Does the new sensitivity analysis feature also work for CATE?

[kirtanp]

Hi,

I saw there is a new sensitivity analysis feature since a few weeks and I am using it for my analysis because I was already using the LinearDRLearner. I have two questions

1. In the line here, is the definition of rho missing?
2. Can I get sensitivity analysis for CATE estimates as well somehow? For instance it's possible to get an ATE interval using est.ate_interval(X=X_subset) where est is a LinearDRLearner instance. Could I get the sensitivity summary for this subset without fitting a separate model on X_subset?

I would offer to work on both of these if they are not done yet and if they make sense.

[kbattocchi]

Unfortunately the link in your first question didn't come through correctly - could you clarify where you're talking about?

For your second question, as far as I've seen, the literature that I've seen in this area is primarily concerned with the sensitivity of single-dimensional estimands; I have some ideas about how to extend this to the multi-dimensional parameters used by some of our CATE estimators (particularly LinearDML and LinearDRLearner), but it won't necessarily be completely straightforward.