Latent growth model with weight variable PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Anonymous posted on Monday, July 08, 2002 - 11:42 am
Dear Linda & Bengt:

I am running Mplus software using LSAY data set. I intend to run a six year logitudinal model including weight variable. However, it seems to me that in LSAY each year has its own weight variable. That is for six time points, there are six weight variables for each year. According to Mplus manual, I can use a weight variable for the whole six year data. I am wondering if you know how to specify these weight variables in Mplus. If you happened to analyze LSAY data set before, would you please inform me which weight variable did you use for a longitudinal analysis? I am a doctoral student and writing up my dissertation by analyzing a latent growth model using LSAY. Your information will be highly appreciated. Thank you very much.

Sincerely,

Chih-chun
 Anonymous posted on Tuesday, July 09, 2002 - 9:17 am
If you can rewrite your model as a univariate type=twolevel you can use time varying weights.
 bmuthen posted on Tuesday, July 16, 2002 - 10:21 am
We have not used weights in our LSAY analyses. I assume that there is also a time-invariant weight variable.
 Anonymous posted on Tuesday, February 15, 2005 - 4:18 pm
Dear Linda and Bengt
I have an LGC model were i have time-varying binary indicators of a "treatment" at each time point. I have calculated an overall Propensity scores model, beforehand thus each individual has a PS score. Hirano et al (2003) use PS scores p(xi), as weights wi, where wi=p(xi)/[1-p(xi)]. What do we know about using the weights wi into an LGC model? Should i trust in the standard errors that MPLUS generates?

Or it is better to incorporate the paths from a PS observed variables predicting the time-varying variable for Treatment effects at each time point.

What are the differences between these two approaches?
 Linda K. Muthen posted on Wednesday, February 16, 2005 - 7:24 am
If you use sampling weights in Mplus, the standard errors are trustworthy. I do not know what the results would be if you use propensity scores as sampling weights. There are several Web Notes on the Mplus website that discuss sampling weights in Mplus.

Because of this, I would use the propensity scores as variables in the model.
 Peggy Clements posted on Wednesday, September 20, 2006 - 1:29 pm
I'm trying to run some latent growth models using sampling weights. I have already successfully estimated these models without weights. Now, even when estimating the simplest unconditional growth model, I receive the message

no convergence. serious problems in interations. estimated covariance matrix non-invertible. check you starting values.

My syntax is:

USEVARIABLES ARE
rc1rdirt rc2rscal rc4rdirt rc5rdirt
rc1mirt rc2mscal rc4mirt rc5mirt;
MISSING ARE blank;
WEIGHT IS c1_5fp0;

ANALYSIS: TYPE = general;
ESTIMATOR = MLR;
MODEL:
read_i read_s | rc1rdirt@0 rc2rscal@1 rc4rdirt@3 rc5rdirt@7 ;
math_i math_s | rc1mirt@0 rc2mscal@1 rc4mirt@3 rc5mirt@7;

The tech5 output indicates that there are 13 iterations.

Thanks.
 Thuy Nguyen posted on Thursday, September 21, 2006 - 11:21 am
This type of question would be better handled through Mplus support. Please send input, output and data to support@statmodel.com.
 Phil Herzberg posted on Thursday, September 28, 2006 - 1:19 am
How is the command that all variables of the input file are saved and not only the USEVARIABLES. It is the id variable that should also be saved.

Thanks
Phil
 Phil Herzberg posted on Thursday, September 28, 2006 - 5:26 am
DATA:
FILE IS "C:\TEMP\m3.txt";

VARIABLE:
NAMES ARE y1 y2 y3 y4 y5 y6 y7;
USEVARIABLES ARE y2 y3 y4 y5 y6;
CLASSES = c(3);

ANALYSIS:
TYPE IS MIXTURE;
LOGHIGH = +15;
LOGLOW = -15;
UCELLSIZE = 0.01;
ESTIMATOR IS MLR;
LOGCRITERION = 0.0000001;
ITERATIONS = 5000;
CONVERGENCE = 0.000001;
MITERATIONS = 5000;
MCONVERGENCE = 0.000001;
MIXC = CONVERGENCE;
MCCONVERGENCE = 0.000001;
MIXU = CONVERGENCE;
MUCONVERGENCE = 0.000001;

MODEL: %OVERALL%

%c#1%
y2 y3 y4 y5 y6;

%c#2%
y2 y3 y4 y5 y6;

%c#3%
y2 y3 y4 y5 y6;


OUTPUT: SAMPSTAT RESIDUAL STANDARDIZED TECH1 TECH4 TECH7;

PLOT: Type is Plot1 Plot2 Plot3;


SAVEDATA:
FILE IS results_3;
FORMAT IS FREE;
RECORDLENGTH = 1000;
SAVE = CPROB;

OUTPUT:

SAVEDATA INFORMATION

Order of variables

Y2
Y3
Y4
Y5
Y6
CPROB1
CPROB2
CPROB3
C

Save file
results_3

Save file format Free

Save file record length 1000


I would like to store also the ID variable y1.

