In https://github.com/amit-sharma/causal-inference-tutorial/blob/master/python/simple_recommender_example.py.ipynb you merely average between strata. This works only because the starta size is equal. In the general case, one needs to multiply by the probability to be in the strata and sum. I think this will confuse students.

Hi Amit, looks like the video was moved elsewhere and stopped working.

• https://mediasite.kellogg.northwestern.edu/Mediasite/Play/8e78dc83c6fb4d20abeeb18028a8f7071d?catalog=1533bdef-0c88-4513-ad97-5fce50c92e62

Any chance that you can update this link in the README? Thanks.

Hello, I am very interesting in this tutorial, could you make the video publicly accessible?

for the R code, for the intervention of recommending algorithem, the effect on CTR is as following

naive_observational_estimate <- function(user_visits){

Naive observational estimate

Simply the fraction of visits that resulted in a recommendation click-through.

est =
summarise(user_visits, naive_estimate=sum(is_rec_visit)/length(is_rec_visit))
return(est)
}

in the beforehand code ,the estimate formulas is
"""
sum(is_rec_visit)/length(is_rec_visit)
"""
but how to derivative and get this formuls?
thanks

The link to the video tutorial redirect me to a Microsoft login page in which I enter my outlook.com email ID and password. I then got the following error that I was asked to send to the administrator for help. Please advise.

Request Id: 9a37ea7e-6327-428c-b1e1-6271be2f2800
Correlation Id: 67f7e30b-5d40-4f07-b150-3d617bd6e213
Timestamp: 2023-04-01T02:57:33Z
Message: AADSTS50020: User account '[email protected]' from identity provider 'live.com' does not exist in tenant 'Northwestern University' and cannot access the application 'https://kellogg-northwestern.hosted.panopto.com/Panopto/Pages/Auth/Login.aspx'(Panopto Kellogg) in that tenant. The account needs to be added as an external user in the tenant first. Sign out and sign in again with a different Azure Active Directory user account.

Very cool repository +1: would like to ask: is it safe to assume in the case of multiple-regression (for example predicting house prices based on location, surface, condition...) that adjusted correlation coefficients obtained as usual are the same as stratified causal coefficients? I think by reading Pearl's book that they should give the same forecasting model, please could you share your thoughts?

Many thanks for your presentation. I have been reading on causal inference and judea pearls do why.

One thing i do not understand is the relationship (if any) between the estimand computed in step 2 and the propensity score matching. Are these two processes linked?

(Also when we say estimand is this what we mean by SCM (structural causal model) .

e.g. i thought that the estimand is what is used to compute the causal effect, and this estimand in turn is computed from the observed data using conditioning. However when we get to step 3 and start calculating 'propensity scores' to get a match between no treatment and treatment where does the estimand come into this? Is the propensity score matching an approximation to the estimand? How is the propensity score calculated in dowhy package... some source code may help me understand better.

Link is not working again, requires signing in 👀

Originally posted by @villasv in #1 (comment)

Your PowerPoints are not opening. PowerPoint says it wants to repair them, but cannot, and they do not open. Can you please clean those up?