Message/Author 


Hello all, I am doing a SEM with a dataset. I was asked to provide a correlation matrix for the latent variables and was provided with the following sample: ****************Sample start************* Variables_______1_______2_______3_____4 1. LatentVarA 1.00 2. LatentVarB 0,768** 1.00 3. LatentVarC 0,561** 0,677** 1.00 4. LatentVarD 0,060__ 0,000__ 0,05__1.00 Notes: * Correlation is significant at the 0.01 level (2tailed), ** = Correlation is significant at the 0.05 level (2tailed), *** = Construct measured through single indicator ****************Sample end************* I learned that I would have to use TECH4 to get the correlation matrix. My question would be  how do I get the significance levels? I would be interested: 1) In the command I have to use. 2) How to interprete the output of that command. 


If they are available, you will get them automatically. They came out in Version 7.1. 


Thanks for the quick reply! Would the significance levels be available in the table as askerisk or do I have to interprete a certain parameter  which one? If they would be not available, what would be the cause? 


We give the estimates and standard errors. You would have to determine the significance yourself by looking at the ratio of the estimate to its standard error which is a zscore in large samples. If they are not available, they have not yet been developed. 


Thanks for the reply. So to see if my two constructs RM and WOM have a significant correlation I got the following output: Correlations ________RM_______RM_SE____WOM____WOM_SE RM______1.000 RM_SE___0.013____1.000 WOM_____0.495___0.029____1.000 WOM_SE_0.011____0.988___0.035__1.000 Now I would devide (a) 0.495/0.029 (b) 0.495/0.011 (if >2.58 => 0.495**) I can't decide if (a) or (b) would be right or if I got you wrong. 


You divide the correlation by its standard error. I can't follow what you are showing. If the ratio is greater than 1.96, it is significant. If this does not answer your question, send the output and your license number to support@statmodel.com. 


Maybe it is a problem with my English. I understand ratio as something / something else and I understood you the way that something = estimate (of the correlations?) something else = standard error. Sorry for me asking such basic questions but I could not find a description of "How to do a correlation matrix of latent variables" anywhere. 


The ratio is correlation/standard error of correlation 


Now I got it  thanks for you patience. 


Dr's Muthen, I am really new to Mplus  I am trying to calculate the significance of the correlations between various latent variables  I get a correlation matrix (TECH4) but not matrix for standard errors  how will I be able to do this? 


Or wont you be able to calculate them because latent variables don't have means? 


Or wont you be able to calculate them because latent variables don't have means? 


The current version of Mplus gives standard errors for TECH4 in most cases. If you don't have this, you will need to define the correlations in MODEL CONSTRAINT. 


Dr Muthen, Thank you for your response  I have tried using WITH statements, but I get errors. Does this possibly mean my model is incorrectly specified? 


Not enough information to go on. If TECH4 doesn't give you what you want, send your output to Support. 


Hi there, I was hoping to include a correlation matrix with latent and endogenous variables for my paper. My question is with v6 of MPlus how can I go about creating this matrix. I have the correlations from Tech4 but how do I get the standard errors so that I can calculate the p values? Do we use the correlations and p values offered by 'with' statements? 


You would use Model Constraint to express the correlations in terms of the labels of model parameters. This gives you the SEs. 


Thank you for your reply. I don't quite understand  would I specify model constraints as for example: family with x1 (a1) x2 (a2) x3 (a3) etc; 


No. Look at the UG ex 5.20 to give you and idea of the general approach. 

Back to top 