|
|
Default Missing Data for LCA |
|
Message/Author |
|
|
In an LCA analysis I am currently implementing in Mplus, there are some missing data points. In my code I have simply written: MISSING ARE ALL (999); since all of my missing values have been assigned as 999. The analysis is TYPE=MIXTURE, and I am also implementing a covariate analysis. My question is, what is Mplus doing with the missing data? I know it isn't deleting it, so it must be using some form of imputation; I have run the code with missing data points deleted and my results change. I would like to know what imputation method is being implemented, or at least find some reference to it in terms of a command in Mplus. Is there like a default setting which can be changed? I would figure it doesn't implement the same methods as CFA. Any help would be really appreciated. Thank you for your time. |
|
|
ML does not need to use imputation of missing values but directly computes the parameter estimates using all available information, taking the missingness into account under the assumption of "MAR"- this is often called FIML. See any missing data book such as the ones by Little & Rubin, Enders, or Chapter 10 of Muthen, Muthen & Asparouhov. All models that use ML use this same general technique. |
|
|
Thank you, this is very helpful. |
|
|
For anyone looking for a quick reference for ML (maximum likelihood) and it's relationship to MI (multiple imputation), I found the following reference to be very helpful: https://statisticalhorizons.com/wp-content/uploads/MissingDataByML.pdf This is just in case anyone else in the future happens upon this thread and is searching for a reference. It discusses the formal differences between MAR, MCAR, and NMAR as well which I found helpful. |
|
|
And if you want an Mplus-specific discussion of these issues including scripts for a variety of missing data models, again, see Chapter 10 of our RMA book: Muthen, Muthen, Asparouhov (2017). Regression and Mediation Analysis using Mplus. |
|
Back to top |
|
|