Message/Author 

Anne Buu posted on Wednesday, March 26, 2008  9:28 am



Q1: Are the factor scores saved by FSCORES calculated by expected a posteriori or maximum a posteriori? Q2: Are the factor scores calculated by assuming the parameters of fixed effects are known (i.e. filling in the estimates of parameters)? Q3: Why EFA does not produce factor scores? Is it "technically" not feasible or just "logically" not favored by the Mplus team? 


1. This varies. For TYPE=GENERAL, categorical dependent variables, and weighted least squares estimation, it is maximum a posteriori. In other cases and for TYPE=GENERAL, categorical outcomes, and maximum likelihood estimation, it is expected a posteriori. 2. Yes. 3. It is basically that if someone asks for factor solutions for one through eight factors, it would end up being a massive file depending on the number of observations. In Version 5.1, there will be a new feature that will allow this in a more limited way. 


Above, you stated that there will be a new feature in Version 5.1 that will allow factor scores for EFA in a more limited way. Is this true in Version 6? Can we get factor scores from an EFA? 


The ESEM feature which was introduced in Version 5.1 is also in Version 6. See Examples 5.24 through 5.27 and the MODEL command in the user's guide. You can get factor scores with ESEM. 


We estimated factor scores from a cfa model and found that the correlations between the obtained factor scores and the manifest scale means (i.e. average scores over the items belonging to each factor) are close to 0 or even negative. We observed this in two independent data sets and respective cfa models. Do you have any idea why this is the case? 


That should only happen if the factor determinacy is very low, making the factor scores poor estimates. 


Hi, I run a CFA model and got the factor score. I thought that the factor score mean is equal to 0. But later I open the saved factor score file by SPSS and found that the mean is 0.0000109, which is not exactly 0. Is it the right case? Why is it not exacctly 0, because of rounding during calcultion? I am just curious about the reason and wondering whether I did something wrong during my calculation. Thank you for your reply in advance. In case you need the model details, here is the model specification, and the factor determinacy is 0.803. Model: factor by a b c d e; d with e; Analysis: estimator = MLM; Output: Standardized; sampstat; modindices(all); FSDETERMINACY; tech4; Savedata: file is fscores.sav; save = fscores; 


Estimated factor scores don't behave exactly like factors. See the FAQ on our website: Factor scores 

Liu BAI posted on Tuesday, October 01, 2019  5:07 am



Dear Mplus Support, I tried to do data reduction for 6 items in Mplus. I know that Mplus could not do PCA, but wonder if I could use CFA and save the 1 Factor score as an alternative way for that? Also, is it okay to use the save factor score as one of the variables in LPA later? My main research question is doing LPA with the variable saved after data reduction and also other observed variables I had. Since I have missing data, I prefer to use Mplus which could handle missing data easily. If the procedure I described above is not theoretically or statistically possible, is there any other way to handle that? Thank you very much! 


I think CFA is a good data reduction technique and unlike PCA it takes into account measurement error. Yes, factor scores can be used for LPA. But you can also do LPA directly on the 6 items and their factor  no need to do this in two steps. 

Liu BAI posted on Wednesday, October 02, 2019  6:47 am



Dear Dr. Muthen, Thanks so much for your quick reply! My issue is that although for this variable I would like to use the factor as a variable for LPA, all my other variables in LPA are just observed variables. Is it still okay to just use the factor in LPA as you mentioned? Is there any syntax I could follow? Thanks! Liu 


Q1: Yes. Q2: See e.g. UG ex 7.17. 

Back to top 