Human land use and sex reversal
We quantified land use in a 500-m wide belt zone around each pond using
geoinformatics software as described in detail in (Bókony et al., 2018).
We divided each belt zone into the following 8 land-use categories:
natural vegetation (e.g. woodlands, non-agricultural meadows), arable
fields, pastures, residential areas, public built areas (e.g. commercial
and industrial areas), roads with vehicular traffic, railroads, and
water; and we calculated the proportion of area falling into each of
these categories (Table S1). Because railroads and water were present
only around 2 and 3 ponds, respectively, and covered very small areas
(Table S1), we omitted these from further analyses. We used two
alternative approaches to quantify the intensity of anthropogenic land
use for each capture site. First, we summed the proportions of arable
land, pastures, residential and public built-up areas, and roads for
each pond; we will refer to this variable as ”total anthropogenic land
cover”. In the second approach, we performed a principal components
analysis (PCA) using the 6 landscape variables, which yielded two axes
with >1 eigenvalue, explaining 82.1% of variation in
total. Urban landscape areas loaded positively on the first axis whereas
agricultural landscape areas loaded positively on the second axis (Table
S4, Figure S4). We will refer to the habitat scores along these two axes
as ”urban PC scores” and ”agricultural PC scores”. We analysed the
relationship between these habitat variables and XX/male ratio of the
adult frogs in generalized linear models with binomial error
distribution, using the brglm function in R package brglm(Kosmidis, 2019). This analysis weights each site by sample size (the
number of phenotypic males in our case) and appropriately handles
separation (i.e. in our dataset, there were no sex-reversed adults at
certain sites) by the maximum penalized likelihood method. One model
contained ”total anthropogenic land cover” as the only predictor,
whereas the other model contained the two PCA score variables
simultaneously.