A comparison of zero-inflated and hurdle models for modeling zero-inflated count data.

Area of Science:

• Biostatistics
• Statistical Modeling
• Health Services Research

Background:

• Count data frequently exhibit excessive zeros, exceeding the capacity of standard distributions like Poisson or negative binomial.
• Zero-inflated (ZI) and hurdle models are commonly applied to such data, but their fundamental differences are not well-understood.
• Understanding these differences is crucial for appropriate statistical analysis in various fields, including healthcare utilization.

Purpose of the Study:

• To review and elucidate the distinct data-generating processes of zero-inflated and hurdle models.
• To compare the performance of zero-inflated and hurdle models using simulation studies.
• To provide guidance on selecting the appropriate model for count data with excess zeros.

Main Methods:

• A comprehensive review of the theoretical underpinnings of zero-inflated and hurdle models.
• Design and execution of simulation studies to assess model performance under various scenarios.
• Evaluation of goodness-of-fit criteria for model selection.

Main Results:

• The study highlights key distinctions in how zero-inflated and hurdle models conceptualize the data-generating process for excess zeros.
• Simulation results indicate differential performance characteristics of the two models depending on data properties.
• No single model universally outperforms the other; selection depends on specific data features.

Conclusions:

• Zero-inflated and hurdle models offer distinct approaches to modeling count data with excess zeros.
• Empirical evaluation through simulations is essential for understanding their practical performance.
• The choice of regression model should be guided by a thorough assessment of goodness-of-fit tailored to the specific dataset.