Message/Author 

Daniel posted on Tuesday, April 13, 2004  10:49 am



Are there any special problems with a SEM that includes a regression path from a latent variable (e.g., perceived global physical competence) to a dependent observed variable (e.g., BMI)? 

bmuthen posted on Tuesday, April 13, 2004  12:15 pm



No, not at all. 

Xia Wang posted on Monday, March 30, 2009  6:26 pm



Hi, I am a new Mplus user, so I apologize for my beginning level questions. I have the following code in which .17708735 is equal to (1alpha)x variance to correct for attenuation caused by measurement error. model: f on treatment age male white hispanic; f BY violentprop@1; violentprop@.17708735; But I got the error message which says that "One or more variables have a variance of zero. Check your data and format statement. Continuous Number of Variable Observations Variance VIOLENTP 502 ********** TREATMEN 502 0.250 AGE 502 0.016 MALE 502 0.083 **WHITE 502 0.000 HISPANIC 502 0.016" My question is how the variances are calculated and why they are different from variances that are calculated in SPSS according to which the variable "white" have a variance of .068. What should I do to fix this error? Thank you very much! 


Variances in Mplus and SPSS will be different because of different samples. SPSS uses all observations that are not missing for each variable so that the sample size for means and variances will differ for different variables. Mplus uses the same sample size for each variable. 

Xia Wang posted on Tuesday, March 31, 2009  1:27 pm



Thank you Linda for your response. But I have doubled checked this issue and made sure that both samples calculated in SPSS and MPLUS only include cases that have valid values on every variable. So the sample size is the same, but variance is still different. And what's your thoughts on the zero variance of the variable "white" when it's not? Do you see anything wrong with the code I had above? Thank you very much. 


It sounds like you have blanks in your data and are reading it with free format. Please send your input, data, output, and license number to support@statmodel.com. 


Hi, I am still new at this, so I apologize for my beginner level. I am doing a simple regression model in MPLUS with a weighted sample. I have one continuous dependent variable and 10 predictors/independent variables (N=930). When I run this model the output tells me that the model estimation terminated normally, but the residual variances for the dependent variable is not calculated. It only gives me ********. What could be the reason for this? Best, Elisabet 


It sounds like your dependent variable has a very large variance. The asterisks mean that the number was too large to fit in the space allocated. You can rescale a variable by dividing it by a constant. See the DEFINE command. We recommend keeping variances between one and ten. If you can't figure this out, send your output and license number to support@statmodel.com. 


Hi Linda, Thank you for your answer. Yes, you are right, the variance is very large. I will try the DEFINE command. Thanks, Elisabet 


Hello All My question is another beginner's one and related to Xia Wang's earlier post. I am running a smallsample (N=115) structural model with three exogenous (a s p), one mediating endogenous (e) and one "ultimate" endogenous variable which is a dummy (action). Given the small sample size, I have formed composite scores of all three exogenous variables and the mediator. Running this as a path model works very well. However, I'd like to account for measurement error if possible. If I have understood correctly, one can introduce measurement error by setting the factor loading (construct to singleitem indicator) equal to the square root of the alpha and the corresponding error term to (1alpha) times the observed variance. My model looks like this: a by att@0.84; s by sn@0.89; p by pbc@0.89; e by ei@0.81; att@0.45; sn@163.25; pbc@0.38; ei@0.67; e on a s p ; action on e a s p ; I cannot get this model to converge. What am I doing wrong? Many thanks for your help! Teemu 


I think what you are trying to do is shown in Slide 44 of the Topic 1 course handout which is on the website. 


Dear Linda Thank you for reacting so quickly. Yes, slide 44 shows exactly what I am trying to do, with the difference that I am fixing the factor loading at the square root of the alpha instead of 1. The problem is that this model does not converge  even if I fix the factor loading at 1 and even if I use standardised variable scores. I am wondering whether  the problem is in the syntax, or  whether the small sample size (115) and the dummy endogenous variable might be the root of the problem? I would appreciate your advice as this is quite an important model (part of an article revision). May I also add that I really enjoy using Mplus as it is not only flexible enough to accommodate pretty much any research setting I may encounter in business studies, but it is also easy to learn and very user friendly. Many thanks Teemu 


Please send your input, data, output, reliabilities for each variable, and your license number to support@statmodel.com. 


Would it be statistically appropriate (or practically meaningful) to derive the correlation between a latent and an observed variable by specifying a model in which they are allowed to correlate, and then taking the correlation between these variables from the Tech4 output? 


That would be ok if that model fits well. 


Hi, I keep coming across the following error message: ERROR: One or more variables have a variance of zero. I have gone back to the original dataset in SPSS and made sure that all values and missing values are in the appropriate format. I have also tried using the standardized and transformed variable. Can you suggest I try anything else, or forget about this particular variable? Any help would be greatly appreciated. Kind Regards 


It is not the original data set you should be looking at. It is the data set that Mplus is reading. If you can't see the problem, send the output, data set, and your license number to support@statmodel.com. 


This discussion confirms that I can use a latent variable (Adversity) to predict an observed outcome (Cortisol). I am interested in a model that includes that path as well as an additional one linking an observed indicator (Instability) of the latent variable (Adversity) to the outcome (Cortisol). My hope is that this model appropriately accounts for the unique impact of the specific indicator (Instability) versus the broader latent construct (Adversity). Is there anything that prohibits such a model? Thank you. 


That should work fine. 


Thank you so much for the prompt reply. 

Back to top 