Effects of Moderators
Given the level of variation in the effect of predators on parasites in
prey we were interested in identifying attributes of the study or study
system that were most important for explaining variation in effect sizes
across studies. To do this, we fit mixed effects models (MEM) to the
prevalence and intensity effect size data sets, including a series of
moderators:predator effect type manipulated (non-consumptive vs. all
interaction types), predator-spreader identity (identified as
predator-spreader or not), and parasite type (macro vs. micro) and all
two-way interactions. Study design (experimental vs. observational) was
also included as a moderator to control for variation in responses but
without a particular hypothesis. We also included study as a random
effect. We note that while we were interested in the distinction between
non-consumptive and consumptive effects, most consumptive effect studies
technically allowed for both non-consumptive and consumptive
interactions due to limitations in experimental design. Therefore we
draw the distinction between studies which manipulate only
non-consumptive interactions and those which include consumptive
interactions (all interaction types). From this model, we generated
candidate sets of all possible MEMs for each data set and used the
Akaike information criterion corrected for sample size (AICc) to compare
model fit. We calculated the importance (on a scale from 0 to 1) of each
moderator as the summed model weights for all MEMs in which a given
moderator occurred. We then fit univariate models for each moderator to
identify the direction of the effect. When reporting the results of
univariate models for the most important variables, we provided the
direction of the effect of the moderator and the results of a test for
residual variation. Because parasitoid studies were uniformly
terrestrial, experimental, and consumptive, we did not fit MEMs with
moderators to these data.