| Message/Author |
|
|
| Ding Yan posted on Tuesday, January 19, 2016 - 3:54 am
|
|
|
I want to analyse the measurement invariance of gender for categorical variable, a four-point Likert-type scale. But the analysis goes wrong with the warnings below when test the invariance of factorial structure. The warning indicates that "[y1-y10]" is wrong, but the code is required for factorial structure invariance. What's the syntax errors ?
INPUT INSTRUCTIONS
TITLE:The MCFA of CESD;
DATA:FILE IS zonghe.dat;
VARIABLE:NAMES ARE diqu age gender y1-y10;
Grouping = gender (1 = Male 2 = Female);
USEVARIABLES are y1-y10;
CATEGORICAL = y1-y10;
ANALYSIS:ESTIMATOR = WLSMV;
MODEL:f1 by y1 y2 y3 y4 y6 y7 y9 y10;
f2 by y5 y8;
MODEL Female:f1 by y1@1 y2 y3 y4 y6 y7 y9 y10;
f2 by y5@1 y8;
[y1-y10];
OUTPUT:Standardized;
Modindices;
*** ERROR
The following MODEL statements are ignored:
* Statements in Group FEMALE:
[ Y1 ]
[ Y2 ]
[ Y3 ]
[ Y4 ]
[ Y5 ]
[ Y6 ]
[ Y7 ]
[ Y8 ]
[ Y9 ]
[ Y10 ] |
|
|
|
|
Categorical variables have thresholds not means. You need three thresholds for a four-category variable.
[y1$1 y1$2 y1$3 y1$4];
You can find the models we recommend for testing for measurement invariance in the usser's guide on the webiste starting on page 482. |
|
|
|
|
When I try running the following model with categorical items I get the error below it. When I tried the parameterization recommendation, states that it is only for MIXTURE models. Can you please advise? Thank you.
CLUSTER=School;
USEOBS=(Role EQ 2);
WITHIN;
BETWEEN;
ANALYSIS: TYPE IS TWOLEVEL;
MODEL:
%WITHIN%
fW BY u1-u4;
u3 WITH u4;
%BETWEEN%
fB BY u1-u4;
u1-u4@0;
u2 WITH u4@0;
*** ERROR in MODEL command
Covariances for categorical, censored, count or nominal variables with other observed variables are not allowed.For covariances between categorical variables, specify PARAMETERIZATION=RESCOV in the ANALYSIS command. Problem with the statement:
u3 WITH u4 |
|
|
|
|
Use
f by u3 u4; f@1;
to capture the u3-u4 covariance in the u4 loading. |
|
|
|
|
Thank you. Is this recommendation for WITHIN and BETWEEN so the model would look like the one below? The model is running, but is taking a long time (still running after 50min).
By the way, in my original message I meant to have the BETWEEN CFA mirror the WITHIN (both with u3 WITH u4, not one with u2 WITH u4 and one with u3 WITH u4).
MODEL:
%WITHIN%
fW BY u1-u4;
fcw u3 u4;
fcw@1;
%BETWEEN%
fB BY u1-u4;
fcb BY u3-u4;
fcb@1; |
|
|
|
|
This should work. On Between you don't need the fcb factor because the factor indicators are continuous (random intercepts) - just use WITH.
Chech Tech8 for negative abs changes - see our Tech8 FAQ. |
|
|
|
|
Thanks for the ABS info. I've checked and there are no negative values. But I get all 1 Estimate values and 0 SE values for the standardized BETWEEN estimates (I do get unstandardized values). Why is that and how would I get standardized values?
For BETWEEN, I used:
%BETWEEN%
fB BY u1-u4;
u3 WITH u4@0;
I fixed the indicator variance to 0 because Mplus required me to do so when I first ran it without fixing it. |
|
|
|
|
| Q1: Perhaps because the residual variances on Between are zero. That gives R-square=1. |
|
|
| Back to top |
