Sample size in LGM? PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Mikael Samuelsson posted on Thursday, March 01, 2001 - 12:36 pm
Hi

I try applying longitudinal growth modeling on the business start up process. It is an exploratory setting were I investigate old theories in a new setting with new techniques, we have no proceeding analyzes and we have a unique random sample of business ideas in Sweden. The purpose with the model is to test for group invariance between two groups. The two-group final model fit the data well Chi-Square Value 77.398 d.f’s 62 P-Value .0899 with 70 free parameters. But in separate analyses one groups model fit the data very well and one only marginally. The model has four time points only continuous normal independent variables /9 time invariant and 1 time varying single item measures /.

The main concern is that the groups’ samples consist of n 34 and n 199, together 233 respondents. I went trough some of the Guidelines for sample sizes in SEM (Boomsma 1982, Bentler & Chou 1987) and through the Semnet archives and found very little help, depending on purpose/ model structure/ conservatism, sample size guide lines seems to vary. I understand that the best case would be a larger sample. Long story short question: are there any guidelines for sample sizes in longitudinal growth modeling, and possible recent references?
 Linda K. Muthen posted on Friday, March 02, 2001 - 9:24 am
The only guideline that I know of is that you need fewer observations for repeated measures than for cross-sectional data. We have done some simulations (not published) that show n=20 can work well in a very simple growth model. You may want to set up your own Monte Carlo study that mimics your real data to see how this situation performs.
 Roxana Dragan posted on Thursday, May 14, 2009 - 12:54 pm
I try to estimate a growth model with multiple indicators. I have 6 indicators and 6 time points on OECD countries (20 observations). The CFA without invariance means I have 6 factors, one for each time point, and 6 indicators for each factor. This CFA without invariance requires at least 5*6=30 loadings, and 6*7/2=21 variances and covariances for the 6 factors. The loadings and var-cov of the factors add up to 51 parameters, which is larger than 20, the number of observations I have for this analysis. Is this a good reason for not attempting such an analysis? Thank you.
 Linda K. Muthen posted on Friday, May 15, 2009 - 9:15 am
Yes.
 Thierno Diallo posted on Wednesday, September 09, 2009 - 8:51 am
Hi,
I am looking for guidelines for sample sizes in longitudinal growth modelind and possible references. I have read the posting on the issue on Mplus discussion. Is there anything knew since your answer in 2001? Thank you.
 Linda K. Muthen posted on Wednesday, September 09, 2009 - 2:38 pm
I can't remember what we said in 2001 but I would say do a Monte Carlo study. See the following paper which is available on the website:

Muthén, L.K. & Muthén, B.O. (2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling, 4, 599-620.
 ywang posted on Friday, October 16, 2009 - 8:07 am
Dear Drs. Muthen:

I have a only limited sample size (n=220, male=110 and females=110) for latent growth modeling with categorical variables and I always have marginal significance in the relationship beteween the independent variable and slope. Is it fine to drop several covariates from the model (e.g. i on covariates, or s on covariates) if these covariates are not significant? In this way, I might get significant findings. Is this way valid?

Thanks!
 Linda K. Muthen posted on Friday, October 16, 2009 - 8:15 am
I would not do this. You should select your covariates based on theory and report the non-significant findings.
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: