A test of the missing data mechanism for repeated categorical data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Categorical Data Analysis

Background:

• Longitudinal data frequently contain missing observations, leading to incomplete and unbalanced datasets.
• Existing weighted least squares methods for incomplete longitudinal categorical data assume missingness is completely at random (MCAR).
• This MCAR assumption can lead to biased results when missingness is related to unobserved responses.

Purpose of the Study:

• To develop a practical and simple test for the missing data mechanism in incomplete longitudinal categorical data.
• To extend existing methods for testing missing data mechanisms to longitudinal categorical settings.
• To provide a robust method for analyzing longitudinal categorical data where missingness may not be completely at random.

Main Methods:

• The study proposes a test for the missing mechanism in incomplete repeated categorical data.
• The test is an extension of Little's (1988) test for missing data.
• It utilizes a test criterion in the general form proposed by Wald.

Main Results:

• The developed test provides a practical approach to assess the missing data mechanism.
• The method can identify when missing observations are related to unobserved responses.
• The test is illustrated with a real-world example from an obesity study.

Conclusions:

• The proposed test offers a valuable tool for researchers working with incomplete longitudinal categorical data.
• It helps to mitigate potential biases arising from non-random missingness.
• Accurate analysis of longitudinal data is crucial for understanding trends and making informed conclusions.