Hi all,
I´m trying to build a Multilevel-Model with -xtmixed- (Stata 14) in order to estimate context effects of classroom-characteristics/ composition on occupational aspirations.
In my data, I have students in classrooms, so I have a nested (2-Level) data structure.
The dependent variable -occupational aspiration- is continious (ISEI-scores).
My contextual variables at the classroom-level are aggregates (class-averages) of individual-level-covariates (school achievement [grades], parents-SES [ISEI], minority-status).
To obtain „context effects“, the continious level-1 covariates are centered at the grand mean and the corresponding level-1 and level-2 covariates are both included in the model.
Until now, I used the whole-class-averages for computing the level-2 covariates, but from theoretical considerations, it is likely that the more relevant reference group for girls resp. boys are the peers of the same sex within the classroom.
So the questions is:
Is it possible and statistically correct, to operationalize a single predictor at the classroom level (level-2) using the individual values of the girls in the classroom for computing girls’ scores and resp. the individual values of the boys in the classroom for boys’ scores?
(i.e. gender-specific classroom-averages, using the mean of the same-sex peers within the classroom, so for all girls resp. all boys in a certain classroom, the value of a level-2-covariate is the same, but differs between the two sexes within the classroom)
Also, if you had some references concerning the issue of sub-group-specific cluster-level covariates, this would be greatly appreciated.
Thanks in advance!
Holger
I´m trying to build a Multilevel-Model with -xtmixed- (Stata 14) in order to estimate context effects of classroom-characteristics/ composition on occupational aspirations.
In my data, I have students in classrooms, so I have a nested (2-Level) data structure.
The dependent variable -occupational aspiration- is continious (ISEI-scores).
My contextual variables at the classroom-level are aggregates (class-averages) of individual-level-covariates (school achievement [grades], parents-SES [ISEI], minority-status).
To obtain „context effects“, the continious level-1 covariates are centered at the grand mean and the corresponding level-1 and level-2 covariates are both included in the model.
Until now, I used the whole-class-averages for computing the level-2 covariates, but from theoretical considerations, it is likely that the more relevant reference group for girls resp. boys are the peers of the same sex within the classroom.
So the questions is:
Is it possible and statistically correct, to operationalize a single predictor at the classroom level (level-2) using the individual values of the girls in the classroom for computing girls’ scores and resp. the individual values of the boys in the classroom for boys’ scores?
(i.e. gender-specific classroom-averages, using the mean of the same-sex peers within the classroom, so for all girls resp. all boys in a certain classroom, the value of a level-2-covariate is the same, but differs between the two sexes within the classroom)
Also, if you had some references concerning the issue of sub-group-specific cluster-level covariates, this would be greatly appreciated.
Thanks in advance!
Holger
