Statistical Analysis
All statistical analyses were performed in R version 3.5.3.
The presence of each parasite was used as the response variable in a
series of binomial generalised linear mixed models (GLMMs). Ecological
variables incorporated in each model were age, sex, Scaled Mass Index
(SMI) as a measure of condition, captures per transect per day (used
here as a proxy for population density), prevalence of microparasites,T. muris , S. obvelata , C. hepaticum and fleas, and
the infection intensity of T. muris, S. obvelata, C. hepaticumand fur mites. Location and trapping session were included as random
effects. Although it is likely that trapping session would have a
significant bearing on parasite infection dynamics, the irregular
spacing of each session means that any seasonal or inter-annual patterns
of parasite distribution are unlikely to become evident, and so it was
not included as a fixed effect.
For parasite species with measures of abundance (gastrointestinal
helminths and fur mites), a series of negative binomial GLMMs were
performed using abundance data, excluding uninfected individuals. As
infection levels of C. hepaticum were measured on an arbitrary
1-5 scale of infection score, the ‘clmm ’ function of the
‘ordinal ’ package in R was used to create a cumulative link
mixed model, taking the infection score as an ordinal factor. For this
model, individuals with infection scores of zero were taken to be
uninfected, and thus were excluded.
As there is missing data for microparasites and C. hepaticuminfection, models in which these parasites were not the predicted
variable were repeated with these species excluded to increase
statistical power. Any significant relationships between pairs of
parasites with complete datasets that are henceforth reported, are taken
from these exclusionary models.
Models of infection level were subsequently simplified and selected
through a model averaging process. The ‘dredge ’ function was
used to compare all possible sub-models, and those with ΔAIC <
2 when compared to the “best model” (with the lowest AIC) were
averaged using the ‘model.avg ’ function (both from R package
‘MuMIn ’) to find optimal models for each parasite species.
Due to the issues surrounding the use of Bonferroni corrections and
statistical power in ecological data (Nakagawa, 2004), the models are
not corrected for multiple testing, but all effects sizes are reported
(Supplementary Materials).