Events Calendar

23 Sep
Biostatistics Seminar: Xu Qin (Pitt) Simulation-Based Sensitivity Analysis for Causal Mediation Studies
Event Type

Lectures, Symposia, Etc.

Topic

Research

Target Audience

Faculty, Graduate Students

University Unit
Department of Biostatistics
Subscribe
Google Calendar iCal Outlook

Biostatistics Seminar: Xu Qin (Pitt) Simulation-Based Sensitivity Analysis for Causal Mediation Studies

Dr. Xu Qin

Assistant Professor of Research Methodology

Department of Health and Human Development

University of Pittsburgh

Abstract: Causal mediation analysis is essential for investigating the mechanisms through which a treatment operates. Causal inference regarding a hypothesized mediation mechanism might be invalidated if there are omitted pretreatment confounders (i.e., confounders preceding the treatment) of the treatment-mediator, treatment-outcome, and mediator–outcome relationships, or in the presence of posttreatment confounders (i.e., confounders affected by the treatment) of the mediator-outcome relationship. However, this has not received enough attention in many empirical studies. After a review of the existing causal mediation analysis methods and the approaches that assess the sensitivity of mediation analysis results to unmeasured pretreatment confounding, we propose a simulation-based sensitivity analysis strategy, assuming no posttreatment confounding. The method has five primary advantages. First, it enables applied researchers to intuitively quantify the strength of an unmeasured pretreatment confounder. Second, by simulating the unmeasured confounder from its conditional distribution and adjusting for it in the analysis, the method accurately reflects the influence of unmeasured pretreatment confounding on both the causal effect estimates and their sampling variability, while most existing sensitivity analysis methods ignore the latter. Third, a convenient tool is provided for the visualization of sensitivity analysis results. Fourth, it is applicable to both randomized experiments and observational studies and to mediators and outcomes of different scales. Fifth, it can assess the sensitivity of results obtained from different causal mediation analysis approaches. We have also developed an R package that implements the proposed method (https://cran.r-project.org/web/packages/mediationsens).

Thursday, September 23 at 3:30 p.m. to 4:30 p.m.

Public Health, A115
130 Desoto Street, Pittsburgh, 15261

Biostatistics Seminar: Xu Qin (Pitt) Simulation-Based Sensitivity Analysis for Causal Mediation Studies

Dr. Xu Qin

Assistant Professor of Research Methodology

Department of Health and Human Development

University of Pittsburgh

Abstract: Causal mediation analysis is essential for investigating the mechanisms through which a treatment operates. Causal inference regarding a hypothesized mediation mechanism might be invalidated if there are omitted pretreatment confounders (i.e., confounders preceding the treatment) of the treatment-mediator, treatment-outcome, and mediator–outcome relationships, or in the presence of posttreatment confounders (i.e., confounders affected by the treatment) of the mediator-outcome relationship. However, this has not received enough attention in many empirical studies. After a review of the existing causal mediation analysis methods and the approaches that assess the sensitivity of mediation analysis results to unmeasured pretreatment confounding, we propose a simulation-based sensitivity analysis strategy, assuming no posttreatment confounding. The method has five primary advantages. First, it enables applied researchers to intuitively quantify the strength of an unmeasured pretreatment confounder. Second, by simulating the unmeasured confounder from its conditional distribution and adjusting for it in the analysis, the method accurately reflects the influence of unmeasured pretreatment confounding on both the causal effect estimates and their sampling variability, while most existing sensitivity analysis methods ignore the latter. Third, a convenient tool is provided for the visualization of sensitivity analysis results. Fourth, it is applicable to both randomized experiments and observational studies and to mediators and outcomes of different scales. Fifth, it can assess the sensitivity of results obtained from different causal mediation analysis approaches. We have also developed an R package that implements the proposed method (https://cran.r-project.org/web/packages/mediationsens).

Thursday, September 23 at 3:30 p.m. to 4:30 p.m.

Public Health, A115
130 Desoto Street, Pittsburgh, 15261

Topic

Research

Target Audience

Faculty, Graduate Students

University Unit
Department of Biostatistics

Powered by the Localist Community Events Calendar ©