Introduction Livestream Discussions to keep track of

Causal Inference Book Club Introduction
1

today in the livestream we had many discussions that were left open and were hoping will be addressed later in the book. Youtube is still processing the chat transcript but in the meanwhile here’s what I remember

Causal assumptions and causal Conclusions

What assumptions are needed to perform causal inference? Once we make those assumptions what type of conclusions can be make?

Differing schools of thoughts in Econometrics

There are different schools of thoughts even within the econometrics group. Someone had pointed out that the two schools mentioned converge to the same thing, but nonetheless we’re curious why there are two to begin with. This book hopefully will give us both in one text.

Randomization, what kind its needed, how much, and is it worth it

Many discussions about randomization as that was one of the key takeaways from the introduction. Does randomization need to be included in a controlled way, or can we take advantage of “natural” randomization that occurs without explicit design?

In this last topic synthetic control was brought up a couple of times.

Is there anything else that I missed?

1 Like
2

Does randomization need to be included in a controlled way, or can we take advantage of “natural” randomization that occurs without explicit design?

AFAICT this relates directly back to the exogenous / IID assumptions - if I can control the experimental assignment, then I can be sure that assumption holds (because I know my dice rolls are fair).

But the error could still be IID w/o assignment, if the data-generating-process is well-behaved. So the assignment is sufficient but not necessary for most regression models. But I am unsure how we can be confident in making that leap.

Counterpoint: In practice, if I ran an RCT, then did a partial dependence plot on the treatment variable and saw a funnel shape typical of non-iid, I think I would still be concerned even if I had faith in my assignment process.

Just like you often hear “correlation does not imply causation”, another (less-common) saying is “no causation without manipulation” - causality - is it always "no causation without manipulation"? - Cross Validated