1
Macroecology and extinction risk correlates of frogs
2
3
Natalie Cooper1,2*, Jon Bielby1,2 , Gavin H. Thomas3 and Andy Purvis1
4
5
1
Division of Biology, Imperial College London, Silwood Park, Ascot SL5 7PY, UK.
6
2
Institute of Zoology, Zoological Society of London, Regents Park, London, NW1 4RY,
7
UK.
8
3
9
Silwood Park, Ascot SL5 7PY, UK
NERC Centre for Population Biology, Division of Biology, Imperial College London,
10
11
*Author for correspondence (natalie.cooper04@imperial.ac.uk)
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1
26
A: ABSTRACT
27
28
B: Aim
29
30
Our aim was to test whether extinction risk in frogs could be predicted from
31
their body size, fecundity or geographic range size. Because small geographic range
32
size is a correlate of extinction risk in many taxa, we also tested hypotheses about
33
correlates of range size in frogs.
34
35
B: Methods
36
37
Using a large comparative dataset (n = 527 species) compiled from the
38
literature we performed bivariate and multiple regressions through the origin of
39
independent contrasts, to test proposed macroecological patterns and correlates of
40
extinction risk in frogs. We also created minimum adequate models to predict snout-
41
vent length, clutch size, geographic range size and IUCN Red List status in frogs.
42
Parallel non-phylogenetic analyses were also conducted. We verified the results of
43
the phylogenetic analyses using gridded data accounting for spatial autocorrelation.
44
45
B: Results
46
47
The most threatened frogs tend to have small geographic ranges although the
48
relationship between range and extinction risk is not linear. In addition, tropical frogs
49
with small clutches have the smallest ranges. Clutch size was strongly positively
50
correlated with geographic range size (r2 = 0.22) and body size (r2 = 0.28).
2
51
52
B: Main conclusions
53
54
Our results suggest that body size and fecundity only affect extinction risk
55
indirectly through their effect on geographic range size. Thus, although large frogs
56
with small clutches tend to be endangered, there is no comparative evidence that this
57
relationship is direct. If correct, this inference has consequences for conservation
58
strategy: it would be inefficient to allocate conservation resources on the basis of low
59
fecundity or large body size; instead it would be better to protect areas which contain
60
many, small geographic range frog species.
61
62
Keywords: extinction risk, independent contrasts, spatial autocorrelation, geographic
63
range size, body size, clutch size, amphibian, conservation.
64
65
A: INTRODUCTION
66
67
Amphibian declines and extinctions have long been a cause of concern to
68
herpetologists (Houlahan et al. 2000) but the publication of the Global Amphibian
69
Assessment (GAA) has brought these concerns to the forefront of conservation
70
biology. The GAA found that 1856 (32.5%) species of amphibian are threatened i.e.
71
within IUCN Red List categories Vulnerable, Endangered or Critically Endangered
72
(Stuart et al. 2004). This value is much higher than that calculated for mammals (23%
73
threatened; Baillie et al. 2004) and birds (12% threatened; Baillie et al. 2004). In
74
addition 1294 (22.5%) amphibian species (compared to 7.8% of mammals and 0.8%
75
of birds; Baillie et al. 2004) are classed as Data Deficient (Stuart et al. 2004), and it is
3
76
likely that many of these are also at risk (Akcakaya et al. 2000). The high proportion
77
of threatened and data deficient amphibian species make understanding the processes
78
behind amphibian extinction risk and population declines a priority in conservation
79
biology research.
80
81
In addition to the high prevalence of threatened species in amphibians, many
82
species have experienced serious population declines (Houlahan et al. 2000).
83
Research has tended to focus on possible extrinsic causes of declines, including
84
habitat modification, climate change, UV-B radiation, chemical contaminants, disease,
85
introduced species, and synergies among these factors (see Diversity & Distributions
86
Special Issue: Amphibian declines, 2003 for reviews). The susceptibility of species to
87
these threatening processes may vary: threat prevalence varies significantly among
88
families (Baillie et al. 2004; Bielby et al. 2006) even at a fine geographic scale where
89
threat severity might be expected to be more homogenous (Bielby et al. 2006). This
90
indicates that biological differences among amphibian families are at least partly
91
responsible for the variation in extinction risk.
92
93
Across a range of taxa, low fecundity, large body size and small geographic
94
ranges are amongst the most commonly cited extinction risk correlates in the literature
95
(McKinney, 1997; Purvis et al. 2000a) regardless of the precise mechanism
96
responsible for increased extinction risk. Each has also been found to correlate with
97
population declines or extinction risk in frogs (Williams & Hero, 1998; Lips et al.
98
2003; Murray & Hose, 2005a; Hero et al. 2005). However, these studies focussed on
99
small or moderate numbers of species in limited areas (n = 60, eastern Australia, Hero
100
et al. 2005; n = 99, Central America, Lips et al. 2003; n = 148, Australia, Murray &
4
101
Hose, 2005a; n = 30, Australian Wet Tropics, Williams & Hero, 1998), so may not
102
reflect global patterns. In addition, only two of these studies (Hero et al. 2005;
103
Murray & Hose, 2005a) attempt to control for the phylogenetic non-independence of
104
species despite the likely presence of a strong phylogenetic signal in these regional
105
datasets (e.g. Hero et al. 2005). We therefore aim to use phylogenetically
106
independent contrasts on a larger, geographically broader dataset to test the following
107
three hypotheses:
108
109
i)
Frogs with low fecundity will be at greater risk than those with high fecundity,
110
because they will be less able to compensate for any increase in mortality
111
(MacArthur &Wilson 1967) and because any recovery from low numbers will
112
take longer, increasing the chance of stochastic extinction prior to recovery
113
(Pimm et al. 1988).
114
ii)
In many taxa, large body size correlates with traits that promote extinction risk
115
e.g. low population density and high rates of exploitation (McKinney 1997;
116
Cardillo et al. 2003). If this is also true in frogs, large species will be more
117
threatened than small ones. In Asia at least, exploitation of amphibians for
118
food is mainly directed towards the larger-bodied species of the family
119
Ranidae (Baillie et al. 2004), indicating the plausibility of this hypothesis.
120
iii)
Frog species with small geographic ranges will be more at risk than those with
121
larger ranges because it is more likely that almost any threatening factor will
122
affect their entire range, or a large proportion of it. They are also more likely
123
to be habitat specialists and such specialisation is associated with local decline
124
in Australian frogs (Williams & Hero 1998). In addition, species with small
125
ranges tend to have small population sizes so demographic stochasticity and
5
126
inbreeding may further enhance extinction risk in the long run (O’Grady et al.
127
2006).
128
129
Small geographic range size tends to be the most important predictor in
130
analyses of extinction risk correlates even when species listed as threatened on the
131
basis of small range size are excluded from the analysis (e.g. in mammals and
132
butterflies: Cardillo et al. 2005a; Koh et al. 2004) . Therefore it is vital to understand
133
why some species have small geographic ranges. Geographic range size correlates
134
positively with body size among 178 Australian frog species (although not strongly: r2
135
= 0.09), Murray et al. 1998) and with clutch size among 40 upland eastern Australian
136
frog species (Hero et al. 2005). However, body size and clutch size may also be
137
positively correlated with each other and with latitude (Crump, 1974; Ashton, 2002),
138
and geographic range size is likely to be strongly correlated with habitat specificity.
