2.4 Statistical analyses
We first tested the effect of four independent response variables (landscape context [i.e., natural forest vs. plantations], trapping method, temperature, and precipitation) on beetle community composition using a Permutational Multivariate Analysis of Variance (PERMANOVA) as implemented in the “vegan” R package (Oksanen et al., 2022). We then used Non-metric Multidimensional Scaling (NMDS) to represent the dissimilarity of beetle communities between natural forest and plantation areas. Subsequently, we compared family richness between natural forest and plantation areas, and between trapping methods, by plotting rarefaction and extrapolation curves with the number of collected individuals as a measure of sampling intensity (‘iNEXT’ package in R v.4.2.1; Hsieh et al., 2019). We used the asymptotic estimators provided by iNEXT as a measure of the total family richness (including unobserved families) in each sampling site.
The drivers of the diversity of beetle assemblages were explored with a linear model using the “lm ” function in R (R Core Team 2022). The model included the asymptotic estimates of richness obtained in each sampling site as a dependent variable, and landscape context, trapping methods, temperature, and precipitation as independent variables. The same approach was used for Shannon and Simpson indices as estimates of the diversity of beetle communities, again using asymptotic estimators from iNEXT.
We used negative binomial regressions, fitted with the “glm.nb ” function in the MASS R package (Venables & Ripley, 2002), to test the effect of the same variables (landscape context, trapping methods, temperature, and precipitation) on beetle abundances. Negative binomial models were used because a Poisson Generalized Linear Model (GLM) that we fitted first showed evidence of overdispersion, and the negative binomial model had a lower AIC that the Poisson GLM. Here, we had to account for the very variable sampling effort that produced the observed variation in beetle abundances; therefore, models also included as an offset the sum of the duration of sampling (the number of days) and the number of traps. The same approach was used first for the total beetle abundance, then for the most abundant beetle families separately: Carabidae, Scarabaeidae, Nitidulidae, Curculionidae, and Chrysomelidae. All statistical analyses were performed in the R platform (version 4.2.1, R development Core Team 2022).