Sensitivity analysis for unmeasured confounding in GRF?

[adeldaoud]

trafficstars

I wonder what procedure you recommend for conducting a sensitivity analysis for unmeasured confounding for GRF models?

E.g., I am looking into the sensemakr package but that seems to require that one specifies the ATE estimate and its SE, but it also requires the degrees of freedom (dF). In ML setting (for non-parametric models), it is unclear what dF is.

See my question in the following link, https://github.com/carloscinelli/sensemakr/issues/51

[erikcs]

Hi @adeldaoud, let's wait and hear what @carloscinelli's thoughts are, this is not something we have spent a lot of time thinking about for GRF.

[carloscinelli]

Hi all, we have recent results that covers ATE estimated using machine learning models, see here: https://arxiv.org/abs/2112.13398

[carloscinelli]

We don't have ready to use software yet, but should be available soon

[adeldaoud]

@carloscinelli thanks. Feel free to ping us in this forum when the software will be available. If you happen to have some tutorial R scripts until the software will be released, then please feel free to share it. I will make sure to cite your arxiv paper.