I am working on a model where I have data from moms and dads and their children. I am trying to estimate the relationship between executive functioning as a latent variable in the parents and their kids, when some of the kids are siblings. I developed a model that fits nicely (CFI=.97) when I fit the parent and child data for executive functioning using type=complex. (I decided to do it this way rather than try a 2-level model, because the number of parameters (76) in the 2-level model was getting high with respect to the number of observations (183)). The problem comes in when I add in IQ as a covariate. The model fit goes bad (CFI=.82) and the standardized regression loading of child IQ on child executive functioning is greater than one, and (therefore) the residual variance of the child executive functioning is negative. Do you have any ideas for what might be causing these problems and how to proceed?
Dear all, I have quite a beginner's question. I am wondering about the standardized coefficients Mplus offers using "Type is twolevel".
I am conductig a simple random intercept model (Y = G00 + G01*(x_mean) + G10*(x) + U0 + R). Mplus with "Type is twolevel" offers standardized coefficents in the output. As far as I know, other software (e.g. HLM) won't offer you standardized coefficients in multilevel analysis (but maybe I am wrong about this). Moreover I can't remember that I have read any paper presenting standardized regression coefficients for multilevel regression results. Thus, I am wondering, if it would be correct, to report the standardized coefficients Mplus offers.
Moreover I'm not sure, if I understand the Mplus User's guide correctly, as I understood "type ist twolevel random" is only needed analyzing random slopes or if I should use this option for random intercept models, too.
For models with random intercepts but not random slopes, Mplus standardizes the coefficients using the within variances for within parameters and the between variances for between parameters. I don't know of any reference for this but I do think it makes sense. Whether you want to report them would be up to you.
You need TYPE=TWOLEVEL; to estimates random intercepts. You need TYPE=TWOLEVEL RANDOM; to estimate random slopes.
Hello Mr.Muthen, i am working on my diploma thesis, which exists of an examination of the BFLPE and BIRGE. I am using the type=twolevel option in Mplus and all continuous variables were standardized(M=0, SD=1)before entering the data in MPLUS. The dependent variable is a continuous variables (self-concept) and the predictors are continuous and binary. So, my question: Which type of standardization should I choose?
If you standardize your variables, the solution is standardized. I would not, however, standardize my variables unless I had a particular reason to do so. You will only obtain correct results with standardized variables if you have a scale free model.
I standardized the variables because I am wanted to have all variables in the same metric, grand-mean centered and to built the aggregated class ability (aggregated z-Values).To test the BFLPE the grand-mean centering has the advantage that the BFLPE could directly been observed by a negative b from the aggregated class ability on self-concept.
I thought that the in Mplus unstandardized ouput with the standardized variables is the same as using the grand mean centering in Mplus and than using sdyx. Is this so?
Hello, i did it and the standardized output in Mplus are nearly the same for z-standardized variables and grand-mean centered variables. Could you tell me what the difference is between stdyx, stdy and std? Thanks a lot!
Hello! Can I be provided a description (any available) on how Mplus produces the standardized solution for a multilevel model (TYPE=TWOLEVEL)?
I saw in the literature that there are various ways that standardization can be done in MLM, such as standardizing the variables on each level before analyzing the model (so, an L-1 variable is not standardized on L-2, but on L-1); then running the model using these variables pre-standardized.
What does Mplus do, in providing the standardized solution (just below unstandardized results) for a two-level model with variables in their original form? Is the standardization done using the variables' variances as these exist on each level (e.g., for L-2 variance components for L-1 variables)?
I appreciate any description you can provide. Thank you.
Linda, I see this answer above, which may answer this question. If there is any other information you can provide, we would appreciate it. For instance, if Mplus or this explanation has changed since 2014, please let me know. Thank you!
Linda K. Muthen posted on Tuesday, August 28, 2007 - 8:37 am For models with random intercepts but not random slopes, Mplus standardizes the coefficients using the within variances for within parameters and the between variances for between parameters. I don't know of any reference for this but I do think it makes sense. Whether you want to report them would be up to you.
Thanks, Bengt! Does Mplus offer the STDY standardized solution, along with STDXY? Only 1 standardized solution came out by default, but I could try manually requesting the STDY? Do you know which came out by default--STDXY? I thought that in simpler models (not MLM), the multiple standardized solutions all are included in output.