Sensitivity Analysis and Bounding of Causal Effects With Alternative Identifying Assumptions.

Area of Science:

• Statistics
• Econometrics
• Causal Inference

Background:

• Causal effect estimation often relies on untestable assumptions, raising concerns about the sensitivity of results.
• Interpreting causal effects requires methods to bound parameters or assess sensitivity to alternative assumptions.

Purpose of the Study:

• To propose a practical method for bounding and sensitivity analysis in causal inference.
• To combine multiple identifying assumptions for constructing tighter common bounds on causal effects.

Main Methods:

• Developed a method to combine competing identifying assumptions that impose different restrictions on the same non-identified parameter.
• Leveraged the cross-translatability between assumptions to integrate information from data.
• Applied the approach to estimate the complier average causal effect (CACE) in a randomized trial with noncompliance and missing outcomes.

Main Results:

• Demonstrated the ability to construct tighter bounds on causal effects by effectively combining information from alternative assumptions.
• Showcased the flexibility of the proposed bounding and sensitivity analysis method.
• Successfully estimated CACE in a challenging real-world trial setting.

Conclusions:

• The proposed method offers a practical and flexible approach to causal inference under untestable assumptions.
• Combining competing identifying assumptions leads to more robust and precise causal effect estimates.
• This technique is valuable for improving the interpretation of causal effects in observational studies and complex trial designs.