Models for three-dimensional contingency tables with completely and partially cross-classified data

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Analyzing three-dimensional contingency tables with both complete and partial cross-classification presents statistical challenges.
• Existing methods may not adequately address situations with missing independent variables in such tables.
• Accurate modeling is essential for reliable inferences in epidemiological studies, particularly in case-control designs.

Purpose of the Study:

• To develop statistical models for three-dimensional contingency tables with mixed (complete and partial) cross-classification.
• To enable independent inference on cell probabilities and probabilities of missing independent variables.
• To apply the developed methodology to real-world epidemiological data, specifically cervical cancer case-control study data.

Main Methods:

• Development of models for three-dimensional contingency tables with dependent and independent variables.
• Utilizing maximum likelihood methods for parameter estimation and hypothesis testing.
• Employing conditional goodness-of-fit test statistics with an additivity property.

Main Results:

• The proposed models effectively handle three-dimensional contingency tables with missing data in independent variables.
• Independent inferences on cell probabilities and missing data probabilities are achievable.
• The methodology demonstrates utility when applied to cervical cancer case-control study data.

Conclusions:

• The developed statistical models provide a robust framework for analyzing complex contingency table data with missing independent variables.
• The ability to make independent inferences enhances the interpretability and applicability of the models in biostatistics and epidemiology.
• This approach offers valuable tools for analyzing epidemiological data, such as the cervical cancer case-control study.