Question: Clarification on Edge properties from FCI algo

[arainboldt]

In the docs the edge properties are explained as:

"""
edges: list. Contains graph’s edges properties. If edge.properties have the Property ‘dd’, then there is no latent confounder. Otherwise, there might be latent confounders. If edge.properties have the Property ‘nl’, then it is definitely direct. Otherwise, it is possibly direct.
"""
The abbreviations seem odd to me. Unfortunately, I can't find the original paper on the internet free and I can't find another reference to these abbreviations. Consequently, I'm inquiring here for clarification.

Is it possible that dd and nl have been switched? I.e. dd means definitely direct and nl means no latent?

Additionally, there is no explanation fro pl. Does it mean possible latent confounders?

I'm guessing that pd means possibly direct.

Thanks for the clarification!

[jdramsey]

These are from Tetrad PAG "graph coloring" and are as follows.

"nl" means "no latent"--i.e., this is a visible edge--i.e., if X->Y is marked as "nl" then there is no latent confounder for X and Y. This is an edge whose coefficient can be estimated.
"pl" means "possibly latent"--i.e., this is possibly not a visible edge.
"dd" means "definitely direct"--i.e., one can expect this edge to actually exist as a direct edge in the true DAG (with latents)
"pd" means "possibly direct"--i.e., this may go through other variables in the underlying DAG (with latents), like X->L1->...->Ln->Y for some latents L1, ..., Ln

[jdramsey]

I checked with Peter about the definition of "possibly direct"; the above is correct sfaik.

[kunwuz]

Thanks so much! This issue has just been solved in this PR by @zhi-yi-huang. Please let us know if there are any further questions.

[arainboldt]

Thanks a lot to both of you!