Meta-analysis

During research, it is possible to come across different sets of publications that suggest different results for the same study, which implies there exists some level of error. Now, these conflicting outcomes are known as mixed results. But because everyone wants to find actual and reliable sources of information, meta-analysis helps us check how factual they are.

Some variations, however, could only be as a result of time difference-because we know most statistics are subject to time. Therefore, such differences are both statistically and philosophically acceptable.

Methods of performing meta-analysis

Even before delving deeper, it's good to note that during meta-analysis, we consider each research study to be a single data point. That is because we are analyzing what multiple research works have brought to the table.

There are two methods of meta-analysis based on the dependent variable in question.

• Combining raw effects: If all the research publications we are considering have the same dependent variable, we can pull them all together during the analysis.
• Using standardized effects: If the datasets have slightly (or completely) different dependent variables, we can use standardized effects.

So, what question does a meta-analysis seek to answer?

Meta-analysis goes beyond just giving a YES or NO answer. Researchers dig deeper to understand patterns, the actual sources of variation, and even question if it is possible to explain them theoretically. Furthermore, researchers can develop theoretical models that describe or correct these differences.

Throughout these processes, it is important to maintain accuracy because meta-analysis is all about obtaining robust findings to address the conflicting situations. And that brings us to the next concept-sensitivity analysis.

Sensitivity analysis

In most research designs, researchers make many assumptions that sometimes are quite arbitrary. For instance, if the research design requires us to choose a numerical age value between 40 and 50, there are ten options available, which can cause disparities in our eventual outcome.

Therefore, the goal of systematic analysis is to correct these unclear assumptions and to develop confidence the value chosen was actually the right one. It is performed after the meta-analysis, where we replace the arbitrary values with correct, desirable values. In a nutshell, it requires us to do a meta-analysis twice-first with all available options and then with what is confirmed correct.

Heterogeneity analysis in meta-analysis

Heterogeneity refers to the variations and inconsistencies we get from the different research outcomes. It could also be termed as ˮnon-compatibilityˮ because of their individual differences or variations.

There are three common techniques used to analyze heterogeneity based on how the variations are presented.

These are:

• Cochranʼs Q test—Itʼs based on chi-square distribution and uses probability to show the disparity between different research outcomes.
• I² statistic—it expresses the inconsistencies in terms of percentage rather than a chance, and it is independent of the number of research projects.
• Meta-regression—Explains the differences in terms of covariates.