You are here

Different means in ACE model

6 posts / 0 new
Last post
pgseye's picture
Offline
Joined: 10/13/2009 - 23:50
Different means in ACE model

Hi,

I'm wondering how to amend an ACE script to allow for different means across zygosity? (means modelling of my phenotype showed poor fits when trying to constrain MZ/DZ means).

Can someone point me in the right direction?

Paul

Ryne's picture
Offline
Joined: 07/31/2009 - 15:12
Paul, You can remove the

Paul,

You can remove the constrain on means by giving the free parameters for the MZ means different labels than the DZ means. Assuming you're setting everything up as a multiple group model (i.e., one mxModel for MZs, one mxModel for DZs, a container model that holds both of them and some constraints across the two groups made with labels), you'll just replace the label that you use for means in both groups with one label for MZs and one for DZs.

ryne

felice gimondi's picture
Offline
Joined: 03/15/2012 - 07:51
across vs within contraints

Hi, I know this is an old topic. But I just applied this and have the problem that now I do not contrain the means within zygosity anymore. So I estimate three more factors. Any idea, what I might do wrong?
thanks!

neale's picture
Offline
Joined: 07/31/2009 - 15:14
Probably need same label for twin 1 & twin 2 mean

Something like
mxMatrix("Full", nrow=1, ncol=2, labels="MZmean", free=T, name="mzmeans")

The label will get re-used and therefore parameters will be equated. For multivariate you could build a list (one twin's number of variables, nVar, long) using, e.g., labels=paste("MZmean",1:nVar,sep="")

tbrick's picture
Offline
Joined: 07/31/2009 - 15:10
The basic idea: instead of

The basic idea: instead of creating one means matrix, create two. Give them each their own free parameters, and apply one to MZ and one to DZ.

It depends on how you lay out your ACE script to begin with. I'm going to use the example from the VIPBG site here: http://www.vipbg.vcu.edu/~vipbg/GE/UnivTwinACEMaRawCon.R but hopefully this will be illustrative enough to show you how to do it in your own code. This model uses the rather popular three-child-model approach: there is a child model called "ACE" which contains most of the algebras, then one child called "MZ" and one called "DZ", which each handle one zygosity group.

In this script, there is one calculation for the mean:

# Matrix & Algebra for expected means vector
        mxMatrix( type="Full", nrow=1, ncol=nv, free=TRUE, values= 20, label="mean", name="Mean" ),
        mxAlgebra( expression= cbind(Mean,Mean), name="expMean"),

Which we'll make into two like so:

# Matrix & Algebra for expected means vectors
# For MZ
        mxMatrix( type="Full", nrow=1, ncol=nv, free=TRUE, values= 20, label="meanMZ", name="MeanMZ" ),
        mxAlgebra( expression= cbind(Mean,Mean), name="expMeanMZ"),
# For DZ
        mxMatrix( type="Full", nrow=1, ncol=nv, free=TRUE, values= 20, label="meanDZ", name="MeanDZ" ),
        mxAlgebra( expression= cbind(Mean,Mean), name="expMeanDZ"),

This is a univariate model, so we only need one free parameter for the mean vector. If your model is multivariate, make sure that you distinguish the labels of the MZ means matrix from the labels of the DZ means matrix so that they are not constrained to be equal.

Now we use them in the individual models. The snippet that defines the existing models currently reads:

    mxModel("MZ",
        mxData( observed=mzData, type="raw" ),
        mxFIMLObjective( covariance="ACE.expCovMZ", means="ACE.expMean", dimnames=selVars )
    ),
    mxModel("DZ", 
        mxData( observed=dzData, type="raw" ),
        mxFIMLObjective( covariance="ACE.expCovDZ", means="ACE.expMean", dimnames=selVars ) 
    ),

We'll change it to read:

    mxModel("MZ",
        mxData( observed=mzData, type="raw" ),
        mxFIMLObjective( covariance="ACE.expCovMZ", means="ACE.expMeanMZ", dimnames=selVars )
    ),
    mxModel("DZ", 
        mxData( observed=dzData, type="raw" ),
        mxFIMLObjective( covariance="ACE.expCovDZ", means="ACE.expMeanDZ", dimnames=selVars ) 
    ),

The result is that you'll have two free parameters, meanDZ and meanMZ, that will allow the individual means to propagate appropriately. When calculating the DZ twins, the DZ mean will be used. When calculating MZ, the MZ mean will be used.

Of course, there are other ways to do it. If you're using a different formulation or want a different parameterization, feel free to ask.

pgseye's picture
Offline
Joined: 10/13/2009 - 23:50
Thanks Guys, that works well.

Thanks Guys, that works well. I wasn't really familiar with what the difference between names and labels were.