@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👇🏻
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 Trends
- 1. #ใครในกระจกEP5 9,339 posts
- 2. #happybdayTYGA N/A
- 3. $NVDA 52.6K posts
- 4. GeForce Season 4,307 posts
- 5. Peggy 36.2K posts
- 6. $LMT $450.50 Lockheed F-35 N/A
- 7. Martha 18K posts
- 8. Sonic 06 3,428 posts
- 9. Saba 10.8K posts
- 10. Jason Crow 1,536 posts
- 11. Comey 51.6K posts
- 12. Poverty 50.2K posts
- 13. Stargate 5,484 posts
- 14. MLB TV 1,988 posts
- 15. Sumrall 1,552 posts
- 16. Halligan 23K posts
- 17. Nvidia 62.9K posts
- 18. NASA 48.1K posts
- 19. Jensen 16.8K posts
- 20. #ComunaONada 5,229 posts
Something went wrong.
Something went wrong.