01.03: When do standard methods fail/succeed?

r-causal / causal-inference-in-R

Causal Inference in R: A book!

https://www.r-causal.org/

181 stars 48 forks source link

Closed malcolmbarrett closed 1 year ago

malcolmbarrett commented 2 years ago

[x] https://github.com/malcolmbarrett/causal-inference-in-R/issues/38
[ ] https://trialsjournal.biomedcentral.com/articles/10.1186/s13063-022-06097-z
[ ] Talk about when RCTs/AB tests don't work well, e.g. unethical, impractical, or too long (many RCTs and AB tests measure outcomes that are proxies and too short of time period)

malcolmbarrett commented 1 year ago

Some notes on @LucyMcGowan's thoughts on why we use causal models:

Why not conditional models? We want marginal effects for some target group to make decisions

Will be the same if no hetergenous treatment effect but we're assuming this is not the case

To get the ATE you don't need to integrate out the confounders, just a single number (propensity score)

Even in an RCT, where distribution of confounders the same, you still assume that the ATE holds only in populations where this is true