Can OpenMx plot predicted y hat of LGC automatically?
Posted on
Veronica_echo
Joined: 02/23/2018
Hi all,
I am conducting latent basis models and have got estimations. I want to plot the functional form of individuals. Is there any way for OpenMx to plot those y hat's for each individual automatically? Or at least can I have those y hat values directly?
Thank you in advance!
Clarification
Hi Veronica
Can you clarify? Do you simply want the expected mean for each row? Or do you also want a function of the estimated factor scores, a la fitted.lme?
For the former, the mxGetExpected() will deliver the model expected means, and its defvar row argument could be apply'd to obtain the predicted means when there are definition variables involved that make the expected means differ between data rows.
Thanks
Mike
Log in or register to post comments
In reply to Clarification by AdminNeale
Thanks
Dr. Neale,
Thanks for your kind reply. I tried the function mxGetExpected() at your suggestion and got mean vectors and variance-covariance matrix of outcomes (i.e., the variable BMI in my case, shown below). It is great to know the model-based mean and var-cov matrix of outcomes, which can be compared to the means and var-cov matrix from the data summary. However, we are more interested in such model-based values for each individual (I understand your point about using "defvar" to specify covariates, but we do not have such covariates/definition variables; we only have repeated outcomes). Can I get such values by some existed functions?
> mxGetExpected(latent.BMI.out, component = "means")
BMI1 BMI2 BMI3 BMI4 BMI5
[1,] 13.45409 16.07842 16.78987 17.20335 16.8839
> mxGetExpected(latent.BMI.out, component = "covariance")
BMI1 BMI2 BMI3 BMI4 BMI5
BMI1 2.0662547 0.4538267 0.4528134 0.4522246 0.4526795
BMI2 0.4538267 2.9940048 1.6378395 1.7846037 1.6712154
BMI3 0.4528134 1.6378395 3.5677893 2.1458102 2.0015591
BMI4 0.4522246 1.7846037 2.1458102 3.9644291 2.1935503
BMI5 0.4526795 1.6712154 2.0015591 2.1935503 3.6539106
Log in or register to post comments
In reply to Thanks by Veronica_echo
It's not entirely clear to me
It's not entirely clear to me what you're trying to do. I'm not very familiar with the kind of LGC you're fitting, so it would help if you post your script and explain what it's meant to do.
Log in or register to post comments
4-line solution
If you start with something like the demo called
LatentGrowthCurveModel_PathRawwhich can be run withdemo(package='OpenMx', 'LatentGrowthCurveModel_PathRaw')
Then you only need a few extra lines to get what you want. The real code here is just four lines, but there's some extra inspection and comments.
fs <- mxFactorScores(growthCurveFit, type='regression')
# There's a message about the space in the model name, but everything is fine.
# See ?mxFactorScores for the structure returned
head(fs[,,1]) # these are the factor scores
# intercept slope
#1 7.304843 1.102765
#2 9.494752 2.357813
#3 9.343139 1.688075
#4 9.335320 1.904806
#5 10.397774 1.942586
#6 10.128553 1.350930
map <- mxEval(A, growthCurveFit)[growthCurveFit$manifestVars, growthCurveFit$latentVars]
# Note that this is the mapping for defvar.row=1. If you have different times of measurement
# for different people, you'll need a different 'map' for each person.
obs <- t(map %*% t(fs[,,1]))
dim(obs) # structured like original data set
#[1] 500 5
matplot(t(obs), type='l')
# quick plot
The general idea here is to (1) get the regression factor scores, (2) get the latent-to-observed variable map (i.e. factor loadings), and then (3) compute the predicted observed scores based on the factor scores and the mapping. Finally, a simple plot is made with
matplot.Log in or register to post comments