The impact of measurement error and omitting confounders on statistical inference of mediation effects and a web tool for sensitivity analysis by Xiao Liu.
To make valid causal inferences from mediation analysis, a number of assumptions are needed. Among the assumptions, two of the most discussed are (1) the input, mediator, and outcome variables are measured without error, and (2) no confounders of the effects in the mediation model are omitted. Impact of violating either assumption alone on statistical inference of mediation has been extensively discussed in previous literature. In practice, however, violations of the two assumptions often co-occur. How the co-occurrence of violation impacts statistical testing of mediation effects has not been formally studied before. Therefore, in the current study, we investigated the combined effects of measurement error and omitting confounders on statistical inference (both point estimates and statistical testing) of mediation effects. Analytical results were obtained to systematically demonstrate the effects. Two simulation studies and two real-data examples were used for illustration. In addition, we developed both R functions and a user-friendly web tool for performing sensitivity analysis in mediation studies. We hope researchers could use the developed tools to assess sensitivity in statistical inference from mediation analysis more conveniently.
QSG meets on Thursdays from 3:30-4:45pm in Corbett 378. All are welcome, and we hope to see you there.
For a complete list of speakers for Spring 2019, please visit this link.