stephankambach / simulatemissingdatainmeta-analyses

Meta-analyses often struggle with missing standard deviation and/or sample sizes in original publications, the following script simulates missingness and correlation patterns in artificial meta-analysis data sets and explores the performance of 14 approaches to treat those missing information

SimulateMissingDataInMeta-Analyses

Ecological meta-analyses often encounter incompletely reported studies that have missing variance measures (e.g. standard deviations, SDs) or sample sizes (SSs) that are necessary for incorporating the study in a weighted meta-analysis. Here, we evaluated 14 different options to treat missing information in simulated data sets of known structures and between 10% and 90% of missing standard deviations and/or sample sizes.

Script 1 (Mainscript)

• creates meta-analysis data sets (group comparisons and correlation analyses) with varying size, range of SD and SS values and correlation structures
• deletes between 10 and 90% of SDs and/or SSs completely at random (MCAR), at random (MAR) or not at random (MNAR)
• applies 14 approaches to treat/impute those deleted values
• saves the obtained grand mean estimates in a folder that must be named "intermediate output"

Script 2(additional functions2 - bias corrected logRR) contains the functions used in script 1

Script 3 (Plotting of results2 - bias corrected logRR) creates figures

• for the literature review
• for comparing the grand mean and confidence intervals obtained from the treatment of missing data with the grand mean and CIs of the complete data sets