You are here

Starting values in a univariate ACE model

2 posts / 0 new
Last post
lior abramson's picture
Offline
Joined: 07/21/2017 - 13:13
Starting values in a univariate ACE model

Hi,
I am very new to twin analyses in general, and even newer to openMX.
I have tried to perform a simple univariate ACE model using paths specifications. I followed the example called "ACE model: A twin analysis" (section 2.5.1) in the openMX user guide.

Could someone please explain what is the meaning of the starting values of 'A', 'C', and 'E' that should be specified in the model? In the example in the user guide, they gave all latent factors the starting value of .5 in the ACE model, and values of .6 to the free paths (A and E) in the alternative AE model. Did they decide on these values based on general assumptions of the model or based on specific characteristics of the data set that they used?

In short:
1) how do you know what to specify as starting values of the estimated latent factors, A, C, and E?
2) in a similar manner, how do you know what to specify as a starting value of the mean of the observed variable (which should also be estimated in the model)?

Thank you so much

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
start values

I refer you to the discussion in thread 4224 about start values. In brief, assuming you're referring to the A, C, and E path coefficients, you can get a decent start value for all three by calculating the phenotypic variance via var() in R, dividing it by 3, and taking the square root. Likewise, you can get a decent start value for the phenotypic mean via mean() in R.

See also: mxTryHard(), mxAutoStart().

Log in or register to post comments