Mean and sd for latent variables PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Laura Paoli posted on Thursday, January 17, 2013 - 8:30 am
Dear Linda and Bengt,

apologies if my question is very basic but I am fairly new to Mplus.

I would like to report some descriptives for my latent variables (esp. my DV) - is it possible to get Mplus to display means and sd for latent variables?

Grateful for your help
 Linda K. Muthen posted on Thursday, January 17, 2013 - 8:43 am
Ask for TECH4 in the OUTPUT command.
 Laura Paoli posted on Friday, January 18, 2013 - 2:34 am
Thanks for your reply!
 Laura Paoli posted on Friday, January 18, 2013 - 8:15 am
I used tech4 but the means in the output are still 0 (it also did not display the standard deviation) - where did I go wrong?
 Linda K. Muthen posted on Friday, January 18, 2013 - 8:28 am
The means of latent variables are zero in cross-sectional model. The standard deviation is the square root of the variance which is on the diagonal of the covariance matrix.
 Laura Paoli posted on Monday, January 21, 2013 - 2:07 am
Thanks a lot Linda!!!
 Sabrina Thornton posted on Wednesday, November 20, 2013 - 9:12 am
Hi I have a model with latent variable interaction, which means that TYPE = RANDOM is enabled, and that TECH4 cannot be acquired. I need standard deviations of the latent variables for calculating standardised coefficients for the interaction. How can I ask for the standard deviations of the latent variables?
 Linda K. Muthen posted on Wednesday, November 20, 2013 - 12:18 pm
See the FAQ Latent Variable Interactions which you can find on the website.
 Sabrina Thornton posted on Wednesday, November 20, 2013 - 12:50 pm
Hi Linda,

I have read the paper that is attached to the FAQ regarding latent variable interaction. I have also checked all the relevant topics on the forum. However, it seems that TECH4 cannot be acquired when MODEL = RANDOM. The question still remains as to how I can ask an output that tells me the SD of the latent variables in the model with interaction term.
 Bengt O. Muthen posted on Wednesday, November 20, 2013 - 1:39 pm
For exogenous latent variables you get their variances in the output. For endogenous latent variables you have to calculate them yourself from model parameters.
 Sabrina Thornton posted on Wednesday, November 20, 2013 - 3:20 pm
Thanks, Bengt. If you could direct me to a book or reference that contains the formula, that would be much appreciated. I am not terribly familiar with manual calculations of statistics, so would really appreciate your help. Thanks in advance.
 Bengt O. Muthen posted on Wednesday, November 20, 2013 - 3:43 pm
General references include Bollen's SEM book or the less advanced Kline's 3rd edition SEM book.

If you want a specific answer, you may want to try SEMNET, or ask a methods-oriented colleague or your stat consulting group.
 Sabrina Thornton posted on Thursday, November 21, 2013 - 5:05 am
I am trying to work out how I can calculate variances of endogenous latent variables following equation (16) in your latent variable interaction paper in FAQ. There are obviously a few parameters that are unknown from the output of a model with latent variable interaction (TECH1 and TECH 8):

V(eta1), Cov (eta1,eta2), V (eta1 eta2) and V (residual variance of eta3).

Can you let me know whether the values of all these parameters have to be calculated by hand from say covariance matrix at the indicator level? Can I specify model in a way that will allow me obtain all the necessary parameters for the calculation of this equation?

Thanks in advance.

I am utterly confused...
 Linda K. Muthen posted on Thursday, November 21, 2013 - 9:59 am
Please send your output and license number to support@statmodel.com.
 Lisa M. Yarnell posted on Thursday, September 25, 2014 - 4:20 pm
Hello, I ran a model assessing simultaneous growth on 11 factors, with covariances between latent Intercepts and Slopes. The model runs, but means for Intercepts and Slopes were not automatically provided. So I requested them by typing, e.g., "[Ment_Slp];"

Without explicitly requesting means, several factors have significant variance in Intercept and Slope. When requesting means, SEs for these parameters increased, so none are significant anymore. The parameter values did not change much, but larger SEs made these values nonsignificant.

Should we consider these significant or not? Why would SEs increase? Below is a portion of the output from these runs.

Variances - w/o requesting means
Estimate S.E. P-Value
NEGL_INT 0.637 0.112 5.685 0.000
PHYA_INT 0.909 0.131 6.921 0.000
PAGG_INT 1.173 0.177 6.645 0.000
PAGG_SLP 0.000 0.000 1.493 0.135
SDIS_INT 0.108 0.017 6.512 0.000
VIOL_INT 0.325 0.039 8.243 0.000
VIOL_SLP 0.000 0.000 1.046 0.296
MENT_INT 1.401 0.310 4.521 0.000
MENT_SLP 0.001 0.000 3.507 0.000

Variances - requesting means
NEGL_INT 0.637 13.800 0.046 0.963
PHYA_INT 0.909 723.666 0.001 0.999
PAGG_INT 1.174 663.860 0.002 0.999
PAGG_SLP 0.000 796.142 0.000 1.000
SDIS_INT 0.108 391.432 0.000 1.000
VIOL_INT 0.326 273.217 0.001 0.999
VIOL_SLP 0.000 39.353 0.000 1.000
MENT_INT 1.403 565.692 0.002 0.998
MENT_SLP 0.001 502.350 0.000 1.000
 Bengt O. Muthen posted on Thursday, September 25, 2014 - 5:18 pm
The large SEs when you request factor means indicate that your model is not identified. You need scalar invariance over time to identify the factor means, and even then the factor means at the first time point need to be fixed to zero. See the setup in the UG for multiple-indicator growth.
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: