Attachment | Size |
---|---|
![]() | 17.87 KB |
Hi all,
I am trying to run a bivariate ACE model. My cross-twin-cross trait correlations are as following:
MZM -.36 (-.44 , -.27) DZM -.23(-.25 ,-.09) DOS -.14(-.21, -.07)
MZF -.24 (-.33, -.14) DZF -.16(-.28 , -.04)
My standardized estimates, however, do not seem to match these correlations. Namely, C is estimated much lower than expected based on the twin correlations (esp. for females).
Males A 0.60(0.38, 0.81) C 0.25(0.07, 0.43) E 0.16(0.14, 0.19)
Females A 0.82(0.48, 1.15) C 0.01 (-.30, 0.30) E 0.18(0.10, 0.27)
I ran the same analyses only for two groups (only males / only females no DOS twins) but the same problem arises.
Any thoughts on what I might have done wrong? Something in my script that I am missing?
Thanks in advance!
I have only skimmed your script. It's difficult to evaluate it without the actual datafile or even the point estimates from each MxModel. But, a discrepancy between crude Falconer-esque estimates and maximum-likelihood estimates of standardized variance components isn't necessarily a sign you did something wrong.
In the saturated model, each group's 4x4 correlation matrix is calculated only from the data in that group. When you actually fit a biometric ACE (or whatever) model, you're estimating biometric variance components from the data in all groups. Therefore, those variance-component estimates will be influenced by things like differing group sizes (i.e., counts of twin pairs) and differing amounts of missing data across groups.
Standardizing variance-component estimates (i.e. turning them into variance proportions) can exacerbate the discrepancy between the two sets of estimates. In the saturated model, each correlation is a covariance standardized relative to variances from that group alone. In the biometric ACE (or whatever) model, each correlation is a covariance standardized relative to "pooled" variances estimated from all groups combined.