Power to Detect E and r(e)

Hi all, I was wondering if anyone had resources for detecting unique environmental variance and unique environmental correlations.

I think I understand why this is not possible using the weighted noncentrality parameter approach of the provided power functions, in that it would involve fitting a null AC model (E=0) which I don’t believe can be done in OpenMx. But, please let me know if my understanding if off and there is a way to do this with these functions.

If it isn’t possible, are there other scripts or approaches you’ve used for these types of estimates? Thanks in advance for any light you can shed on this.

Hi Colin - I don’t think any software could fit an AC model by maximum likelihood. This is because the MZ pairs expected correlation would be 1.0, so the expected covariance matrix could be inverted to calculate the likelihood. For a specific amount of E in an ACE model, you could simply examine the standard errors or the likelihood-based confidence intervals of the parameter e (or better VE if working with variance components) to get an idea of the precision with which it can be estimated. But essentially the likelihood ratio of an AC vs an ACE model is undefined.


1 Like

Thank you, Dr. Neale, that is very helpful. Right, I see that AC is not feasible with maximum likelihood estimation which makes the wncp approach not viable for power estimation for E and r(e). Really appreciate your response!

I’m curious if estimating power to detect unique environmental correlations is feasible some other way, perhaps with simulation.