Testing level 2 relationship with lev... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Sven Lohrey posted on Saturday, February 21, 2015 - 8:26 am
Dear All,

This question is not only addressed at the Mplus support team, but all of you: I want to test a relationship between two group-level variables (leadership and group climate), while controlling for a number of individual-level variables (gender, age, personality). Optimally, I want the individual-level controls to be group-mean centered, while the group-level predictor should be grand-mean centered. I came up with three different options to do this and would value your input.

1) I could run this as a single-level model, in which I include the aggregated versions of my individual-level controls. I could then group-mean center the aggregated versions of the controls, and grand-mean center leadership.

2) I could run a multilevel model in which I regress the individual-level controls on the individual-level version (i.e. the unaggregated version) of group climate, and then regress leadership on the between-part of the same unaggregated version of group climate. The problem I see with this model is that I cannot grand-mean center group climate on the between level.

3) I could run a multilevel model where I use the unaggregated and group-mean centered version of group climate on the within-level, and the aggregated and grand-mean centered version of group climate on the between level, and specify them beforehand as within/between variables, respectively.

What do you think is the best option?
 Sven Lohrey posted on Saturday, February 21, 2015 - 10:57 am
Self-correction: Of course I cannot group-mean center aggregated variables in option 1) and the outcome will remain uncentered in option 2)... also, 2) would give me a cross-level effect, ruling out this option. Maybe I should not work on Saturdays...
 Bengt O. Muthen posted on Sunday, February 22, 2015 - 3:02 pm
I like option 3). If small group sizes I would consider replacing the between-level group climate with its latent counterpart, doing the Mplus decomposition discussed in the Ludtke et al article on our website.

We request that posts are limited to one window.
 Sven Lohrey posted on Monday, February 23, 2015 - 2:45 am
Thank you very much! This is very 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