Systematic literature survey
In the compiled dataset of 505 published ecological meta-analyses, 35%
used log response ratios, 36% used standardized mean differences, 24%
used correlation coefficients and 5% used a combination of the three
investigated effect size measures. At least 64 % of the investigated
ecological meta-analyses encountered missing variance measures or sample
sizes in the primary literature (Fig. 1). Most often, the exact number
of incompletely reported primary studies was not stated. Forty-five
percent of meta-analyses just noted that they included only completely
reported primary studies. The highest percentage of missing data was
reported for those studies that summarized response ratios. For 10% of
the studies, we could not determine whether they were affected by
missing data. Most studies simply omitted incompletely reported studies
from their analyses (complete-case analysis). A minor fraction of
imputed missing data and only two percent of the reviewed meta-analyses
(9 out of 505) applied multiple imputations or Bayesian models to
account for imputation uncertainty. The proportion of meta-analyses that
omitted incompletely reported studies versus those that imputed missing
data did not change with the publication year (Fig. 1).