Treatment Assignment, Treatment Assignment and Treatment Assignment

As promised, I am excited to share what I learned during the SGPE Summer School. In the coming posts, I will cover what I learned throughout the summer school. This post will reflect on the four intensive days and highlight the most crucial takeaway for me (TLDR at the end if you want to skip).

But first, a sincere thank you to our instructor, Scott Cunningham, for all his effort in teaching. It was a constant for him throughout four days, and he definitely took a beat every day from all of us with conversations but somehow managed that passion throughout. It shows his passion for what he is doing, and you are truly inspiring. Also, a shoutout to Hector Rufrancos for great management.

Key Takeaways: It’s All About Treatment Assignment

For most of us, we have always been taught that causal inference is “Identification, Identification and Identification”. And it definitely is and remains true. But what I learned from Scott is that identification is a result of something else, knowing treatment assignment. If you understand the nature of this treatment assignment mechanism, the identification part is simply choosing the appropriate method. So what does it look like to know treatment assignment? It very closely resembles the following statement from the Journal of Economic Perspective paper by Angrist and Krueger. And this is the same with all the other methods. Understanding how things work in real life dictates the treatment assignment.

This statement is true for instrumental variables, DID, unconfoudedness, RDD and all the others. So whichever topic we are working on, knowing the context, working mechanisms, decision makers, how the treatment is assigned, who decides treatment assignment is exceptionally important for causal inference.

I leave today’s post with the current project that Scott is involved in, which showcases what it looks like in real life to understand treatment assignments. He is concerned about the suicide and self harm in prisons, which apparently is a huge problem due to the scale of prisons in the US. To understand what is truly going on he visited hospitals and met people day in and day out. After a few meetings, he got to know that whenever an individual was brought to prison, he/she would be subject to a mental health test and would receive a score. So his causal lamp lit up, a regression discontinuity approach. He kept these conversations going on, he found that the score were from 0 to 4, and anyone with score of 3 and 4 would go to mental health court, and others to normal procedings. I won’t go into the nitty gritty of the problem. So, regression discontinuity was off the table. He kept meeting and asking questions. He found that the way these mental health tests were assigned was that all of the tests to be done was arranged in alphabetical order, and then it went to mental health examiners. Meaning that there was randomness. For those who have not yet studied or heard this, this is a perfect setup for a lineancy design. It is an instrumental variable method (More on this in future posts). Using this method, he can estimate the impact of being labeled as mentally ill on suicide in US jails. He is also doing a lot of other important work for prisons, which I wont go into.

Subscribe now

TLDR / Conclusion

1. Context is King

2. Real-World Understanding (Get out of your data)

3. Decision Makers Matter (How is treatment decided? Who decides who gets the treatment? How do they decide how someone gets treated?

4. Keep Talking (Conversations with decision makers, stakeholders, and your friends)

Flexibility is key, and persistence to understand true mechanisms pays off. There is so much you can do if you learn how the variables of your interest interact. Knowing this mechanism will allow you to make a clear identification. Had he not kept talking, he would not have known that there was randomization within the system. And as researchers, our job is to uncover these mechanisms through persistent questioning, reasoning, and keen observations.

So, treatment assignment, treatment assignment, and treatment assignment. Do share your thoughts. Next week, we will look at the Potential Outcomes framework, a new way of looking into causality and causal inference.

Refer a friend

This post was originally published on Substack. If you enjoyed this content, consider subscribing to my newsletter for regular updates.