Hi again! I've run into some identification problems on a multi-group CFA. I've got three indicators of one latent variable. Bollen suggests that one way of achieving identification is to constrain the mean of the latent variable to be equal to the intercept of one of the indicators; I don't want to do this, since I'm interested in comparing the means of the latent variable across groups. So...can I constrain the thresholds of these indicators to be equal across groups to achieve identification? What kind of consequences are there to doing this? I guess my basic question is: what do you recommend? Thanks so much for all your help!
I would have to see your output to see why you are having identification problems. You would need to send it along with your license number to firstname.lastname@example.org.
Sarah Ryan posted on Tuesday, September 20, 2011 - 2:13 pm
I am modeling an SEM model with several LF's, all with observed ordinal or binary indicators, and an ordinal outcome using WLSMV.
The standard errors for the estimated thresholds of the u* underlying these categorical variables are relatively large (ranging from about .15 to .25 in most cases), although most of the threshold estimates are significant.
Is the magnitude of SE's for these thresholds important to attend to?
And there is this statement: "A threshold is the expected value of the latent variable or factor at which an individual transitions from a value of 0 to a value of 1.00 on the categorical outcome variable when the continuous underlying latent variable's score is zero."
Is this right? Should the expected value be for the continuous underlying latent variable of the categorical indicator, not the factor? Also by "continuous underlying latent variable's score is zero" do you mean the factor score is zero?