11 - Causal Discovery from Interventions

[@akwstr]

These videos are extremely helpful. Thanks so much for making them!

[@slippedgrey]

hi, can you please specify what necessity and sufficiency mean in these graphs? thank you!

[@gwillis3323]

In your 3 variable example, is it not the case that from the correlations in the raw data, we know that the graph is fully connected (as all variables are correlated to one and other, and also no two variables are independent of one and other, conditioned on the 3rd), and thus we can get to the answer by only intervening on B (and it's actually not necessary to additionally intervene additionally on C, as I feel was implied) ?

[@gwillis3323]

Upon revisiting this, I'm pretty sure the answer is no. If we just intervene on B, we have no way of directing the A-B edge. I think when I originally asked the question, I had confused the collider at C with an immorality. Because it's a collider but not an immorality, there's nothing we learn from intervening on the empty set other than that this is a fully connected DAG.

[@Grouahh]

16:37 I guess you wanted to say 'from B to C'.

[@BradyNealCausalInference]

Good catch. I think I meant to say "there's no edge from C to B."