I am unclear as to how Mplus treats variables in a dataset that are named in the USEVARIABLES option but are not included in the MODEL command. I have heard from some colleagues that this technique has the effect of "controlling" for a variable in the analysis model, but they have not been able to provide a definitive explanation of how this would work.
I have not found any information on this technique in the user guide. According to previous posts in these discussion boards, I see that "any variable on the USEVARIABLES list is used in model estimation whether it is mentioned in the MODEL command or not." Can you provide additional explanations or insights into how these variables are treated and how they affect the rest of the model?
All variables on the USEVARIABLES list are used in the analysis. If they are not mentioned in the MODEL command, means and variances are estimated for them. You will receive the following message in the output:
*** WARNING in MODEL command Variable is uncorrelated with all other variables: Y4 *** WARNING in MODEL command Variable is uncorrelated with all other variables: Y5 *** WARNING in MODEL command Variable is uncorrelated with all other variables: Y6
Sara Babad posted on Tuesday, April 07, 2020 - 7:18 pm
Hi Drs. Muthen,
I created a new variable Resil, which is a composite of other variables, using the DEFINE command. This new variable is the outcome variable in the MODEL command. The variables that were used to create this new variable are not part of the MODEL command. When I run the syntax, I get an error message for all variables that were used to create the composite outcome variable (Resil) but were not included in the MODEL command.
*** WARNING in MODEL command Variable is uncorrelated with all other variables: DSMPTSD
Does this mean that there is something wrong, overall, with how my model was run or is this just letting me know that, because some variables were mentioned under USEVARIABLES, but did not appear in the MODEL command, they are uncorrelated with other variables (which is to be expected)?
Relatedly, is there a way to indicate that variables should be "used" because they are needed to define a new variable, but that they won't be appearing in the MODEL command?
The USEV list refers to variables that are to be used in the Model command. It should have only the variables that are used in the Model command. The Define statement can operate with variables that are not on the USEV list - it finds them in the NAMES list.
Sara Babad posted on Sunday, April 12, 2020 - 8:46 pm
Thank you very much, Dr. Muthen!
If that's the case, then I am unsure of the following - For the analyses I am running, one define command creates a new variable (var1) which is then used to compute a second variable (var2). In the model command, only var2 is used. When I remove the variables used to compute var1 from the USEV list, the define commands do not run properly and my sample size shrinks by 1/3. Could you explain why this might be happening?