Effect size calculations
Even though we have used a standardized protocol, the sampling was conducted by several researchers in different biomes. Thus, we used a meta-analysis (meta-regression) approach to control for biases in sampling (Gurevitch, 2013). The magnitude of the leaf shelter effect on arthropod community structure (measured as abundance, species richness, biomass and body size) was calculated at the site level (n = 52 sites) using Hedges’ d (standardized mean difference; Rosenberg et al ., 2013) following the formula:
\(Hedges^{\prime}\ d\ =\ \frac{Xe-Xc}{\text{SDpooled}}\ \),
where Xe and Xc are the mean community measures for plants that had their leaves rolled and control plants, respectively, andSDpooled is the pooled standard deviation of the two groups. Positive effects indicate that leaf rolls had higher arthropod species abundance, richness, biomass or mean body size relative to the control leaves. Values around 0.2, 0.5 and 0.8 are considered small, medium and large effect sizes, respectively (Rosenberg et al., 2013). Larger effect size values mean larger differences in occupation of rolled compared to control leaves. Although we initially classified the surveyed arthropods into different feeding guilds (i.e., predators, herbivores, detritivores, omnivores and parasitoids), the sample size was only sufficiently large for separate analyses of predators and herbivores.