@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 เทรนด์
- 1. Raiders 81.8K posts
- 2. #WWERaw 171K posts
- 3. Cowboys 51.5K posts
- 4. Pickens 21.2K posts
- 5. Gunther 21.8K posts
- 6. Geno 15.8K posts
- 7. #WickedForGood 8,378 posts
- 8. Chip Kelly 2,389 posts
- 9. Jeanty 7,050 posts
- 10. Pete Carroll 3,511 posts
- 11. Roman 74.8K posts
- 12. Jlexis 8,361 posts
- 13. #Dragula N/A
- 14. Grok 4.1 1,307 posts
- 15. Mark Davis 1,509 posts
- 16. #GMMTV2026 204K posts
- 17. Becky 54.2K posts
- 18. Brock 21.4K posts
- 19. Dolph 43.1K posts
- 20. Sigourney N/A
Something went wrong.
Something went wrong.