Path analysis - variable with a follo... PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Sébastien Brouillette-Alarie posted on Wednesday, August 17, 2016 - 1:17 pm
Dear Dr. Muthen,

I am a Ph.D. student working on an etiological model of criminal recidivism. Therefore, I want to do a path analysis where the main dependent variables are three types of criminal recidivism, each defined as a dichotomous variable (no/yes) with a specific follow-up time.

In a regular regression analysis, I would use Cox regression. But because I want to do an etiological model, I must use path analysis (or SEM). My question is the following: is there a way to take time-to-event into account in a path analysis?

Here's a simplified version of my syntax (without time):

TITLE:
...
DATA:
FILE IS "...";
VARIABLE:
NAMES ARE delsex delgen youthstr sexrec viorec nsnvrec;
USEVARIABLES ARE delsex delgen youthstr sexrec viorec nsnvrec;
CATEGORICAL IS sexrec viorec nsnvrec;
MODEL:
sexrec ON delsex delgen youthstr;
viorec ON delgen youthstr;
nsnvrec ON delgen youthstr;
OUTPUT:
standardized;

If I wanted to take the time elapsed until recidivism into account, how would I proceed? The important thing to remember is that each recidivism type (sexrec, viorec, nsnvrec) has its own follow-up time.

Thanks a lot for your time, your software is fantastic!
 Sébastien Brouillette-Alarie posted on Wednesday, August 17, 2016 - 1:21 pm
P.S. I already have the "time" variables in my database, formatted for SPSS use (Cox regression). What I want to know is how to integrate them in the path analysis syntax.

Thanks again!
 Bengt O. Muthen posted on Wednesday, August 17, 2016 - 2:19 pm
This modeling is possible. The User's Guide has several examples of Cox regression (ex6.20 - ex6.22).

I don't think the results are different when running each of your 3 DVs together compared with 1 at a time.
 Sébastien Brouillette-Alarie posted on Friday, August 19, 2016 - 1:44 pm
Thanks for the fast answer!

Then, my question becomes: how would I modify the 6.20 example to account for multiple time-dependent DVs?

In the manual, we have:
VARIABLE: NAMES = t x tc;
SURVIVAL = t (ALL);
TIMECENSORED = tc (0 = NOT 1 = RIGHT);
ANALYSIS: BASEHAZARD = OFF;
MODEL: t ON x;

For multiple DVs, would it look like this?
VARIABLE: NAMES = t1 t2 t3 x tc1 tc2 tc3;
SURVIVAL = t1 t2 t3 (ALL);
TIMECENSORED = tc1 tc2 tc3 (0 = NOT 1 = RIGHT);
ANALYSIS: BASEHAZARD = OFF;
MODEL: t1 t2 t3 ON x;

Thanks again.
 Bengt O. Muthen posted on Friday, August 19, 2016 - 2:26 pm
That looks right - try it.

Perhaps you want to add a factor measured by the 3 outcomes to make them correlate beyond their common dependence on x.
 Sébastien Brouillette-Alarie posted on Friday, August 19, 2016 - 3:48 pm
Absolutely! This is just an oversimplified version of my model to make sure I got the part about multiple time-dependent DVs right.

In my real model, there are 3 IV predicting criminal recidivism, each with their own developmental antecedents.

Thanks again, I'll let you know how the syntax worked!
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: