One of the best videos on Causal Inference. Just one request, Please have some videos on "Quasi Experiments" and their implementation in R/Python. There's so much of theory about it and so less of how it's being implemented practically.
Here are some hands-on tutorials on causal inference in Python/R: * Causal Inference Logit Propensity Score Matching (PSM): medium.com/grabngoinfo/causal-inference-logit-propensity-score-matching-psm-c290fd522bb8 * OLS Treatment Effects Estimation Using Python Package Causal Inference: medium.com/grabngoinfo/ols-treatment-effects-estimation-using-python-package-causal-inference-393306d53940 * 8 Matching Methods for Causal Inference Using R: medium.com/grabngoinfo/8-matching-methods-for-causal-inference-using-r-3c32c6aeb498 * One-to-one Matching on Confounders Using Python Package Causal Inference: medium.com/grabngoinfo/one-to-one-matching-on-confounders-using-python-package-causal-inference-5cce5f348863 * Time Series Causal Impact Analysis in Python: medium.com/grabngoinfo/time-series-causal-impact-analysis-in-python-63eacb1df5cc More step-by-step tutorials on causal inference are on GrabNGoInfo Medium page (medium.com/@AmyGrabNGoInfo/list/causal-inference-633898947606) and TH-cam channel (th-cam.com/video/BCGFul1FEAQ/w-d-xo.html)
Thank you so much for this great video! I always find Causal Inference to be extremely useful. Wonder if you could introduce more contents on how we could discover the causal relationships (or how to draw out the DAG) out of observational data in the future?
Great topic - would love to see more examples of this! Maybe exploring what drives certain metrics. Like what variables affect "Demand" more and testing Price, Month of Year, Lag of Demand, etc.
I wonder if the explanation could be more well structured.... It's like jumping from one to another concept.... But the cases given are very insightful thank you
Thanks for the explanation! I have a question about the example of propensity score in the video 18:58 . The p(exposure) means the probability to be exposed to harmful contents and other variables are features we used in the prediction model. If so, user 3 and user 9 should have the same prediction result in my opinion because they have same values in all variables(whether actually being exposed to harmful contents shouldn't be taken into consideration). Why are their p(exposure) are different?
Great video, a quick question, how can we select the training dataset/label of the propensity score model? If the label is biased, the propensity score will be impacted as well
Awesome video as usual, and great timing as more companies find the limitation of A/B testing and look for alternatives. Just curious, in this video, regression and matching are two techniques that can be used either or, not something we need to combine together within the same study?
Propensity matching seems like a really complex and hard to get right model. How much does it get causality correctly and how do you know you can rely on this method?
Appreciate your content. How about Doubly Robust Estimation which combines both regression and propensity? I am curious which counter factual techniques you might talk about next? I liked reference blog which explains doordash's actual problem, but I don't think it may work for many use cases. I would really appreciate if you can reference all techniques different companies use for counter factual in practice. Also, I face similar problems while evaluating contextual bandit model offline and not 100% sure if IPS (Inverse Propensity Score ) would be good enough technique.
![ออกพรรษาน้ำตาหล่น - แอน อรดี [OFFICIAL MV]](http://i.ytimg.com/vi/PFZtOcY9ONA/mqdefault.jpg)
One of the best videos on Causal Inference. Just one request, Please have some videos on "Quasi Experiments" and their implementation in R/Python. There's so much of theory about it and so less of how it's being implemented practically.
