Problems using a continuos covariate ...
Message/Author
 Thomas C. Mack posted on Tuesday, October 08, 2013 - 5:59 pm
I'm attempting to use continuous standardized reading data as a covariate in a latent class analysis where the dependent variables are standardized longitudinal math scores from 3rd to 6th grade. However, when I run the analysis I end up w/ all participants in the same latent class. When I run the same analysis with the reading scores transformed into ordinal categories (e.g., failing, proficient, advanced) the analysis seems to run fine. I'm not sure if I need change the syntax when I am using a continuous vs an ordinal covariate?

Variable: names are R6 M6 R5 M5 R4 M4 R3 M3 ID S G C R3L IEP S1871 S1874 S1875 S1876 S1877 IEP6 IEP5 IEP4 IEP3;
usevar = R3 M3 M4 M5 M6;
missing = all (-9);
CLASSES = c(3);
Analysis: type = MIXTURE;
Model: %OVERALL%
i s | M3@0 M4@1 M5@2 M6@3;
i-s@0;
i s ON R3;
c#1 ON R3;
Output: sampstat standardized TECH1 TECH8 TECH11 TECH14;
PLOT: SERIES = M3-M6 (s);
TYPE = PLOT3;

Output from categorical variable analysis:

Latent Classes

1 666.00269 0.55500
2 372.30052 0.31025
3 161.69678 0.13475
 Linda K. Muthen posted on Wednesday, October 09, 2013 - 10:19 am
You should never use a standardized outcome in a growth model. The model is not scale free. I would keep the variable in its original metric.
 Thomas C. Mack posted on Saturday, October 12, 2013 - 3:01 pm
Thank you for your response Dr. Muthen. Just to be clear the math and reading scores I am using are not z or transformed scores. The high-stakes reading and math scores have a M = 1300 and SD = 100. I do have the raw scores as well. What you are saying is to use the raw scores instead of the standard scores? Thanks.
 Bengt O. Muthen posted on Saturday, October 12, 2013 - 3:33 pm

Seltzer, Frank, Bryk (1994). The Metric Matters: The Sensitivity of Conclusions About Growth in Student Achievement to Choice of Metric.
Educational Evaluation and Policy Analysis Spring 1994, Vol. 16, No. 1, pp. 41-49
 Thomas C. Mack posted on Monday, October 14, 2013 - 12:34 pm
Thanks for the reference Dr. Muthen. I created a new data set with the raw scores as suggested but am still getting the same result. After the syntax this is what appears:

*** WARNING in MODEL command
All continuous latent variable covariances involving I have been fixed to 0 because the variance of I is fixed at 0.
*** WARNING
Data set contains cases with missing on x-variables. These cases were not included in the analysis. Number of cases with missing on x-variables: 214
2 WARNING(S) FOUND IN THE INPUT INSTRUCTIONS
 Linda K. Muthen posted on Monday, October 14, 2013 - 2:28 pm
Without a variance, you cannot have a covariance. Is this what you are asking aobut?
 Thomas C. Mack posted on Wednesday, October 16, 2013 - 5:50 pm
No, I am just puzzled as to why the model predicts all members of the sample to belong to a single latent class when I am asking for 2 or 3 class solution. I get the result in my first post when I use an ordinal version of the reading covariate, but when I use the continuous version, using either standard or raw scores, all participants (N =1200) belong to the same class. I know that the variance of the reading scores is not zero, so I am not sure why the model is giving me so much trouble. I don't know if it is a coding error on my part or if I can use a continuous covariate in the latent class growth model?

I am sorry about the confusion. I'm new with MPlus and I am not sure if I am framing my questions with sufficient clarity.
 Linda K. Muthen posted on Thursday, October 17, 2013 - 8:26 am