139
We need to understand the relationships among these traits more fully if we are to
140
understand variations in geographic range size. We therefore intend to investigate the
141
correlations among geographic range size, habitat specificity, clutch size, body size
142
and latitude. These macroecological patterns are interesting in their own right since
143
the few studies of frog macroecology in the literature have looked at a smaller number
144
of species within an area or genus (e.g. n = 178, Australia; Murray et al. 1998; n =
145
148, Australia, Murray & Hose, 2005b; n = 60, Eastern Australia, Hero et al. 2005).
146
147
It can be difficult to determine the biological meaning of correlations involving
148
latitude since it is a complex surrogate for a number of environmental variables. To
149
counteract this problem we use several environmental variables (temperature,
150
precipitation and annual evapotranspiration rate) as well as latitude. If any of these
6
151
variables are better predictors than latitude in the models, they may be the real
152
variable of interest. These environmental variables may also be linked directly to
153
extinction risk in frogs since these taxa are very sensitive to changes in the
154
environment (Pounds et al. 2006) which can also increase their susceptibility to other
155
threats, for example the chytrid fungus which has an optimum growth temperature of
156
17-25ºC and is apparently more lethal under moist conditions (Piotrowski et al. 2004).
157
Therefore we also add these environmental variables to our models of extinction risk.
158
159
A: METHODS
160
161
B: Data
162
163
Extinction risk assessments, in terms of IUCN Red List status, came from the
164
GAA (GAA, 2004) and were recoded on a six point scale following Purvis et al.
165
(2000a) as follows: 0 = Least Concern (LC), 1 = Near Threatened (NT), 2 =
166
Vulnerable (VU), 3 = Endangered (EN), 4 = Critically Endangered (CR), and 5 =
167
Extinct in the Wild (EW) or Extinct (EX). Data deficient (DD) species were not
168
included in the analyses (see Bielby et al. 2006). IUCN Red List status provides the
169
best available comparable estimates of species extinction risk (Rodrigues et al. 2005).
170
171
Geographic ranges (in km2) and habitat breadths, as the number of habitats (defined
172
using the IUCN Habitat’s Authority file, Baillie et al. 2004) a species has been
173
recorded in, were also taken from the GAA (GAA, 2004). Habitat breadths were used
174
throughout as a measure of habitat specificity. Absolute latitudes were taken from the
175
geometric centre of each species’ geographic range. To extract the other
7
176
environmental variables needed we overlaid shape files of species’ geographic ranges
177
with a 30 seconds grid. We then recorded the median grid cell value within the
178
geographic range for datasets of annual actual evapotranspiration (AET) (mm)
179
(Willmott & Kenji, 2001), mean annual precipitation (mm) and mean annual
180
temperature (ºC) (Hijmans et al. 2004).
181
182
We compiled a data set of species values from the literature (including peer
183
reviewed literature, guide books, internet databases and reference books; see
184
Appendix S2 in Supplementary Material). Mean snout-vent length (mm) and mean
185
clutch size (eggs per clutch) were collected from a large number of sources (see
186
Appendix S3 in Supplementary Material). Where we had more than one value of a
187
variable for a species we used the mean. We biased our collecting effort towards
188
threatened species to increase the number of comparisons between threatened and non
189
threatened species, and to compensate for the fact that the literature is biased towards
190
widespread, abundant (and hence often low risk) species. Therefore, the dataset
191
contained an approximately equal number of threatened (VU, EN and CR) and non-
192
threatened (LC and NT) frogs. We used the taxonomy of Frost (2005) throughout. The
193
final dataset contained 527 frog species, representing 31 of 32 families, and every
194
continent on which frogs are found.
195
196
We normalised the distribution of all continuous variables using log
197
transformation prior to analysis. Collinearity amongst predictor variables can lead to
198
unreliability in model parameter estimates. Since we expect correlations among our
199
variables, prior to model building we checked the predictors for multicollinearity
200
(following the method of Belsley et al. 1980). For all models, variance inflation
8
201
factors are lower than three and condition indices are lower than nine which indicates
202
that no undesirable levels of multicollinearity are present. We used R version 2.1.1 in
203
all analyses (R Development Core Team, 2005).
204
205
Many variables in comparative data sets show phylogenetic pattern, with close
206
relatives tending to have similar values (Harvey & Pagel 1991). Snout-vent length
207
and clutch size are likely to show strong phylogenetic structure because species tend
208
to maintain the ecological niche occupied by their ancestors (Wiens 2004). The same
209
argument also applies to climatic variables and latitude, which might additionally
210
show structure if closely-related species live near to each other. Geographic range
211
size is not inherited by daughter species at speciation, but is shaped by traits that do,
212
so often shows phylogenetic structure (Purvis et al. 2005). Lastly, extinction risk is
213
shaped by the interaction between such phylogenetically-patterned traits and human
214
impacts, which will also tend to show geographic and hence phylogenetic pattern
215
(Purvis et al. 2005); it is therefore unsurprising that extinction risk is phylogenetically
216
non-random in frogs (Baillie et al. 2004; Bielby et al. 2006). If variables show
217
phylogenetic structuring in this way, statistical analyses that treat species as
218
independent points will run the risk of pseudreplication (Harvey & Pagel 1991).
219
Before conducting the main analyses, we therefore used analysis of variance
220
(ANOVA) to test whether each variable differed significantly among taxonomic
221
families, which would indicate phylogenetic structure in the data. Every variable in
222
our dataset showed highly significant phylogenetic structure even under this
223
conservative test (see Appendix S1, Table A1 in Supplementary Material)
224
demonstrating that species values are not independent. Therefore, we performed all
225
analyses using independent contrasts (Felsenstein 1985) generated using Pendek v
9
226
1.03 (Purvis et al. unpublished; available from A.P.) and the R package APE (Paradis
227
et al. 2005). Independent contrasts are calculated for each bifurcation of the
228
phylogeny by subtracting one observed value of the variable of interest from the other
229
and then standardising these values by dividing by the square root of the branch
230
lengths connecting the two taxa. In regressions of independent contrasts, each species
231
is presumed to fit the relationship Y = a + bX, therefore, because contrasts are
232
calculated by subtracting the values from two species, the equation of the line for a
233
contrast involving species 1 and 2 simplifies to Y1-Y2 = b(X1-X2). This equation for
234
the line has no intercept term so all regressions using independent contrasts must be
235
forced through the origin (Garland 1992).
236
237
We used the genus-level phylogeny of Frost et al. (2006), the largest and most
238
applicable phylogenetic analysis of living amphibians to date, with their taxonomy as
239
a surrogate for phylogeny within genera since their tree did not contain all of our
240
species. The Frost et al. (2006) phylogeny contains estimates of branch lengths but,
241
since we were using taxonomy as a surrogate for phylogeny at the species level, we
242
did not have suitable branch lengths for every branch. Therefore we set all branch
243
lengths to one unit. Independent contrasts have been shown to have acceptable power
244
and type I error rates with this procedure (Purvis et al. 1994).
