| Attachment | Size |
|---|---|
 R script | 4.88 KB |
 CSV data file | 2.39 KB |
Hi there,
I am running a multivariate meta analysis and running into an error code I cannot seem to resolve. In prior error messages related to this content, others have noted negative or zero values in the covariance matrix, but I'm not seeing any from the output. Any/all help you could provide would be amazing!
Context: Looking at two types of memory tasks (forward and backward) comparing deaf and hearing participants using Hedge's g effect size.
The error message I'm receiving is: The job for model 'Multi MA' exited abnormally with the error message: fit is not finite (The continuous part of the model implied covariance (loc2) is not positive definite in data 'Multi MA.data' row 8. Detail:
covariance = matrix(c( # 2x2
0.332955234678017, 0.6
, 0.6, 0.298097818087639), byrow=TRUE, nrow=2, ncol=2)
In response to running the multivariate meta code:
fit0a <- meta(y=cbind(FWD_G, BWD_G),
v=cbind(FWD_SE,FBcov,BWD_SE),
data=Verbal_ForMultivariate_HedgesG,
model.name="Multi MA")
Do you have any advice or help for resolving this issue?
Script and data attached here. Thank you!
Printer-friendly version
There were some errors in the R code. First, some data were incorrectly read as strings. Second, the meta() expects the sampling variances, whereas the code uses the standard errors. Finally, the sampling correlation (0.5) was incorrectly treated as the sampling variance.
Attached is the revised version that works okay.