activities that are most crucial for performance and the
long-term maintenance of the species.
Similarly, research demands a judgment about which
functional traits are most decisive in competition and the
interaction milieu, and about which physical gradients
those traits interact with most. These functional traits
might be the same as those that decide fundamental niche,
depending on competition in the particular community
(Box 1). In a distinct preference niche organization, the
most predictive traits will be those that relate to resource
acquisition along the niche dimensions (e.g. beak size and
shape). However, in a shared preference niche organization,
traits related to competitive dominance ranking,
such as height and aggression, will often be
most informative.
This process of prioritization produces several questions.
For a given system (Box 1), which performance
currencies are most predictive of long-term success? For
these performance currencies, which traits and environmental
variables most affect performance? Which traits
modify the effects of the interaction milieu the most? To
prioritize factors is to hypothesize about which fundamental
and realized niche processes are most important in
a given system. Progress of the research program can be
thought of as testing and improving those hypotheses by
revising or confirming the ranking list. These hypotheses
can be objectively tested using measures of predictive
power, such as r2 (Box 2).
An additional approach to dealing with the many
potential factors to explore depends on the fact that,
often, many separate traits or environmental variables
can be reduced to one or two axes of variation that capture
a large proportion of the original variation in traits,
because of allocation and life-history tradeoffs. For
example, a study [14] of O2500 plant species at 175 sites
showed that a single axis of variation explained almost
75% of the variation in six leaf-trait dimensions
(Figure 1e). Between prioritization and reduction, we are
optimistic that the many potential factors can be reduced
to a manageable level.
Contrast with other approaches
How does the functional trait approach compare with
other community ecology approaches? Three common
approaches (mainstream empirical studies of species
interactions, community-matrices and neutral theory)
stand in strong contrast to what we propose.
Past empirical studies of species interactions covers a
range of work, some of which fits well into the functional
trait research program [5,27,30,38,40–42]. However, we
suggest that much of this work differs from the functional
trait-based approach in subtle but important ways. First,
many empirical studies have traditionally been nomenclatural
in nature, studying typically two–four species
with no measurement of the traits that distinguish them.
Second, when studies are conducted at multiple sites, the
variation among sites has usually been treated as noise or,
more recently, as an example of ‘historical contingency’
with little effort to find systematic variation in performance
and outcome as a function of explicit environmental
factors, such as temperature, that vary between sites.
Finally, most empirical studies rely on ANOVA with a
focus on statistical significance, whereas we emphasize
predictive power measured by effect sizes and r2 (Box 2)
Meanwhile, the dominant paradigm over the past 40
years for theoretical ecologists is based on population
dynamics built up into community matrices (paralleling
the emphasis on pairwise species interactions). This
approach is sensible, but, with hindsight, population
dynamics are most successful in modeling only one- or
few-species systems. Larger numbers of species lead to
modeling challenges, such as chaotic dynamics and
difficult parameterization [20].
Population dynamics could move in directions that
incorporate our four themes [7,24], but has generally
moved in the opposite direction. Thus, the most-discussed
population-dynamic community model of the past few
years, neutral theory [43–45], is explicitly predicated on
an assumption that the differences in traits between
species and the environmental variation along gradients
have no effect on population dynamics, diametrically
opposite to what we propose.
Two less-dominant paradigms (community resource
models and macroecology) have some overlap with the
functional trait approach. Consumer resource models
(CRM) [1,12,46–49] avoid the proliferation of pairwise
interaction coefficients and approach a ‘milieu’ concept of
competition. They use measurable numbers (the slope of
impact vectors and R*, defined as the external resource
concentration level below which a species cannot sustain
its population size) to predict community structure.
Recent CRM work [12] emphasizes gradients and niches
as central organizing principles. Specifically, CRM
explores gradients in nutrient availability, along which
R* delimits the fundamental niche, while the impact
vector converts this to and delimits the realized niche. As
such, CRM can be thought of as a specific example of the
functional trait research program (Box 1), hypothesizing
that resource uptake and survival at low resource levels
are the most important factors in the system; one can test
for which systems these prioritization hypotheses are true
[47]. It is not clear how CRM would be applied to gradients
of nonconsumable factors, such as temperature, that we
believe to be important (but see [12]). Thus, CRM is
limited to systems in which its mechanism of resource
competition dominates [47], whereas the functional traits
approach lays out a more general research program. The
two approaches also differ on what types of variable
should be measured. R* functions like our performance
currency, but is measured differently; it is a highly
abstract, integrative measure incorporating the state of
an environmental variable at which a population dynamic
measure reaches a particular level (growth rate is zero).
By contrast, we believe that traits, performance currencies
and the environment are three distinct factors that
should be measured independently, using concrete morphological
and physiological features of individuals for the
first two. Work has been done on how traits (sensu our
definition) such as root allocation correlate with R* [12].
We suggest that our approach will be more operational
and predictive and will apply to a broader range of
systems, but hope that this will be evaluated empirically.