245
246
B: Macroecology
247
248
We first used bivariate regressions through the origin (Garland et al. 1992) of
249
independent contrasts to investigate the effects of latitude, AET, mean annual
250
precipitation and mean annual temperature on snout-vent length, clutch size and
10
251
geographic range. We were mainly interested in correlates of range size in frogs but
252
we also looked at correlates of snout-vent length and clutch size since both are
253
thought to correlate with geographic range size. We also performed bivariate
254
regressions through the origin of independent contrasts to look for relationships
255
between snout-vent length, geographic range and clutch size, and between geographic
256
range size and habitat breadth. The relationship between habitat breadth and
257
geographic range size was predicted to be strong so it could lead to spurious
258
correlations between geographic range size and other variables also strongly
259
correlated with habitat breadth. Therefore, we repeated all regressions involving
260
geographic range size controlling for habitat breadth.
261
262
Most hypothesis tests using independent contrasts assume that the relationship
263
being tested for significance is linear, and may have greatly reduced power to detect
264
nonlinear relationships (Quader 2004). Unrecognised nonlinearities can also
265
undermine multiple regression models, which we go on to fit below. This occurs
266
because multiple regression assumes the model has been specified correctly so if Y is
267
actually a non linear function of X1 as well as of X2, Y may be estimated using X1in
268
one part of the graph and X2 in another, which can lead to spurious significant
269
relationships.
270
271
When testing for a correlation between Y and X, we therefore computed
272
contrasts in both X and X2 and used multiple regression. If the X2 term was not
273
significant, we dropped it from the model. If it was significant, we additionally tested
274
the significance of contrasts in X3 in the multiple regression model.
275
11
276
We then created minimum adequate models (MAM; Crawley 2002) for
277
predicting snout vent length, clutch size and geographic range size in frogs. Initially
278
we computed contrasts for a maximal model, which included all four environmental
279
variables, plus snout-vent length in the clutch size MAM, and both snout-vent length
280
and clutch size in the geographic range MAM, as well as any significant quadratic or
281
cubic terms. We then deleted the term with the highest p-value and recalculated the
282
contrasts. We continued to drop the term with the highest p-value until all terms were
283
significant (p<0.05). We tested the robustness of the MAM by adding deleted terms
284
back into the model one by one to check they were not wrongly omitted (Cardillo et al.
285
2005a).
286
287
If the pattern of trait variation among species departs strongly from the
288
random-walk model assumed in the calculation of independent contrasts, the contrasts
289
will show heterogeneity of variance, with some points having too much influence on
290
the results of regressions (Diaz-Uriarte & Garland 1996). Because our interest is in
291
general relationships across the dataset, rather than those driven by atypical species or
292
contrasts, we investigated the effects any highly influential points (i.e. those with a
293
studentised residual exceeding ±3; Jones & Purvis 1997), by deleting them and
294
repeating our analyses. Deletion of points only made a qualitative difference in two
295
cases (latitude against snout-vent length and latitude against geographic range size) so
296
we only report the results after deletion to improve the clarity of our tables.
297
298
B: Correlates of extinction risk
299
12
300
Some species are listed as threatened because of their small geographic range
301
size, so any relationship between extinction risk and geographic range could be
302
circular. To avoid this problem we omitted 160 species listed as threatened due to
303
their small geographic range size (IUCN Red List criteria B and/or D without A
304
and/or C; Baillie et al. 2004) leaving 229 non threatened species and 138 threatened
305
species in the extinction risk correlates analyses. This meant that 37.6% of the species
306
in this analysis were threatened, which lies midway between 32% (the overall
307
proportion of amphibian species considered threatened) and 41% (the proportion of
308
non-data-deficient amphibian species considered threatened).
309
310
We regressed IUCN Red List status against each variable in turn, using
311
independent contrasts. Then we repeated the analyses controlling for geographic range,
312
the most important predictor identified in the single predictor analyses. Finally we
313
created a MAM for predicting IUCN Red List status in frogs, as described above,
314
including all variables as model terms. Cardillo et al. (2005a) found significant
315
nonlinear terms contributed to models of extinction risk so we tested for nonlinearity
316
as above and included any significant non linear terms in the MAM. Cardillo et al.
317
(2005a) also found interactions significantly contributed to models of extinction risk
318
so we also included the interactions between the three life history variables as model
319
terms.
320
321
On finding a non linear relationship between geographic range size and extinction risk
322
in the MAM we performed sliding window analyses on the data (sensu Cardillo et al.
323
2005a). These analyses involved fitting a model of geographic range size against
324
IUCN Red List status for the contrasts within a geographic range size window (width
13
325
of the window = 2 units along the log geographic range size (x) axis). The window
326
was then moved up the geographic range size axis in increments of 0.25, with the
327
model being refitted for each geographic range size window. The analysis produces a
328
sliding window plot where the midpoint of the geographic range size window is on
329
the x axis, and the slope of geographic range size against IUCN status is on the y axis.
330
Autocorrelation is inherent in these plots so they should not be overinterpreted but
331
they provide a way of visualising the effects of non linear terms. We also found a non
332
linear relationship between mean annual temperature and extinction risk but we did
333
not investigate this further because mean annual temperature explained less than 5%
334
of the total explained variance for the extinction risk MAM.
335
336
Most previous studies on correlates of extinction risk in frogs have not used
337
independent contrasts, and their use in this kind of analysis has been disputed in the
338
past (Putland, 2005; but see Cardillo et al. 2005b), so we also carried out cross-
339
species analyses where we treated species as independent data points and the results
340
are available for comparison (see Appendix S1 in Supplementary Material, Tables A2
341
& A3).
342
343
B: Spatial autocorrelation
344
345
Spatial autocorrelation is present when the value of a variable at any one point in
346
space is dependent on the values at the surrounding points. This was likely to be a
347
problem with the analyses above since many factors in amphibian biology vary
348
regionally, including geographic range size and extinction risk. We therefore repeated
349
our phylogenetic analyses using gridded data. All variables were mapped as vectors
14
350
and converted to an equal area grid at a resolution of 193 km; this is equivalent to 2º
351
longitude at 30º latitude. We controlled for spatial autocorrelation using normal error
352
generalized least squares modelling in the R package nlme (Pinheiro et al., 2006). We
353
fitted exponential spatial covariance structures with the latitudinal and longitudinal
354
cell centroids as spatial variables. The maximum geographic distance (range
355
parameter , in degrees) over which spatial autocorrelation occurred was estimated
356
simultaneously with model fitting. The fitted values were verified by visual inspection
357
of the semi-variogram of residuals from the corresponding ordinary least squares
358
(OLS) models. The results did not differ qualitatively whether a spherical or
359
exponential correlation structure was used. This corrected for spatial non-
360
independence in the data but not phylogenetic non-independence. Currently methods
361
that control for both phylogenetic and spatial autocorrelation are in their infancy.
362
Since our main focus is on the factors that make a species, rather than an area, more
363
prone to extinction we compared our spatial results to our phylogenetic results but
364
report only the phylogenetic analyses in the main text with spatial results reported
365
only where they differ qualitatively from the significant phylogenetic results. The full
366
set of spatial models results are available in the appendix (see Appendix S1 in
367
Supplementary Material, Tables A4-A7).
