120 Bayesian Networks: With Examples in R
Furthermore, even when dealing with interventional data collected from a
scientific experiment (where we can control at least some variables and ob-
serve the resulting changes), there are usually multiple equivalent BNs that
represent reasonable causal models. Many arcs may not have a definite direc-
tion, resulting in substantially different DAGs. When the sample size is small
there may also be several non-equivalent BNs fitting the data equally well.
Therefore, in general we are not able to identify a single, “best”, causal BN
but rather a small set of likely causal BNs that fit our knowledge of the data.
An example of the bias introduced by the presence of a latent variable
was illustrated by Edwards (2000,