Is it possible to change the precision of the numerical integration for binary FIML? I don't get the same likelihood in classic versus OpenMx.
To the extent that we're open-source, anything is customizable if you want it to be, but I haven't the slightest idea. There's nothing in ?mxOption, and the NPSOL manual isn't much help. Consider this a "bump."
To anyone reading this thread, a full set of options to control integration precision are now available. See thread http://openmx.psyc.virginia.edu/thread/3868
As of build 1515, it is not possible to specify precision on numerical integration. This is marked as a TODO, but hasn't been implemented quite yet.
You should still get results very close to what you got in classic--I believe we're using almost, if not precisely, identical settings.
Would you want this to be a global mxOption? An argument to mxFIMLObjective? Both?
Thanks for the feedback!
Copyright © 2007-2017 The OpenMx Project