Open maxbiostat opened 2 years ago
I am not sure of how efficient his method is if it still needs to use sampling methods to obtain the posterior.
For variance-based sensitivity, which is defined here, and simple inference models with closed form solutions, we can indeed save computations by computing the variance from the parameters of the posterior, and using that to calculate first order or higher order sensitivities.
Can't think of a non-trivial example, though...
I am not sure of how efficient his method is if it still needs to use sampling methods to obtain the posterior.
I understand. But since we are going to use sampling methods (in Stan) to get the posterior, anyways, it might be worth a shot.
Another nice paper to have a look: https://arxiv.org/pdf/2203.15184.pdf
Yet another tool: https://corymccartan.github.io/adjustr/
This paper proposes a way of formally assessing prior-data conflict and sensitivity. Might be a very nice plus once all of the building blocks are in place.