that were placed in the exact same locations in both years (i.e., 75 locations
in 2010 and 2011 for a total of 150 locations). We ran models where the prey
covariate was included as a single variable, summing the total number of
detections for all six primary prey species at each camera trap (29).We used this
method rather than including the total number of detections from each of the
six prey species as covariates, because two prey species (i.e., gaur and hog deer)
were not detected outside the park. Because spotted deer comprised 75%of all
prey detections (Figs. S2 and S3), we also ran each model just using the total
number of spotted deer detections at each camera trap as the prey covariate.
This method produced similar results to the results produced using the combined
prey covariate; therefore, we only report models using the combined
prey covariate.Models were ranked according to their second-order Akaike’s
information criterion (AICc), with higher-ranked models having lower AICc
values. Because several models with different combinations of covariates
performed comparatively well (i.e., ΔAICc < 4), we averaged model results
(i.e., covariate coefficients, detection probability, and occupancy) from the
top-ranked models using standard methods (i.e., multimodel inference) (57).
Model-averaged coefficient estimates were considered significant if their unconditional 95%CIs did not include zero. We used kernel density estimation
to estimate the probability density function of the activity patterns (i.e., density
of activity) of tigers and types of human presence. Then,we used the procedures
described in ref. 34 to measure the extent of overlap between them.