Visualisation of the simulation results
In Figures 3-6 we show the results of treating increasing proportions of
missing SDs and/or SSs on the grand mean of the three investigated
effect sizes (log response ratio, Hedges’ d and Fisher’sz ). The different figures 3-6 correspond to the four
deletion/correlation scenarios (MCAR, MAR, MNAR, corMCAR) and are
similarly organized in the style of a row-by-column matrix. The 14 rows
correspond to the 14 options to treat missing data (labelled on the
right and described in Table 1). The seven columns correspond to the
three effect sizes (log response ratio, Hedges’ d or Fisher’sz ) and the type of data deleted (only SDs, only SSs or both).
Fisher’s z was weighted by SSs alone and thus only those could be
deleted. Within each figure, every cell corresponds to a specific
combination of the effect size investigated, the type of data deleted
and the treatment applied. Within each cell, we show how an increasing
proportion of deleted data (from 10% at the top to 90% at the bottom)
leads to deviations of the grand mean (solid coloured line) and one side
(i.e. 50%) of its corresponding confidence interval (dotted lines) from
the estimates of a fully informed weighted meta-analysis (solid black
line). Coloured lines that match the solid black line indicate that the
respective treatment of missing data leads to grand means and confidence
intervals that strongly resemble those from fully informed weighted
meta-analyses.