CausalML Vignettes

Click Here for three vignettes on using Causal Forests for Causal Inference

Hello class,

I’ve uploaded 3 videos with coding vignettes on how to use the Causal Forest model in the EconML python package. I used Causal Forests to learn the conditional average treatment effect of advertisting a brand of orange juice on orange juice sales, conditional on the median income of people living near the grocery store that advertised the sale.

Note

During the last video, I get quite confused about the results I am seeing. That is because I had a bug in the original version of the official vignette. I corrected while going through the live vignette, but forgot to re-run the official vignette so didn’t realize I needed to expect different results. The bug was that I had left the ‘price’ variable out of the confounder matrix. This was because I had been using price in another analysis where it was the treatment variable (like in your homework). Because it is such an important variable, leaving it out of the analysis made it look like store median income had a large impact on the effect of the ‘feat’ variable! The effect disappeared when I introduced price back into the confounders. In hindsight, it would have been most interesting to look at price as the modifier variable to ‘feat’.

Code:

To follow along in code:

• Official, well commented notebook
• Live Notebook

Videos:

• Causal ML Vignette 1: Simulating Data
• Causal ML Vignette 2: Recovering True Effect in Simulated Data
• Causal ML Vignette 3: Quality Checks and Fit to Data