|
Post by PJ on May 27, 2014 13:53:51 GMT
Hello,
I am trying to run a bivariate GCTA analysis for a series of traits. Most analyses run correctly, but a few give the following error message:
Error: the information matrix is not invertible.
What does this mean? I also did univariate analysis for all traits, and that worked without problem. That said, the problematic traits did have very high (i.e. non-significant) p-values in the univariate analysis. I am guessing it has to do with lack of power, but any advice would be appreciated.
Thanks! PJ
|
|
|
Post by PJ on Jun 4, 2014 13:32:54 GMT
Apologies for bumping this thread, but I still haven't been able to determine why the analysis fails for some traits and succeeds for others. Can anyone tell me what the error messages means?
|
|
|
Post by Jian Yang on Jun 5, 2014 11:44:51 GMT
It usually happens when variance components become negative so that the V matrix becomes non-positive definite. It often happens to data with small sample size.
|
|
|
Post by JJ Gao on Aug 26, 2014 20:47:01 GMT
I also got the error report "Error: the information matrix is not invertible." bu tI don't think it's the sample size problem. We have 5000+ Cases Vs 5000+ Controls. Would you plz help me out?
|
|
|
Post by JJ Gao on Aug 27, 2014 17:24:56 GMT
An update about the error report "Error: the information matrix is not invertible." I changed to analyze on another unix server, the error disappeared. So I guess it may be caused by improper installation of GCTA or different versions. My work was done V1.22 but not V1.21. Hopefully this can be some helpful.
|
|
|
Post by Zhihong Zhu on Sept 2, 2014 12:53:56 GMT
v1.24 was released. Another possible reason is that estimate of hg^2 is really small and negative, ie. as Jian suggested, negative variance component -> V matrix is non-positive definite
|
|