,
3 tweets,
3 min read
Read on Twitter
@PHuenermund @yudapearl @fuzzydunlop123 Okay, to briefly summarize, suppose that I am trying to rule out backdoor paths from an instrument to a dependent variable. What I would do is ask, "Why does the instrument vary? What evidence can be brought to bear to verify that this is why it varies?"
@PHuenermund @yudapearl @fuzzydunlop123 Additionally, I would conduct placebo tests to try to verify that there was no correlation between Z and Y when the hypothesized channel was turned off.
@PHuenermund @yudapearl @fuzzydunlop123 And I would conduct balance tests to show that the relationship between Z and observables fits the assumed relationship between Z and unobservables.
