Abstract:
|
In this talk we discuss two novel sensitivity analysis measures for regression coefficients suited for routine reporting: (i) the robustness value describes the minimum strength of association unobserved confounding would need to have, both with the treatment and the outcome, to change the research conclusions; and (ii) the partial R2 of the treatment with the outcome shows how strongly confounders explaining all the residual outcome variation would have to be associated with the treatment to eliminate the estimated effect. These measures do not require assumptions about the treatment assignment mechanism, can be used to assess sensitivity to multiple (non-linear) confounders and are easily computable from usual regression software output. We show how the routine reporting of these measures make the discussion of sensitivity to unobserved confounding both more accessible and standardized, allowing researchers to easily report the sensitivity of their estimates in standard regression tables, and also enabling readers and reviewers to easily initiate the discussion about sensitivity to unobserved confounders when reading papers that did not formally perform sensitivity analysis.
|