Geoscience Reference
In-Depth Information
property (
Barnes et al., 2010; Brose et al., 2006a
). Further, analyses to
systematically investigate and compare the determinants of PPMR among
different definitions have never been conducted.
Here, we analyse the twomajor effects of predator species identity and body
mass on PPMR to estimate possible determinants of PPMRs. Through the
selection of 11 sampling sites containing multiple predator species from the
dataset (i.e. sites 01, 02, 03, 09, 12, 14, 15, 16, 18, 19, and 20), we developed
and tested the following four statistical models for each PPMR definition
and each sampling site: (i) a null model assuming that PPMR is common
among all individuals of all species (i.e. log
10
(PPMR)
), (ii) a taxonomic
model accounting for species-specificity of PPMR (i.e. log
10
(PPMR)
¼a
¼a
þb
(predator species identity)), (iii) an allometric model accounting for
size dependence of PPMR (i.e. log
10
(PPMR)
log
10
(predator body
mass)), and (iv) a combined model, including both the effects of species
identity and body mass (i.e. log
10
(PPMR)
¼aþb
¼aþb
1
(predator species identi-
ty)
log
10
(predator body mass)). PPMR and predator body mass are log trans-
formed to improve normality in the statistical analysis. Note that body mass
represents the averagedmeasurements for species-averaged and link-averaged
PPMRs, while individual mass is used for individual-predator and individual-
link PPMRs. We do not consider prey species identity as an explanatory
factor. Yet, this decision is not because prey species identity is not expected
to explain PPMR. Rather, this is simply because the information about prey
species identity is often absent from the datasets. Following
Barnes et al.
(2010)
, we use linear mixed models for individual-predator PPMR by includ-
ing individual identity as a random factor. Therefore, it should be noted that
individual-predator PPMR in this study is slightly different to that shown in
Figure 1
C, but the basic concept is still same, since we regard a predator
individual as a basic unit. On the other hand, linear models are applied for the
other three definitions of PPMR. By comparing the Akaike information
criteria (AIC), we determine the best statistical model. All analyses are per-
formed in the statistical environment R (
R Development Core Team, 2010
).
Mixed model analyses are conducted by using 'nlme' package of R (
Pinheiro
et al., 2009
).
þb
2
log
10
(predator body mass)
þb
3
(predator
species
identity)
B. Results
For species-averaged and link-averaged PPMRs, different models were se-
lected among the sampling sites and the best model was not clear (
Table 2
).
This may be partly due to the limitation of sample number (i.e. species
number). Taxonomic and combined models performed optimally for the
datasets, where multiple prey species were pooled into a single category
