













n 
No missing data
or hidden variables



n 
Let G be a
causal graph then:



n 
Causal
Markov Assumption




n 
Data is
generated according a Bayesian Network with


structure G



n 
E.g., Can’t have
G=(X Y) and X dependent on Y



n 
Faithfulness
Assumption



n 
No probabilistic
independencies other than those



represented by G




n 
E.g., Can’t
have G=(X®Y)
and X independent of Y