368
369
B: Geographic range size data quality and range size distributions
370
371
The way in which the GAA geographic ranges were constructed limits their accuracy.
372
If the ecological conditions between two known localities of a species were
373
appropriate for that species, the GAA included the intervening area in its range map.
374
However, the GAA did not permit extrapolation beyond known locations, thus in
15
375
many cases these maps may be minimum estimates of species geographic ranges. In
376
addition, data quality varies across the Anura (some poorly known species have
377
unrealistic triangular or square geographic range polygons) and between geographic
378
regions. Any analysis which involved geographic range size would therefore be prone
379
to error.
380
381
We attempted to assess geographic range data quality by repeating the analyses of
382
Murray and Hose (2005b) using the GAA geographic range data instead of their range
383
data. If the results are not significantly different, they are robust to the reduced data
384
quality of the GAA range data which suggests that our results may also be robust to
385
data quality issues. We regressed geographic range size (from the GAA) against SVL
386
for the species from Murray and Hose (2005b) then compared the results with those
387
published in their study using analysis of covariance (ANCOVA) to determine
388
whether the slopes of the two relationships were significantly different.
389
390
Regional variation in geographic range data quality may also be a problem in our
391
analyses. We dealt with this by controlling for spatial autocorrelation (see below). We
392
also investigated the range size distributions of frog species in each region, and each
393
IUCN Red List category by comparing geographic range size in each region or Red
394
List category using ANOVAs and Tukey’s honestly significant difference (HSD) tests.
395
396
A: RESULTS
397
398
B: Macroecology
399
16
400
Snout-vent length was only weakly predicted by latitude, with larger species
401
being found at higher latitudes (t175 = 2.32, slope = 0.08, r2 = 0.03, p < 0.05). Clutch
402
size was strongly negatively correlated with mean annual precipitation (t107 = -4.94,
403
slope = -1.45, r2 = 0.19, p < 0.001) and also depended more weakly on AET (t105 = -
404
2.34, slope = -1.27, r2 = 0.05, p < 0.05). Geographic range size was weakly correlated
405
with latitude (t178 = -2.24, slope = -0.52, r2 = 0.03, p < 0.05), and non-linearly
406
correlated with mean annual temperature (temperature & temperature2: t176 = -3.36 &
407
3.75, slope = -14.2 & 3.12, r2 = 0.09, p < 0.001; temperature & temperature2 &
408
temperature3: t175 = 1.46 & -1.97 & 2.43, slope = 23.5 & -14.0 & 2.45, r2 = 0.12, p <
409
0.05.
410
411
There was a weak positive correlation between snout-vent length and
412
geographic range size (t180 = 3.51, slope = 1.70, r2 = 0.06, p < 0.001). There were
413
strong positive correlations between snout-vent length and clutch size (t107 = 6.47,
414
slope = 2.06, r2 = 0.28, p < 0.001; Fig. 1a), and between clutch size and geographic
415
range size (t106 = 5.50, slope = 0.85, r2 = 0.22, p < 0.001; Fig. 1b).
416
417
There was a strong positive correlation between geographic range size and
418
habitat breadth. When the regressions above were repeated controlling for habitat
419
breadth the results remained qualitatively similar apart from the correlation between
420
clutch size and geographic range size which became non-significant (t105 = 0.27, slope
421
= 1.61, r2 = 0.36, p = 0.08.
422
423
424
The macroecological MAMs showed that mean snout-vent length was weakly
correlated with absolute latitude (t175 = 2.32, r2 = 0.03, p < 0.05). Mean clutch size
17
425
increased with body size and mean annual temperature but decreased with mean
426
annual precipitation (Table 1) however, temperature only accounted for 0.96% of the
427
explained variance in the MAM (compared to 34.7% explained by mean annual
428
precipitation and 64.6% explained by snout-vent length). Mean geographic range size
429
increased with clutch size, decreased with mean annual precipitation and was non
430
linearly affected by mean annual temperature so that the largest ranges were at
431
medium to high temperatures (Table 1). Temperature accounted for 48.3% of the
432
explained variance in this MAM, clutch size 21.1% and mean annual precipitation
433
30.6%.
434
435
All of the significant results above remained significant in the same direction when
436
we controlled for spatial autocorrelation (see Appendix S1 in Supplementary Material,
437
Tables A4 and A5).
438
439
B: Correlates of extinction risk
440
441
All results below excluded species listed as threatened on the basis of small
442
range size. In single-predictor regression models, geographic range size (nonlinearly)
443
(Fig. 2; r2 = 0.50; 0.53; 0.57) mean annual temperature (nonlinearly) (r2 = 0.04; 0.08;
444
0.11), and habitat breadth (r2 = 0.10) were found to negatively correlate with
445
extinction risk in frogs (Table 2). In multiple regressions with geographic range as a
446
fixed predictor, low mean annual precipitation (r2 = 0.51) and large body size (r2 =
447
0.52) were associated with high risk (Table 2).
448
18
449
The MAM for predicting IUCN Red List status in frogs contained geographic range
450
size and mean annual temperature, including some non-linear terms (Table 3) with
451
geographic range size accounting for 96.6% of the explained variance and mean
452
annual temperature only 3.33%. Extinction risk tended to be highest when geographic
453
range size was small, and temperature was low. Sliding windows plots showed that
454
the relationship between geographic range size and extinction risk is positive at very
455
small geographic range sizes (approximately <20 km2) but generally negative at larger
456
geographic range sizes (Fig. 3).
457
458
The results of the non-phylogenetic analyses are shown in Appendix S1 in
459
Supplementary Material, Tables A2 & A3). In single predictor analyses all variables,
460
except AET, mean annual precipitation and mean annual temperature, are
461
significantly negatively correlated with extinction risk. All variables except clutch
462
size are also significantly non-linearly correlated with extinction risk. In multiple
463
regressions with geographic range size as a fixed predictor all variables (except
464
absolute latitude and some non-linear terms) are significant predictors of IUCN Red
465
List status. The MAM for predicting extinction risk is qualitatively similar to that
466
produced using independent contrasts except it also contains snout-vent length
467
(including non-linear terms) as a predictor.
468
469
All of the significant results above remained significant in the same direction when
470
we controlled for spatial autocorrelation (see Appendix S1 in Supplementary Material,
471
Tables A6 and A7).
472
473
B: Geographic range size data quality and range size distributions
19
474
475
There was no significant difference between the slopes of the relationships
476
between snout-vent length and geographic range size using the range data of Murray
477
and Hose (2005b) and the range data from the GAA (t292= -1.17 , p = 0.24). This
478
suggests that our results are robust to geographic range size data quality issues.
479
480
Species geographic range sizes differed significantly between regions (F5,487 =
481
23.7, p < 0.001) and IUCN Red list categories (F5,351 = 122.8, p < 0.001 excluding
482
species classed as threatened on the basis of their small range size). European and
483
North American frogs had larger ranges than Asian frogs, and they had larger ranges
484
than Australasian, Latin American and African species. Species classed as LC on the
485
IUCN Red List had the largest geographic ranges followed by species classed as NT,
486
VU and EN. CR and EX species had the smallest geographic ranges.
