@soboleffspaces I have a question: given a DAG that is compatible with an observational dataset. But I have an extra variable, that is not yet part of this DAG, for which I do not have the data yet. What can be said about the stability of this DAG if later on this var is added?
1) DAG represents knowledge, rather than data. 2) The variable's inclusion in the graph is often determined by the level of abstraction. 3) Our assumption is that the dependence of pa(X) -> X remains unchanged when we modify the graph.
Does this mean that the existing edges would not change due to the inclusion of this new variable? Only new edges could be add, but the old ones would not change?
Well, DAG may evolve as we learn more. But adding a node shouldn’t affect the existing arrows, unless of course you aggregate some nodes👇🏻
Not me. It’s the graph by David Eppstein commons.m.wikimedia.org/wiki/User:Davi…
This is exactly what I mean! So I expect if I learn the DAG from obs data without the extra var and then learn a new DAG’ on the data with the extra var, that DAG’ has exactly all the edges in original DAG, perhaps with extra edges in DAG’ coming from the extra var. Do you agree?
#DependencyGraph represents knowledge, not data. We use #CausalDiagram (dependency graph with designated cause and effect nodes) to identify non-causal paths and classify them as either open or blocked. This leads an adjustment set. #causaltwitter sobolevspace.com/teaching/f/sho…
United States 趨勢
- 1. Falcons 29.2K posts
- 2. Bears 58.3K posts
- 3. Bryce Young 14.7K posts
- 4. Josh Allen 22.2K posts
- 5. Chiefs 29.8K posts
- 6. Vikings 29.3K posts
- 7. Chargers 20.7K posts
- 8. JJ McCarthy 11.4K posts
- 9. #KeepPounding 5,527 posts
- 10. Panthers 25.7K posts
- 11. Packers 40.1K posts
- 12. Ramsey 9,890 posts
- 13. Bengals 23.6K posts
- 14. Bucs 12.8K posts
- 15. Hyatt 4,576 posts
- 16. Troy Franklin 1,142 posts
- 17. Bills 110K posts
- 18. Raheem 7,141 posts
- 19. Mahomes 6,287 posts
- 20. Steelers 37.8K posts
Something went wrong.
Something went wrong.