Congratulations for making it through the textbook. Our final interview will be with the man himself. Scott Cunningham will come onto the livestream and answer questions.
Be sure to list them here!
We may also do one other livestream on Jan 8th to cover the conclusion chapter
I’ll be skiiing but I’ll do my best to attend! Some questions:
- Whats your favourite CI method which isn’t in the book?
- I always see Twitter spats between (mainly) Judea Pearl and experimental design statisticians, usually taking the form of each side saying that other doesn’t read their literature at that their side is right on how to do CI. What side of the argument do you lie if any?
- Has Scott looked at incorporating any Bayesian methods into his CI analyses?
Thank you for carrying out! Here are my question:
- What is the role of Bayesian Inference in Causal Inference research? Is it’s importance rising or constant?
- Do you yourself use Bayesian Inference?
- Are there methods which are more associated with specific research areas? If so why?
- What are Milestones achieved in the past years in Causal Inference? What has to be resolved?
- With which method presented in the book, are you the most familiar with? Has your view of any of the methods changed as a result of writing the book?
This has been moved up to January 8th! I’ll post the details here and over email as well
My question is perhaps a bit specific, but here it is:
In difference-in-differences, specifically regression-based twoway fixed effects with differential timing, what method do you currently see as the best way to deal with heterogeneity in the treatment effects over time?
Has there been more development with matrix completion and the use of machine learning to avoid this issue?
More generally, what is the future of the Difference-in-Differences quasi-experimental design?
In doing some research prior to this interview found this article. Does this help and are there specific questions you’d have for Scott after reading what’s here?
This is a great resource! Thanks for finding it. It’s very helpful. I suppose my question would then be:
In difference-in-differences, specifically regression-based twoway fixed effects with differential timing, there are a number of methods that have been developed to deal with heterogeneity in the treatment effects over time. Are there any general guidelines for which methods are more appropriate for a given situation? Is there a ‘best method’ for dealing with heterogeneity? Does the matrix completion/machine learning method offer any advantages over the other available methods?