Post by paino on Oct 19, 2016 15:06:53 GMT
Hello,
I am trying to calculate the standard error of sum(Vg)/Vp using the variance-covariance matrix but I can't get the same answer that GCTA gives me. I need to do this as I want to estimate the SNP-h2 with GREML-MS using related individuals. However the output of this gives me the sum of V(G)/Vp for all variance components including the pedigree component. I have put an example of the output below where the first 6 variance components are based on SNPs and the 7th is the pedigree data.
##########################################################################
Summary result of REML analysis:
Source Variance SE
V(G1) 0.026530 0.027493
V(G2) 0.004974 0.022275
V(G3) 0.002882 0.025628
V(G4) 0.000001 0.018022
V(G5) 0.014042 0.022646
V(G6) 0.000001 0.019295
V(G7) 0.389158 0.048547
V(e) 0.558619 0.025787
Vp 0.996208 0.017715
V(G1)/Vp 0.026631 0.027586
V(G2)/Vp 0.004993 0.022359
V(G3)/Vp 0.002893 0.025725
V(G4)/Vp 0.000001 0.018090
V(G5)/Vp 0.014096 0.022728
V(G6)/Vp 0.000001 0.019369
V(G7)/Vp 0.390640 0.047650
Sum of V(G)/Vp 0.439254 0.025919
Sampling variance/covariance of the estimates of variance components:
0.000755838 -0.000101952 -3.4164e-05 -1.80735e-05 -1.46089e-05 -3.6496e-06 -0.000563145 -4.58624e-06
-0.000101952 0.000496164 -0.000115337 -2.21975e-06 -4.98142e-06 -2.29137e-05 -0.000244251 -1.09126e-06
-3.4164e-05 -0.000115337 0.000656776 -7.45752e-05 -5.57259e-05 -5.16293e-05 -0.00032578 1.17509e-06
-1.80735e-05 -2.21975e-06 -7.45752e-05 0.000324779 -6.35102e-05 -3.59988e-05 -0.000127065 -1.07094e-06
-1.46089e-05 -4.98142e-06 -5.57259e-05 -6.35102e-05 0.000512861 -0.000145029 -0.000219005 -6.46825e-07
-3.6496e-06 -2.29137e-05 -5.16293e-05 -3.59988e-05 -0.000145029 0.000372309 -0.000111192 3.81222e-07
-0.000563145 -0.000244251 -0.00032578 -0.000127065 -0.000219005 -0.000111192 0.0023568 -0.000572675
-4.58624e-06 -1.09126e-06 1.17509e-06 -1.07094e-06 -6.46825e-07 3.81222e-07 -0.000572675 0.000664947
##########################################################################
I understand that each diagonal element is the variance of each component. So here, the SE of V(G1) is sqrt(0.000755838) = 0.027493.
I understand the standard error of the sum of V(G) = square root of the average variance. So here, the SE of Sum of V(G) is sqrt(mean(0.000755838, 0.000496164, 0.000656776, 0.000324779, 0.000512861, 0.000372309, 0.0023568, 0.000664947)) = 0.02749251
However, how do I then convert this to the SE of sum of V(G)/Vp? Simply dividing it by Vp doesn't give me the same answer i think because it doesn't take into account the SE of Vp?
Many thanks for any help. As you can probably tell, I don't have a strong background in statistics!
Oliver
I am trying to calculate the standard error of sum(Vg)/Vp using the variance-covariance matrix but I can't get the same answer that GCTA gives me. I need to do this as I want to estimate the SNP-h2 with GREML-MS using related individuals. However the output of this gives me the sum of V(G)/Vp for all variance components including the pedigree component. I have put an example of the output below where the first 6 variance components are based on SNPs and the 7th is the pedigree data.
##########################################################################
Summary result of REML analysis:
Source Variance SE
V(G1) 0.026530 0.027493
V(G2) 0.004974 0.022275
V(G3) 0.002882 0.025628
V(G4) 0.000001 0.018022
V(G5) 0.014042 0.022646
V(G6) 0.000001 0.019295
V(G7) 0.389158 0.048547
V(e) 0.558619 0.025787
Vp 0.996208 0.017715
V(G1)/Vp 0.026631 0.027586
V(G2)/Vp 0.004993 0.022359
V(G3)/Vp 0.002893 0.025725
V(G4)/Vp 0.000001 0.018090
V(G5)/Vp 0.014096 0.022728
V(G6)/Vp 0.000001 0.019369
V(G7)/Vp 0.390640 0.047650
Sum of V(G)/Vp 0.439254 0.025919
Sampling variance/covariance of the estimates of variance components:
0.000755838 -0.000101952 -3.4164e-05 -1.80735e-05 -1.46089e-05 -3.6496e-06 -0.000563145 -4.58624e-06
-0.000101952 0.000496164 -0.000115337 -2.21975e-06 -4.98142e-06 -2.29137e-05 -0.000244251 -1.09126e-06
-3.4164e-05 -0.000115337 0.000656776 -7.45752e-05 -5.57259e-05 -5.16293e-05 -0.00032578 1.17509e-06
-1.80735e-05 -2.21975e-06 -7.45752e-05 0.000324779 -6.35102e-05 -3.59988e-05 -0.000127065 -1.07094e-06
-1.46089e-05 -4.98142e-06 -5.57259e-05 -6.35102e-05 0.000512861 -0.000145029 -0.000219005 -6.46825e-07
-3.6496e-06 -2.29137e-05 -5.16293e-05 -3.59988e-05 -0.000145029 0.000372309 -0.000111192 3.81222e-07
-0.000563145 -0.000244251 -0.00032578 -0.000127065 -0.000219005 -0.000111192 0.0023568 -0.000572675
-4.58624e-06 -1.09126e-06 1.17509e-06 -1.07094e-06 -6.46825e-07 3.81222e-07 -0.000572675 0.000664947
##########################################################################
I understand that each diagonal element is the variance of each component. So here, the SE of V(G1) is sqrt(0.000755838) = 0.027493.
I understand the standard error of the sum of V(G) = square root of the average variance. So here, the SE of Sum of V(G) is sqrt(mean(0.000755838, 0.000496164, 0.000656776, 0.000324779, 0.000512861, 0.000372309, 0.0023568, 0.000664947)) = 0.02749251
However, how do I then convert this to the SE of sum of V(G)/Vp? Simply dividing it by Vp doesn't give me the same answer i think because it doesn't take into account the SE of Vp?
Many thanks for any help. As you can probably tell, I don't have a strong background in statistics!
Oliver