Abstract

Modeling count data using generalized linear mixed models (GLMMs) is a common practice in ecological research. However, datasets such as bird vocalization counts often display overdispersion, where the variance exceeds the mean, violating Poisson assumptions and potentially leading to biased inference. This study investigates the call counts of owl nestlings by comparing Poisson and Negative Binomial (NB) GLMMs to assess the impact of overdispersion on model performance. The models incorporate a random intercept for nest identity, thereby accounting for the hierarchical structure of the data and the correlation among observations within the same nest. Parameter estimation was performed using the Laplace approximation and the Adaptive Gauss–Hermite Quadrature (AGHQ) method to evaluate likelihood-based inference under different estimation approaches. Although the overdispersion diagnostic indicated extra-Poisson variation, model comparison based on the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and log-likelihood showed that the Poisson GLMM provided a better overall fit than the NB-GLMM. These findings demonstrate that the presence of overdispersion alone does not necessarily justify replacing a Poisson GLMM with a NB-GLMM and highlight the importance of combining distributional diagnostics with likelihood-based model selection when analyzing hierarchical ecological count data.

IPC Classification

Keywords