|
|
 |
| Factor Mixture Model vs LPA w/ Factor... |
 
|
| Message/Author |
|
|
| Todd Jensen posted on Wednesday, May 11, 2016 - 9:15 am
|
|
|
I have four hypothesized factors, each with a set of observed ordinal-level indicators.
I originally planned on using factor mixture modeling to estimate both a measurement/common-factor model and a mixture model simultaneously. However, this approach will not allow me to use the WLSMV estimator to handle ordinal-level indicators (which produces good model fit). Also, it appears I would not be able to use the WEIGHT and CLUSTER functions, both of which are suitable for the data being analyzed.
I am considering the following alternative approach:
1) To handle measurement error, estimate the measurement model using the WLSMV estimator, sampling weights, and clustering variable.
2) Generate factor scores for each case (and potentially standardize scores for ease of interpretation).
3) Use factor scores as continuous indicators in subsequent latent profile analysis.
I understand that this alternative approach would not allow me to assess measurement invariance between latent profiles (as discussed in Lubke & Muthen, 2005), although I believe invariance can be assumed/justified in my case.
Any feedback on this approach would be greatly appreciated. |
|
|
|
|
| Mixture modeling requires ML or Bayes estimation. ML with mixtures can handle weights and clustering. |
|
|
| Back to top |
|
|
