Hi,
I am trying to detect and identify bivariate outliers in a dataset using OpenMx, in order to see whether specific outliers have significant contribution. Preferrably the output would be like that of %p in old Mx.
(i.e. 8 columns with:
1) -2lnL,
2) Mahalanobis,
3) estimated Z,
4) number of observations in data set,
5) number of data points in vector,
6) optimization details,
7) whether or not likelihood was calculable, and
8) model number if there are multiple models)
I have already tried using the vector=TRUE argument in FIMLobjective (as suggested in https://openmx.ssri.psu.edu/thread/584), however ID number is not included and I am unsure whether the output is in the same order as in the data file.
Any help would be appreciated, thanks.
Printer-friendly version
I think this would be useful to extract, and it would be possible to write an R function to extract the information after the fact. I think that the likelihoods come back in the same order (a little experimentation would confirm this quite easily). If you take a look at:
http://openmx.psyc.virginia.edu/docs/OpenMx/2.0.0-3838/FIML_RowFit.html#row-fit-example
You can see that there are functions that tell you which variables are present/missing etc. An mxEval ought to be able to extract what you want - but I think you have to be using vector=TRUE in the first place. Or at least, post-model fitting without it you would need to specify vector=TRUE and use the solution you had found without it, perhaps with all parameters fixed to save time.
Do let us know if you write a function that can do all this! It was, I think, a useful feature of classic Mx.
NB at present, mixture distribution components have to be specified in different mxModel's so you would need to run the function separately on the different mxModels, unless you wrote a function to walk the tree of all sub MxModels, looking to see which ones are vector=T.