Message/Author 

R. Jackson posted on Thursday, January 29, 2009  5:52 pm



This is a very basic question. I am having difficulties reading the MPlus output. I am unclear on what values are actually the path estimates. For example, I ran a CFA model with two continuous latent variables and 6 continuous observed variables. I created a visual model and want to add all of the path values and correlation values. Under Model Results, I see the estimates column, but the estimates are all greater than 1. I don't think those would be the path coefficients. I used the STANDARDIZE output option and noticed under STDY Standardization, the estimates are all less than 1. Are those the values I should use? Thanks in advance. 


The raw coefficients are given in the first column under Model Results. These coefficients can be greater than one. I believe that most people report standardized coefficients on path diagrams. In this case, report STDYX for continuous covariates and STDY for binary covariates. See Chapter 17 where the Mplus output is described. 


I have this message, what should I do ? thank you WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES. CHECK THE TECH4 OUTPUT FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE SMSM. 


If you can't see the problem as described in the message, you should send the output and your license number to support@statmodel.com. 


Hi  I am running a growth model that includes linear and quadratic terms, and i get the same error as above: WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES. CHECK THE TECH4 OUTPUT FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE S. I looked at the tech4 output, and the correlation between i and s is greater than 1 (1.22). what does this mean and how do i address this issue? thanks! aprile 


You can try adding residual covariances of adjacent time points. Otherwise, it may be that the model is not suitable for your data. 


Dr. Muthen, I am creating latent factors using categorical variables. In the Standardized STDYX results two items have a correlation greater than one, however, in the Standardized STD results they correlate 0.391. Which should I be using? Thanks 


See the correlations in TECH4. 


In TECH4 I can only see latent factor correlations (which are fine), but the correlation greater than one is between two of the items that load onto the latent factor. Can I continue? Many Thanks 


Std gives a residual covariance. StdYX gives a residual correlation. If it is greater than one, the model is inadmissible. 

QianLi Xue posted on Tuesday, November 09, 2010  1:16 pm



In the MPlus user's guide on Page 656, although the text says "The TECH4 option is used to request estimated means, covariances, and correlations for the latent variables int the model," the TECH 4 output includes variable X, which I assume is an observed variable. why is that? 


Sometimes latent variables are put behind observed variables such that the observed variables are equivalent to the latent variables. This has no impact on model estimation. 


Hello! I am running growth curve models and receive the "Latent covariance"...error message When I fix the slope to 0, and the quadratic when I have it in the model to 0, I no longer receive this message. When I look at the TECH4 output, the covariance matrix indeed shows that the S by S value is negative...so I can understand why I would need to fix it to 0. While this is ok for a single growth model, for the models that I have 2 growth models, this defeats the purpose of my analyses: to determine whether these processes covary over time....so that I was hoping to assess the correlation between the two correlations. Any thoughts on how to proceed? thanks! 


You should first fit each process separately and find a wellfitting model for each before you put them together. 


Related to previous posts, I am running a SEM with both latent and observed variables. In the standardized solution of the structural model, a few of the path coefficients in are greater than 1. Looking in Tech 4, however, there are no correlations greater than 1. What am I missing here? I thought that the paths, in terms of standard deviations, have to be less than 1. Many thanks 


See the FAQ on the website regarding standardized regression coefficients greater than one. 


Hi Dr. Muthen, I did a CFA test of a 4 factor model (binary data) and got the warning in the output: WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE. THIS COULD INDICATE A NEGATIVE VARIANCE/RESIDUAL VARIANCE FOR A LATENT VARIABLE, A CORRELATION GREATER OR EQUAL TO ONE BETWEEN TWO LATENT VARIABLES, OR A LINEAR DEPENDENCY AMONG MORE THAN TWO LATENT VARIABLES. CHECK THE TECH4 OUTPUT FOR MORE INFORMATION. PROBLEM INVOLVING VARIABLE PR. This is my TECH4 output: TECHNICAL 4 OUTPUT ESTIMATED CORRELATION MATRIX FOR THE LATENT VARIABLES SP SA FP PR _______________________________ SP 1.000 SA 0.685 1.000 FP 0.960 0.823 1.000 PR 0.828 0.734 0.955 1.000 Q1. Is it because the high correlation between these three latent variables (SP, FP, PR), it indicated a linear dependency among more than two latent variables? When I combined the items of these three factors into one factor, then did a CFA test of two factor model, the warning was not there anymore. Q2. Even there is a warning with my 4 factor model, the output still shows the model fit indices, information of thresholds and factor loadings. Does that mean these results are no longer reliable because of the warning? Many thanks! 


It seems that the high correlations generate the message. I would not necessarily make one factor. Try an EFA to see what is suggested. That message cannot be ignored. You should not interpret the results. 

Back to top 