More in this series 👇 Propensity Scores: th-cam.com/video/dm-BWjyYQpw/w-d-xo.html Do-operator: th-cam.com/video/dejZzJIZdow/w-d-xo.html DAGs: th-cam.com/video/ASU5HG5EqTM/w-d-xo.html Regression techniques: th-cam.com/video/O72uByJlnMw/w-d-xo.html Intro to Causality: th-cam.com/video/WqASiuM4a-A/w-d-xo.html Causal Inference: th-cam.com/video/PFBI-ZfV5rs/w-d-xo.html Causal Discovery: th-cam.com/video/tufdEUSjmNI/w-d-xo.html
Hi Shawhin, thanks for the valuable information in the video. I have one question. What if we have more than 1 treatment effect in the post-period? Let's think about a campaign & sales scenario. We were using 3 campaigns and then we launched 2 more campaigns at a certain time (became totally 5). In the case of 1 newly launched campaign, I was planning to use a causal model to learn its effect but 2 campaigns together will create noise. How can I distinguish their effect from each other ? Do you know any alternative method for it?
That’s a good question. Applying this stuff to the real world is often non-trivial, so it’s hard to say what would be best for your specific use case. But here are a couple thoughts. For flavor of causal effects I talk about here, it’s critical to have a control group to which you can compare all other treatment groups. For a sales/marketing case where you have multiple campaigns which impact a handful of kpis over time, it is more difficult to define the causal effect. However it is possible if you are able to reasonably model how variables interact. I find drawing out a causal graph to be a helpful first step.
Correction @5:26 - I meant to say "average treatment effect" 😅
More in this series 👇
Propensity Scores: th-cam.com/video/dm-BWjyYQpw/w-d-xo.html
Do-operator: th-cam.com/video/dejZzJIZdow/w-d-xo.html
DAGs: th-cam.com/video/ASU5HG5EqTM/w-d-xo.html
Regression techniques: th-cam.com/video/O72uByJlnMw/w-d-xo.html
Intro to Causality: th-cam.com/video/WqASiuM4a-A/w-d-xo.html
Causal Inference: th-cam.com/video/PFBI-ZfV5rs/w-d-xo.html
Causal Discovery: th-cam.com/video/tufdEUSjmNI/w-d-xo.html
Thank youuuuuuuuuu! I was really confused before I discover your channel by sudden.You really are helping me prevent failing:)) Merci
This is a one million times better explanation than my professor. Thank you sir!
Glad it made sense :)
Thank you so much for your great explanations
Happy to share!
Thank you! Great content
Thanks 😁
Hi Shawhin, thanks for the valuable information in the video. I have one question.
What if we have more than 1 treatment effect in the post-period?
Let's think about a campaign & sales scenario. We were using 3 campaigns and then we launched 2 more campaigns at a certain time (became totally 5). In the case of 1 newly launched campaign, I was planning to use a causal model to learn its effect but 2 campaigns together will create noise. How can I distinguish their effect from each other ? Do you know any alternative method for it?
That’s a good question. Applying this stuff to the real world is often non-trivial, so it’s hard to say what would be best for your specific use case. But here are a couple thoughts.
For flavor of causal effects I talk about here, it’s critical to have a control group to which you can compare all other treatment groups.
For a sales/marketing case where you have multiple campaigns which impact a handful of kpis over time, it is more difficult to define the causal effect. However it is possible if you are able to reasonably model how variables interact. I find drawing out a causal graph to be a helpful first step.
@@ShawhinTalebi Thank you :)
Great video!