3.1 Quantification of specialization metrics
Over the course of four non-sequential years, at five different
locations, we quantified the diet of 1,520 scat samples.
We successfully determined the
sex of the depositor for 1,145 of those scats (75% success rate)
(Voelker et al. 2018). The number of scat with successful sex
determination varied by location and month (Table 1). Samples with
successful sex determination were then binned by the factors Sex,
Location, Month, and Year to form unique groups for analysis (Table S1).
After eliminating samples without sex determination and with small
sample sizes (< 5 samples), we were left with 1,083 samples in
89 groups. Only these 1,083 samples were used in further analyses.\(\text{PS}_{i}\) was calculated relative to the samples in a specific
group, and the average \(\text{PS}_{i}\) across all samples and groups
was 0.399 (95% CI = 0.026, R = 100,000). The \(\text{PS}_{i}\) values
of the 1,083 samples were not normally distributed (kurtosis = 2.66,
skewness = 0.65, Figure S2). Therefore, a logit transformation was used
to adjust the variance distribution (kurtosis = 5.21, skewness = 1.01,
Figure S2). These transformed \(\text{PS}_{i}\) values were used to run
the GLMMs. Additionally, the range of theoretical minima across the 89
groups was 0.027 – 0.2 (average = 0.103, median = .091); there was also
a correlation between average \(\text{PS}_{i}\) and theoretical minimum\(\text{PS}_{i}\) (rho = -0.231, p = 0.03). This potential bias is
addressed in the discussion.