Message/Author 


Dear Mplus users, In the multilevel SEM model that I am testing I am encountering the following issues: 1. Is there any default centering method for the twolevel analysis in v6 of Mplus or do I still need to define the centering myself? 2. In my model, I have decided not to define withinlevel variables (but only betweenlevel) because this is the only way I can repeat all my withinlevel paths at both levels. I need the withinlevel part of the model to be tested at both levels. This means I cannot use groupmean centering for my withinlevel predictors as Mplus only permits that for withinlevel variables, which I have not specified. Is this a problem? Should I use grandmean centering for all predictors? 3. To plot one interaction that is part of my model, I need the intercept of my dependent variable. The answering scale of this variable is 15, but the unstandardized intercept in the output is 25.86 and the standardized one is 50.15. How is that possible? I look at the "Intercepts" column in the between part of the output. Thank you in advance. Kind regards, Paris Petrou 


1. No centering is the default. 2. See top of page 243 of the Version 6 User's Guide, showing that when an x variable has both a latent within and a latent betweenlevel part, there is an implicit latent groupmean centering of the latent withinlevel covariate. 3. The intercept is not the mean of the variable. 


Thank you very much Bengt. 2. Does that mean that I can leave all my withinlevel variables (which repeat at both levels) uncentered and only apply centering to my betweenlevel variables (which appear only at the betweenlevel)? Or leave all my variables uncentered? 3. I need to report my interaction plot and the y axis will have a range between 25 and 35. That happens because of the high interecept. I wonder if that will look strange for the reader, because the y variable ranges through a scale of 15. Kind regards, Paris Petrou 


2) You can leave all your covariates uncentered, but you can also center the betweenlevel covariates. 3) The estimated mean for your DV should not be 2535 if your DV has a sample mean of 15. Either the model is set up wrong or is misinterpreted. 


3. By "estimated mean" do you mean intercept? Can I trust an unstandardized estimate of 19.09 for the intercept of a DV which ranges through a scale of 15? When I ask for sampstat this DV has a mean of 0.00 at the within level and 2.02 at the between level. Thank you Paris Petrou 


No, the intercept is not the estimated mean. Take for example a regression of y on x: y= a + b*x + e so that the mean of y, E(y), is E(y) = a + b E(x), where a is the intercept and b is the slope. Mplus estimates a and b, not E(y), but you can express and compute E(y) using a, b, and the sample mean of x. If this doesn't clear it up, please send input, output, data, and license number to support@statmodel.com. 


I managed to change my model in a way that I don't get these big intercepts any more. But now I get this warning: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NONPOSITIVE DEFINITE FIRSTORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.422D34. PROBLEM INVOLVING PARAMETER 1. Parameter 1 is the alpha for one of my withinlevel predictors. One way to stop getting this warning is not centering my predictors, which I do not find a good idea since this is multilevel analysis. I will send my data to the support email, thank you. Paris 


Please send your output and license number to support@statmodel.com. 


Dear Linda, Returning to point 3 of my first post (see above) I need to find some parameters in my multilevel SEM model (TECH1)and then get some values of the covariance matrix (TECH3) and use them so as to plot a significant interaction effect. In particular, I need the parameters that correspond to the intercept of one dependent variable of my model. Is this the ALPHA? In the Mplus guide you say that ALPHA stands for means and/or intercept of the latent variables. But I want the parameter for intercept (and not mean). I am puzzled because in my output I get ALPHAs also for variables which are only treated as predictors and not outcomes (but are correlated with other variables at the between level). Kind regards, Paris Petrou 


All mean and intercept parameters are in either alpha or nu. It does not matter which matrix they are in. 

Back to top 