Rules Comm posted on Monday, May 03, 2010 - 3:18 am
I am thinking to calculate a score for the latent variable. but I do not know the right way to do it.
The reason I decide to do this:
I worked out a pretty good CFA; but when I put the same observed and latent variables in a SEM, the SEM cannot produce a good result. Even after I have adjusted the model specification, SEM just does not work out. The major reason is that I have many I.V. in the SEM and I cannot delete any of them out.
so, I decide to give up the SEM, but turn to calculate a score for the latent variable in the CFA.
I have read some books about CFA. Sounds nobody is talking about this.
My questions: 1. Is this idea reasonable? 2. If so, what is the appropriate way to do this.
The way I can imagine now is as follows: 1, There are 3 observed variables. I use the three StdYX to calculate three scores of latent variables, e.g. dividing the score of observed variable by the related StdYX, and do this for the three observed variables (but I am doubting on this method, because I did not include the error term). 2. then I add the three values of calculated latent variables together and divide the sum by 3.
I don't think it is a reasonable idea. Instead you should investigate why your SEM does not fit the data when you add your IVs. If the measurement part of the model (the CFA part) fits well, the SEM problem sounds like you have some direct effects of the IV's on the factor indicators.
Creating factor scores and then moving on, merely hides the problem, but it is still there.
Rules Comm posted on Monday, May 03, 2010 - 2:20 pm
Thank you, Dr. Muthen.
I am trying to figure out the the direct influence of IVs on the factor indicators.
But I still think of some methods of assigning scores to the latent variable. Do you have any recommendation for me to read or try?
BTW, this is actually a one-factor CFA, with only three indicators. I thought I can address it very soon. Never imagined it is so complicated, especially for a beginner like me.