You are here

Resampling strategies for small-sample ACE model

4 posts / 0 new
Last post
chl's picture
chl
Offline
Joined: 09/08/2010 - 06:16
Resampling strategies for small-sample ACE model

Hi all,

I was wondering whether there exists relevant papers about the use of ACE model in small sample studies. More specifically, I am working on data collected on about 15 to 20 subjects (for each group of MZ/DZ) in a neuroimaging studies. The outcome of interest is actually related to morphometric quantitization, but this may evolve toward fMRI eventually.

Obviously, with such a small sample, we cannot make strong inference nor rely on asymptotic LR tests; so I use a permutation scheme (around zygoticity labels) to estimate sampling fluctuations of A/C/E components and get a p-value. I wonder if this makes sense in this context, and if you know of any papers that deal with this kind of underpowered studies and focus on robust methods to estimate A/C/E components and appropriate tests of ACE against AE/CE models. The only study relying on resampling that come to my mind is a technical report by Chiang et al., Mapping genetic influences on brain fiber architecture with high angular resolution diffusion imaging, although a different technique is used in Chiang et al. Genetics of brain fiber architecture and intellectual performance. The Journal of neuroscience 29(7): 2212-2224.

tbates's picture
Offline
Joined: 07/31/2009 - 14:25
You can use library(boot) to

You can use library(boot) to do resampling, but why not use mxCI to show the confidence intervals on your estimates?

neale's picture
Offline
Joined: 07/31/2009 - 15:14
I would expect that the

I would expect that the desirable properties of MLE's - the asymptotically unbiased one in particular would not be met, so the estimates could be biased. Second, I suspect that the likelihood ratio test would not have sufficient sample size for it to have asymptotic properties, and that therefore the likelihood-based CI's would be inaccurate.

Ben Neale and I kicked around permutating zygosity, maintaining T1-T2 pairs which provides a test of heritability. If zygosity is ignored, and one permutes individuals, breaking T1-T2 pairs, this provides an overall test of familial resemblance (which can be done within zygosity also). However, we didn't publish it, nor do I know of anyone who has - though I've not search the literature for it.

chl's picture
chl
Offline
Joined: 09/08/2010 - 06:16
Thanks for the answer. I'll

Thanks for the answer. I'll continue checking for papers that deal with this issue. I'm not very confident with generalizability of the results published so far in neuroimaging studies where sample size is < 60. Anyway, a permutation approach should be at the very least still better than an MLE one.