LCA with different indicator variables PreviousNext
Mplus Discussion > Categorical Data Modeling >
 Ceyda Oksel posted on Thursday, August 03, 2017 - 8:24 am

Iím working with longitudinal data coming from multiple cohorts. The indicator variables represent presence/absence of wheeze in a child from birth to 16 years of age. I'd like to assign children into different groups with similar patterns of wheezing. The problem is that data collection frequency and time points are quite different in each cohort. I performed LCA using common ages that are approximately shared across all cohorts with and without a covariate (cohort type) but that way thousands of data points are ruled out (eg. in one cohort, data available at month 6, 12, 18 but in others there is only month12 data so month6 and 18 data are excluded).

Do you have any suggestions as to how I can make use of all existing data during LCA without extensive data imputation?

Many Thanks, Ceyda
 Bengt O. Muthen posted on Thursday, August 03, 2017 - 5:33 pm
Perhaps UG ex 6.18 would be suitable for you.
 Ceyda Oksel posted on Friday, August 04, 2017 - 3:02 am
I've checked it but I don't think it's suitable for me because what I precisely need is a model that can accommodate multiple subgroups, binary indicators, categorical latent variables and subgrouply-varying times of observations.
 Bengt O. Muthen posted on Friday, August 04, 2017 - 2:29 pm
You can take the analogous approach by using Knownclass for the subgroups together with (unknown) latent classes. Binary indicators are also handled analogously.
 Ceyda Oksel posted on Monday, August 07, 2017 - 9:13 am
Thanks Bengt. I am trying to run a multilevel model with the following codes but it seems the model does not converge (estimation doesn't stop).

TITLE: wheezing up to age 15

DATA: FILE IS Wheezegroup.dat;

VARIABLE: NAMES ARE age1 age2 age5 age10 age15 cohort;
USEV ARE age1-age15;
CATEGORICAL = age1-age15;
CLASSES =c(4);

STARTS = 10 5;

C#1; C#2; C#3; C#1 WITH C#2 C#3;
C#2 WITH C#3;

Any thoughts?

Thanks, Ceyda
 Bengt O. Muthen posted on Monday, August 07, 2017 - 3:15 pm
Check the output to see how many dimensions of integration you have (I assume 3) - you find that in the initial Summary.

Ask for TECH8 and check if you have any negative ABS changes in which case you need more precision in your integration - this is done by asking for say integration = 30 or MonteCarlo integration (5000); see UG.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message