Cluster and Covariates
Message/Author
 Ulrich Schroeders posted on Saturday, August 15, 2009 - 4:42 am
Dear Linda,
I want to examine the relationship between listening and reading comprehension with school grades (g1/ g2). In the sample (N ~= 500) there are two different school types (st1/ st2) and two grades (9th/ 10th grade). (However, one cell in this pattern is missging, i.e., no 10th grade in st2.) I wonder what would be the most elegant way to handle the fact that grading is at least school type and grade specific? Put differently, how to account for clustering of the covariates only (without influencing the covariance structure of the measurement part of the model)?

Cordially,
Ulrich
 Linda K. Muthen posted on Saturday, August 15, 2009 - 10:01 am
In this situation you can examine only the main effects of grade and school type but not the interaction between them.
 Ulrich Schroeders posted on Saturday, August 15, 2009 - 10:47 am
Dear Linda,

Thanks for your quick reply. In principle that's true, of course. However, I'm not primarily interested in the effects you mentioned and, luckily, I can put the clusters in a definite order of increasing ability:
1 = 9th grade/ st1
2 = 9th grade/ st2
3 = 10th grade/ st2
That is my variable "cluster".

Is it correct to model the effect the following way?

c BY u1, u2, u3, u4, u5 ;
cluster ON g1 g2 ;
g1 g2 WITH c ;

(with c = reading)
Thanks for your help.

Cordially,
Ulrich
 Linda K. Muthen posted on Saturday, August 15, 2009 - 11:16 am
This is another option. You are assuming that the variable is an interval scale.
 Ulrich Schroeders posted on Sunday, August 16, 2009 - 12:13 pm
Thanks for your patience.