Path analysis and categorical data PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Jay Wilkins posted on Thursday, November 12, 2009 - 2:56 pm
I would like to use ordinal data in a path analysis.

In previous correspondance I learned that I can use tetrachoric correlations and WLS.

In order to run the analyses do I use a "correlation" matrix with tetrachroic correlations. Or do I have to start with raw data?
 Linda K. Muthen posted on Thursday, November 12, 2009 - 3:11 pm
You need to use raw data. You would need a weight matrix in addition to the tetrachoric correlation matrix to correctly estimate a model using WLS. Mplus requires raw data in this case.
 rongqin posted on Thursday, May 31, 2012 - 11:54 am
Dear Professors,

I would like to do a path analysis on categorical data. Independent and dependent variables are involvement in romantic relationship (1 or 2) and delinquency (1 or 2).

My model looks like this:

usevariables are

vdel wdel xdel ydel zdel vaip waip xaip yaip zaip sexegoed;

CATEGORICAL ARE waip xaip yaip zaip wdel xdel ydel zdel;



MISSING ARE ALL (99999);


analysis:

parameterization = theta;


Model:
vdel;
vaip;

wdel on vdel;
xdel on wdel;
ydel on xdel;
zdel on ydel;

waip on vaip;
xaip on waip;
yaip on xaip;
zaip on yaip;

vdel with vaip;
wdel with waip;
xdel with xaip;
ydel with yaip;
zdel with zaip;

wdel on vaip;
xdel on waip;
ydel on xaip;
zdel on yaip;

waip on vdel;
xaip on wdel;
yaip on xdel;
zaip on ydel;

OUTPUT: SAMPSTAT MODINDICES(4) STANDARDIZED TECH4 CINTERVAL;

The model could run, however, I always get a warning:

WARNING: VARIABLE VDEL MAY BE DICHOTOMOUS BUT DECLARED AS CONTINUOUS.

WARNING: VARIABLE VAIP MAY BE DICHOTOMOUS BUT DECLARED AS CONTINUOUS.

Do I also have to specify independent variable as categorical? Many thanks for your help.

Yongking
 Linda K. Muthen posted on Thursday, May 31, 2012 - 5:39 pm
Because you mention the variances of vdel and vaip in the MODEL command, they are treated as dependent variables. The message is telling you they are dichotomous dependent variables not on the CATEGORICAL list. Means, variances, and covariances of observed exogenous variables are not parameters in a regression model. Those parameters should not be mentioned in the MODEL command.
 rongqin posted on Friday, June 01, 2012 - 6:25 am
Thanks for your answer. However, even I deleted the first two lines:
vdel;
vaip;

I still get warning the same warning.
Then I delete the correlation between

vdel with vaip;
wdel with waip;
xdel with xaip;
ydel with yaip;
zdel with zaip;

the model terminated normally without warning.

However, how do i also estimate the relation between these two categorical data in my model?

Thanks for your answer,
Yongking
 rongqin posted on Friday, June 01, 2012 - 11:25 am
add to my previous questions.

although the model terminated normally, the standardized model results were not estimated.

the output only present MODEL RESULTS. For STANDARDIZED MODEL RESULTS, the output only present StdYX Estimate and Std Estimate, but not for the other two ( S.E. Est./S.E. P-Value).

Do you know why is that?
is that because the model results are already standardized?

Thanks a lot for your answer,

Yongking
 Linda K. Muthen posted on Friday, June 01, 2012 - 12:01 pm
If you want to know the correlation between the two observed exogenous covariates, do a TYPE=BASIC.

With WLSMV and a conditional model, we don't give full standardized results.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: