Does the dag functionality account for Direct vs. Indirect channel effects? #1837
Replies: 2 comments
-
I have the same question. I ran it with and without the dag and the results are exactly the same. Not sure if I am doing something wrong |
Beta Was this translation helpful? Give feedback.
-
Hey guys, let me try to add clarity! Short answer ✌🏻 Imagine a DAG like A → B → Y and A → Y.
The DAG method will help you to identify this relationships, and decide if you need more than one using pymc. Here you can take a look to an article where based on a DAG using PyMC-Marketing, a few models are build and those can recover direct and indirect effect. The main point in the article it's not recover indirect effect, but can be as easy as take one effect minus the other. Again, the DAG its build for identification, as mentioned in the notebook example. How to estimate the indirect relationships its another story which can be kick-off once we identify they exist, and what are the nodes to control (that last part it's where we are helping, second part it's easy). |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
I'm interested in the
dag
functionality to pymc marketing, but looking through it's not clear how direct vs. indirect channel effects are handledSay for instance, increasing TV spend increased the effectiveness of paid search spend for the following few weeks - that would be an indirect effect, and that contribution should get moved away from paid search to TV.
If I specify the channel relationship in a dag input (
MMM(..., dag=causal_dag)
), does this get accounted for? Or does it only account for immediate causation within the same time periods?Beta Was this translation helpful? Give feedback.
All reactions