# How to get the P value of heritability in ACE model

2 posts / 0 new
Offline
Joined: 04/11/2017 - 23:56
How to get the P value of heritability in ACE model
AttachmentSize
10.22 KB

Hello all,

I am very new with ACE twin modeling as well as R, therefore I need some help.

I am trying to run a univariate ACE model using the attached script (UnivACE.R) in OpenMx. It would be very grateful to know whether the script does not have any mistakes. And one more thing, what modifications are needed in the script to get the p-values of the estimated h2,c2 and e2 in ACE model? Furthermore, when using the same dataset to run ACE model, Mplus reported different CIs for the path coefficients compare to that in OpenMX results. Which one is correct, and why?

Another problem is how to set the start value, since I found this sometimes has a great influence on the final results.

Thanks a lot for your help!
Best Regards,

Chenxi

Offline
Joined: 01/24/2014 - 12:15
It would be very grateful to know whether the script does not have any mistakes.

Does it run and give sensible answers?

what modifications are needed in the script to get the p-values of the estimated h2,c2 and e2 in ACE model?

You're getting confidence intervals for those three quantities, so you don't really gain any additional information from p-values. But anyhow, you can get p-values for the a² and c² coefficients via mxCompare(), e.g.

mxCompare(univACEFit, univCEFit)

would provide a likelihood-ratio-test p-value for the heritability in its output. Likewise, a comparison of 'univACEFit' and 'univAEFit' would test the null hypothesis that c² is zero. A comparison of 'univACEFit' and 'univEFit' would test the joint null hypothesis that a² and c² are both zero.

Furthermore, when using the same dataset to run ACE model, Mplus reported different CIs for the path coefficients compare to that in OpenMX results. Which one is correct, and why?

How different are we talking here? Also, how large is your sample? I don't know how MPlus calculates confidence intervals, but it might be doing "Wald"-type intervals, whereas OpenMx does profile-likelihood intervals. The two kinds of confidence interval are asymptotically equivalent, but profile-likelihood intervals have superior theoretical properties. If you take each estimate of the path coefficient and and add and subtract 2 times its standard error, do you get something close to MPlus' confidence limits?

Another problem is how to set the start value, since I found this sometimes has a great influence on the final results.

The start value for each free parameter is its corresponding value of the 'values' argument to mxMatrix(). If you need to change the start values of an entire MxModel object, use omxSetParameters().