| Message/Author |
|
|
| ljc posted on Friday, February 07, 2014 - 7:45 pm
|
|
|
I am trying to use esem multiple group analysis like one-way anova to do bivariate analyses with my factors. I have about 20 categorical variables that I want to do bivariate analyses on. Later I would like to do multivariable analyses with the factors and the classification variables.
The problem is that each grouping variable gives me a different version of the factors. I have to make different modifications to achieve partial invariance with each grouping variable. With some grouping variables I can’t achieve partial invariance at all and with other grouping variables the interpretation of the factors is different than the interpretation in the total sample.
Am I using the technique appropriately? Any suggestions on a better way to approach this? Or should I just conclude that my factors are too unstable to work with? |
|
|
|
|
| It seems your factors may be too unstable. |
|
|
| ljc posted on Monday, February 10, 2014 - 7:22 am
|
|
|
One of the criticisms of EFA is that you get a different answer with every sample.
Do you know of any papers or unpublished studies in which an EFA survived more than a few multiple group analyses without falling apart. Thank you. |
|
|
|
|
I don't think such a critique of EFA is valid. It may instead be that EFA is often applied to measurements that have not been well designed and piloted to measure a certain set of hypothesized constructs.
I am sure the literature contains many studies where the measurement structure holds up. Perhaps you find some in the work by Marsh et al at
http://www.statmodel.com/ESEM.shtml
You can also try out the Holzinger-Swineford data on our web site and study school, gender and age differences - you will find that the EFA structure holds up well. |
|
|
| Back to top |
