Estimation method

1 reply [Last post]
Jean's picture
Offline
Joined: 11/10/2009

Hi,

Can we choose estimation method when conducting SEM, such as OLS, GMM besides ML, FIML?
FIML assumes errors have a multivariate normal distribution. If the errors are not normally distributed, the FIML method might produce poor results, while 2SLS and 3SLS do not assume a specific distribution for the errors. Do we have a option of estimation method in OpenMx?

tbrick's picture
Offline
Joined: 07/31/2009
OpenMx is designed to make it

OpenMx is designed to make it easy for users to put together whatever estimation method they would like to use.

At present, we only have "built-in" optimization for the most common flavors the dev team expected our user base to want, which are FIML and ML. Our feature set at any given time is based on the original specs we expected people to want and the requests we get from the community. The variety of functions available is likely to expand as time goes on and the community decides it wants more variety.

But OpenMx is open-source. If you're comfortable with the math behind other techniques, you can always build your own objective function. The easiest way to do this is to use either an mxRObjective or an mxAlgebraObjective to perform the objective calculation for you. These only require you to know R or OpenMx and the formula to be calculated. The steps are available at http://openmx.psyc.virginia.edu/wiki/howto-make-objective-function and the dev team is happy to provide what support we can while you develop it.

If you do build your own, we invite you to share it with the rest of the community so that they have access to it as well. That way, everybody benefits.

If you're not feeling up to doing it yourself, you can add a feature request at http://openmx.psyc.virginia.edu/node/add/project-issue/openmx by creating an issue with the category "feature request" and it'll go on the list for the dev team.