Showing strength of dependency in DAG (causal inference)

Question

I'm using DAG for causal inference, the code is written in python. But when I obtain data by implementing GOLEM or IACLiNGAM I wish also to see the strength of the dependency on the graph. Is it possible to somehow evaluate or show that? Or maybe it's only possible to get from feature importance after applying the ML model to the data?

Answer 1

You can obtain the weights of the edges through regression on the causal parents

If the weights of the causal relationships (assuming a linear additive noise model) are not returned directly by the algorithm, you can obtain them by regressing each variable on its causal parents in the graph.

A big NOTE OF CAUTION:

GOLEM and LiNGAM make completely different and opposing assumptions about the data generating process (Gaussian vs. Non-Gaussian noise). Moreover, GOLEM is sensitive to the data scale, meaning that it cannot be expected to perform well in many real-world contexts where the data scale is arbitrary. More generally speaking, causal discovery of DAGs from data is a tricky business and requires careful consideration of which assumptions and methods may be adequate.