SEM for mixed desgin

[LIU Lei]

My questionnaire is novle. 100 listeners listen to 10 sound files and then rate each file on a 12-items likert type scale. So I have 100 participants but 1000 cases. Now my theory is 3 factors can be extracted from 12 items. And I want to use SEM to model the repeated-measured data in lavaan. The general Method factor is added for controlling the common method bias. The model looks like this

md = '
F1 =~ x1 + x2 + x3 + x4
F2 =~ y1 + y2 + y3 + y4
F3 =~ z1 + z2 + z3 + z4
Method =~ x1 + x2 + x3 + x4 + y1 + y2 + y3 + y4 + z1 + z2 + z3 + z4
Method ~~ 0*F1
Method ~~ 0*F2
Method ~~ 0*F3
'

My question is , is it right to model the repeated-measured data like this? Or other advanced SEM models, such as multilevel SEM?

[Terrence Jorgensen]

Your question is more about method than software, so you will probably be better off looking for guidance on SEMNET, a more general SEM forum:

https://listserv.ua.edu/cgi-bin/wa?SUBED1=semnet&A=1

You could certainly use MLSEM, treating subjects as the clustering variable.  If everyone listed to the same 10 sound files, and you had hypotheses about differences among the sound-file conditions, then you might be motivated to treat the repeated measures as multivariate data (wide format, like longitudinal CFA), but then you would have 10 copies of each variable in your current syntax, with only N=100 observations.  If you do not have particular hypotheses about the differences among sound-file conditions, but subjects nonetheless listened to the same sound files, then what you have is a cross-classified design: observations are nested within both subjects and sound files, each of which has its own random effect(s).  So lavaan cannot yet accommodate your study design.  Mplus can handle cross-classified MLSEMs.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[LIU Lei]

Thank you, Jorgensen. As you suggested, I find some methods to deal with cross-classified data for SEM, such as https://link.springer.com/article/10.1007%2Fs11336-015-9485-6.

But they are very different from ordinary method and little literature to reference. So can the general method factor deal with the repeated-measures data enough?

在 2020年6月14日星期日 UTC+8上午4:12:35，Terrence Jorgensen写道：