,
3 tweets,
2 min read
Read on Twitter
@pablogerbas @yudapearl I am not claiming applied researchers trained in SCM list the 150,000 variables. Often, they content themselves with (not very defensible) backdoor strategies. But that is an error. I'm saying that instead that should often search for quasi-experiments.
@pablogerbas @yudapearl No one is disagreeing that once you have a quasi-experiment, a DAG can be a good way of clarifying it and determining how to transform it into an estimation strategy (although I think @yudapearl sometimes overstates the absolute necessity of doing this).
@pablogerbas @yudapearl The question is whether the search for the quasi-experiment should come first, pre-DAG. And it seems to me that it must. The reason I emphasize this is because I think this pre-DAG step is *the* critical step in credible causal inference.
