You are here

TSSEM_stage2: issue with CIs. And a multiple effect sizes for the same variable

3 posts / 0 new
Last post
khaliduni's picture
Offline
Joined: 05/24/2018 - 22:31
TSSEM_stage2: issue with CIs. And a multiple effect sizes for the same variable
AttachmentSize
Binary Data Metasem_final_Data&Scritp.R6.61 KB

Hi all,

First of all, I'd like to say thank you to Professor Mike for his efforts in helping all of those who are using his incredible metaSEM package.

Recently, I successfully managed to conduct TSSEM. But, I have some issues with the CIs. The results of stage 2 don't show some of CIs . So, I'm not sure if I'm missing something. (R script, including data, is attached).

Also, I have another question regarding studies with multiple effect sizes for dependent/independent variables.
How can I import the correlation matrices for the studies that report multiple effect sizes? Should I follow the bivariate meta-analysis approach by taking the average?

Regards,
Khaled

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Hi Khaled,

Hi Khaled,

I suggest using diag.constraints=FALSE, which is usually more robust. Since the metaSEM package uses the OpenMx package to construct the confidence intervals, there is not much I can do. I would suggest checking the recent discussion on the confidence intervals in OpenMx. If it still does not work, you may go back to the Wald confidence intervals.

Regarding the non-independent effect sizes, there is not much research on MASEM. If the amount of non-independent effect sizes is small, it seems reasonable to use the average in the analysis. If there are lots of non-independent effect sizes, either the three-level model or the robust standard error approach may be helpful. The average correlation matrix may still be used with the wls() function in the metaSEM package.

Regards,
Mike

## Use diag.constraints=FALSE rather than TRUE
random2 <- tssem2(random1, Amatrix=A, Smatrix=S, 
                  diag.constraints=FALSE, intervals="LB")
random2 <- rerun(random2)
summary(random2) 
## Inspect why it does not work
summary(random2$mx.fit, verbose = TRUE)
 
## Use Wald CI
random2b <- tssem2(random1, Amatrix=A, Smatrix=S, 
                   diag.constraints=FALSE, intervals="z")
summary(random2b) 
khaliduni's picture
Offline
Joined: 05/24/2018 - 22:31
Thanks Mike

Thanks Mike

I really appreciate your help

Log in or register to post comments