Modeling change trajectories with count and zero-inflated outcomes: Challenges and recommendations

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Analyzing longitudinal count data, such as daily alcohol intake, presents unique statistical challenges.
• These challenges include discrete outcomes, excess zeros, overdispersion, and clustered data structures.
• Selecting appropriate time metrics and distributional forms is crucial for accurate modeling.

Purpose of the Study:

• To describe statistical models for examining change over time in longitudinal count outcomes.
• To address common challenges associated with count data, including zero-inflation and overdispersion.
• To provide guidance on selecting appropriate models and distributional forms for analyzing such data.

Main Methods:

• Overview of generalized linear models (GLMs), generalized estimating equations (GEE), and generalized latent variable (mixed-effects) models.
• Focus on Poisson, negative binomial, zero-inflated, and hurdle distributions for count data.
• Demonstration using longitudinal alcohol intake data from the National Longitudinal Survey of Youth 1997.

Main Results:

• The study outlines methods to handle discrete, overdispersed, and zero-inflated longitudinal count data.
• Recommendations are provided for identifying appropriate individual change trajectories and distributional forms.
• Model fitting and selection processes are illustrated with real-world adolescent alcohol consumption data.

Conclusions:

• Appropriate statistical models are essential for accurately analyzing longitudinal count data.
• Addressing data characteristics like zero-inflation and overdispersion is key to robust findings.
• The presented models and recommendations offer a framework for researchers studying changes in count outcomes over time.