(A) Comparison of models of the relationship between total annual mortality and human-caused mortality for wolves in North America.

1Expanded model descriptions:(i) Generalized linear model (binomial errors with identity link) that allowed different slopes and intercepts for the relationship between total mortality and human offtake for two regions (wolves in the Northern Rocky Mountains (NRM) recovery area and wolves in previously-studied populations),(ii) General additive model that allowed regional differences, fit in the ‘mgcv’ package of R with cross-validation used to determine the optimum amount of smoothing. GAM models allow curvilinear functions if the data support curvature.(iii) Generalized linear model (binomial errors with identity link) that allowed the slope to change at a breakpoint and allowed regional differences,(iv) Generalized linear model (binomial errors with identity link) with no regional effect.(v) Generalized linear model (binomial errors with identity link) that allowed the slope to change at a breakpoint with no regional effect,(vi) Constant total mortality (no effect of human offtake on total mortality).2Number of parameters in the model (non-integer values are expected for general additive models).3*QAICc calculated using c-hat = 4, the estimated overdispersion value obtained from a quasi-binomial model and using the number of mortality rates (N = 48) as the sample size.4Akaike model weight.