shurtler

Fantastic! Any tips on tools one can use to parse PDF including their structure? Much appreciated.

You will benefit from reading Pearl/Glymour/Jewell: Causal Inference in Statistics. A Primer

(The contrast to Imbens and Rubin is crazy)

This is one the most thought-provoking things I've read on markets and economic policy in a very long while.

Note there's an exploding literature that reads these models as causal models: http://ftp.cs.ucla.edu/pub/stat_ser/r350.pdf

Central proposal on p. 98: "What if, instead, knowledge was unbundled from the start and embedded in how researchers actually develop their knowledge? This is essentially what happens with computer code , where programmers are encouraged to use GitHub for their own repository and versioning processes. Imagine that instead of code, it was the experiments, observations, calculations, and so on, that researchers stored. They could then release those units for “forking” and development by others — all with attribution and updating built in. From that base, they could then write their journal articles linking straight back to the primordial elements of their own and others’ research. It is easy at a high level to think about how knowledge could be unbundled, but once a framework is developed, then graduate students who were learning and reading past knowledge would be encouraged to translate their own information into the new framework. The knowledge could be freed from the bounds of journals without undermining all the curation and attribution work that goes with them. And at the same time, a searchable database that is open by design would exist not for articles, pages, or PDFs, but for the knowledge itself."

"I really like penalized maximum likelihood estimation (which is really empirical Bayes) but once we have a penalized model all of our frequentist inferential framework fails us. No one can interpret a confidence interval for a biased (shrunken; penalized) estimate."

Not sure I completely understand it, but good point.

Hey, the key assumption (the "identification" assumption) is in 2.4 of the paper you referred to. I do not have time to go through the notation, but it seems to be like a fairly standard "no unmeasured confounders" assumption - so you need to ASSUME that nothing influences both the treatment and the outcome you are interested in. Then, of course, you do not need to randomize, but can use observational data.

I know that Petersen and van der Laan are well-respected researchers in causality that know exactly what they are doing. E.g., Petersen has a course (I think at Berkeley) that uses causal graphs which were pioneered by Judea Pearl (see comments below). I can only second the recommendation to dig into his work.