Thanks
 Bengt O. Muthen posted on Sunday, October 01, 2006 - 12:52 pm
You define y1 to be an IDVARIABLE (see UG).
 Annie Desrosiers posted on Thursday, February 08, 2007 - 12:32 pm
Hi!

I'm running a growth model on 3 time points for 3 groups of persons. I want to know if the trajectories of my 3 groups are the same. So I put 2 dicho variables in the model (tcseul and toseul) and I find that it’s different. But, I know that in one of my groupe I have more female and I whant to be sure that the difference that I found was not because of the inequal proportion of female. How can I control this?

Thank you and there is my input

variable: names are id sexe age tdah
groupe tc1 tc2 tc3 opp1
opp2 opp3 toseul tcseul tcxs
oppxs;

usevariables are tcseul toseul opp1-opp3;

missing = . ;

analysis: type = missing H1;

model: i s | opp1@0 opp2@1 opp3@1.325;
i s on tcseul;
i s on toseul;

output: sampstat;
 Linda K. Muthen posted on Friday, February 09, 2007 - 8:40 am
You can add gender as a covariate:

i s ON tcseul toseul gender;
 Simone Croft posted on Tuesday, September 02, 2014 - 9:51 am
Hello,

I am having difficulty running my LGM with categorical variables. I have a 2nd order model, am using the WLSMV estimator and TYPE = GENERAL. I do not have the space here to show you what I want to model, however I am getting the following condition number: -0.145D-14

Regards,
Simone
 Linda K. Muthen posted on Tuesday, September 02, 2014 - 9:56 am
Please send the output and your license number to support@statmodel.com.
 Mike Zyphur posted on Sunday, February 07, 2016 - 8:51 pm
Regarding the initial purpose of this thread--the use of weights in SEM when different variables have different weights--is there any creative potential to setup such a model in Mplus 7.4? In my case, I do not have a time-invariant weight variable and my model of interest cannot be setup as TWOLEVEL.

Any thoughts are greatly appreciated.

Thank you,
Mike
 Tihomir Asparouhov posted on Monday, February 08, 2016 - 1:53 pm
You can also try multiple group. Most models can be setup as two-level - maybe add time specific dummy variables.

Also I would recommend questioning the weight non-invariance. Weight should be inverse proportional to the probability of selection - so there should be just one weight. Time specific weights could be constructed for some cross-sectional estimation but should not be used with a longitudinal study.
 Mike Zyphur posted on Monday, February 08, 2016 - 5:09 pm
Hi Tihomir,
Thanks for your thoughts!

Unfortunately, in my case, two-level models produce biased estimates with longitudinal data and autoregressive effects (Nickel, 1981). The within-group centering in two-level models causes negative autocorrelation (Arellano, 2003; for an intro see Bond, 2002). So, two-level models can't be used unless the bias caused by within-group centering is reduced, but I can't think of a way to do this. A wide-format procedure using latent variables in SEM works (Allison, 2014; Moral-Benito, 2013). If you can think of a way to solve the problem in two-level models, I'd love to hear about it! (Allowing between-level covariance among the lagged and non-lagged variables does not seem to address the issue.)

Thanks again!
Mike

Allison, P. D. (2014). Maximum likelihood for dynamic panel models with cross-lagged effects. Unpublished manuscript.

Arellano, M. (2003). Panel data econometrics. Oxford: Oxford University Press.

Bond, S. R. (2002). Dynamic panel data models: A guide to micro data methods and practice. Portuguese Economic Journal, 1, 141-162.

Moral-Benito, E. (2013) Likelihood-based estimation of dynamic panels with predetermined regressors. Journal of Business & Economic Statistics, 31, 451-472.

Nickell, S. (1981). Biases in dynamic models with fixed effects. Econometrica, 49, 1417-1426.
 Tihomir Asparouhov posted on Wednesday, February 10, 2016 - 1:23 pm
Perhaps this article is helpful.

Conditions for the Equivalence of the Autoregressive Latent Trajectory Model and a Latent Growth Curve Model With Autoregressive Disturbances by ELLEN L. HAMAKER, SOCIOLOGICAL METHODS & RESEARCH, Vol. 33, No. 3, February 2005 404-416
 Mike Zyphur posted on Wednesday, February 10, 2016 - 7:42 pm
Thanks, Tihomir.
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: