Model selection, zero‐inflated models, and predictors of primate abundance in Korup National Park, Cameroon

Abstract

Determining the ecological and anthropogenic factors that shape the abundance and distribution of wild primates is a critical component of primate conservation research. Such research is complicated, however, whenever the species under study are encountered infrequently, a characteristic of many taxa that are threatened with extinction. Typically, the resulting data sets based on surveys of such species will have a high frequency of zero counts which makes it difficult to determine the predictor variables that are associated with species abundance. In this study, we test various statistical models using survey data that was gathered on seven species of primate in Korup National Park, Cameroon. Predictor variables include hunting signs and aspects of habitat structure and floristic composition. Our statistical models include zero-inflated models that are tailored to deal with a high frequency of zero counts. First, using exploratory data analysis we found the most informative set of models as ranked by Δ-AIC (Akaike's information criterion). On the basis of this analysis, we used five predictor variables to construct several regression models including Poisson, zero-inflated Poisson, negative binomial, and zero-inflated negative binomial. Total basal area of all trees, density of secondary tree species, hunting signs, and mean basal area of all trees were significant predictors of abundance in the zero-inflated models. We discuss the statistical logic behind zero-inflated models and provide an interpretation of parameter estimates. We recommend that researchers explore a variety of models when determining the factors that correlate with primate abundance.