Journal of the Royal Statistical Society Series B (Statistical Methodology) · 2019 · 973 citations · 35 references
Public HealthBiostatisticsCausal InferenceMeta-analysisRisk AnalysisBias ControlCausal Effect EstimationMissing Data HandlingSensitivity AnalysisTreatment EffectsConfounding AdjustmentConfounding BiasConfounding ControlResidual ConfoundingRobustnessRobustness To Unmeasured ConfoundingSensitivity Analysis Of Model ParametersUnobserved Confounding
The authors extend the omitted variable bias framework by introducing a suite of sensitivity analysis tools that do not rely on functional‑form assumptions, accommodate multiple potentially nonlinear confounders, and provide novel measures and graphical diagnostics for routine reporting. They define a robustness value and a partial R² metric to quantify the strength of unobserved confounding needed to overturn conclusions, present graphical tools and an extreme‑scenario analysis, critique standard benchmarking, and propose a formal procedure to bound confounder strength, illustrated with an example on violence exposure and peace attitudes.
Summary We extend the omitted variable bias framework with a suite of tools for sensitivity analysis in regression models that does not require assumptions on the functional form of the treatment assignment mechanism nor on the distribution of the unobserved confounders, naturally handles multiple confounders, possibly acting non-linearly, exploits expert knowledge to bound sensitivity parameters and can be easily computed by using only standard regression results. In particular, we introduce two novel sensitivity measures suited for routine reporting. The robustness value describes the minimum strength of association that unobserved confounding would need to have, both with the treatment and with the outcome, to change the research conclusions. 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. Next, we offer graphical tools for elaborating on problematic confounders, examining the sensitivity of point estimates and t-values, as well as ‘extreme scenarios’. Finally, we describe problems with a common ‘benchmarking’ practice and introduce a novel procedure to bound the strength of confounders formally on the basis of a comparison with observed covariates. We apply these methods to a running example that estimates the effect of exposure to violence on attitudes toward peace.
35
Let's Take the Con Out of Econometrics
Edward E. Leamer · American Economic Review · 1982 · 2.3K citations