You are here

Issues with confidence interval estimation for Cholesky Decomposition ACE model

3 posts / 0 new
Last post
lyra9826's picture
Offline
Joined: 10/20/2021 - 12:29
Issues with confidence interval estimation for Cholesky Decomposition ACE model
AttachmentSize
Binary Data twoACE_Cholesky.R8.57 KB

I am estimating a Cholesky decomposition ACE model and deriving the rA, rC, and rE estimates. The confidence intervals are not estimating for many of these parameters. There appear to be varying diagnoses when I use verbose=T (i.e., active box constraint, alpha level not reached infeasible non-linear constraint). I have tried choosing a different optimizer, re-scaling the phenotypes, using mxTryHard, and changing start values. The univariate variance components as well as their correlations are significant. I would appreciate any guidance regarding how to obtain these confidence intervals.

Please note we will also look at the correlated ACEs with direct variance but this is in response to a reviewer request for the Cholesky.

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
output?

Could you provide your verbose summary() output?

AdminNeale's picture
Offline
Joined: 03/01/2013 - 14:09
Invariance to sign

Hi

You may be running into issues with a path-specified ACE model where an estimate of -.5 for a results in the same fit as +.5. In these circumstances, the lower CI may end up at negative the upper CI. To avoid this you can bound a c and e parameters to be positive. It is also worth noting that Cholesky is not a recommended multivariate model, because it has implicit constraints (to keep things positive definite) and incorrect Type I error rates when, e.g., dropping the C matrix. See Verhulst et al 2019 for an alternative specification that does not have these problems.