487
488
A: DISCUSSION
489
490
This study aimed to investigate whether extinction risk in frogs could be
491
predicted from their body size, fecundity or geographic range size and, since
492
geographic range size is such an important predictor of extinction risk, to explore why
493
some frogs have smaller geographic ranges than others. The resulting MAMs suggest
494
that frogs with small geographic ranges are most at risk; and that tropical species with
495
small clutches have the smallest geographic ranges. Mean annual temperature also
496
affects extinction risk but this effect is very weak.
497
20
498
Our results corroborate the findings of several previous regional, and more
499
taxonomically restricted, studies which also found that small geographic and
500
elevational range was a correlate of population decline or extinction risk in frogs (Lips
501
et al. 2003; Hero et al. 2005; Murray & Hose 2005a), although in our study the
502
relationship between geographic range size and extinction risk is much stronger (r2 =
503
0.50 compared to r2 = 0.20 in Murray & Hose 2005a. Lips et al. 2003 and Hero et al.
504
2005 do not contain r2 values). This may be because regions show similar patterns so
505
when these are combined in a global model the predictive power is greater, or may
506
reflect the greater influence of other variables on extinction risk at smaller scales.
507
508
Frog species with small geographic ranges may be more threatened than those with
509
larger geographic ranges for several reasons. Species with small ranges are more
510
likely to be exposed to a threatening process throughout their entire range, which may
511
threaten global extinction. For example, several narrow range endemic amphibian
512
species of Central America have experienced population declines leading to increased
513
Red List status (e.g. Atelopus spp), and possibly even extinction (e.g. Bufo periglenes).
514
At the same time, other species are exposed to the same threats in the same locations
515
may have declined locally but not declined sufficiently throughout their range to
516
qualify for an increase in extinction risk as measured by the IUCN Red List.
517
Australian frogs with small geographic ranges have been shown to have low
518
abundance (Murray et al. 1998). If this correlation is generally true in frogs it could
519
make small geographic range species more susceptible to stochastic events or to
520
elevated mortality. Finally, restricted range species are often habitat or environmental
521
specialists. In our analyses species with smaller habitat breadths, and hence higher
522
habitat specificity, had smaller geographic ranges and were more at risk than those
21
523
with lower habitat specificity. Habitat specialisation was also associated with local
524
population declines in Australian rainforest frogs (Williams & Hero 1998) while
525
narrow tolerance to environmental conditions make a species highly susceptible to
526
climatic change or fluctuations (Murray & Hose 2005a).
527
528
Whilst our overall results suggest that frogs with smaller geographic ranges
529
are the most threatened, the non-linear relationship between geographic range size and
530
extinction risk showed that, among species with very small ranges, risk is higher for
531
species with larger (but still small) ranges. This is probably a sampling artefact; if the
532
tendency to underestimate the extinction risk of a species increases as its range size
533
decreases (because small range species are often rare, so assessing their IUCN Red
534
List status can be difficult), then the extinction risk of very small range species would
535
apparently increase with their geographic range size.
536
537
Although large body size has been found to correlate with population declines
538
in some non-phylogenetic analyses (e.g. Lips et al. 2003), other studies have found no
539
relationship (Murray & Hose 2005a). Our results show no direct effect of body size on
540
extinction risk in frogs, though we note that snout vent length was a predictor in the
541
extinction risk MAM in non-phylogenetic analyses (see Appendix S1 in
542
Supplementary Material). We also find no direct link between clutch size and
543
extinction risk in frogs, once geographic range size is controlled for, although
544
fecundity has been shown to correlate with decline rates in previous studies (Williams
545
and Hero 1998; Hero et al. 2005). The differences between our results and those of
546
previous studies may reflect regional patterns (previous studies only looked at either
547
Australia e.g. Hero et al. 2005; Murray & Hose 2005a; Williams & Hero, 1998; or
22
548
Central America e.g. Lips et al. 2003), or phylogenetic non-independence (only Hero
549
et al. 2005 and Murray & Hose 2005a, controlled for phylogeny, and see our cross-
550
species results), but equally may be due to the correlation between fecundity and
551
geographic range size. Additionally, the majority of previous studies focussed on
552
groups of species that are probably threatened by a single process - infectious disease.
553
As such, they were looking at correlates of a particular threat mechanism. The
554
biological traits underlying increased extinction risk/decline can and often do vary
555
according to the particular threat involved, the environment of the location of study,
556
and the particular species involved (Bennett & Owens 2000). Our analyses explore
557
the generalities that exist despite these differences, but will miss some of the specific
558
regional correlates. These threat process and region specific responses may also
559
explain why our MAM explained just over half of the variation in extinction risk
560
between the frog species in our dataset leaving much of the variation unaccounted for.
561
Some of this variation may be also explained by additional general variables which
562
were not in our dataset but it is likely that much variation in extinction risk is
563
idiosyncratic, which may pose problems for conservationists trying to conserve frogs
564
on anything but a species by species basis.
565
566
Our results also show that geographic range location as well as geographic
567
range size may influence extinction risk in frogs: mean annual temperature of an area
568
affects threat levels. Although this relationship was weak at best, temperature still
569
appears in the MAM for predicting IUCN Red List status, so appears to have some
570
effect on extinction risk in frogs. As well as direct effects of high temperatures, this
571
correlation might reflect other factors associated with the regions in which high
572
temperatures are found, such as thermal seasonality, which is lower in the tropics
23
573
making the environmental conditions there more stable. Amphibian species and
574
populations living at higher, more stable temperatures commonly have a faster life
575
history (i.e. more eggs per clutch and more clutches per year) than those living at
576
lower, more variable temperatures (Morrison & Hero 2003) which could explain why
577
our results indicate that tropical species are less threatened than temperate species.
578
579
As expected, small geographic range size was the most significant predictor of
580
extinction risk in our phylogenetic analyses, so understanding geographic range size is
581
vital to the understanding of amphibian extinction biology. The correlations among
582
temperature, precipitation and geographic range size suggest that large geographic
583
ranges may be more easily maintained in drier regions with medium to high
584
temperatures. Such regions fall in the temperate and subtropical zones, where there is
585
a much greater area for large geographic ranges, and fewer mountain ranges and
586
archipelagos to restrict species dispersal than in the warm, wet tropics. Clutch size is
587
also a significant component of the geographic range size MAM. This relationship
588
between clutch size and geographic range has been previously reported by Hero et al.
589
(2005) for eastern Australian frogs but seems to be much more general and among the
590
strongest predictors of geographic range size yet found in any group. These results
591
suggest that more fecund species may have greater potential to spread to new areas
592
and occupy large geographic ranges. Geographic range can be linked to both dispersal
593
ability and colonisation success in amphibians (Green 2003) suggesting that dispersal
594
ability may have a part to play in the formation of this pattern. Dispersal ability was
595
also the main determinant of range size in Sylvia warblers (Bohning-Gaese et al.
596
2006). Although many previous studies show that body size is correlated with
597
geographic range size (Gaston & Blackburn 1996), our MAM suggests this
24
598
correlation in frogs may be the result of intercorrelations among clutch size, snout-
599
vent length and geographic range size. Geographic range size in Australian frogs has
600
also been found to correlate with egg size and local abundance (Murray & Hero
601
2005b). These variables, which were not available for many species in our dataset,
602
also need to be taken into account before we can have a proper understanding of frog
603
geographic range size especially since much of the variation in geographic range size
604
is not captured by our MAM (r2 = 0.36).