Here are some hands-on tutorials on causal inference in Python/R:
* Causal Inference Logit Propensity Score Matching (PSM): medium.com/grabngoinfo/causal-inference-logit-propensity-score-matching-psm-c290fd522bb8
* OLS Treatment Effects Estimation Using Python Package Causal Inference: medium.com/grabngoinfo/ols-treatment-effects-estimation-using-python-package-causal-inference-393306d53940
* 8 Matching Methods for Causal Inference Using R: medium.com/grabngoinfo/8-matching-methods-for-causal-inference-using-r-3c32c6aeb498
* One-to-one Matching on Confounders Using Python Package Causal Inference: medium.com/grabngoinfo/one-to-one-matching-on-confounders-using-python-package-causal-inference-5cce5f348863
* Time Series Causal Impact Analysis in Python: medium.com/grabngoinfo/time-series-causal-impact-analysis-in-python-63eacb1df5cc
More step-by-step tutorials on causal inference are on GrabNGoInfo Medium page (medium.com/@AmyGrabNGoInfo/list/causal-inference-633898947606) and TH-cam channel (th-cam.com/video/BCGFul1FEAQ/w-d-xo.html)
Great! I just uploaded a video on causal inference too haha. Glad it's getting at least some exposure :)
This explanation of the framework is as clear as crystal. Thank you for the effort. Appreciate it!
Great content! Also, thank you for attaching the sources for learning in the description. I am glad I subscribed 🙂
Great overview of the techniques and their motivations
watching this vid after completing the course on coursera - really helped a bunch
So glad to hear you found it helpful, Rikki! 😊
Awesome content! Some references to real world implementation would be super helpful. Looking forward to the next video on this topic.
Great content. Very interested in causal inference.
Thank you so much for this great video! I always find Causal Inference to be extremely useful. Wonder if you could introduce more contents on how we could discover the causal relationships (or how to draw out the DAG) out of observational data in the future?
so helpful, thank you and Yuan so much!
Great topic - would love to see more examples of this! Maybe exploring what drives certain metrics. Like what variables affect "Demand" more and testing Price, Month of Year, Lag of Demand, etc.
Thanks for the suggestion! 😊 Really appreciate your feedback!
I wonder if the explanation could be more well structured.... It's like jumping from one to another concept.... But the cases given are very insightful thank you
Thanks for the explanation! I have a question about the example of propensity score in the video 18:58 . The p(exposure) means the probability to be exposed to harmful contents and other variables are features we used in the prediction model. If so, user 3 and user 9 should have the same prediction result in my opinion because they have same values in all variables(whether actually being exposed to harmful contents shouldn't be taken into consideration). Why are their p(exposure) are different?
yeah same question here, @Emma do you mind confirming?
I don't think the toy table showed all the features to calculate p(exposure)
Loved this video! Thanks! :)
great video! When would the second half of causal inference be released?
Good job!
Great video, a quick question, how can we select the training dataset/label of the propensity score model? If the label is biased, the propensity score will be impacted as well
Awesome video as usual, and great timing as more companies find the limitation of A/B testing and look for alternatives. Just curious, in this video, regression and matching are two techniques that can be used either or, not something we need to combine together within the same study?
Really helpful.This actually helped me in of one my project.Also can you made videos on Linear Programming and Multibanded solution
Absolutely, I'm working on creating more content, stay tuned!
Propensity matching seems like a really complex and hard to get right model. How much does it get causality correctly and how do you know you can rely on this method?
Appreciate your content. How about Doubly Robust Estimation which combines both regression and propensity?
I am curious which counter factual techniques you might talk about next?
I liked reference blog which explains doordash's actual problem, but I don't think it may work for many use cases.
I would really appreciate if you can reference all techniques different companies use for counter factual in practice. Also, I face similar problems while evaluating contextual bandit model offline and not 100% sure if IPS (Inverse Propensity Score ) would be good enough technique.
2:11 4:50 7:55
Fantastic interview. I have a question - PSM seems very similar to logistic regression. Is psm an extension of it?
How can we know what variable ? Just check the coefficient of the variable that would affect y keeping the confounders fixed ?
11:32 How can we use Structural Equation Modeling (SEM) to detect this causal relationships ?
Causal inference 很重要。。面试也会问。。谢谢分享~~。。能不能再讲一下confounding variable怎么处理~~
Hey Emma, I don't know if it's just me, but the links in your promotional emails seem to be broken. Just wanted to let you know
Question： Can I use the Propensity score matching technique in a regression model? I mean the "treatment" is a continuous variable?