Related Experiment Videos
Application of pattern-mixture models to outcomes that are potentially missing not at random using pseudo maximum
1Division of Biostatistics, School of Medicine, Indiana University, 1050 Wishard Boulevard RG R4101, Indianapolis, IN 46202, USA. chashen@iupui.edu
Biostatistics (Oxford, England)
|March 18, 2005
Summary
This study introduces a robust method for analyzing data with missing outcomes, particularly when missingness is not at random (MNAR). The approach offers reliable estimation for regression parameters and handles various data types effectively.
Area of Science:
- Statistics
- Biostatistics
- Epidemiology
Background:
- Missing data poses significant challenges in statistical analysis, especially when data are missing not at random (MNAR).
- Traditional methods may yield biased results when assumptions about missing data mechanisms are violated.
- Pattern-mixture models offer a flexible framework for addressing MNAR data.
Purpose of the Study:
- To develop and evaluate a novel statistical method for analyzing data with potentially MNAR outcomes.
- To provide consistent and efficient estimators for regression parameters in the presence of MNAR data.
- To apply the method to real-world data, such as an epidemiologic cohort study on cognitive decline.
Main Methods:
- Fitting pattern-mixture models using pseudo maximum likelihood estimation based on exponential families.
- Estimating identifiable regression parameters with correct mean structure specification for consistency and variance structure for efficiency.
- Developing a hypothesis test for model simplification to enhance efficiency.
- Conducting simulation studies to compare the proposed method with existing approaches.
Main Results:
- The proposed pseudo maximum likelihood method provides consistent estimators for regression parameters under correct mean structure specification.
- The method is adaptable for both continuous and discrete outcomes.
- Simulations demonstrate the performance of the proposed estimation procedure compared to other methods.
- The approach, combined with sensitivity analysis, allows for parsimonious semi-parametric modeling of MNAR outcomes.
Conclusions:
- The developed method offers a reliable approach for analyzing data with MNAR outcomes, applicable to diverse data types.
- The method facilitates robust statistical inference and model simplification.
- Application to an elderly cohort study demonstrates its utility in examining cognitive decline.