You are here

Weighting and heterogeneity

3 posts / 0 new
Last post
smh1003's picture
Offline
Joined: 04/22/2019 - 11:47
Weighting and heterogeneity

Dear Dr. Cheung,

I wrote a manuscript using OSMASEM and submitted it to a journal.
I recently got reviewer's comments asking about the weighting and heterogeneity of OSMASEM.
My understanding is that the method use sample size weighting, and heterogeneity can be explained by Tau2, which is between-study variances.
I was wondering if I could have your advice to deal with these comments.
I have read articles related to OSMASEM several times, but I wasn't able to get a clear answer for them due to my lack of understanding:(

The specific comments from the reviewers were,
1. When conducting the meta-analysis, were any weights taken into account based on the variances and/or sample sizes across the studies? If the pooled analysis was done, it will be more informative to include the weights for the seven studies.

For this comment, how can I include the weights for each study? As OSMASEM is different from what usual meta-analysis should include, such as forest plot.

2.It seems that there is no result showing the heterogeneity across the studies in meta-analysis. If yes, please include it and also specify the criterion used for detecting the
significance.

I intend to use Tau2, but how can I justify the criterion for detecting the significance?

I will look forward to having your answer at your earliest convenience.

Regards,
Mason Sur

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Dear Mason Sur,

Dear Mason Sur,

1) The weight (w_i) in the ith study equals the inverse of the sampling variance (1/v_i) in a fixed-effect meta-analysis with one outcome variable. The weight becomes (1/(v_i+tau^2) in a random-effects meta-analysis, where tau^2 is the estimated heterogeneity. In a multivariate meta-analysis with 5 effect sizes, the weight is a 5x5 matrix. You may calculate it from V_i and T^2 (both are standard notations in the multivariate meta-analysis, MASEM, and OSMASEM). However, I doubt the usefulness of presenting these matrices.

2) Yes, T^2 is the heterogeneity matrix. We may test whether the population heterogeneity matrix is 0. But its usefulness is questionable, especially, the test is on the boundary (variance cannot be negative). In a univariate meta-analysis, we may also report the I^2, a relative index of heterogeneity. The situation becomes more complex when T^2 is a matrix.

Mike

smh1003's picture
Offline
Joined: 04/22/2019 - 11:47
Appreciation

Dear Dr. Cheung,

I appreciate your clear and helpful answers to my questions!
I will look forward to reaching out to you with my article once it is published.
Have a great day

Regards,
Mason Sur