Unbalanced data PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Abderrahim Oulhaj posted on Friday, February 03, 2006 - 7:27 am
Dear Muthen,

How can I say in my Mplus program for a growth curve model that individuals have different number of observations

Thanks a lot,

 Linda K. Muthen posted on Friday, February 03, 2006 - 8:24 am
I assume that you mean some observations have missing data for certain measurement occasions. You would assign individuals a missing value flag for those occasions and use TYPE=MISSING;
 Abderrahim Oulhaj posted on Friday, February 03, 2006 - 9:48 am
Suppose that I have three individuals. The first one has 3 observations, the second one has 4 observations and the third one has 5 observations as follows:

ID Y1 Y2 Y3 Y4 Y5 t1 t2 t3 t4 t5
1 y11 y21 y31 NA NA t11 t12 t13 NA NA
2 y21 y22 NA y24 y25 t21 t22 t23 . t25
3 y31 y32 y33 y34 y34 t31 t32 t33 t34 t35

where ID identify subjects, Y1-Y5: The five measures (with possibility of missing) made in times t1-t5.
So I can use my latent growth model by:
TSCORES = t1-t5
Model i s | y1-y5 at t1-t5

IS it like that that I should write my program? and how to tell to the program that my missing values are represented by NA's?


 Linda K. Muthen posted on Friday, February 03, 2006 - 10:17 am
See the MISSING option in the Mplus User's Guide to see which values can be used as missing value flags. NA cannot be used.
 Susan Pe posted on Wednesday, September 05, 2012 - 4:42 pm
Dr Muthen,
I have unbalanced panel data NOT due to missing data. Rather, companies died due to bad performance. How do I treat this in order to conduct a CFA? Thank you.
 Bengt O. Muthen posted on Wednesday, September 05, 2012 - 6:26 pm
When you use the ML estimator, the default missing data option in Mplus is MAR (see e.g. the Enders, 2010, book). So for instance if your CFA outcomes are related to performance, MAR will be a good approach to use.
 Wen-Hsu Lin posted on Thursday, January 14, 2016 - 4:54 am
Dear Mplus team:
I have one question regarding unbalance design. I have a repeated measure across 7 time points; however, at each time point I have different n (attrition is different).Some individuals come and go. Can I estimate LGM with different n at different time point? If I turn on listwise, I will have too few cases. If I use FIML, the n is based on wave 1. Thank you so much.
 Bengt O. Muthen posted on Thursday, January 14, 2016 - 6:12 pm
This is not a problem. Use FIML.
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