Limitations of Causal Graphs
Causal graphs do not depict the size or quantify the strength of the relationship
There is always a possibility of uncontrolled confounding in DAG, which can not be estimated
A DAG is dependent on the information provided. Lack of information can lead to wrong analysis.
In most cases, we construct multiple DAG, without being able to choose the correct one.
A variable can be a mediator, collider, and confounder simultaneously, which can be interpreted differently in different research questions leading to different results.
If we interpret a collider as a confounder and condition on it, it can lead to a false causal relationship.
Last updated