Post by mmagoo6 on Sept 1, 2016 22:34:08 GMT
Hello All,
I am trying to estimate the phenotypic variance explained by SNPs using GCTA. I have a sample of 1903 with 400 events. I was concerned from this paper and the FAQs that this might be too small of a sample, but from the output, I'm not sure that is the case. I would have expected an error ("Error: the variance-covaraince matrix V is not positive definite" or ""Log-likelihood not converged”) if the sample size were too small, or a non-zero estimate with a large standard error.
Instead, I get an estimate of exactly zero with a regular-sized standard error:
Source Variance SE
V(G) 0.000000 0.017752
V(e) 0.166104 0.018566
Vp 0.166104 0.005386
V(G)/Vp 0.000001 0.106875
The estimate of variance explained on the observed scale is transformed to that on the underlying scale:
(Proportion of cases in the sample = 0.210194; User-specified disease prevalence = 0.210000)
V(G)/Vp_L 0.000002 0.213321
logL 752.426
logL0 752.426
LRT 0.000
df 1
Pval 0.5
n 1903
The zero value makes me particularly suspicious. I have re-checked my inputs, and they seem right. This analysis was is also a subset of a larger sample of 3000 people, and when I run the analysis on the larger sample, I get heritability estimates that are in line with what I expect.
Can anyone comment on how to think of a V(G) that is exactly zero? Does it suggest that the sample size is too small, or some error on my end?
Thank you!
I am trying to estimate the phenotypic variance explained by SNPs using GCTA. I have a sample of 1903 with 400 events. I was concerned from this paper and the FAQs that this might be too small of a sample, but from the output, I'm not sure that is the case. I would have expected an error ("Error: the variance-covaraince matrix V is not positive definite" or ""Log-likelihood not converged”) if the sample size were too small, or a non-zero estimate with a large standard error.
Instead, I get an estimate of exactly zero with a regular-sized standard error:
Source Variance SE
V(G) 0.000000 0.017752
V(e) 0.166104 0.018566
Vp 0.166104 0.005386
V(G)/Vp 0.000001 0.106875
The estimate of variance explained on the observed scale is transformed to that on the underlying scale:
(Proportion of cases in the sample = 0.210194; User-specified disease prevalence = 0.210000)
V(G)/Vp_L 0.000002 0.213321
logL 752.426
logL0 752.426
LRT 0.000
df 1
Pval 0.5
n 1903
The zero value makes me particularly suspicious. I have re-checked my inputs, and they seem right. This analysis was is also a subset of a larger sample of 3000 people, and when I run the analysis on the larger sample, I get heritability estimates that are in line with what I expect.
Can anyone comment on how to think of a V(G) that is exactly zero? Does it suggest that the sample size is too small, or some error on my end?
Thank you!
