Assessing the Sensitivity to Unmeasured Confounding in Observational Studies
Event details
| Date | 24.08.2021 |
| Hour | 16:00 › 18:00 |
| Speaker | Sina Akbari |
| Category | Conferences - Seminars |
EDIC candidacy exam
exam president: Prof. Ruediger Urbanke
thesis advisor: Prof. Negar Kiyavash
co-examiner: Prof. Patrick Thiran
Abstract
Causal inference with observational data is highly motivated for estimating associations without unknown mechanisms. However, inference with observational data usually relies on assumptions that are not testable from data such as unconfoundedness. It is therefore crucial to assess the sensitivity of the inference to such assumptions. In this report, we discuss `sensitivity analysis' as a tool to evaluate the robustness of causal effect estimates to unmeasured confounding. Sensitivity analysis allows for details that can be given and discussed about the degree of confounding that would invalidate the causal conclusions. The expert then can make judgments about the plausibility of such strong confounders in the observational study, within a formal and precise framework.
Background papers
exam president: Prof. Ruediger Urbanke
thesis advisor: Prof. Negar Kiyavash
co-examiner: Prof. Patrick Thiran
Abstract
Causal inference with observational data is highly motivated for estimating associations without unknown mechanisms. However, inference with observational data usually relies on assumptions that are not testable from data such as unconfoundedness. It is therefore crucial to assess the sensitivity of the inference to such assumptions. In this report, we discuss `sensitivity analysis' as a tool to evaluate the robustness of causal effect estimates to unmeasured confounding. Sensitivity analysis allows for details that can be given and discussed about the degree of confounding that would invalidate the causal conclusions. The expert then can make judgments about the plausibility of such strong confounders in the observational study, within a formal and precise framework.
Background papers
- Making Sense of Sensitivity: Extending Omitted Variable Bias (Carlos Cinelli and Chad Hazlett)
- Sense and Sensitivity Analysis: Simple Post-Hoc Analysis of Bias Due to Unobserved Confounding (Victor Veitch and Anisha Zaveri)
- Flexible Sensitivity Analysis for Observational Studies Without Observable Implications (Alexander M. Franks, Alexander D’Amour, and Avi Feller)
Practical information
- General public
- Free
Contact
- edic@epfl.ch