Hello friends, how many paper should include in a meta analysis??
How it is different from systematic review??
the more, the better; but it is allowed to be at least 2 studies for the bare minimum.
Responding to your first query the number of studies to be included depends on the heterogeneity of the study type you choose, there is no clear cut rule of thumb that can be relied on for example, in a study by Davey et al*. a sample of 22453 meta-analyses, show that the number of studies in a meta-analysis is often relatively small, with a median of 3 studies (Q1-Q3: 2–6), and only 1% of meta-analyses containing 28 studies or more. To compensate the heterogeneity of the studies rigorous review of the methods of the studies should be conducted and to reduce errors statistical methods such as DerSimonian and Laird approach and Hartung, Knapp, Sidik and Jonkman can be used for random effects meta analysis.
*Article Characteristics of meta-analyses and their component studies...
To add on Saleh Idrisnur's reply: it is indeed important to be aware of between-study heterogeneity underestimation upon conducting a random-effects meta-analysis with only a limited number of studies available. In that case, the between-study heterogeneity is commonly downward biased towards 0, which underestimates the uncertainty in the pooled evidence and may even lead to false positives in significance testing. See the following article for more details:
Article Avoiding zero between-study variance estimates in random-eff...
A Knapp-Hartung modification with ad-hoc variance correction may give you a better coverage probability, but its 95% confidence intervals may be overly wide in the extreme case of N = 2 studies. See the following article for more details:
Article Hartung-Knapp-Sidik-Jonkman approach and its modification fo...
In case of limited available studies, Bayesian meta-analyses could provide superior results over the conventional frequentist framework, as we can impose a nonzero between-study variance in our priors. But these methods are more complex to implement and they require prior knowledge as well.
https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/bayesian-ma.html
To conclude, meta-analyses can already be performed with as little as N = 2 studies, but they require more complex and critical examinations to counter biases. I would recommend to only perform meta-analyses with a small N if the results can be rigorously defended from a methodological standpoint.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers