Uncertainty intervals for unobserved confounding of direct and indirect effects with extensions to censoring and truncation
(English)Manuscript (preprint) (Other academic)
When performing a mediation analysis, i.e. estimating direct and indirect effects of a given exposure on an outcome, strong assumptions are made about unconfoundedness. These assumptions are difficult to verify in a given situation and therefore a mediation analysis should be complemented with a sensitivity analysis to assess the impact of violations. Lindmark et al. (2016) proposed a sensitivity analysis method for parametric estimation of conditional and marginal direct and indirect effects when the mediator and outcome are binary and modeled using probit models. In this paper we extend this to include cases with continuous mediators and outcomes and suggest extensions to the cases when the continuous outcome variable is censored or truncated. Three sensitivity parameters are used, consisting of the correlations between the error terms of the mediator, outcome and exposure assignment mechanism models. These correlations are incorporated into the estimation of the model parameters and sampling variability is taken into account through the construction of uncertainty intervals.
mediation, sensitivity analysis, unmeasured confounding, sequential ignorability, uncertainty intervals, censoring, truncation
Probability Theory and Statistics
Research subject Statistics
IdentifiersURN: urn:nbn:se:umu:diva-125930OAI: oai:DiVA.org:umu-125930DiVA: diva2:973916