Message/Author 


Dear Linda and/or Bengt, I have a data set with 1042 subjects that I am using for a CFA of a measure with 23 categorical items. We also have 4 grouping variables in the data for doing invariance analyses. When I run any model other than a unifactorial model with imputation of missing data, Mplus tells us there are 1042 observations as it should and without missing data it tells us there are 1012 observations (the number of cases with complete data). However, when we try to run a unifactorial model, with imputation of missing data, Mplus tells us there are only 521 observations, and without missing data imputation it tell us there are 503 observations. Also, when we ask for a difference test between the unifactorial model and a 3factor oblique model or (equivalently) a higherorder model with one secondorder factor and 3 firstorder factors, Mplus will not give us a difference test if we impute missing data (telling us the models aren't nested) but it will do so if we don't impute missing data (though it tells us that the number of observations is 503). Even stranger (to me at least), in our multiple group CFAs (testing invariance) the unifactorial model runs just fine including the difference tests and with the correct number of observations. Any help you can give us will be much appreciated as we are trying to get the revision of ms ready to resubmit. best, Rick Zinbarg 


I am not sure what you mean by running "a unifactorial model with imputation of missing data". Mplus does not do imputation of missing data  perhaps you mean running with Type=Missing? Analyzing the same set of observed variables using Type = Missing should give the same sample size irrespective of the model. It sounds like it is best if you send this to support. 


yes, I meant with Type=Missing, thanks. I will send my data and input files to support. Thanks again 


Just for the record, in case anyone is following this particualr discussion thread other than me, this was not a problem with Mplus but rather the result of a stupid syntax error on my part. Rick Zinbarg 


I wanted to ask that can we use the MLR estimator in a confirmatory factory analysis with binary indicators, provided the sample size is large (>4000) to accommodate for the missing data through FIML. 


Yes, just put the binary indicators on the Categorical list. 


Thanks Dr. Muthen. I now wanted to ask that can the continuous variables on different scales such as : FastingTriglycerides, mmol/l FastingLipoA1, g/l FastingLipoB, g/l FastingHomocysteine, µmol/l FastingRF, IU/ml FastingCRP mg/l Be used as observed variable for a single latent variable? Secondly,can MLR also be used as estimator in a CFA models with latent variables consisting of both binary and continuous observed variables (although binary and continuous observed variables are used separately for the each latent variable).(model N=4691) 


Q1: Yes Q2: Yes 


Thanks a lot Dr. Muthen. regards J 


Dear Dr. Muthen following error was reported for my first attempted CFA: Mplus VERSION 8 MUTHEN & MUTHEN 04/27/2018 5:33 PM INPUT INSTRUCTIONS TITLE: First CFA with continuous and categorical factor indicators INPUT READING TERMINATED NORMALLY First CFA with continuous and categorical factor indicators *** FATAL ERROR THERE IS NOT ENOUGH MEMORY SPACE TO RUN Mplus ON THE CURRENT INPUT FILE. THE ANALYSIS REQUIRES 7 DIMENSIONS OF INTEGRATION RESULTING IN A TOTAL OF 0.17086E+09 INTEGRATION POINTS. THIS MAY BE THE CAUSE OF THE MEMORY SHORTAGE. YOU CAN TRY TO REDUCE THE NUMBER OF DIMENSIONS OF INTEGRATION OR THE NUMBER OF INTEGRATION POINTS OR USE INTEGRATION=MONTECARLO WITH FEWER NUMBER OF INTEGRATION POINTS SUCH AS 500 OR 5000. 


7 dimensions of integration is needed by ML but gives very heavy computations. You can try to follow the advice in the message about Monte Carlo integration. Or you can use Estimator  WLSMV or Bayes. 


I have used the Montecarlo integration with total number of integration points = 5000, but i guess this will take more than one hour at least, as evident from current situation, I wanted to ask: 1)Is it the best method in my situation (handle the missing data and CFA with different types of categorical and numerical observed vraiables)? 2)Or will Bayes as a estimator also generate the results with same accuracy in the current situation? 3)Lastly, if both the estimators (Montecarlo with MLR and Bayes) have equal performance in estimations and missing data handling which is more efficient and less time consuming? 


Bayes and ML are equally good at handling missing data. Their relative timings depend on many aspects as described in the FAQ on our website: Estimator choices with categorical outcomes 


Can you please prescribe a way of viewing the figures generated from Mplus after you have generated and saved them, especially for persons who dont have Mplus installed on their computers and those who have Mplus installed but still are not able to view diagrams after they have saved them and closed the analysis which generated the diagram? 


Exporting the diagram to PDF allows you to share the diagram with those who don't have Mplus. But edited diagrams can be saved in MDG format and then reopened in the Mplus Diagrammer again. 

Back to top 