Measuring multivariate trait ‘syndromes’ to explain ecological outcomes
Organisms’ responses to their environment are governed by complex suites of correlated traits that confer important about performance under specific sets of environmental filters (Figure 1). Strong correlation among trait types that can recur across unrelated species—trait ‘syndromes’ or ‘typologies’—underpin trait-based community assembly theory for plants (Grime 1988) and behavioural syndromes in animals (Sih et al. 2004). Yet single trait-type studies make up roughly half of the research we reviewed (387 papers) compared with multi-trait (i.e. 3 or more traits) studies (176 papers, 21%), and we estimate that the true ratio of single vs. multi-trait studies to be more significantly skewed towards single-trait studies. Our review is likely conservative in assessing their prevalence because an unknown number of studies have likely been excluded from this review because they do not self-identify as trait-based investigations of ecological processes, and rather simply identify the particular trait investigated in relation to species distributions or abundances. In considering single traits at a time, as a function of a species’ or ecological community’s response to gradients of change, we risk overlooking the combined effect that a range of traits may have in explaining those responses to change (e.g. multivariate traits for restoration design to resist invasion; Funk et al. 2008, and plant trait typologies along ecological gradients; Diaz et al. 2004). While there has been a trend towards identifying a subset of traits that are strong indicators for ecosystem processes (e.g. Hausner et al. 2003), increasingly sophisticated statistical tools and computing power exist to deal with the greater model complexity that comes with multi-trait based analyses of ecological relationships. For example, multi-matrix modeling solutions enable the simultaneous assessment of relationships between species abundances and/or distribution data, environmental gradients of change and the role that traits play in mediating changes in biodiversity and assemblages in the face of environmental change (Dray et al. 2014, Wang et al. 2012, Brown et al. 2014). Comparatively simple techniques available in standard multivariate statistical tool kits and packages include parametric and non-parametric techniques for variable reduction and synthesis of trait combinations. Traditionally, techniques such as principal component analysis, non-metric dimensional scaling and clustering are most commonly used for identifying trait groupings and trait typologies, while systematically reducing variables included in models, and these remain powerful techniques fit for that purpose (Legendre & Rogers 1972, Legendre & Legendre 1998).