Unable to run meta-SEM due to missing data

Posted on
No user picture. smh1003 Joined: 04/22/2019
Forums

First of all, I am grateful for asking my question to this forum where gurus of meta-SEM are helping people like me.
I am currently conducting meta-SME with the theory of planned behavior by following the example of Scalco17 in this website.
https://cran.r-project.org/web/packages/metaSEM/vignettes/Examples.html#scalco17
It seems my dataset work pretty well, but something is messing up the analysis.

Some data show 'NA' when I conducted the test for positive definite matrix using 'is.pd', and I think it is the reason for unreliable result because all results seems good when i conduct the analysis after eliminating data with NA from is.pd.

Would someone help me to figure this issue out?
How can I deal with missing data or NA from is.pd like this?

Replied on Tue, 04/23/2019 - 23:02
Picture of user. Mike Cheung Joined: Oct 08, 2009

The error is likely related to the fact that two of the variances (Tau2_5 and Tau2_9) are near zero.


> VarCorr(mx.fit0)
Tau2_1 Tau2_2 Tau2_3 Tau2_4 Tau2_5 Tau2_6
Tau2_1 0.00680293 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_2 0.00000000 0.0006061789 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_3 0.00000000 0.0000000000 0.009556938 0.000000000 0.000000e+00 0.000000000
Tau2_4 0.00000000 0.0000000000 0.000000000 0.003327514 0.000000e+00 0.000000000
Tau2_5 0.00000000 0.0000000000 0.000000000 0.000000000 1.847617e-11 0.000000000
Tau2_6 0.00000000 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.006505271
Tau2_7 0.00000000 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_8 0.00000000 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_9 0.00000000 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_10 0.00000000 0.0000000000 0.000000000 0.000000000 0.000000e+00 0.000000000
Tau2_7 Tau2_8 Tau2_9 Tau2_10
Tau2_1 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_2 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_3 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_4 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_5 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_6 0.000000000 0.00000000 0.000000e+00 0.00000000
Tau2_7 0.001114616 0.00000000 0.000000e+00 0.00000000
Tau2_8 0.000000000 0.01365853 0.000000e+00 0.00000000
Tau2_9 0.000000000 0.00000000 1.223589e-20 0.00000000
Tau2_10 0.000000000 0.00000000 0.000000e+00 0.03870517

Rerunning the model seems to work.


mx.fit0 <- rerun(mx.fit0, extraTries=50)

Replied on Wed, 04/24/2019 - 10:14
No user picture. smh1003 Joined: Apr 22, 2019

In reply to by Mike Cheung

I really appreciate your quick and helpful answer!
So, in this case, should I report the result as is?
Since the summary of mx.fit0 shows below message,
** Information matrix is not positive definite (not at a candidate optimum),
does it mean the overall fit of the model is not trustworthy and the result of meta-analysis is not credible?

Replied on Thu, 04/25/2019 - 07:55
Picture of user. Mike Cheung Joined: Oct 08, 2009

Does rerunning the model work? It worked fine in my R.

If it still does not work, you may increase the number of trials, e.g., 100, and see if it works better.