What This Document Is
This document presents a focused exploration of statistical methods for analyzing count data, specifically within the context of correlated observations. It’s part of the Biostatistics 411 course at UCLA, designed for students developing advanced analytical skills. The material delves into techniques beyond standard approaches, addressing scenarios where data points aren’t independent – a common challenge in many biological and health-related studies. It builds upon foundational knowledge of generalized linear models and introduces more sophisticated modeling strategies.
Why This Document Matters
Students and researchers working with data that represents counts – occurrences of events, frequencies, or quantities – will find this resource particularly valuable. This is especially true when dealing with longitudinal studies, repeated measures, or clustered data where observations within a subject or group are likely to be related. Understanding these methods is crucial for drawing accurate inferences and avoiding misleading conclusions. If you’re encountering challenges with overdispersion or correlation in your count data analyses, this material offers potential solutions and deeper insights.
Topics Covered
* Poisson Regression and its assumptions
* Addressing Overdispersion in Count Data
* Negative Binomial Models
* Poisson Random Effects Models for Correlated Data
* Interpretation of Coefficients in Random Effects Models
* Population-Averaged Effects
* Modeling Random Intercepts and Slopes
* Link Functions and Linear Predictors
What This Document Provides
* A framework for understanding the nuances of correlated count data.
* Discussion of the theoretical underpinnings of advanced modeling techniques.
* Guidance on interpreting model coefficients and understanding their implications.
* Connections to relevant statistical software (SAS Glimmix and Genmod are referenced).
* References to specific chapters within a core course textbook (ALA) for supplemental reading.
* A foundation for applying these methods to real-world research questions.