Appendix S2: Consideration of categorical and continuous covariates.
Across different dominant plant community types, fertilization and herbivore exclusion increased
community biomass and there were no interactive effects, although relative response strengths
varied within and among producer community types (Supplementary Fig. 1A). Across systems,
producer community type explained some variation in LRRF (1-way ANOVA on 5 types, F4,186 =
3.23, p = 0.013) and LRRH (F4,186 = 3.79, p = 0.005), but not LRRI (F4,186 = 0.31, p = 0.868).
Multiple comparisons of all pairwise combinations of producer community types demonstrated
that phytoplankton and periphyton differed in their responses to herbivores (Tukey’s HSD
adjusted p = 0.005), and phytoplankton differed from macroalgae in their response to
fertilization (adjusted p = 0.035).
Most studies (n = 159) manipulated invertebrate herbivores, although 16 studies
manipulated vertebrates primarily, and 16 others removed or excluded both vertebrates and
invertebrates (Supplementary Fig. 1B). The average LRRH for invertebrates was significantly
positive (LRRH = 0.87, CI = 0.66 to 1.11), but studies including vertebrate herbivores showed
weaker and nonsignficant effect sizes, but these were highly variable and overlapping with those
from invertebrates alone (vert. only: LRRH = 0.43, CI = -0.12 to 1.04; both: LRRH = 0.64, CI = 0.05 to 1.42). The non-significant overall result for terrestrial LRRH was not explained by
whether vertebrates or invertebrates were the target of manipulations (vert. [n = 9]: LRRH = 0.30,
CI = -0.23 to 1.0; invert. [n = 6]: LRRH = 0.18, CI = -0.59 to 0.82).
Studies using experimental enclosures or exclosures did not differ from other habitat
trends (Supplementary Fig. 1C), but in removal experiments (largely by chemical pesticides),
fertilization effect sizes were larger (LLRF = 2.17, CI = 1.59 to 2.77) and the interaction effect
size was signficantly negative (LRRI = -0.18, CI = -0.36 to -0.02). In comparisons among field (n
= 157) and lab (n = 34) experiments (Supplementary Fig. 1D), LRRF and LRRH trended towards
larger effect sizes in the lab, but LRRI did not deviate from zero in either case.
LRRs were invariant as a function of latitude (absolute value), with the lone exception
that LRRH showed stronger effects with increasing latitude within freshwater systems
(Supplementary Fig. 2A; LRRH = -0.06 + 0.022x; p = 0.012, df = 114, adj. R2 = 0.046).
Moreover, there were no linear correlations between LRR effect sizes and the area of
experimental plots, within or across systems (Supplementary Fig. 2B).
However, the overall fertilization effect across all studies declined significantly with logtransformed experimental duration in days (Supplementary Fig. 2C; LRRF = 1.76 - 0.159x; p =
0.0049, df = 188, adj. R2 = 0.036). This effect was confounded by the differing characteristic
time scales used to capture autotroph dynamics in different systems and habitats. Typical
terrestrial studies ran for multiple years (mean ± SE duration in days, 993.8 ± 255.7), freshwater
studies averaged less than one month (27.7 ± 6.6), and marine studies were intermediate (81.1 ±
17.5; 1-way ANOVA: F2,187 = 83.245, p << 0.0001). Analyzed within systems, herbivore effects
increased with experimental duration in freshwater habitats (LRRH = 0.087 + 0.34x; p = 0.0001,
df = 113, adj. R2 = 0.115), and the interaction effect showed a modest trend from net positive to
net negative with increasing duration in marine studies (Supplementary Fig. 2; LRRI = 0.56 0.109x; p = 0.058, df = 58, adj. R2 = 0.058). These results, in turn, may be explained by varying
experimental durations in habitats within systems. Within the freshwater subset, lake studies
concentrating on pelagic planktonic interactions ran for much shorter durations (4.3 ± 1.0) than
benthic studies focused on periphyton or macrophytes in lakes (68.3 ± 31.7) or streams (44.4 ±
6.0; 1-way ANOVA: F2,113 = 186.53, p << 0.0001). Analysis of covariance, with habitat
classification included as a fixed factor with three levels, eliminates the covariate, ln-transformed
study duration, as explanatory for LRRH (AIC, duration only = 379.38; AIC, duration plus
habitat factor = 377.42, duration n.s.). However, study durations across habitat types in marine
systems were not significantly different (Rocky reef: 133.1 ± 51.4; Coral: 67.8 ± 15.4; Soft
bottom: 52.2 ± 8.8; 1-way ANOVA: F2,56 = 0.205, p = 0. 0.815), and this factor did not explain
the weak negative correlation of study duration with LRRI.
LRRs showed no relationship with ambient total or available nitrogen (analyses restricted
to marine and freshwater systems, Supplementary Fig. 3A). However, LRRI was weakly,
negatively related to standardized phosphorus availability within aquatic systems overall (LRRI
= 0.21 - 0.15x, df = 123, p = 0.007, adj. R2 = 0.05), and freshwater (LRRI = 0.17 - 0.16x, df = 85,
p < 0.05, adj. R2 = 0.035) and marine (LRRI = 0.32 - 0.13x, df = 36, p < 0.07, adj. R2 = 0.063)
systems individually (Supplementary Fig. 3B).
Figure S1 LRRX of fertilization (white squares), herbivore absence (black triangles) and their interaction (grey squares) on autotrophs
as a function of: (A) dominant producer community type (phytoplankton, periphyton, macroalgae, herbaceous, woody); (B) consumer
type (invertebrates, vertebrates, or both); (B) experimental method (enclosures, exclosures, or chemical/mechanical removal); and (C)
experimental venue (field or lab). Sample sizes for each category are given across the top panel. An LRR is statistically significant
when the boot-strapped 95% confidence intervals do not overlap the dashed line of zero effect, and is statistically distinct from other
LRRs when 95% CIs do not overlap.
Figure S2 Producer log response ratio (LRR) effect sizes for fertilization (LRRF), herbivore absence (LRRH), and their interaction
(LRRI) as a function of (A) latitude (absolute value), (B) area of herbivore manipulation replicates (m2, ln-transformed), (C) study
duration in days (ln-transformed). Symbols are plotted by system (freshwater: open circles; marine: grey triangles; terrestrial: black
squares). Significant system-level linear regressions are plotted and labeled by system and significance level (*0.05 > p > 0.01; **
0.01 > p > 0.001; *** 0.001 > p).
Figure S3 Producer log response ratio (LRR) effect sizes for fertilization (LRRF), herbivore absence (LRRH), and their interaction
(LRRI) as a function of (A) available N and (B) available P. Nutrient variables were standardized within systems by dividing by
systems means. Symbols are plotted by system (freshwater: open circles; marine: grey triangles; terrestrial: black squares). Linear
regressions of P availability with LRRI were statistically significant overall (R2 = 0.06, df = 123, p = 0.006) and marginally significant
within marine (R2 = 0.08, df = 36, p = 0.069) and freshwater (R2 = 0.05, df = 85, p = 0.046) systems individually.