Effect Sizes
Prior to calculating the effect size for each case, variances that were
not already expressed as a standard deviation, were transformed. For
cases where means were zero, variances were assumed to also be zero when
missing. In two publications, non-zero (< 0.1) standard
deviations had been rounded to zero, these were set to 0.01 (in one
publication the abundances were < 0.1 individuals
m2x103, and in the other publication
abundances were between 0 and 50 individuals). Cases that were missing
non-zero variances in either the control or treatment were removed from
the analysis, as standard deviations are needed to calculate the effect
size used.
For each case in the dataset, Hedges’ g , a standardised mean
difference, was calculated using the metafor package . A standardised
mean difference is required when different cases are on different scales
, and this is a common effect size calculated in ecological
meta-analyses . In addition, Hedges’ g is appropriate when zeros are
present within either the control or treatment means (n = 127 zeros in
control mean, and n = 159 zeros in treatment mean) .
Cases that measured anything other than taxonomic (species, genus)
richness, biomass, abundance, or Shannon diversity, were removed, to
ensure adequate data across the four metrics. Richness, biomass,
abundance, and Shannon diversity accounted for 95.78% of the cases.
Four effect sizes were removed from the database due to being extreme
outliers of the dataset. The median effect size of the full dataset was
-0.11 and the median variance was 0.42. Three of the removed data points
had effect sizes smaller than -50 (ranging from -95 to -233), with
variances > 400 (ranging from 455 to 4538). The fourth
datapoint that we removed had an effect size of 184, and variance of
568. Removal of the four data points had no impact on results. All four
outliers were caused by relatively extreme changes in abundances and
richness between the control and treatment.