New in β 3.0:
The most exciting new feature is the ability to edit forest plots after they have been generated. For details, see the help. We have also introduced automated subgroup analyses. Other updates include the ability to paste data directly from Excel, improved data importing, custom labels (columns) for inclusion in forest plots, as well as myriad small bug fixes.
β 2.0 release
The biggest two updates to this version from the last are: 1) exhaustive testing of many of the available methods (more on this below) and 2) integration of the BUGS library for Bayesian meta-analyses. Additionally, here is a selected list of new features/updates added since the last (&beta 1) release:
- Import data from Microsoft Excel and .csv files
- New outcome metrics available in continuous analyses: Cohen's D and Glass Delta
- Forest plots scaling improved, also introduced arrows for (outlying) very large confidence intervals
- Leave-one-out forest plots on log scale when appropriate
- Rename graphics titles and axis labels via right-click menu
- Lots of miscellaneous small bug fixes, UI quirks, resolved
- We now have a document outlining the statistical methods used in Meta-Analyst.
Testing
Since the last release, we have completed a thorough testing of many of the statistical routines used in MetaAnalyst. In particular, we verified our output over roughly 12,000 continuous meta-analyses and 7,000 binary data meta-analsyes (using all metrics and methods) against Stata (using metan) for the 'simple' (i.e. non-Bayesian) routines. The following tables summarize the datatypes and models currently available in MetaAnalyst. Further, the Testing Level column indicates how thoroughly the respective methods have been tested. This is a somewhat subjective scale, but roughly:
- Gamma (Γ) tested: These methods have been thoroughly tested; thousands of meta-analyses have been run and verified against Stata.
- Beta (β): This means MetaAnalyst is using BUGS to conduct this analysis. BUGS is itself a mature, tested program, and so we are relatively confident in these results.
- Alpha (Α): Still 'experimental'. Any help testing these methods (i.e., making sure the output is correct) would be very much appreciated.
| Model | Testing Level |
|---|---|
| Fixed (Peto, Inverse Variance, Mantel-Haenszel) | |
| Random (DerSimonian-Laird) | |
| Expectation Maximization | |
| Bayes (Linear, Quadratic and 'None' control rates) |
| Model | Testing Level |
|---|---|
| Fixed | |
| Random (DerSimonian-Laird) | |
| Expectation Maximization | |
| Bayes (Linear and 'None' control rates) |
| Model | Testing Level |
|---|---|
| Fixed | |
| Random (DerSimonian-Laird) | |
| Expectation Maximization | |
| Fixed SROC | |
| Random SROC | |
| Bivariate | |
| Bayes |