6.8. Meta Analysis

• Meta Analysis Data Preparation
• Meta Analysis Input Data Types
• Meta Analysis Summary of Effect Sizes
• Meta Analysis Output Options
• Meta Analysis Examples

Meta analysis is used to combine and analyse the results of several independent studies on a particular research topic. Often, different research results on the same topic are reported in terms of different statistical entities such as t-tests, correlations, risk ratios, etc. with different dispersion measures such as standard errors, p-values or confidence intervals. The aim of meta analysis is to translate these statistics back to a common measure (the effect size) and then to aggregate them taking into consideration the weight of each individual study.

Here we assume that a systematic review has been carried out and possible sources of bias are well understood. UNISTAT’s intuitive user interface allows you to select many different types of data and mix them as input for any meta analysis. This is a powerful feature that needs to be used with caution. The user should determine beforehand which types of data can be combined in a meta analysis in a meaningful way.

It is possible run one study removed (OSR), cumulative (CUM) or subgroup analyses. You can select one or both of fixed effect (with inverse variance (IV) or Mantel-Haenszel (MH) weights) or random effect models.

The output includes a summary table with confidence intervals, significance tests and a forest diagram, Cochran’s Q and I-square heterogeneity tests, Begg-Mazumdar Rank Correlation and Egger regression tests for publication bias and forest, funnel and precision plots.