parameter theta is identified if it is a unique functional of the observed data distribution.
got it, thanks!
Judea also calls it the ‘structural causal model.’
if you are curious, this is the paper Eli is referring to:
these sides comments are awesome
semi-parametric inference: estimating parameters without having to specify a parametric likelihood (which will lead to problems if it’s wrong). Search is trying to use data to help you figure out what graph to use (which is a precondition for work Eli discussed).
Thanks, Eli! Great talk!
Data fusion is important to deal with for these global problems like COVID-19 — where there is no one model that faithfully represents different jurisdictions (even though they have commonalities).
In other words, NYC and Italy are similar, but it’s probably not reasonable to have a single causal model — so how do we use multiple models?
Elias Barenboim is a leading causal researcher on this problem (although we are starting to think about this stuff also).