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

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Non-monotone missing data poses challenges in statistical analysis.
• Existing models may not fully capture complex missingness mechanisms.

Purpose of the Study:

• Introduce and define randomized monotone missingness (RMM) models.
• Evaluate the generalizability of RMM models for non-monotone ignorable data.
• Assess the implications of RMM model misspecification for data analysis.

Main Methods:

• Development of the randomized monotone missingness (RMM) framework.
• Theoretical analysis of ignorable missing data processes.
• Application of RMM models to a case-control study on radiation and breast cancer.

Main Results:

• RMM models represent a general mechanism for non-monotone ignorable data.
• Identified ignorable missing data processes that are not RMM representable.
• Demonstrated potential inappropriateness of assuming ignorable missingness when RMM is rejected.

Conclusions:

• The RMM framework offers a more comprehensive approach to modeling non-monotone missing data.
• Statistical tests rejecting RMM warrant caution when assuming ignorable missingness.
• RMM models provide a valuable tool for analyzing complex missing data in epidemiological studies.