Overview
Meta Analysis is the analysis of a set of similar statistical studies in order to compile an updated set of statistics or to explain the differences in the results of the studies.
Effects
The primary thing that a meta analysis studies is termed an effect. This comes from studies that attempt to show the effect of a drug, or other intervention, in the course of disease or other health issue. In order to be able to claim that the intervention had an effect, the study is partitioned into two groups, a group with the intervention and a control group without the intervention.
The word effect is used to describe situations in which the thing being measured wouldnt normally be called an effect. For instance, a study could study the average difference in heights between men and women, and this difference would be termed an effect. This is done in order to keep terminology consistent when doing a meta analysis.
Regardless of the terminology, most studies try to measure the relationship of two random variables. For example, the health impact of the group receiving the intervention versus the health of the control group. Or, the height of men versus women in the second example.
In most examples, the studies under question will have been compiled in order to test a hypothesis. for example, the hypothesis that the intervention leads to better health outcomes, or that on average, men are taller than women.
Typical Effects Measures
Studies generally report their outcomes in some type of aggregated measure. The typical types of aggregation are listed here.
- Mean and Standard Deviation (see moments) for continuous outcomes
- Odds/Risk ratios for binary outcomes
The goal of the meta analysis is to take the reported statistics and produce a statistic that incorporates those from each study. Additionally, if the difference in reported statistics is larger than would be expected from randomness in each study, the analyst may try to explain the difference due to some heterogeniety in the studies.