| Message/Author |
|
|
| JM posted on Tuesday, March 31, 2009 - 7:04 am
|
|
|
| Hi, I'm in the first step of multiple indicator LGM doing the EFA of indicators across time points. I've not specified any growth yet, I'm going to run the factor loading equality next. My question is that the EFA fit the data okay, but MI suggested intercorrelations between indicator error would improve fit...can I specify that to improve fit or does that impact factor loading and intercept invariance analyses later on? |
|
|
|
|
| EFA is done to determine whether the same number of factors is seen across time. I would not add residual covariances unless there is a strong substantive reason to do so, for example, for the same variable across time. |
|
|
| JM posted on Wednesday, April 01, 2009 - 6:15 am
|
|
|
Thanks. I've a few follow-up questions:
On your LGM handout for computing multiple indicator CFA, it does not specify what syntax to include between the Title and the Model:
Q1. So, as I have non-normal and missing data, would I just specify TYPE = MISSING, and ESTIMATOR = MLR....what about TYPE = MEANSTRUCTURE?
Q2. I read that the first indicator should be fixed to zero. This is not covered in the handouts so I can't tell if this should be specified in the syntax also.
THANK YOU!! |
|
|
|
|
Between TITLE and MODEL, you would specify whatever is relevant to your data. TYPE=MISSING MEANSTRUCTURE has been the default since Version 5.
I have never heard that the first indicator should be fixed to zero. Are you referring the the time scores for the growth factor where the value zero defines the intercept growth factor? |
|
|
| JM posted on Tuesday, April 07, 2009 - 6:54 am
|
|
|
| Maybe, i read that the meanstructure aspect of the measurement model maybe identified by either fixing the latent mean to zero or assigning the factor mean to take on the mean of one of its indicators (by fixing the intercept of one indicator to zero). I'm confused if that is the same or different? |
|
|
|
|
| You can fix an intercept to zero and free a factor mean. All this does, however, is move the parameter estimate from nu to alpha. It does not change the results. |
|
|
| Back to top |
