You are here

Going back to OpenMX 1: Confidence intervals of ACE model not showing

4 posts / 0 new
Last post
Rolandgem's picture
Offline
Joined: 10/24/2015 - 10:02
Going back to OpenMX 1: Confidence intervals of ACE model not showing
AttachmentSize
Binary Data Openmx1.R11.05 KB

I am trying to get the confidence intervals for the ACE model, but for some reason they are not showing up in OpenMX version 2. I have been advised to try and run the model in OpenMX version 1.4 so I have been changing some functions in my code to make it compatible.
So far I changed mxExpectationNormal to mxFIMLObjective; and mxFitFunctionAlgebra with mxAlgebraObjective based on what i've found online.
I am still getting one error as follows:
"Error: could not find function "mxFitFunctionML"
Could you please advise how I should edit my code for it to work in OpenMX version 1.4?

Thanks!

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
I am still getting one error
I am still getting one error as follows:
"Error: could not find function "mxFitFunctionML"

mxFitFunctionML() didn't exist in OpenMx version 1. Instead, "objectives" carried out the roles that are now split between "fitfunctions" and "expectations." Just delete all the instances of mxFitFunctionML() from your code; they're extraneous.

I am trying to get the confidence intervals for the ACE model, but for some reason they are not showing up in OpenMX version 2

Could you describe this problem further?

Rolandgem's picture
Offline
Joined: 10/24/2015 - 10:02
Thank you very much for your

Thank you very much for your quick reply RobK. I apologize for the late reply, I was just moving back and forth between the different OpenMX versions.

The problem was that when trying to generate confidence intervals for the sub models AE and CE, I would get an error (NA !!!) in some of the upper bounds, and when I expand the results I would get an upper bound greater than 1. This problem occurs when I run these sub models in the new OpenMX with the following code:

CEModel <-mxModel( AceFit, name="CE")
CEModel <-omxSetParameters( CEModel, labels="a11", free=F, values=0 )
CEFit <-mxRun(CEModel, intervals=T)
(CESum <-summary(CEFit, verbose=T))

round(CEFit@output$estimate,4)
round(CEFit$Est@result,4)
mxEval(MZ.A, CEFit)
mxEval(MZ.C, CEFit)
mxEval(MZ.E, CEFit)

The results for the confidence intervals are as follows:

confidence intervals:
bound estimate ubound note
MZ.A[1,1] NA 0.0000000 NA !!!
MZ.C[1,1] 0.3630855 0.4230395 0.5282773
MZ.E[1,1] 0.4717223 0.5769605 NA !!!

CI details:
parameter value side fit BaTH BsTH BpTH BparTH Th11 c11 e11
1 MZ.A[1,1] 0.0000000 upper 1887.154 -0.01503094 -0.2214035 0.01372319 -0.3057541 1.977253 0.6504149 0.7595791
2 MZ.A[1,1] 0.0000000 lower 1887.154 -0.01503094 -0.2214035 0.01372319 -0.3057541 1.977253 0.6504149 0.7595791
3 MZ.C[1,1] 0.5282773 upper 1890.996 -0.01514985 -0.2152637 0.01637458 -0.2903185 1.930830 0.7268269 0.6868207
4 MZ.C[1,1] 0.3630855 lower 1890.995 -0.01265185 -0.2719947 0.02252237 -0.2785094 1.937280 -0.6025657 0.7980692
5 MZ.E[1,1] 1.1683689 upper 1937.110 -0.01503094 -0.2214035 0.01372319 -0.3057541 1.977253 0.2751512 1.0809111
6 MZ.E[1,1] 0.4717223 lower 1890.996 -0.01514503 -0.2153946 0.01639248 -0.2906954 1.931692 0.7268271 0.6868205

I tried running this on OpenMX V1 with the following code and results:

CEModel <-mxModel( AceFit, name="CE")
> CEModel <-omxSetParameters( CEModel, labels="a11", free=F, values=0 )
> CEFit <-mxRun(CEModel, intervals=T)
> (CESum <-summary(CEFit, verbose=T))

Confidence intervals:

                lbound  estimate    ubound

MZ.A[1,1] 0.0000000 0.0000000 0.0000000
MZ.C[1,1] 0.3084175 0.4230232 0.5283611
MZ.E[1,1] 0.4716388 0.5769768 0.6915826

Thanks again for you help!

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
Difference in optimizers?

I see. So when you were using OpenMx version 2, the upper confidence limit for the standardized E component was (nonsensically) greater than 1, and more importantly, corresponded to a worsening in fit vastly different from the target of 3.841. For the latter reason, OpenMx version 2 didn't report such an obviously wrong confidence limit in summary() output.

In OpenMx v2.2.x and later, the on-load default optimizer is SLSQP, whereas in OpenMx version 1.x, NPSOL was the only available optimizer. I don't see in your script that you switched optimizers anyplace. I'm wondering, if you stuck mxOption(NULL,"Default optimizer","NPSOL") in your script, after you load packages, would you then get the same results that you got from OpenMx version 1?