Table 1.

Performance of tiger tolerance models with and without landscape covariates or measures of risk

Model and covariates	AICc	ΔAICc	Log-like	K
Social predictors + geographic profile: Kill_tiger + Protect_tiger + Injunctive + Descriptive + BadGood + DangHarm + Spirit + Health + Env + TrustB + Scenario + Age + Sex + Ethnicity + GP	3760.59	0	−1846.30	34
Social predictors + ensemble risk probability: Kill_tiger + Protect_tiger + Injunctive + Descriptive + BadGood + DangHarm + Spirit + Health + Env + TrustB + Scenario + Age + Sex + Ethnicity + Prob_conf	3766.97	6.38	−1848.79	34
Social plus landscape covariates from ensemble risk model: Kill_tiger + Protect_tiger + Injunctive + Descriptive + BadGood + DangHarm + Spirit + Health + Env + TrustB + Scenario + Age + Sex + Ethnicity + dis_for + dis_rds + dis_riv + connect + farmers + pop_grd + occupancy	3767.41	6.82	−1839.70	44
Social predictors only: Kill_tiger + Protect_tiger + Injunctive + Descriptive + BadGood + DangHarm + Spirit + Health + Env + TrustB + Scenario + Age + Sex + Ethnicity	3767.55	6.96	−1851.77	32

Models are presented in order of performance according to Akaike’s information criterion corrected for small sample sizes (AICc). The ΔAICc indicates the difference in AIC relative to the top performing model. Two measures of encounter risk were explored: an ensemble model combining the outputs of three presence–absence algorithms (Prob_conf), and a geographic profile (GP). A third model incorporated the landscape predictors utilised in the ensemble predictor of risk. Social covariates were identical throughout. All social covariates and risk scores (probability of conflict, or geographic profile) were entered as fixed effects.Data sources, covariate abbreviations and analyses are described in the Methods and Supplementary Table 3