In SAS it is possible to extract “latent variable score regression coefficients”. In SAS these coefficients are used to calculate a factor score. Is it also possible in MPLUS to extract “latent variable score regression coefficients”?
I am trying to compare the way SAS and MPLUS calculate factor scores. Both use the regression method and the parameter estimates and fscoefficients are the same for both programs. In both programs the fscoefficients are multiplied with standardized observations. To mean it seems that in SAS observations are standardized as ((obs-means)/sd)), whereas in MPLUS they are standardized as(obs-means). Do you know what the advantages are of the different standardization methods?
Your understanding of Mplus is correct. In principle it doesn't matter what standardization is used as long as it implies the correct factor score (posterior mean). However from your information above it seems that you would get different results in the two programs. You should make sure that the factors scores you get agree with those computed by Mplus using the savedata command. savedata: save=fs; file=...;
Thanks for your explanation. Bij default sas uses a prior distribution to calculate a factor score. If I understand you well Mplus uses a posterior distribution to calculate factor scores.
With the "prior distribution approach" I used the following strategy. 1) I calculated a factor score for controls. 2) I calculate a factor score for cases using the factor score coefficients for controls multiplied by observations which were standardized according to the distribution in controls (so I used mean + sd values of controls to standardize the distribution)
Is such an approach also possible when using a posterior distribution? If yes, how can I extract the posterior mean + sd from MPLUS?
Here are a couple of clarifying points. Mplus also uses a prior distribution - a normal distribution. The standard regression method of computing factor scores gets the estimates from the peak of the posterior. So in this regard, there seems to be no differences between SAS and Mplus.
I don't understand your 1) and 2) steps. You mention "controls" - is that a certain group and you also have other groups? If you have several groups, it seems like there is a better approach that can be used.
benedetta posted on Tuesday, January 20, 2015 - 4:36 am
I would like to compare factor score estimates for two different CFA models, the first using WLSMV estimator, the second MLR with Monte Carlo integration. I run the analysis and saved the factor scores for each model SAVEDATA: FILE IS CFA_montecarlo.sav; SAVE IS fscores; FORMAT IS free; I did not get any warning message, but apparently Mplus does not save the factor scores for the second model. Can it depend on the fact that I am using Monte Carlo integration?
I am running a CFA with 15 categorical variables, extracting 4 latent factors, using the WLSMV estimator.
I am trying to save the factor scores as such:
SAVEDATA: FILE IS fscores.dat; SAVE=FSCORES;
A .dat file is being produced but actual factors scores are not being extracted, instead the .dat file contains the values of the 15 categorical variables that make up the latent factors. When I specified the factor names by adding FSCORES= F1 F2 F3 F4; I was asked to run the analysis with the BAYES estimator. Once I run this analysis I received the following error message: 'The syntax for the FILE option has changed. Please refer to the Mplus User's Guide for available options.'
I am not quite sure how to proceed and would appreciate your help very much.
I'm new to confirmatory factor analysis and was hoping to learn more about how MPlus calculates factor scores for an analysis with categorical indicators. Are factor scores standardized to a normal distribution? The factor scores exported from my analysis are not perfectly normal. This isn't problematic for our project, but I am worried I've done something wrong in my analysis.
The factor scores are not standardized to a normal distribution bit their computation builds on a model that specifies a normal distribution for the factors.
ZHANG Liang posted on Wednesday, July 17, 2019 - 8:36 pm
Dear Muthen: I found something strange to me. I modified the model 5.6 in User's guide example V7.0, like this:
MODEL: f1 by y1-y3; f2 by y4-y6; f1 with f2 @0; SAVEDATA: file = factorscore.dat; save = fscores;
(estimator is the default ML)
F1 and F2 were set to be orthogonal, uncorrelated with each other. However, I found their factor scores were still significantly correlated (.28 ***), though not as strong as when "f1 with f2@0;" was freed (.34 ***).
How should I understand this contradiction? Thank you!