Non-response models for the analysis of non-monotone non-ignorable missing data

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Missing data present challenges in statistical analyses, particularly when data are non-monotone and the missingness mechanism is non-ignorable.
• Existing methods may yield biased estimates if assumptions about the missing data process are violated.

Purpose of the Study:

• To introduce a novel class of statistical models for non-ignorable, non-monotone missing data.
• To provide a framework for assessing the sensitivity of statistical estimates to untestable assumptions regarding missing data.

Main Methods:

• Development of a new class of statistical models designed to handle complex missing data patterns.
• Application of these models to analyze data from a case-control study investigating radiation exposure and breast cancer risk.

Main Results:

• The newly developed models allow for robust sensitivity analyses concerning the missing data mechanism.
• Application to the case-control study demonstrated the utility of the models in evaluating the impact of missing data assumptions on effect estimates.

Conclusions:

• The proposed models offer a valuable tool for researchers dealing with non-ignorable, non-monotone missing data.
• These models enhance the transparency and reliability of findings in epidemiological and biostatistical research by explicitly addressing data limitations.