Model selection for wild turkey hunter success rates using small‐area estimation methods

Abstract

Wildlife management agencies conduct hunter surveys to obtain information useful for making management decisions. Often these surveys are designed to collect data, such as hunter success rates, with reliable precision at statewide or regional scales. However, at finer spatial resolutions, such as at the county level, sample sizes are often small, making estimates either impossible to calculate or unreliable using traditional statistical procedures. We used Bayesian hierarchical modeling methods to overcome this smallarea estimation problem for the 1996 Missouri, USA, spring wild turkey (Meleagris gallopavo) hunter survey. We explored 8 potential models based on 3 general hypothesized models. These families of general models included a binomial-beta-gamma model, 5 variations of spatial dependency models with and without covariates, and 2 variations of models with covariates but without accounting for spatial dependency. We used Bayes factors to select among models. We selected a conditional autoregressive binomial logistic linear mixed model as the most appropriate and used the posterior predictive distributions from this model to assess its fit. This small-area estimation model provided precise hunter success-rate estimates at the county level even though this hunter survey was designed to provide reliable estimates at the regional-scale. This approach provides estimates at smaller scales than are possible using traditional statistical analyses, allowing managers to identify when small-scale management decisions are warranted. Ó 2018 The Wildlife Society.