What is MetaWin?


MetaWin 2.0 will allow one to summarize the results of multiple independent studies using meta-analytic procedures. Version 2.0 is much more general than its predecessor, and allows for greater flexibility in both the effect sizes that can be used as well as the statistical models for summarizing meta-analytic data. This version can calculate both fixed effects models and random effects models, and can be used for a variety of meta-analytic data structures, including no data structure, categorical (grouped) data, and continuous (regression) data. MetaWin now comes with its own spreadsheet; data can be entered directly into the spreadsheet or can be read from a text file, a Microsoft Excel file, or a Lotus 1-2-3 file. A variety of commonly used meta-analysis effect sizes can be calculated, including Hedges' d, response ratio, odds ratio, risk difference, relative risk, and Fisher's z- transform. From these (or other) effect sizes, cumulative mean effects, their confidence intervals, and various heterogeneity statistics can be calculated. The total heterogeneity, QT, is calculated for each analysis, and, for categorical and continuous data models, the heterogeneity explained by the model, QM, and the residual error heterogeneity, QE, are calculated as well.

MetaWin 2.0 also allows one to refine the analysis by removing certain studies or groups of studies from the analysis without having to alter the data file. A variety of exploratory data analyses can be performed, including various tests to evaluate potential publication bias. Cumulative meta-analyses can also be performed, in order to investigate changes in the cumulative mean effect size as new studies are added to the model. In addition, one can graphically explore the data through histograms, normal quantile plots and funnel plots. Scatter plots, regression plots, radial plots, and plots of cumulative mean effect sizes can also be generated.

Finally, because data may violate the underlying assumptions of meta-analysis, it may be useful to evaluate the significance of meta-analytic statistics using resampling methods (Adams et al., 1997). Therefore, this program will allow one to incorporate resampling tests into the meta-analysis. In particular, confidence intervals for cumulative mean effect sizes can be generated using two different bootstrap procedures (bootstrap confidence intervals and bias-corrected bootstrap confidence intervals). MetaWin also allows one to test the significance of the heterogeneity explained by the model, QM, using a randomization test.


Revised by Dean C. Adams, dcadams@iastate.edu, 7 December, 1999.