Main Effect and Publication Bias
We analyzed effect size data for each of the three categories of our
data (prevalence, intensity, parasitoid) according to the following
scheme. First, we fit a random effects model (REM) to estimate the
overall effect of predators on parasites in prey. We report the size and
direction of the overall effect as well as I2, a
measure of heterogeneity that can be interpreted as the proportion of
total variation that is due to between study variation
(Higgins & Thompson
2002). We also used these models to diagnose publication bias in the
data by visualizing the relationship between effect size and variance
with a funnel plot and testing for a significant correlation between
these traits using a rank-order correlation test. If significant
correlation was detected, we used the trim-and-fill method
(Duval & Tweedie
2000) to determine whether introduction of studies to balance the
diagnosed bias would alter the main effect.