Implement fully Bayesian approach for sensitivity analysis

Kuan-Liu-Lab / causens

An R package for causal sensitivity analysis methods

https://kuan-liu-lab.github.io/causens/

MIT License

1 stars 0 forks source link

Implement fully Bayesian approach for sensitivity analysis #5

Closed larryshamalama closed 5 months ago

larryshamalama commented 1 year ago

Title of paper: A comparison of Bayesian and Monte Carlo sensitivity analysis for unmeasured confounding

Link: https://pubmed.ncbi.nlm.nih.gov/28386994/

larryshamalama commented 6 months ago

Modelling of the unmeasured confounder leads to non-identifiable models, although this may not be a problem with longitudinal outcomes and a time-invariant unmeasured confounder.

Instead, the key idea would be to: "reparametrize the model in such a way that the distribution of the data is determined completely by a smaller collection of parameters for which $\sqrt{n}$-consistent estimation is expected".