605
606
Clutch size tends to be larger in larger species living in dry places, and may
607
also be affected by mean annual temperature although this explains very little (<1%)
608
of the variance in our models. Our results represent the first evidence of the positive
609
correlation between body size and clutch size in frogs on a broad taxonomic and
610
geographic scale, confirming results from earlier studies which looked at just a few
611
species in a single area or genus (e.g. Crump 1974). Latitudinal gradients in body size
612
and clutch size have previously been used to explain this pattern (Morrison & Hero
613
2003) but this mechanism cannot explain our results because clutch size did not
614
correlate with absolute latitude. Another possibility is that large species have more
615
energy available to put into reproduction because they can access more resources than
616
small species (Preziosi et al. 1996) and have less stringent energy requirements
617
(Brown & Maurer 1989). Larger species also have larger abdomens than small species,
618
into which they can physically fit more eggs: such a mechanism has been
619
demonstrated in snakes (Bonnett et al. 2000).
620
621
Phylogenetic comparative analyses of extinction risk have not previously considered
622
that comparative data can contain spatial as well as phylogenetic non-independence.
25
623
Spatial autocorrelation in variables such as geographic range size could cause
624
significant, if weak, correlations in any analysis (whether phylogenetic or not) that
625
does not explicitly consider this form of non-independence. However, every one of
626
the significant correlations from our analysis of independent contrasts was also
627
significant in a fully spatial (but non-phylogenetic) model, indicating that they are true
628
patterns, not artefacts of particular analytical approaches. There is an urgent need for
629
methods which deal with both phylogenetic and spatial sources of non-independence
630
to be developed more fully.
631
632
Our dataset contains representatives of 31 of 32 frog families, from every
633
continent on which frogs are found, and has a balance of threatened and non-
634
threatened species. Nonetheless, although it contains far more species than in previous
635
studies, this is still only around 10% of frog species. A more inclusive dataset might
636
reveal additional correlations, though we suggest our data should be sufficient to
637
detect any strong signals.
638
639
Although we attempt to address data quality issues it is possible that the
640
quality of the GAA range data may have affected our results, especially those where
641
correlations are weak. There are substantial regional and taxonomic differences in
642
data quality with well sampled regions (e.g. North America) and common species
643
having more accurate range maps than less well sampled regions (e.g. Africa) and
644
rarer species. Amphibians also tend to be patchily distributed due to water and habitat
645
requirements, a fact which is not taken into account in the GAA maps. This means the
646
range maps may not accurately represent the actual distributions of the species.
647
However, the amount of error in the range maps is likely to be proportional to range
26
648
size (maps of larger ranges probably overestimate true ranges by more than maps of
649
small ranges) so the amount of error in small and large ranges will be equal when we
650
log range size. Therefore, the relative sizes of the ranges will be correct, which is all
651
that is needed for hypothesis testing, so the patchiness of species distributions should
652
not significantly alter the results presented here. Ideally all the geographic ranges
653
would have been built using point locality data from which geographic ranges could
654
be extrapolated. However, this was unfeasible since point locality data does not exist
655
for the majority of species in our data set.
656
657
According to our models, the most threatened frog species tend to have small
658
geographic ranges and may also be affected by high mean annual temperatures,
659
although our evidence for a temperature effect is weak. In addition, species with small
660
clutch sizes, living in wet regions with medium to high mean annual temperatures (i.e.
661
tropical regions) have the smallest geographic ranges. Together these results tell an
662
interesting story. The fecundity and body size of frog species are not directly linked to
663
extinction risk, but their current distribution (geographic range size) is. The
664
distributions of the species in our analyses are probably influenced by both biological
665
and environmental factors. Many of these are correlated with extinction risk in single
666
predictor analyses, but not once geographic range size was entered into the model,
667
indicating that their effects on risk are indirect (i.e., through geographic range size)
668
rather than direct. This is further suggested by the fact that neither fecundity nor body
669
size are model terms in the extinction risk MAM. Since fecundity, body size and the
670
environment in which a species finds itself are the result of its evolutionary and
671
biogeographic history, this suggests that, in frogs, it is the history of the species that
672
shapes their conservation status rather than their size or fecundity. This contrasts with
27
673
mammals, within which particular biological traits including body size do contribute
674
independently to extinction risk (Cardillo et al. 2005a). If correct, the inference that
675
size and fecundity do not directly shape extinction risk has consequences for
676
conservation strategy. Neither large frog species nor those with low fecundity appear
677
be at particular risk of extinction for a given geographic range area, suggesting that it
678
would be inefficient to allocate conservation resources on the basis of these life
679
history traits. It would be better to protect areas which contain many, small
680
geographic range species instead.
681
682
A: ACKNOWLEDGEMENTS
683
684
We would like to thank the following for comments on previous versions of
685
this manuscript and assistance in data analysis: Marcel Cardillo, David Orme, Kate
686
Jones, Shai Meiri, Owen Jones, Susanne Fritz, and Nicola Toomey. We would also
687
like to thank Brad Murray for providing us with his geographic range size data.
688
689
A: SUPPLEMENTARY MATERIALS
690
691
The following material is available online at www.blackwell-synergy.com/loi/geb
692
693
B: Appendix S1: Supplementary tables
694
B: Appendix S2: Dataset
695
B: Appendix S3: Sources of snout-vent length and clutch size data
696
697
28
698
A: REFERENCES
699
700
Akcakaya, H.R., Ferson, S., Burgman, M.A., Keith, D.A., Mace, G.M., & Todd, C.
701
2000. Making consistent IUCN classifications under uncertainty. Conservation
702
Biology ,14, 1001-1013.
703
704
Ashton, K.G. 2002. Do amphibians follow Bergmann’s rule? Canadian Journal of
705
Zoology, 80, 708-716.
706
707
Baillie, J. E. M., Hilton-Taylor, C., & Stuart, S. N. 2004. 2004 IUCN Red List of
708
Threatened Species. A Global Species Assessment. IUCN, Gland, Switzerland and
709
Cambridge, UK
710
711
Belsley, D.A., Kuh, E., & Welsch, R.E. 1980. Regression Diagnostics. John
712
Wiley, New York.
713
714
Bielby, J., Cunningham, A., & Purvis, A. 2006. Taxonomic selectivity in
715
amphibians: ignorance, geography or biology? Animal Conservation, 9, 135-143.
716
717
Blaustein, A.R., & Dobson, A. 2006. A message from the frogs. Nature, 439: 143-
718
144.
719
720
Bonnett, X., Naulleau, G., Shine, R., & Lourdais, O. 2000. Reproductive versus
721
ecological advantages to larger body size in female snakes, Vipera aspis. Oikos,
722
89, 509-518.
29
723
724
725
726
Bohning-Gaese, K., Caprano, T., van Ewijk, K., & Veith, M. 2006. Range size:
Disentangling current traits and phylogenetic and biogeographic factors, American
Naturalist, 167, 555-567.
727
728
Brown, J.H. & Maurer, B.A. 1989. Macroecology – the division of food and space
729
among species on continents. Science, 243, 1145-1150.
