Group level DV
Message/Author
 Jaimi Greenslade posted on Wednesday, September 20, 2006 - 7:59 pm
I have not used MPLUS before and am trying to find out whether it is appropriate to analyse my data. I have a model that looks at the relationship between group perceptions and group performance. I only have level 2 data for my DV (i.e., one rating of group performance for each group) but have about five ratings for each group for each of my IV's (group perceptions). I am interested in examining the relationship between group level variance (while partialling out individual level variance) in the IV and relating this to my group level DV. Does MPLUS handle a level 2 DV such as this?
 Bengt O. Muthen posted on Sunday, October 01, 2006 - 12:04 pm
Yes. In the Between part of the Mplus 2-level model you let the between-group part of the IV's influence the level 2 DV, where the level 2 DV is put on the Between = list.
 Kees van Schluten posted on Tuesday, December 12, 2006 - 4:49 am
Hello,

may I ask a follow-up question of understanding:

(1) Do I understand this correctly that Jaimi's model/Bengt, your answer refers to the problem of predicting higher-level variables from lower-level variables?

(2) If (1) is correct - which statistical literature illustrates the approach (as used in Mplus) to predict higher-level variables from lower-level variables?

Thanks
 Linda K. Muthen posted on Tuesday, December 12, 2006 - 8:48 am
1. Yes.
2. Bengt has a note on this. You can request it from bmuthen@ucla.edu.
 Jim Shaw posted on Sunday, April 22, 2007 - 3:28 pm
You indicated that Dr. Bengt Muthen wrote a paper on the subject of modeling grouped data from individual data. What is the reference for this paper?

I am analyzing data from a clinical trial. Patients were put on a drug at baseline and then followed monthly at a clinic for 3 months. At each clinic visit, patients completed a questionnaire asking about their functional status during the preceding month. There are 3 sets of questionnaire data for each patient. Between visits, each patient made diary entries recording his ability to perform a certain health-related act. Each time this act was carried out, another diary entry was made.

I want to model the data collected at the clinic visits as a function of the diary data collected during the preceding visit interval, i.e.,

Y_ijk = X_ijk + ... + e_ijk

where i=1,...,N indexes the patient, j=1,...,3 indexes the visit interval, k=1,...,K_ij indexes the kth diary entry for the ith patient during the jth visit interval, and Y_ijt = Y_ijs for arbitrary s and t.

Based on your previous post, I believe I could fit this model using ordinal logistic regression. A robust variance estimator could be applied to account for arbitrary within-patient correlation. Alternatively, I could fit a multilevel model to the data. I am sending you my Mplus code by e-mail.

Do you have any suggestions for how I should model my data?