Exploration of simulation results
A summary of the findings regarding the effects of different options to
handle missing SDs and/or SS in meta-analysis data sets are listed in
Table 2. As a general observation, the deviation introduced by the
omission of studies with missing SDs and/or SSs (with regard to fully
informed weighted analyses) mostly exceeded the deviation from all other
options to treat those missing data. Unweighted analysis yielded grand
means and confidence intervals similar to fully informed weighted
analyses except for the case of a correlation between effect sizes and
effect size precision. The same holds for the sample-size weighted
analysis. Imputing missing data introduced the least deviation in the
log response ratio dataset, followed by the correlation coefficient
dataset and the strongest deviation in the Hedges’ d dataset.
Missing SDs introduced larger deviations than missing SSs with regard to
fully informed weighted analyses. Imputing data missing not at random
(MNAR) in the Hedges’ d dataset lead to deviations that are
similar to those from the omission of studies with missing SDs and/or
SSs.
Compared to all other imputation methods, mean, median and random sample
imputation yielded the largest deviation in grand mean estimate and
Bayes predictive mean matching yielded the largest increase in the
confidence interval. Imputation via bootstrap expectation maximization
and additive regression and bootstrap predictive mean matching
frequently failed above a threshold of ca. 60% of missing data.