A comparison study on modeling of clustered and overdispersed count data for multiple comparisons

Area of Science:

• Biostatistics
• Statistical Modeling
• Bioinformatics

Background:

• Count data is prevalent across scientific disciplines.
• Analysis of clustered and overdispersed count data presents statistical challenges.
• Ignoring these factors can inflate Type I error rates in multiple comparisons.

Purpose of the Study:

• To compare the performance of generalized estimation equations (GEE) and generalized linear mixed models (GLMM) for analyzing clustered, overdispersed count data.
• To evaluate multiple contrast tests under various data settings, focusing on coverage and rejection probabilities.
• To highlight the impact of overdispersion on statistical analyses in biological research.

Main Methods:

• Simulation studies were conducted using generated overdispersed, clustered count data in small sample sizes.
• Parameter estimates were obtained using both GEE and GLMM.
• Multiple contrast tests were compared based on coverage and rejection probabilities.

Main Results:

• GEE demonstrated superior performance over GLMM when the variance-sandwich estimator was correctly specified.
• GLMM exhibited convergence issues in specific data configurations, though alternative implementations exist.
• Ignoring strong overdispersion in genetic data analysis led to significant inferential problems.

Conclusions:

• GEE is a robust method for analyzing clustered and overdispersed count data, particularly in biological settings.
• Careful model specification and consideration of overdispersion are crucial for accurate statistical inference.
• The study underscores the importance of appropriate statistical methods for handling complex biological data structures.