Does OpenMx support SEM analysis using ordinal/binary indicators and sampling weights

3 replies [Last post]
mari's picture
Offline
Joined: 09/26/2013

Hi,

I'd like to know if it's possible to estimate a sem model with ordinal and binary indicators using a WLS estimator based on the raw data including the sampling weights. The data set contains a weight column with an individual weight for each case and I don’t need multilevel modeling.

I am currently using the R package lavaan for SEM analysis. As lavaan does not support this functionality, I’m considering changing to OpenMx.

Looking into feature lists and forum topics I found that OpenMx supports ordinal indicators as well as sampling weights, but I couldn’t figure out if the combination of both is also supported.

Any hint is very appreciated. If anything is unclear I'm happy to give further information.

Best,
Markus

Ryne's picture
Offline
Joined: 07/31/2009
Not yet, but soon. WLS

Not yet, but soon.

WLS estimation is one of the next features to come to OpenMx, if not the next feature. Users who build from source can see our current version, which still needs extensive testing prior to a beta release.

Thanks,
Ryne

mari's picture
Offline
Joined: 09/26/2013
Thanks

Thanks for your reply Ryne,

Apparently the correct combination of sample weights and ordinal indicators is currently only available in commercial software packages. I'm really looking forward to a new release of OpenMx, which may contains these features.

Thanks a lot for your work on these open packages.

Best,
Markus

neale's picture
Offline
Joined: 07/31/2009
Can do this with FIML

This can be done with FIML. Also, if the covariance matrix and weight matrix can be estimated with weights to begin with (say using FIML) then WLS can be used in OpenMx for fitting subsequent models.

How is a bit more involved... rather similar to mixture distributions is the short answer.