Sensitivity analysis for m-estimates, tests, and confidence intervals in matched observational studies.

Area of Science:

• Statistics
• Biostatistics
• Robust Statistics

Background:

• Huber's m-estimates offer controlled influence for observations, including least squares and maximum likelihood.
• Maritz proposed permutation inference methods for m-tests and estimators, applicable in randomized studies.
• Observational studies present challenges due to potential unobserved covariate differences between matched subjects.

Purpose of the Study:

• To develop a sensitivity analysis method for m-tests, m-intervals, and m-estimates in observational studies.
• To assess the impact of biases from nonrandom treatment assignment on statistical inferences.
• To provide a framework for evaluating the robustness of findings in matched observational studies.

Main Methods:

• Developed a sensitivity analysis applicable to matched pairs and matched sets in observational studies.
• Applied the method to m-tests with Huber's weight function and other robust weight functions.
• Illustrated the approach using studies on DNA damage and tuberculosis drug side effects.

Main Results:

• The method quantifies how biases due to nonrandom assignment alter inferences from m-tests and m-estimates.
• Sensitivity analyses were demonstrated for various procedures including the permutational t-test, sign test, and Noether's test.
• The approach is applicable to both paired and set-matched observational data.

Conclusions:

• The developed sensitivity analysis is valuable for interpreting results from m-tests and m-estimates in observational research.
• It enhances the reliability of statistical inferences by accounting for potential biases.
• The method provides a quantitative measure of how findings might change under different bias scenarios.