Summary of the selection process for the model for mammal identification.

We modeled the frequency of identifications at a particular level using a Poisson (link = log) GLMM, with village and respondent nested within village included as random effects (not shown). We investigated the effect of respondent gender, respondent age, and species abundance (extant or extirpated) and their interactive effects on the ability of people to correctly name species at two levels (overall (group+species level) and specific (species vs group levels)). Full model details are given in the online supplementary material Table S3. Starting with the null model, we added and subtracted parameters by hand and assessed the impact of a factor by comparing AIC values. K = number of model parameters. ΔAICc = difference between AICc of the top ranked model and current model.