed variables.

The assumption is almost always false in practice, yet it is rarely tested because testing it would require measuring the unmeasured. This epistemic asymmetry makes causal sufficiency less an assumption than an article of faith, and it raises the question of whether causal discovery is possible at all in open systems where the set of relevant variables is never fully known.