BOX 3. STATISTICAL METHODS FOR COMPARATIVE ANALYSES OF LIFE HISTORIES
A set of measurable, correlated traits can be described using a smaller number of emergent, orthogonal variables representing dominant axes of life history variation, such as the fast-slow continuum. Principal component analysis (PCA) is one common statistical method to achieve this goal [3–5,7,9,26,27,29,30,33,34,74,75]. Although some -largely arbitrary- guidance exists regarding how many dimensions explain meaningful amounts of variation [48], it is less clear how to compare dominant axes across independent analyses with data sets comprising different traits, taxa, and sampling methodologies.
It is also possible to explore relationships between two different multivariate data sets (e.g ., climatic and life history data). Canonical Correlation Analysis (CCA) yields two sets of emergent uncorrelated variables by calculating axes with highest correlation between variable sets. This approach recognises that dominant axes of variation (e.g. , fast-slow) may not yield strong relationships with other ecological or evolutionary processes.
Unlike both PCA and CCA, Factor Analysis (FA) [47,76] treats measured variables as functions of latent variables, with associated measurement and/or residual error. As FA does not require latent variables to be uncorrelated, the approach offers solutions to some problems of PCA and CCA [47]. We suggest that FA is better suited to the testing of hypothetical, rather than data-driven, axes of variation.
Dimension-reduction analyses fit axes through data, rather than measure the multidimensional boundaries of life history variation. Cluster Analyses (CA) may help understand, after standardising for the species-specific ranking on the fast-slow continuum (by usinge.g. generation time), which and why certain life history strategies do not exist, as much as the clustered patterning of those that do. Hierarchical CA has previously been used to identify substructure in life history variation [77] and could prove a useful tool to apply much more widely in comparative life history theory [78–82].
Depending on the researcher’s perspective, a given environmental, phylogenetic, or morphological variable may drive life history variation, or be a nuisance covariate to deal with statistically [83]. Comparative analyses must account for the non-independence from shared evolutionary history [9,47]. Phylogenetic methods exist for PCA [83], but not yet for FA or CA, forcing researchers to use phylogenetically independent contrasts. Correlates of life history traits, particularly organism size, cause further challenges: including these correlates in analyses risks deriving life history axes defined by non-life-history traits. Moreover, using residuals in multivariate analyses drawn from regressions of life history against body size can introduce statistical biases [84].