730
731
Cardillo, M. 2003. Biological determinants of extinction risk: why are smaller
732
species less vulnerable. Animal Conservation, 6, 63-69.
733
734
Cardillo, M., Mace, G.M., Jones, K.E., Bielby, J., Bininda-Emonds, O.R.P.,
735
Sechrest, W., Orme, C.D.L. & Purvis, A. 2005a. Multiple causes of high
736
extinction risk in large mammal species. Science, 309: 1239-1241.
737
738
Cardillo, M., G.M. Mace, and A. Purvis. 2005b. Problems of Studying Extinction
739
Risks: Response. Science. 310: 1276-1278.
740
741
Crawley, M.J. 2002. Statistical computing: an introduction to data analysis using
742
S-plus. John Wiley and Sons Ltd.: Chichester.
743
744
Crump, M.L. 1974. Reproductive strategies in a tropical anuran community. Univ.
745
Kansas Museum of Natural History Miscellaneous Publications, 61, 1-68.
746
30
747
Diaz-Uriarte, R., and T. Garland. 1996. Testing hypotheses of correlated evolution
748
using phylogenetically independent contrasts: Sensitivity to deviations from
749
brownian motion. Systematic Biology. 45:27–47.
750
751
Diversity & Distributions Special Issue: Amphibian declines. 2003. 9, 89-167.
752
753
Felsenstein, J. 1985. Phylogenies and the comparative method. American
754
Naturalist, 125, 1-15.
755
756
Frost, Darrel R. 2004. Amphibian Species of the World: an Online Reference.
757
Version 3.0, 2004. http://research.amnh.org/herpetology/amphibia/index.html.
758
American Museum of Natural History, New York, USA. (Accessed May-July
759
2005).
760
761
Frost, D.R., T. Grant, J. Faivovich, R.H. Bain, A. Haas, C.F.B. Haddad, R.O. De
762
Sa, A. Channing, M. Wilkinson, S.C. Donnellan, C.J. Raxworthy, J.A. Campbell,
763
B.L. Blotto, P. Moler, R.C. Drewes, R.A. Nussbaum, J.D. Lynch, D.M. Green,
764
W.C. Wheeler. 2006. The amphibian tree of life. Bulletin of the American
765
Museum of Natural History. 297: 1-291.
766
767
GAA 2004. IUCN Global Amphibian Assessment. IUCN Species Survival
768
Commission, Conservation International, Biodiversity Centre for Applied
769
Biodiversity Science and Nature Serve. http://www.globalamphibians.org.
770
(Accessed May-July 2005).
771
31
772
Garland, T., Harvey, P., & Ives, A.R. 1992. Procedures for the analysis of
773
comparative data using phylogenetically independent contrasts. Systematic
774
Biology, 41, 18-32.
775
776
Gaston, K.J. and Blackburn, T.M. 1996. Global scale macroecology: interactions
777
between population size, geographic range size and body size in the Anseriformes.
778
Journal of Animal Ecology, 65: 701-714.
779
780
Green, D.M. 2003. The ecology of extinction: population fluctuation and decline
781
in amphibians. Biological Conservation, 111, 331-343.
782
783
Harvey, P.H. & Pagel, M. 1991. The comparative method in evolutionary biology.
784
Oxford University Press: Oxford.
785
786
Hero, J.M., Williams, S.E., & Magnusson, W.E. 2005. Ecological traits of
787
declining amphibians in upland areas of eastern Australia. Journal of Zoology
788
London, 267, 221-232.
789
790
Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. & Jarvis, A. 2004. The
791
WorldClim interpolated global terrestrial climate surfaces. Version 1.3. Available
792
at http://biogeo.berkeley.edu/.
793
32
794
Houlahan, J.E., Findlay, C.S., Schmidt, B.R., Meyer, A.H., and Kuzmin, L. 2000.
795
Quantitative evidence for global amphibian population declines. Nature, 404, 752-
796
755.
797
798
Jones, K.E. & Purvis, A. 1997. An optimum body size for mammals?
799
Comparative evidence from bats. Functional Ecology, 11, 751-756.
800
801
Jones, K. E., Purvis, A., & Gittleman, J. L. 2003, Biological correlates of
802
extinction risk in bats. American Naturalist, 161, 601-614.
803
804
Koh, L.P., N.S. Sodhi, and B.W. Brook. 2004. Ecological correlates of extinction
805
proneness in tropical butterflies. Conservation Biology. 18: 1571-1578.
806
807
Lips, K.R., Reeve, J.D., & Witters, L.R. 2003. Ecological traits predicting
808
amphibian population declines in Central America. Conservation Biology, 17,
809
1078-1088.
810
811
MacArthur, R.H. & Wilson, E.O. 1967. The theory of island biogeography.
812
Princeton University Press.
813
814
McKinney, M.L. 1997. Extinction vulnerability and selectivity: combining
815
ecological and paleontological views. Annual Review of Ecology & Systematics,
816
28, 495-516.
33
817
818
Morrison, C. & Hero, J. M. 2003. Geographic variation in life-history
819
characteristics of amphibians: a review. Journal of Animal Ecology, 72, 270-279.
820
821
Murray, B.R., Fonseca, C.R., & Westoby, M. 1998. The macroecology of
822
Australian frogs. Journal of Animal Ecology, 67, 567-579.
823
824
Murray, B.R., & Hose, G.C. 2005a. Life-history and ecological correlates of
825
decline and extinction in the endemic Australian frog fauna. Austral Ecology, 30,
826
564-571.
827
828
Murray, B. R. & Hose, G. C. 2005b. The interspecific range size-body size
829
relationship in Australian frogs. Global Ecology & Biogeography, 14, 339-345.
830
831
O’Grady, J.J., Brook, B.W., Reed, D.H., Ballou, J.D., Tonkyn, D.W., & Frankham,
832
R. 2006. Realistic levels of inbreeding depression strongly affect extinction risk in
833
wild populations. Biological Conservation, 133: 42-51.
834
835
Owens, I.P.F. & Bennett, P.M. 2000. Ecological basis of extinction risk in birds:
836
habitat loss versus human persecution and introduced predators. Proceedings of
837
the National Academy of Sciences, 97: 12144-12148.
838
839
Paradis, E., Strimmer, K., Claude, J., Opgen-Rhein, R., Dutheil, J., Noel, Y. &
840
Bolker, B. 2005. APE: analyses of phylogenetics and evolution. R package v. 1.6.
34
841
842
Pimm, S.L., Jones, H.L., & Diamond, J. 1988. On the risk of extinction. American
843
Naturalist, 132, 757-785.
844
845
Piotrowski, J.S., Annis, S.L., & Longcore, J.E. 2004. Physiology of
846
Batrachochytrium dendrobatidis, a chytrid pathogen of amphibians. Mycologia,
847
96: 9-15.
848
849
Pounds, J.A., Bustamante, M.R., Coloma, L.A., Consuegra, J.A. Michael P. L.
850
Fogden, M.P.L., Foster, P.N., La Marca, E. Masters, K.L., Merino-Viteri, A.
851
Puschendorf, R., Ron, S.R., Sánchez-Azofeifa, G.A., Still, C.J., & Young, B.E.
852
2006. Widespread amphibian extinctions from epidemic disease driven by global
853
warming. Nature, 439: 161-167.
