version of the EMIB, but are not necessarily
inconsistent with equilibrium theories.
Several authors have developed more formal
approaches to address the habitat effects that con-
cerned Whitehead and Jones. One such study is
Buckley’s (1982, 1985) habitat-unit model of island
biogeography. The standard method used by most
authors has simply been to include indices of habi-
tat diversity, along with area, isolation, altitude,
and other geographical factors, in multiple regres-
sions of species number (e.g. Ricklefs and Lovette
1999; Morand 2000). Buckley took a different
approach, dividing islands into definable habitats
and floristic elements, and relating species richness
to area and isolation independently for each unit.
His analysis was based on a series of small islets in
inshore waters off Perth, Western Australia. He dis-
tinguished three terrain units: limestone, white cal-
careous sands, and red sands, and constructed
separate species–area relationships for each. The
values were then summed for the whole island. He
found a significantly better fit for his habitat-unit
model in predictions of actual species richness than
derived from whole-island regression (Buckley
1982). This is unsurprising given the increased
complexity of the analysis. He subsequently
deployed the approach in a study of 61 small islets
on bare hypersaline mudflats in tropical Australia,
consisting of 23 shell, 19 silt, and 19 mixed-composition
substrates (Buckley 1985). The flora comprised
three elements: ridge species, salt-flat species, and
mangrove species. In this second study, area was
found to be the primary determinant of total
species number. Moreover, elevation above mudflat
was a better predictor of plant species richness than
was substrate composition, probably through its
effect on the soil salinity profile (cf. Whitehead and
Jones 1969). Underlying these results, different
floristic elements were found to behave differently
as, not surprisingly, both island area and maximum
elevation above mudflats were more important
determinants of numbers of ridge species than of
salt-flat species.
A similar approach to the habitat-unit model has
been developed by Deshaye and Morrisset (1988).
Working on plants in a hemiarctic archipelago in
northern Quebec, they found that habitats worked
as passive samplers of their respective species
pools, but that at the whole-island level species
number was controlled by both area and habitat
diversity. They forcibly expressed the view that
data on habitat effects are crucial to interpretations
of insular species–area relationships (cf. Kohn and
Walsh 1994). In response to this need, Triantis et al.
(2003) have proposed a simple approach by which
to incorporate habitat diversity within species
number modelling for island data sets. They
introduce the term choros(K), apparently an
ancient Greek word that describes dimensional
space.Kis calculated by simply multiplying the
area of the island with the number of different
habitat types present. Species richness is then
expressed as a power function of the choros: i.e.
ScKz, which is of course directly analagous to the
more familiar ScAz. They compared the perform-
ance of both these models for 22 data sets of
varying taxa (e.g. lizards, birds, snails, butterflies,
beetles, plants), finding significantly improved fits
in all but two cases. They further suggested that
improvements in fit might be particularly evident
when dealing with smaller islands, and the
so-called ‘small-island effect’ (below). There are
two points to be made concerning this approach,
the first of which is the likely sensitivity of the out-
come to how habitats are defined (Triantis et al.
2005). Secondly, that the choros model includes an
additional parameter (the calculation is really
Sc(H*A)z(where His habitat number and Ais
area), and so an improvement in model fit is in
many respects unremarkable. The approach may
well prove useful in developing more powerful
fits to particular data sets, but the problems of
standardization of Kacross studies means that it
will be less easy to incorporate in comparative
analyses of multiple data sets.Area is not always the first variable in the model
Not all studies identify area as the primary variable.
Connor and McCoy (1979) speculate that non-
significant correlation coefficients between species
number and area are probably published less often
than discovered, because they may be perceived by90 SPECIES NUMBERS GAMES: THE MACROECOLOGY OF ISLAND BIOTAS