2.5. a) Data sources for distribution-wide assessment of dhole
pack size:
Through google scholar we searched for scientific literature on pack
size of dholes, using the keywords “Cuon alpinus”, “Dhole”,
“Average”, “Mean”, “Pack-size”. Our search resulted in 34
scientific assessments from 1973 to 2018 that had reported average pack
size of dholes. These 34 assessments belonged to 24 unique protected
areas across dhole ranging countries in South and South-east Asia. 18 of
these unique sites were also a part of the recently published dhole diet
review (Srivathsa, Sharma, & Oli, 2020). Subsequently, snowball
sampling approach was used (Handcock & Gile, 2011), by using dhole pack
size as baseline information from the previously conducted assessments.
Literature cited within these assessments were referred to collate data
on tiger density along with prey density (of the closest or same
assessment year) and size of the protected area. We obtained data on the
topography of 24 protected areas through google earth engine (Gorelick
et a., 2017).
b) Analytical methods:
We used generalized linear models to examine correlates of dhole pack
size reported from 24 unique sites across dhole distribution range. We
used only those studies (n=29) for which data on all the predictor
variables were available i.e., tiger density and ungulate density, size
of the protected area (PA), elevational heterogeneity and terrain
ruggedness of the PA. We checked for correlations among predictors and
dropped the correlated ones (r > 0.6), prior to analysis.
After screening for normal distribution of response variable, we used
”gaussian” family for the analysis. We tested for model parameters based
on our hypothesis, and compared them to null model (Intercept only).
Model fits were compared using Akaike’s Information Criterion corrected
(AICC), and the effect of parameters was gauged based on
direction and statistical significance of corresponding β-coefficients.
We used ”MuMIn” package for model selection and averaging. Model
selection was based on difference between AIC models, (ΔAIC <
2). Further, model selection was done using Royall’s 1/8 strength of
evidence and 95% cumulative weight criteria. Model averaging was
carried out for parameters based on top model selection. All analyses
were performed in program R (R Development Core Team, 2014).