Evidence of publication bias
More negative effects of stressors in studies with lower precision
suggest that publication bias may partially explain our results for both
Q1 and Q2 (Fig. 4). We confirmed these negative relationships between
effect size and precision using a two-step modified Egger’s test (Table
S8). We thus adjusted meta-analytic estimates for analyses in Q1 and Q2
by including variance as an additional moderator in both models.
Some results in Q1 differed qualitatively after adjusting for small
study effects. Specifically, effects of endogenous environmental
stressors and pollution became non-significant when variance was
included as a moderator (Table S9; Fig. S9). Moreover, the effect of
resource limitation on survivorship changed direction after the
small-study adjustment. However, we note that this effect was
indistinguishable from zero in both unadjusted and adjusted models and
was based on few studies (n = 8).
In Q2, our qualitative results remained essentially unchanged after
adjusting for publication bias. Overall, effects of endogenous
environmental stressors reduced host survival and increased both
infectivity traits (Table S10; Fig. S10). As in our primary analysis,
resource limitation in the adjusted model negatively affected fecundity,
but the meta-analytic effect on intensity was marginally non-significant
(Table S10). Finally, the negative impact of chemical pollution on host
survival and prevalence in our primary analysis (Table S5; Fig. 3)
became indistinguishable from zero in the adjusted model (Table S10;
Fig. S10). However, we caution that this result was based on a
relatively small number of experiments (n = 9).