Supporting Information

Files in this Data Supplement:

Table 2. Studies included in metaanalysis with full citation, ecosystem type, the presence of factorial and nonfactorial data on evenness and richness, the study type (laboratory or field), and the number of independent studies obtained from each paper

Yin L, Cai Z, Zhong W (2006) Crop Protection 25:910-914.

Table 3: Correlation between the integrated productivity index and the log-transformed values for each of the six resource variables

SI Text

We defined a study as a temporally and spatially distinct experiment with treatments and corresponding controls. Multiple studies could be reported from within one publication if, for instance, the same experimental treatments were performed in multiple independent settings. When multiple measures were reported over time, we used the last temporal sample to avoid phases of transient dynamics. Exceptions were made if some unusual disturbance affected some or all of the treatments or replicates, which occurred for only three studies. In one fertilization study (Hobbs et al. 1988, see SI Table 4), the invasion of gophers destroyed the fertilization setup in the second year, so we took first year data. In a second study (Osem et al. 2002, see SI Table 4), the second to last year was used because the last year was extremely dry and not representative for the period. In one periphyton study (Hunter & Russell-Hunter 1983, see SI Table 4), data were the mean of the last three samplings. We used studies that manipulated the availability of resources or the presence of consumers or both. Almost all studies enriched mineral nutrients; three studies (Steinman et al. 1989, Wellnitz et al. 1996, and Hillebrand et al. 2004, see SI Table 4) manipulated light levels. Almost all studies manipulated herbivore presence directly; only one study used fishes to reduce herbivory by zooplankton (Moss et al. 2003, see SI Table 4).

Each study had a unique study identifier linked to the citation of the publication. We categorized the system as marine, terrestrial, or freshwater. Studies in wetlands and salt-marshes were operationally defined: studies addressing submersed or floating macrophytes or periphyton were classified as aquatic (marine or freshwater), whereas studies on above-water rooted plants were considered terrestrial. For each study site, we noted the latitude from -90 (S) to + 90 (N). We categorized the study type as factorial or nonfactorial. Main effects of herbivory and fertilization on producer richness and evenness did not differ significantly between factorial and nonfactorial experiments [group contrast metaanalysis (hereafter MA) on study type, P > 0.1], allowing us to analyze the different types of experiments and effect sizes together. We recorded three aspects of the temporal and spatial scale of the experiment: the duration of the experiment (days), and the experiment size, both as the replicate area used for fertilization manipulation (fertilization area), and the area over which consumers were manipulated (consumer area). To characterize the community structure of the plants, we also used the control (ungrazed, unfertilized) levels of evenness and ln-transformed species richness for each of the studies. Finally, treatment effects on plant biomass were calculated as ln response ratios of plant biomass to herbivory and fertilization for factorial and nonfactorial experiments in the same way as described in Methods.

We used resource availability as a proxy for ecosystem productivity. Because the studies differed in the resources they reported, it was not possible to use any single resource to characterize ecosystem productivity. Instead, we characterized productivity by calculating an integrated index of resource availability (for similar rationale, see additional refs. 1 and 2). In a first step, we retrieved as many concentrations of relevant resources as possible: We recorded the unmanipulated nutrient levels as available and/or total N as well as P (mmol g^-1 for terrestrial and mmol ml^-1 for aquatic studies). For terrestrial studies, we obtained a measure of water availability (precipitation in mm per year). We used the latitude and longitude of the experimental site (only field experiments) to obtain measures of ambient photosynthetically active radiation (PAR, W m^-2) for all field experiments (URL: http://www.ngdc.noaa.gov/seg/cdroms/ged_iib/datasets/b05/pl.htm; see additional refs. 3 and 4). In a second step, we calculated the database-wide mean for each of the six resource variables (light, water, available P, total P, available N, total N), and expressed each value as a proportion of that mean. Finally, integrated ecosystem productivity was then expressed as the ln-transformed average of all standardized resource values obtained from that study. This aggregated index was strongly linearly correlated to the ln-transformed values for each of the six nutrient variables and thus successfully summarized the productivity of the system (see SI Table 2).

We also distinguished field and lab studies, where lab studies also include outdoor mesocosms or streamside channels or other contained experiments, which lack direct contact with the original environment. The types of consumer manipulation included exclosures, enclosures, removal, gradient, multiple, and none. Exclosures are experiments that restrict herbivore access to plants, enclosures are experiments where stocked herbivores are kept with their forage, and removal experiments exclude herbivores by means other than physical boundaries, e.g., insecticides. A few studies used a natural gradient of herbivore presence/absence, or used a "multiple" setup, representing a combination of the manipulations mentioned above or a split-plot design. Neither the type of consumer manipulation nor the distinction between field and lab studies resulted in any significant difference (metaanalysis, P > 0.05) in average effect sizes. Therefore, we used finally six continuous variables for bivariate metaanalyses (MA) and the general linear model (GLM): latitude, productivity, treatment effects on biomass, producer community evenness or richness, experiment size, and experiment duration. The variables showed very moderate levels of collinearity (see SI Table 3). Only experiment size and duration were strongly correlated (r = 0.878), others moderately (r < 0.5).

1. Hillebrand H (2005) J Ecol 93:758-769.

2. Hillebrand H (2002) J N Am Bentholog Soc 21:349-369.

2. Pinker RT, Laszlo I (1992) J Appl Meteorol 31:194-211.

3. Pinker RT, Laszlo I (1992) J Climate 5:56-65.