Message/Author 


Hello, In the CFA model below, I am trying to save factor scores of latent variables to use them as independent variables in OLS regression analysis using STATA. My questions are following: 1) When I open 'FEVS 2012+2.sav' by SPSS, this data seemed to be not the same as an original one used for the CFA model here (e.g., more missing values, disappearance of variable names). Is there anything wrong with the commands below? 2) I am trying to save factor scores for the following three latent variables: re, pro, and account. Do I need to add more commands below to get the three scores? Your help will be appreciated!  ...... TYPE = general; ESTIMATOR = mlr; ITERATIONS 1000; CONVERGENCE = 0.00005; MODEL: re BY Q6* Q16 Q50 Q22 Q25 Q31 Q33 Q23 Q24; pro BY Q37* Q61 Q56 Q58 Q30 Q63; re@1 pro@1; account BY re@1 pro@1; account@1; re WITH pro; ...... SAVEDATA: FILE IS FEVS 2012+2.sav; SAVE IS fscores; FORMAT IS free; 


1. The order and format of the saved variables is described at the end of the output where the data were saved. Mplus does not save variable names. The missing value flag is as asterisk. 2. This looks correct. 


Thank you Dr. Muthen. MODEL: re BY Q6* Q16 Q50 Q22 Q25 Q31 Q33 Q23 Q24; pro BY Q37* Q61 Q56 Q58 Q30 Q63; re@1 pro@1; account BY re@1 pro@1; account@1; re WITH pro; I would like to hear your opinion whether there is any problems with constraining path coefficients and variances of the three latent variables (re, pro, and account) at 1 in this model. I am wondering whether I constrain something that should be constrained at 1. 


I don't think this model makes sense. Not only do you have only 2 indicators (1storder factors) of a 2ndorder factor  but you also add a residual correlation between these 2 indicators. I don't see how this model could be useful. 

Hongseok Lee posted on Wednesday, February 18, 2015  11:59 am



Thank you for your reply. Based on the theory I refer to, I came with two indicators (1storder factors) under a 2ndorder factors. In this case, can I constrain path coefficients of both 1storder factors (re and pro) and variance of 2ndorder factor (account) at 1 to address an identification issue and others (e.g., convergence)? 


You have 3 elements in the 1st order factor covariance matrix so you can identify 3 parameters in the 2ndorder model, such as 2 residual variances and one covariance, where the covariance can be captured by say the factor variance being free and the loadings fixed at 1. But you can't correlate the residuals of the 2 1storder factors. However, a 2ndorder model with only 2 indicators (1storder factors) isn't a very strong model, i.e. it doesn't have much content. 


Thank you Dr. Muthen. 

Back to top 