Cluster averages PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Marko Lüftenegger posted on Tuesday, December 18, 2007 - 3:11 am
We have a quick question about data structure. We are working with a two-level model and need to calculate cluster averages for level two. (we only have data on the individual level but have to use class-aggregates)
Is this possible with mplus or do we have to use spss for this? But if we have to use SPSS we have the problem with missing datas...
Thank you!
 Linda K. Muthen posted on Tuesday, December 18, 2007 - 9:23 am
Mplus does not currently have a function to create cluster averages automatically. This is something we are currently adding.
 christine meng posted on Sunday, March 11, 2012 - 3:35 am
I have a question about the cluster averages. Assuming a two-level analysis (level 1 is the individual level and level 2 is the group level), are the cluster averages calculated from level one or level two variables? I am also not clear how the cluster averages are calculated when you have multiple predictors at both levels.

 Linda K. Muthen posted on Sunday, March 11, 2012 - 7:31 am
If you are referring to the CLUSTER_MEAN option, level 1 variables are used to create a level 2 variable where the level 2 variable is the average of the level 1 variable for each cluster.
 christine meng posted on Sunday, March 11, 2012 - 12:11 pm
I was not aware of the CLUSTER_MEAN option. I am using CHILDID as the cluster variable, school aggression as the dependent variable, parenting and social skills as the level-one predictors, and child gender and family structure as the level-two covariates. (1) Would I use parenting and social skills to calculate two separate cluster averages? (2) How is the CLUSTER-MEAN option different from (1)? And (3) why do we need to calculate cluster averages when cluster averages are not used in the MODEL statements?
 Linda K. Muthen posted on Sunday, March 11, 2012 - 12:34 pm
See Example 9.1 to see how individual-level variables are handled in the between part of the model for TYPE=TWOLEVEL. There are two options - creating a cluster-level variable or latent variable decomposition.
 christine meng posted on Sunday, March 11, 2012 - 1:33 pm
Thank you, example 9.1 has answered my questions above. I have two more questions: How is missing data handled in the first option (creating a cluster-level variable)? Is the second option (latent variable decomposition) a better way at handling missing data?
 Linda K. Muthen posted on Monday, March 12, 2012 - 1:13 pm
In creating a cluster variable, all non-missing values within each cluster are averaged and assigned to each cluster member. If a cluster contains all missing values, each cluster member is assigned a missing value.

With latent variable decomposition of a dependent variable, all observations are used. With latent variable decomposition of an independent variable, observations with missing data on the variable are excluded from the analysis.
 christine meng posted on Monday, March 12, 2012 - 5:22 pm
Thank you, Linda. This has been really helpful.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message