Data treatment
Using a multivariate approach, we explored and described our data using
a three-step procedure, taking into account the type of variables
(quantitative or qualitative) considered. As quantitative variables, we
determined (i) species abundances (i.e. number of individuals trapped)
by trap, room, locality and group of localities, and (ii) capture rates
(i.e. number of individuals of a given species divided by trapping
effort) for each locality. As qualitative variables, we considered the
type of trap (wire mesh or Sherman) at the trap scale, and the presence
of food, the type of room and nature of the floor (mostly concrete vs.
clay, aka “banco”), walls (mostly concrete vs. clay), and ceiling
(mostly concrete or corrugated iron vs. straw) at the room scale. Note
that variables noted at the scale of the rooms were aggregated in
percentages for each locality. We first performed i) a centred Principal
Component Analysis (cPCA) on the localities x species table (using
square roots of trapping success as data); ii) a fuzzy Correspondence
Analysis (fCA), on the localities x room characteristics (using numbers
of each modality for each variable); iii) a K+1 analysis coupling the
previous two analyses (Bougeard et al., 2011; Bougeard & Dray, 2018),
with the aim of describing the relationships between these two types of
data (rooms treated through partial least squares (PLS) regressions and
mammals described through a cPCA). The method is a multiblock PLS
regression (mbpls) applied to the particular case of a single response
dataset. Block of response variables are explained by a large number of
explanatory variables which are divided into K meaningful blocks. All
the variables – explanatory and dependent – are measured on the same
localities. The main results are summarized using overall graphical
displays. All data were analysed using ade4 R package (Chessel et al.,
2004; Dray et al., 2007; R Core Team 2022).
Then, a Pearson’s Chi-squared test was realised on the contingency table
enumerating the numbers of captures of the seven main species of the
small mammal community in the room types recorded, in order to evaluate
whether room types may explain the local distribution of each species.
The habitat component of the ecological niche of each species was
further evaluated using its distribution in the different room types
(considered as integrative descriptors of microhabitat) recorded.
Following Pianka (1973), we used two indices to characterize each
species niche and their overlap between species pairs:
Niche breadth quantified using Simpson’s index of diversity B = 1 /
∑pi2, where pi is
the proportion of the ith room type actually used by
the species.
Niche overlap based upon Levin’s (1968) index Oij =
∑pijpik / √
(∑pij2∑pik2),
where pij and pik are the proportions
of the ith room type used by the
jth and the kth species,
respectively
Finally, we examined co-occurrence patterns through the analyses of
presence–absence matrices with “null model” randomization tests of
marginal row and column totals (Gotelli, 2000; Gotelli & Ulrich, 2010)
using pairs software (Ulrich, 2008). Aggregated / random /
segregated pattern of co-occurrence of species pairs was inferred from
the p value associated with the Z–score for each pair of
species, either using the global dataset (from all 49 localities) or
local datasets (per locality and per district in large cities). We used
the “fixed row–fixed column” and “fixed row–equiprobable column”
randomization algorithms to generate randomized matrices that serve as
null models as advised by Gotelli (2000), and ran the models with 10,000
iterations.