Hi, when examining longitudinal measurement invariance, comparing equal form, factor loading, and indicator intercept models, I understand that chi-sq diff testing is done to check as to whether the nested models differ compared to the comparison model; however, should I also be looking at estimates in the output to check for specific changes or non-changes in standardized or understandarized (not sure which) co-efficients?
Sean Tucker posted on Wednesday, August 26, 2009 - 8:46 pm
I have a three wave longitudinal dataset with equal time intervals. Following Chan (1998) I am trying to test for measurement invariance but I canít seem to find an applied example that shows the relevant MPlus syntax (including setting free vs. equal factor loadings, free vs. fixed error variances, etc) and guide to understanding the related output. Would you be so kind as to direct me to a simple applied example or provide the syntax in a reply? Iím relatively new to MPlus and I have a weak background in stats.
See the Topic 4 course handout starting with slide 77.
Sean Tucker posted on Thursday, August 27, 2009 - 11:14 am
Many thanks, Linda.
I received an error message (i.e., "*** ERROR in MODEL command Unknown variable(s) in a BY statement: (1)- ") when running the model with factor loading invariance:
V1 BY v11 v12 v13 v14 v15 v16 v17 (1)-(6); V2 BY v21 v22 v23 v24 v25 v26 v27 (1)-(6); V3 BY v31 v32 v33 v34 v35 v36 v37 (1)-(6);
Pages 532-533 of the user manual describe the syntax for fixing parameter values. Would you be so kind has to describe what (1) Ė (6) means in expanded form. Should I be using different command for fixing the parameter values?
Partial measurement invariance requires relaxing the equality constraints of some items which. Modification indices were used to determine which items. The Topic 1 course video covers testing for measurement invariance. You should listen to that and then the Topic 4 course video for a full explanation.
Pamela May posted on Thursday, August 14, 2014 - 3:14 pm
I am testing measurement invariance for a latent difference model. I have the same construct, activity frequency, being tested at two time points. I am confused by the fact that I have the same degrees of freedom, for when I fix the factor loadings (chi-square = 27.30, df = 18), as well as when the factor loadings and the intercepts are fixed (chi-square = 350.02, df = 18). Would there be any reason that these two models have the same degrees of freedom?
!Fixing factor loadings CA08 BY dv kv (1) lv (2) pv (3); CA12 BY Ev Mv (1) Nv (2) Rv (3); dv with Ev; kv with Mv; lv with Nv; pv with Rv; CA08 with CA12;
!Fixing factor loadings and intercepts (same syntax as before plus this below) !Indicators are categorical, with five levels
I get the error message: *** ERROR in MODEL command There are more equality/parameter labels in the following statement than there are parameters. Problem found in the statement: EXT3 BY y3ext7 y3ext8 y3ext9 y3ext10 y3ext11 (L8-L18)
Are you able to advise on what I am doing wrong? Many thanks
ERROR in MODEL command There are more equality/parameter labels in the following statement than there are parameters. Problem found in the statement: cbcl3 BY y3ext3 y3ext4 y3ext5 y3ext6 y3ext7 y3ext8 y3ext9 y3ext10 y3ext11 (L1-L18). As I have 18 variables I don't know why this is happening. Could you help me identify where I have gone wrong?
Kurt Beron posted on Tuesday, September 08, 2015 - 10:46 am
Hi Drs. Muthen,
I am trying to check for longitudinal measurement invariance of constructs over a number of time periods. The observed indicators of the constructs, though, are a combination of count-type variables (Poisson or negative binomial and, perhaps, inflated).
I understand the steps for the series of constraints necessary to go from least restrictive to most restrictive with continuous or categorical data but have not found the restrictions for, say, a (zero-inflated) negative binomial or the Poisson versions.
Can you suggest an approach in Mplus for this, or direct me to where this might be documented?
I haven't seen this discussed, but I would simply hold all the parameters invariant: intercept, slope, dispersion, inflation. Although I guess strictly dispersion wouldn't have to be invariant since it only influence the variance, not the mean. I don't think one can view it like for continuous outcomes where invariance for loadings only makes the SEM part comparable.