The k-ZIG: flexible modeling for zero-inflated counts.

Area of Science:

• Statistics
• Biostatistics
• Environmental Science

Background:

• Count data frequently exhibit an excess of zero values, posing challenges for standard statistical models.
• Traditional zero-inflated models like ZIP and ZINB, and hurdle models, often struggle with very high proportions of zeros (e.g., >80%) and covariate identification.

Purpose of the Study:

• To introduce a novel class of k-zero-inflated Gaussian (k-ZIG) models designed for flexible analysis of count data with excess zeros.
• To enhance the modeling of both zero-inflation and non-zero counts, allowing for interplay between these components.
• To address limitations of existing zero-inflated models, particularly in scenarios with extreme zero incidence.

Main Methods:

• Development of the theoretical properties of the k-ZIG model class.
• Reparameterization of the models to a natural link function for improved interpretability.
• Implementation of Bayesian inference for straightforward model fitting.
• Illustration using simulated data and a real-world forest seedling dataset from the USDA Forest Service.

Main Results:

• The proposed k-ZIG models provide a more flexible framework for handling excess zeros in count data.
• The models demonstrate improved performance in identifying important covariates, even with extreme zero proportions.
• Bayesian fitting allows for efficient and robust estimation of model parameters.

Conclusions:

• The k-ZIG models represent a significant advancement for analyzing count data with excess zeros, outperforming traditional methods.
• This new class of models offers greater flexibility and better covariate identification, particularly in challenging datasets.
• The methodology is practical and applicable to ecological and other fields dealing with zero-inflated count data.