Some applications of categorical data analysis to epidemiological studies

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Traditional epidemiological analyses often rely on tests of independence for categorized data.
• These methods may not fully capture the complexity and nuances within epidemiological datasets.
• A need exists for more sophisticated analytical frameworks to derive deeper insights.

Purpose of the Study:

• To demonstrate that fitting statistical models provides more informative analyses than standard tests of independence for epidemiological data.
• To present a unified conceptual framework for these advanced analyses.
• To illustrate the calculation of key statistical outputs including parameter estimates, variances, and chi-squared tests.

Main Methods:

• Application of weighted least squares (WLS) as a unified method for model fitting.
• Development and illustration of techniques for calculating point estimates of parameters.
• Calculation of asymptotic variances and asymptotically valid chi-squared tests for model assessment.

Main Results:

• Successful application of the WLS framework to diverse epidemiological data types.
• Demonstration of enhanced analytical capabilities for relative risk estimation from 2x2 tables.
• Illustrative examples include life table analysis, synthetic life table construction, and dose-response curve analysis.

Conclusions:

• Model fitting, particularly using weighted least squares, offers a powerful and unified approach for analyzing categorized epidemiological data.
• This methodology yields more informative results than conventional tests of independence.
• The presented methods facilitate robust statistical inference for a range of epidemiological research questions.