Message/Author |
|
|
Hi, I just read the paper by Todd Little (A Non-arbitrary Method of Identifying and Scaling Latent Variables in SEM and MACS Models, 2006) and would like to model a curve-of-factors model by using the effects coding method Little and colleagues suggest to scale the latent factors. So I would like to know what would be the Mplus syntax to constrain each factor's loadings to average one. Thanks you very much for you guidance. Ana |
|
|
You can do this using MODEL CONSTRAINT. Let's say you have labelled two factor loadings in the MODEL command p1 and p2 and you want the average of p1 and p2 to be one. In MODEL CONSTRAINT, you would say and you want the average of p1 and p2 to be one: MODEL CONSTRAINT: p1 = 2 - p2; See the user's guide for more details about how to label parameters. |
|
deana desa posted on Thursday, July 28, 2011 - 1:43 pm
|
|
|
Hi Dr. Muthen, I've question about two indicators identification. My model is as below: model: g by *item1 item1-item2; numb by *item1 item1-item15; alg by *item11 item11-item22; geom by item24-item25@1; datch by item2(a)@1; item2(e)@0; I received error such that: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING PARAMETER 55. How can I specify the two-indicator factor to be identified? |
|
|
I'm not sure how your input is being interpreted as it is not specified correctly. Please send the full input and your license number to support@statmodel.com so I can see. |
|
|
Hi, I am trying to apply effects coding (Little et al. 2006) in a multi-group framework in order to be able to estimate factor means for both of my two groups. However, I have difficulties to free the factor mean in the first group. I have used the following syntax: [FA1*]; But still the factor mean in the first group is zero (and apparently the factor mean in the second group is the difference between factor means). So how I can free the factor mean in the first group? Thank you very much! |
|
|
I don't see what you gain by trying to identify the factor mean in both groups. But, if you do that, you have to free it in the group-specific statement (e.g. Model G1) - and you have to fix the measurement intercept for e.g. the first factor indicator. |
|
|
Well, at the moment I'm just experimenting and trying to fit the similar model as in Little et al. (2006). In their effects-coding approach: "The identification and scale setting conditions can be met (for e.g. three indicators) with the constraints, lambda1r =3 – lambda2r – lambda3r and tau1r =0 – tau2r – tau3r". So, no measurement intercept is fixed in this approach. Can such a model be fitted in Mplus? |
|
|
Yes, you add those constraints in Model Constraint. |
|
|
Below is the code I am using. Seems that the line "[F*]" has no effect and the factor mean in the first group remains at zero (and as a reference group). MODEL: F BY agg* (l1) dom (l2) moo (l3); [agg] (i1); [dom] (i2); [moo] (i3); [F*]; MODEL males: MODEL CONSTRAINT: l1=3-l2-l3; i1=0-i2-i3; |
|
|
You have to put [F*] in Model males, so right after that line. |
|
|
That doesn't change a thing - the F mean in the first group is still zero (and everything else is the same). Deleting that "[F*]" from the first model command doesn't make any difference either. |
|
|
So maybe Males isn't your first group, in which case you have to say (presumably): Model Females: [F*]; |
|
|
Got it! Many thanks! |
|
|
Effect coding done using MODEL CONSTRAINT does not seem to work with ESTIMATOR = BAYES. Is it so or have I made some mistake? |
|
|
With Bayes you are not allowed to place constraints on the model in Model Constraint, only reparameterizations. |
|
Back to top |