854
855
Preziosi, R. F., Fairbairn, D. J., Roff, D. A., & Brennan, J. M. 1996. Body size and
856
fecundity in the waterstrider Aquarius remigis: A test of Darwin's fecundity
857
advantage hypothesis. Oecologia, 108, 424-431.
858
859
Purvis, A. Gittleman, J.L., & Luh, K.H. 1994. Truth or consequences: effects of
860
phylogenetic accuracy on two comparative methods. Journal of Theoretical
861
Biology, 167: 293-300.
862
863
Purvis, A., Gittleman, J. L., Cowlishaw, G., & Mace, G. M. 2000a. Predicting
864
extinction risk in declining species. Proceedings of the Royal Society London
865
Series B: Biological Sciences, 267, 1947-1952.
35
866
867
Purvis, A., Cardillo, M., Grenyer, R., & Collen, B. 2005. Correlates of extinction
868
risk: phylogeny, biology, threat and scale in Phylogeny and Conservation. Purvis,
869
A., Brooks, T. M., & Gittleman, J. L. eds. Cambridge University Press.
870
871
Putland, D. 2005. Problems of studying extinction risks. Science, 310, 1277.
872
873
Quader, S., K. Isvaran, R.E. Hale, B.G. Miner, and N.E. Seavy. 2004. Nonlinear
874
relationships and phylogenetically independent contrasts. Journal of Evolutionary
875
Biology. 17: 709-715.
876
877
R Development Core Team. R: A language and environment for statistical
878
computing. 2005. Vienna, Austria, R Foundation for Statistical Computing.
879
Available at http://www.R-project.org.
880
881
Rodrigues, A. S. L., J. D. Pilgrim, J. F. Lamoreux, M. Hoffman, and T. M. Brooks.
882
2006. The value of the IUCN Red List for conservation. Trends in Ecology and
883
Evolution. 21: 71-76.
884
885
Stuart, S. N., Chanson, J. S., Cox, N. A., Young, B. E., Rodrigues, A. S. L.,
886
Fischman, D. L., & Waller, R. W. 2004. Status and trends of amphibian declines
887
and extinctions worldwide. Science, 306, 1783-1786.
888
889
Wiens, J.J. 2004. Speciation and ecology revisited: phylogenetic niche
890
conservatism and the origin of species. Evolution, 58, 193-197.
36
891
892
Williams, S.E., and Hero, J.M. 1998. Rainforest frogs of the Australian wet
893
tropics: guild classification and the ecological similarity of declining species.
894
Proceedings of the Royal Society London Series B: Biological Sciences, 265, 597-
895
602.
896
897
Willmott, C. J., & Kenji, M. 2001. Terrestrial Water Budget Data Archive:
898
Monthly Time Series (1950-1999). Available at
899
http://climate.geog.udel.edu/~climate/ html_pages/README.wb_ts2.html.
900
901
A: BIOSKETCH
902
903
Natalie Cooper is a Ph.D. student at Imperial College London and The Institute of
904
Zoology. This work formed part of her MSc thesis. She is currently working on
905
character evolution and community assembly in mammals.
906
Jon Bielby is a Ph.D. student at Imperial College London and The Institute of
907
Zoology. His research is focussed on patterns of extinction risk and population
908
declines in amphibians.
909
Gavin H. Thomas is a postdoctoral researcher at Imperial College London. He has
910
research interests in the evolutionary patterns of avian diversity and in the
911
phylogenetic and phenotypic structure of ecological communities.
912
Andy Purvis is Professor of Biodiversity at Imperial College London where he uses
913
phylogenetic approaches to research macroevolution and conservation biology.
914
915
37
916
Tables
917
918
Table 1. Minimum adequate models (MAM) predicting snout-vent length, clutch size
919
and geographic range size in frogs. SVL = Snout-vent length (mm), Precipitation =
920
Mean annual precipitation (mm), Temperature = Mean annual temperature (ºC). For
921
clarity only significant terms are included. *p<0.05, **p<0.01, ***p<0.001.
922
Clutch size
Geographic range size
df
105
103
r2
0.45
0.36
predictors
coefficient
t
coefficient
t
Precipitation
-1.40
-5.54***
-2.45
-4.06***
Temperature
0.71
2.06*
37.4
2.66**
Temperature2
-
-
-20.7
-3.14**
Temperature3
-
-
3.46
3.52***
1.95
6.87***
-
-
0.45
2.78**
SVL
Clutch size
923
924
925
926
927
38
928
Table 2. Results of one and two predictor regressions of independent contrasts for
929
predicting IUCN Red List status in frogs. Species listed as Threatened based only on
930
their small geographic range size been omitted. Precipitation = Mean annual
931
precipitation (mm), Temperature = Mean annual temperature (ºC). Cubic and
932
quadratic terms are only included where significant. *p<0.05, **p<0.01, ***p<0.001
933
With
As single predictor
geographic
range
predictor variable(s)
Geographic range size
df
134
Geographic range size &
t
-11.5***
t
-
-
-
-
-
-
-5.51***
133
Geographic range size2
3.18**
Geographic range size &
1.71
Geographic range size2 &
df
132
Geographic range size3
-2.98**
3.41***
Snout-vent length
134
-0.38
133
2.26*
Clutch size
84
-2.79**
83
0.72
Habitat breadth
134
-3.87***
133
0.13
Latitude
133
-0.20
132
-0.91
AET
132
-1.70
131
-1.95
Precipitation
133
0.45
132
-2.13*
Temperature
133
-1.53
132
-0.96
Temperature &
132
2.27*
131
1.09
39
Temperature2
-2.57*
-1.26
Temperature &
-1.52
-1.33
Temperature2 &
Temperature3
131
1.89
-2.22*
130
1.53
-1.69
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
40
954
Table 3. Minimum adequate models (MAM) predicting extinction risk in frogs.
955
Species listed as Threatened based only on their small geographic range size been
956
omitted. Temperature = Mean annual temperature (ºC). For clarity only significant
957
terms are included. *p<0.05, **p<0.01, ***p<0.001.
958
MAM
df
129
r2
0.59
coefficient
t
Geographic range size2
-0.07
-8.15***
Geographic range size3
<0.01
6.57***
Temperature2
0.69
2.05*
Temperature3
-0.18
-2.16*
predictors
959
960
961
962
963
964
965
41
966
Figure legends
967
968
Figure 1: Regression of independent contrasts through the origin showing the positive
969
correlations between (a) snout-vent length (SVL) and clutch size (t107 =6.47, p <<
970
0.001, r2 =0.28), and (b) clutch size and geographic range size in frogs (t106 =5.50, p
971
<< 0.001, r2 =0.22).
972
973
Figure 2: Regression of independent contrasts through the origin showing the
974
negative correlation between extinction risk and geographic range size in frogs (t134 =
975
-11.5, p << 0.001, r2 =0.50). Frog species listed as Threatened based only on their
976
small geographic range size been omitted.
977
978
Figure 3: Slope of geographic range size against extinction risk at different range
979
sizes. Frog species listed as Threatened based only on their small geographic range
980
size been omitted. The solid line is a lowess smoother fitted through the points with a
981
span of 0.5.
982
983
984
985
986
42