Suitable data were available from 46 studies, measuring 362 individual traits across 150 species that were assigned into 75 species pairs (see Appendix S1). Within a given study, one or more pairs consisting of an invasive and non-invasive species were generated to maximize the phylogenetic relatedness within pairs. It was not possible to pair species across studies because the environmental conditions that were manipulated, and the traits that were measured differed greatly among studies. The pairing of species allowed us to investigate differences in plasticity between species that can more readily be attributed to invasiveness status because we have partially removed variation in phenotypic plasticity that is due to systematic differences among studies (e.g. methodology, exact levels of treatments). A database of all species pairs was created.
To construct the database of all species pairs, we first categorized the growing conditions into eight ‘resource treatment’ types based on analysis of the most common types used in the available studies. These were manipulation of: (1) nutrients (including different elements, e.g. nitrogen, phosphorous, etc.), (2) light, (3) water, (4) competition or density, (5) disturbance, (6) CO2 enrichment, (7) presence/absence of climbing substrate and (8) presence/absence of soil biota and/or mycorrhiza. Second, we categorized the response traits for plasticity measures into 11 categories: (1) water use efficiency (WUE), (2) photosynthetic rate, (3) biomass, (4) relative growth rate, (5 and 6) tissue nitrogen or phosphorous content, (7) root biomass, (8) shoot length, (9) specific leaf area (SLA), (10) root to shoot ratio (R:S) and (11) nitrogen use efficiency (NUE). Third, we noted the extent of phylogenetic relatedness between the pair of non-invasive and invasive species, which we defined as: (1) congeners, (2) confamilial or (3) less closely related. Fourth, we noted whether the growth form of the invasive was: (1) herb, (2) grass, (3) shrub or (4) tree or (5) vine. And fifth, we divided the invaded habitat into eight types: (1) grassland, (2) forest, (3) scrub-, shrub- or heath-land, (4) tropical forest, (5) wetland, (6) desert, (7) tree-shrub mix and (8) disturbed land or agriculture.
As described above, if we had data on trait plasticity for more than one invasive and one non-invasive species in a given study, species were paired to provide the closest phylogenetic matches between the invasive and non-invasive species. In each study, a species was only represented in a single species pair. Of the 46 available studies, five had two species pairs and eight had more than two species pairs. We treated species pairs as independent data points for the purposes of most analyses. In so doing, we make the reasonable assumption that there is no systematic variation among studies in the likelihood that they will report greater plasticity in invasive than non-invasive species. Biologically, this assumption is well justified because plants of different species were grown separately and were not in competition. The only exception is seven field-based studies in which natural competition occurred.
Likewise, when two or more types of environmental manipulations were imposed on the same species pair, we calculated separate effect sizes for each resource treatment. Of the 46 studies, 14 studies measured responses to two treatments and 10 studies to more than two treatments. Finally, we calculated separate effect sizes for each response trait for each species pair for each treatment type. Thirteen studies recorded two response traits per species pair per treatment and 23 recorded three or more response traits.
Plasticity was not explicitly measured or specifically reported in the majority of the studies used in our meta-analysis as the traits were originally measured for other purposes. This has the advantage of avoiding any publication bias directly associated with our main hypothesis (i.e. towards only publishing positive results). This should reduce any associated ‘file drawer’ problems (Rosenthal 1979). It does not, however, avoid the issue of a ‘research bias’ whereby data are more often collected from certain species (Gurevitch & Hedges 1999). For example, the available species pairs might over-represent commercially important invasive species (although inspection of our species list does not support this claim) or be biased towards species which are more easily grown in greenhouse experiments (e.g. short-lived herbs and grasses). As with most research fields, these caveats about the availability of data in the current literature should inform interpretation of our meta-analyses (Jennions et al. in press).
The available studies rarely used clones or full-siblings, so genotypes per se were imperfectly replicated across experimental environments. Phenotypic difference between treatments could therefore be due to both genetic differences in plants assigned to each treatment and phenotypic plasticity. If, however, plants from a given population (or species) express consistently different phenotypes in the different environments, phenotypic plasticity can still be analysed. Clearly, the resultant plasticity estimates will have greater uncertainty than those based on measurement of replicated genotypes (see Funk 2008). On the other hand, however, the use of only a few genotypes could result in a poor sample of the available mean level of plasticity within a species if there are moderate to high levels of genotypic variation in phenotypic plasticity within a species. Perhaps most importantly, however, there are no obvious bias in how individual plants were assigned to growing treatments (i.e. no propensity to assign certain genotypes to specific treatments), so there should be no systematic bias in the resultant measure of the difference in plasticity.