Message/Author |
|
ME posted on Wednesday, May 27, 2009 - 6:43 am
|
|
|
How would you investigate DIF with factor mixture analysis models? Would you use modindices as in the case of LCA? |
|
|
Yes. |
|
|
Dear Muthens, When estimating DIF with a MIMIC model using an observed variable, we include the direct effect, for example, u1 on Z where u1 is the item response variable and Z is the observed grouping variable. If I am using a factor mixture model with a latent continuous factor and a latent grouping variable, can I measure the DIF effect by using: u1 on C, where C represents the latent class variable. I tried this in the overall section of the model and got an error message. If this doesn't work, how do I detect DIF with latent classes. Thanks. |
|
|
No. You would test the equality of the the intercepts or thresholds across classes. |
|
|
Thanks! So am I correct in saying that I would have to use a multigroup approach rather than a MIMIC approach? Also, would you have an example posted of how you would specify such a model in Mplus? |
|
|
If you are only testing intercepts or thresholds, MIMIC and multiple group are the same. So I wouldn't make a big deal of which approach you are using. The advantage of multiple group over MIMIC is that you can test the equality of factor loadings and other measurement parameters. If these parameters are not part of the model, it is not an issue. A small example for one threshold is: MODEL: %OVERALL% %c#1% [u1$1] (1); %c#2% [u1$1] (1); versus a model without the equality. |
|
|
Final question. Does this analysis provide a way of testing the significance of the differences? |
|
|
Yes, you do it by difference testing of the two models I mentioned. See pages 400-401 of the user's guide for a brief description of difference testing. |
|
Back to top |