24 Jérôme Chave
establishment of some species. Tuomistoet al.
(2003a) reported that the large-scale distribution
of melastomes was positively correlated with soil
Al content (r=0.26,P<0.05, Mantel test), as
expected given the status of most melastomes as
Al-accumulators (Jansenet al. 2002). Soil texture
(fraction of sand, silt, and clay) and the availabil-
ity of base-metal, or exchangeable, cations (Ca^2 +,
Mg^2 +,Na+,K+) have been investigated in other
studies. Soil texture was found to be a signifi-
cant predictor for understory plants (Tuomisto
et al. 2003b, Vormistoet al. 2004) but not for
canopy trees (Phillipset al. 2003). Although this
findin gdeserves further scrutiny, it also confirms
the naïve prediction that large-statured plants are
more tolerant to variation in soil texture, per-
haps because this variable primarily controls the
water holdin gcapacity. On the last correlation
emphasized by Sollins (1998), a significant partial
correlation between floristic diversity and base-
metal cations was indeed supported by data from
Phillipset al. (2003), Tuomistoet al. (2003a), and
Vormistoet al. (2004).
DISCUSSION
Recent progress in the study of tropical plant
beta-diversity has been greatly facilitated by
the establishment of ambitious field sampling
protocols (Gentry 1988, Duivenvoorden 1995,
Terborghet al. 1996, Pitmanet al. 2001, Pyke
et al. 2001, Balvaneraet al. 2002, Tuomisto
et al. 2003a, Vormistoet al. 2004). These works
have sought general explanations for the observed
patterns in beta-diversity, and have made exten-
sive use of correlative approaches, served by
the recent biostatistical literature (Legendre and
Legendre 1998). All too often, however, patterns
are described and post-hoc explanations are pro-
posed without explicit reference to a theoretical
framework. This has led to a tension between
results and their interpretation and calls for a
tighter connection between empirical work and
ecological theory (Hubbell 2001, Chaveet al.
2002, Chave and Leigh 2002, Ricklefs 2003,
2004).Beforeaskingwhetheracorrelationshould
be sought between tree floristic diversity across
sites and geographical distance between these
sites or, say, the difference in the soil concentra-
tion of base-metal cations, one should provide a
conceptual focus with which a priori hypotheses
can be tested. In a model of isolation by distance,
Chave and Leigh (2002) have shown that a corre-
lation was expected between one measure of floris-
tic diversity and the logarithm of the geographical
distance. Similar models should be developed to
justify searchin gfor correlation between floristic
diversity and abiotic environmental features.
There are other reasons why it might be diffi-
cult to relate the results of correlative approaches
to a theoretical framework. Ideally, one environ-
mental variable would predict the variation in
floristic composition, and this variation would
then be interpretable physiologically. However,
an existin gcorrelation between plant species
occurrence and environmental variation may
alsofailtobedetected,duetodispersal-relatedpro-
cesses: trees may be present in places where the
species is not perfectly adapted to the local con-
ditions, just because a large population is present
nearby (Shmida and Wilson 1985, Pulliam 1988,
Cannon and Leighton 2004). This source–sink
phenomenon may cause correlative approaches
to overestimate the importance of dispersal over
niche-assembly processes. Alternatively, a species
adapted to the environmental conditions of a site
may fail to be encountered in this site because
it may have been unable to disperse there, or
unable to invade in the absence of a facilitat-
in gspecies (Law and Morton 1996). This leaves
room for less-adapted species, but also confuses
the interpretation of any correlative analysis. Of
course, one might argue that simple correlative
approaches are better than mechanistic theo-
ries that make no, or patently false, predictions
(Currie 1991). However, it is unlikely that sim-
ple correlative approaches will predict patterns
of biodiversity at all scales (Latham and Ricklefs
1993, Qianet al. 2005). Although Francis and
Currie (2003) demonstrated a stron gcorrela-
tion between family-level richness in angiosperms
worldwide and environmental variables (temper-
ature, potential evapotranspiration, rainfall), the
abovementioned studies at smaller spatial scales
show that historical factors and complex environ-
mental gradients also play an important role, not
captured in a simple energy-based theory. At this