In the meta analysis the Q-value was 255 & the df=25 & P-value=0.00 and the conclusion there is a substantial dispersion.
Is the above result good or bad? If i want to estimate the effect size depending on the correlation by using sample size and t-test.
Dear Rasha,
Nothing is bad or good in science. It is pure science. First you should calculate how large the heterogeneity is by I square index. Then, you should find a way such as subgroup analysis, sensitivity analysis or meta-regression to explain the heterogeneity.
If you could explain the heterogeneity by a variable such as age, gender, injection type, dose, etc, the heterogeneity concern is not that important. However if the source of heterogeneity is not clear, you should mention it in the limitation based on how larger the heterogeneity is by I square index).
typically you want to report the estimated true effect (mu, z score, p value) and heterogeneity (tau^2, Q statistic, p value), which will suggest a significant mean effect and heterogeneity among the true effects, respectively. so your statistics imply significant heterogeneity among the effect sizes in your study. from here, you should consider quantifying potential for publication bias (funnel plot analysis, Egger test) and performing a moderator analysis (essentially multiple regression) to explain the variation in effect sizes. the earlier comment about reporting I square is also good.
Dear Rasha,
Nothing is bad or good in science. It is pure science. First you should calculate how large the heterogeneity is by I square index. Then, you should find a way such as subgroup analysis, sensitivity analysis or meta-regression to explain the heterogeneity.
If you could explain the heterogeneity by a variable such as age, gender, injection type, dose, etc, the heterogeneity concern is not that important. However if the source of heterogeneity is not clear, you should mention it in the limitation based on how larger the heterogeneity is by I square index).
As mentioned previously, the use of sub-group analysis (http://handbook.cochrane.org/chapter_9/9_6_2_what_are_subgroup_analyses.htm) may be a viable approach in your meta-analysis to help 'explain' some of the heterogeneity if you think that it could be attributed to more study-level characteristics (i.e., sample size, male-female ratio, country income level).
I recommend looking over the Cochrane handbook on how to identify and measure heterogeneity: http://handbook.cochrane.org/chapter_9/9_5_2_identifying_and_measuring_heterogeneity.htm
Also, in Stata you have a command for meta-regression (metareg - http://handbook.cochrane.org/chapter_9/9_6_4_meta_regression.htm) which allows you to have multiple sub-groups (if you do a sub-group analysis in your forest plot, you're only able to choose one explanatory variable).
Good luck!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size