You are here

tssem1 random effects model: diag vs. symm

2 posts / 0 new
Last post
k.corker's picture
Offline
Joined: 10/05/2015 - 18:51
tssem1 random effects model: diag vs. symm

Hi Mike,

Thanks very much for your excellent metaSEM package. I have successfully wrangled my data into the necessary format and run a fixed effect model and then a random effects model with the "diag" option. The data is 37 samples (ranging in size from 28 to 1589) of 8 variables (and none of the samples have missing data; all samples provide all 28 correlations). I can't get the random effects model to run with the "Symm" option, and it is driving me nuts as to why this is the case. I get Open Mx error code 6, and all of the parameters come out NA.

As a second question, I was wondering if you had any resources on the similarities and differences between metaSEM's tssem1() and metafor's rma.mv (the multivariate meta-analysis). I've been trying to compare and contrast results in both programs, and I'm not sure I'm giving either program the proper inputs. They are running, and giving somewhat similar results (typically +/- .02 correlation units), which makes me feel like I'm on the right track.

Thank you!
Katie

P.S. As a side note - I was trying to run analyses in metaSEM on my mac running OS 10.8.5, and I couldn't get any commands to run (I got a cryptic c++ error). After switching to my laptop running OS 10.9.5, everything seems to work OK (except the aforementioned REM).

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Hi Katie, The first question

Hi Katie,

The first question is related to the number of parameters. Since there are 8 variables, there are 87/2=28 correlation coefficients. There will be 2829/2=406 elements in the variance component of the random effects. Thirty-seven samples do not seem to be enough to estimate all parameters.

The mathematical models between tssem1() and metafor's rma.mv() are the same. Both of them are standard multivariate meta-analysis. tssem1() uses ML estimation. If you are also using ML estimation in rma.mv(), the results should be the same.

Regarding the Mac issue, I am sorry that I rarely use Mac. Maybe other Mac users may comment on whether this is due to Mac's version or other issues.

Mike