@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 shouldnโt affect the existing arrows, unless of course you aggregate some nodes๐๐ป
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 DAGโ on the data with the extra var, that DAGโ has exactly all the edges in original DAG, perhaps with extra edges in DAGโ 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/shoโฆ
Part 3 is here ๐๐ป sobolevspace.com/teaching/f/shoโฆ
United States Trends
- 1. Texans 24.4K posts
- 2. #MissUniverse 51.9K posts
- 3. James Cook 4,076 posts
- 4. #TNFonPrime 1,371 posts
- 5. Sedition 226K posts
- 6. Davis Mills 1,555 posts
- 7. Will Anderson 2,560 posts
- 8. Christian Kirk 2,322 posts
- 9. Shakir 3,033 posts
- 10. Lamelo 9,445 posts
- 11. Treason 120K posts
- 12. Prater N/A
- 13. Cheney 104K posts
- 14. #htownmade 1,177 posts
- 15. Paul George 1,968 posts
- 16. Seditious 126K posts
- 17. TMNT 6,193 posts
- 18. #BUFvsHOU 1,410 posts
- 19. Nico Collins N/A
- 20. Jalen Suggs N/A
Something went wrong.
Something went wrong.