Message/Author |
|
|
Hello all, I'm new to Mplus and SEM so please bear with me. I wanted to create a latent variable called 'benefit perception', but there are only two indicators (personal and societal level)that are used to measure the latent construct. As I was taught, latent should have at least 3 indicators. What should I do with this situation? I am thinking of using the mean of the two indicators (like creating a composite variable) as a new indicator variable. Would it work? If so, what the appropriate syntax be? Thank you so much! |
|
|
2 indicators isn't great (sensitive to mispecification), but ok in a bigger model. You can use the MEAN option ymean = MEAN(y1 y2); |
|
|
Thank you so much Dr. Muthen for your clarification! Another question I have is that, is it doable to add multiple control variable to the SEM? Say if I want to control for age (C1), gender (C2), income (C3), and education levels (C4) in the model, do I just regress the control variables on the outcome variables? Like Y1 on C1 C2 C3 C4? Thank you so much! |
|
|
Right. |
|
|
Thank you Dr.Muthen. I am also wondering, does MLR handle both normally distributed and non-normally distributed missing data? In other words, MLR is robust enough to handle both situations only that the SEs would be different? Thank you! |
|
|
MLR is not totally robust to missing data where the normality assumption behind the missing data handling, as in regular FIML, doesn't hold. See articles by Savalei, e.g. in the SEM journal. |
|
|
Got it. Thank you very much Dr. Muthen for the clarification! |
|
|
Hi Drs. Muthen, I recently discovered that My Mplus returns different results from the same dataset and the same syntax. I used SAS to merged two files together as one dataset (D1) was so cleaned that ID and demographics were gone, so I needed to merge the info to info back to D1 in case I need to run control variables. After merging the two, I output the merged file as .txt (D3) for Mplus to read in. I used the same syntax I used before merging and had only added ID and demo to the Variables line. So nothing changed in Use Variables. The first time I ran it, the model fit was good and was the same as I used the data file before merging. but then Mplus kept returning poor model fits that were very different from previous ones. How could I fix it? Thank you so much! |
|
|
Use Savedata for both runs to check that the 2 data sets are really the same. If yes, send both outputs to Support along with your license number. |
|
|
Thank you Dr. Muthen for all of your help! I have some problems with the fit indices I got from the two models, one is measurement the other one is the full model. Measurement model X2 df CFI TLI RMSEA[90% C.I.]" P SRMR 878 296 0.946 0.936 .045 [.042, .049] 0.991 0.046 Full model 834 234 0.933 0.921 .051 [.048, .055] 0.259 0.080 The measurement model was fine but the p value of the full model dropped a lot. And the SRMR is right on the border line. Could you please let me know what are some of the possibilities for the p value to drop this much and is a SRMR=.08 acceptable when other fit indices were fine? Thank you so much! |
|
|
Try Modindices. |
|
|
Thank you Dr. Muthen, I requested modindices and that some of the suggestions by Mplus did help with improving the model fit. But if the suggestions don't theoretically make sense, what should I conclude? |
|
|
Good question for SEMNET. |
|
Back to top |