| Attachment | Size |
|---|---|
 Data | 984 bytes |
 Script for metafor | 721 bytes |
| Script for metaSEM | 2.96 KB |
Hi all!
I hope everyone is well.
I was hoping someone could help me with an issue I am having. I am running a model, but first I wanted to explore the correlation matrix as you would normally do. For this, I created the correlation matrix using the metafor package. After this, I run the model using the metaSEM package, and when I checked the tssem1 results, I realised that the correlation results were very different in magnitude (significance levels did not vary).
As an example, in metafor, the correlation between two variables is -.167, and in metaSEM is -.223
I am wondering now what is the difference between the two approaches, and when reporting, which one should I use. I have read the technical material, but some of the things are a bit too advanced for my knowledge.
For ease, I ran both models with uncorrected correlations and random effect models. In the case of metafor, I used the HS method.
I am attaching both scripts, together with an extract of the sample I am using.
Any help or orientation is much appreciated
Printer-friendly version
Hi Maximiliano,
I have checked the detail. But it seems to me that you are using Hunter-Schmidt approach in metafor and ML in metaSEM. Moreover, the first one is based on a univariate approach, whereas the second one is based on a multivariate approach.
It is not unreasonable to see different values.
Mike
Hi Mike!
Thank you very much for your reply! I understand now why I got the differences. Now I wonder, however, if I were to report the "correlation matrix" for my metasem model in the same way as one reports a normal correlation matrix, which one would it be better? I guess the multivariate approach carries more information as it takes into account more variables, but most MA studies are based on the univariate approach.
Looking forward hearing your thoughts on the matter!
Regards,
Max
Hi Max,
It depends. If you are using a TSSEM, it makes sense to report the multivariate approach so that the results are more consistent.
Best,
Mike
Hi Mike,
That's what I thought as well. I just wanted to have a second (informed) opinion!
Thank you very much for your help and time!
Cheers,
Max