|
Message/Author |
|
Tim Seifert posted on Thursday, December 07, 2006 - 7:13 am
|
|
|
Most would probably test gender differences using a dummy coded regression. But I'd like your opinion about whether or not the following analysis makes sense. Sample -- ~23000 13 year olds, a nationally representative sample, students within schools within provinces Measures -- a latent construct (interest) and an observed variable (pass) Analysis -- an LCA with 2 categories (gender) with gender as a knownclass. A fixed factor structure, with means/thresholds allowed to vary across groups. MPlus syntax: Variable: weight is studentwt; Usevariables are interest1 interest2 import pass; Categorical are interest1 interest2 pass; Cluster is schlid; classes = gender (2); knownclass = gender (sex=0 sex=1); Analysis: type=mixture meanstructure complex; estimator=mlr; algorithm=integration; Model: %Overall% f1 by interest1@1.0 interest2@-1.0; f1@.30; %gender#1% [f1 pass$1]; |
|
|
This seems to make sense. It is multiple group factor analysis done using a KNOWNCLASS variable instead of a GROUPING. These two approaches are the same. |
|
Back to top |
|
|