I'm designing a questionnaire which has a 'self' and 'other-rater' version.I conducted an EFA for the self-version and I've run a CFA on 2 groups (red people and blue people) who rated the person who completed the self version - if that makes sense?
So I ran a multigroup CFA on the 'other-rater' version comparing blue raters and red raters.
But a reviewer has come back and said (quite rightly) that because the raters are clustered i.e. they are rating the same person, then it should be multilevel?
But I'm confused as to whether the CFA would be multi-level because I don't want to CFA the self-data (as its already EFA'd) I just want to CFA the 'other raters' data.
in that case do I just use the CLUSTER = command, but not TYPE = TWOLEVEL, or should it also be multilevel?
It sounds like the non-independence of observations comes about by students having two ratings. It sounds like you never analyze data for the two ratings together. If this is the case, you would not need to worry about non-independence of observations.
Basically I'm trying to see if my factor structure 'holds' in the 'other' data i.e. in the blue people and the red people.
So person 1 might have 4 blue people rating them and 3 red people rating them.
My understanding is that I can't just analyse them just in two groups, instead, I need to control for the fact that several red/blue people might be rating the same person by clustering them using the variable "person_rated"
So my original idea was to do a multi-level CFA on both groups.
But now I'm not sure whether I need to multi-level it because I don't quite understand what my %between% commands would be.
I'm thinking it would just be clustered using TYPE=COMPLEX to account for the fact that several people rate the same person. and then analyse this across both red and blue groups (i.e. multi-group & clustered CFA using type=complex)
I'm really trying to get to grips with this (tried to read the guide and all the info on the site) but failing massively - it appears I'm not a natural mathematician! any help would be very much appreciated
I did manage to run something - not sure if its right!
I found an old post that would mean I could just run the following
CLUSTER = ratee;
ANALYSIS: TYPE = TWOLEVEL; MODEL: %WITHIN% FW1 BY q12 q41 q87 q3 q19; FW2 BY q13 q49 q89 q91 q98; FW3 BY q93 q29 q100 q28; FW4 BY q15 q7 q10 q59; FW5 BY q34 q38 q70;
and leaving the %between% function blank. I was then thinking I could run this once for blue raters and once for red raters and compare the models for the two groups. Does this look like it will work. It has run - although I had to up the H1iterations, but obviously running and actually being appropriate are two separate things.