Results
A total of 290 experiments could be fitted with both piecewise regression and exponential decay models (Fig. S4). The remaining 18 could not be fitted with piecewise regression, which most frequently occurred because the minimum value of D (i.e. complete acclimation) was reached at the second observation, in which case multiple breakpoints can yield an identical fit. Of these 290 experiments, comparisons of residual standard errors from models showed that the data in 62% (n=180) of them were best explained by exponential decay models. Furthermore, superior fit by piecewise regressions (N = 110 experiments) was observed primarily in experiments where thermal tolerance was measured at relatively few time points following exposure of the study organisms to the new temperature (Fig. S5). Thus, for more comprehensive experiments (i.e. greater number of measurement points in time), the shape of the plasticity rate response is better described by an exponential decay function. Thus, in all further analyses we use the estimates of λE (Table S2).
For the data included in the statistical analyses presented here, species-specific mean estimated λE was 0.0342 h-1. The variation in estimated meanλE across species was considerable, with SD = 0.0305 (i.e. CV = 89%). Species-specific mean λEranged from 0.0009 to 0.1892 h-1. Variation inλE was best explained by a model that included taxonomic class, acclimation temperature, and the slope of the estimated exponential decay function at tn , while evidence for an effect of body mass was weak (Table 1). Model coefficients of the best ranked model show that plasticity rates were highest among amphibians and reptiles, lowest among fishes and crustaceans, and intermediate in insects (Table 2). This pattern was also evident when examining distributions of λE among these classes without correcting for the covariates fitted in the best ranked model (Fig. 1). We also observed that plasticity rates increased with acclimation temperature (Table 1, 2, Fig. 2). Finally, plasticity rates were observed to be higher when the slope of the estimated exponential decay function at tn was shallower (Table 1). In other words, experiments where complete acclimation was more likely to have been obtained were associated with higher estimated plasticity rates. This pattern was mainly driven by the amphibian data, which had a large number of experiments with relatively steep slopes attn (Fig. S6). This is in the opposite direction to what might be predicted if a relationship between this slope and estimated acclimation rate is a statistical artefact (Fig. S1). Rather, the causality of this relationship is likely in the opposite direction, i.e. experiments that had a low acclimation rate were stoppedbefore complete acclimation to the new temperature had occurred. Nevertheless, to evaluate if this anomaly could have influenced our results, we repeated the model comparison summarized in Table 2 but including only experiments where the slope of the estimated exponential decay function at tn was larger than -0.002 and -0.001, respectively. For both cases, results remained qualitatively identical and showed only minor quantitative changes (Table S3 - S6).