All,
if you carry out a confirmatory factor analysis (CFA) on responses from a survey that you know are not representative can you generalise the results to the rest of the population?
I am working with psychologists who are using questions to represent latent variables. They are likert scale questions. One of their tests is to carry out a CFA on the questions that 'relate' to a specific latent variable to check that the majority of the variation falls within the same factor.
The survey sent out was purposely unrepresentative in order to keep the numbers for the pilot low. (i.e. the population can be split into 2 groups; group 1 is known to have a higher response rate. We require a certain number back in order to detect a moderate effect. Therefore, we forced the sample to select more from group 1 and our sample has an unrepresentative ratio of group1:group2). The sample was randomly selected within group.
Therefore, in order to determine whether the questions do (or do not) represent the latent variable in terms of the population, should we be using weights to correct the group1:group2 ratio? Or do weights have no affect in CFA?
I hope this makes sense.
Thanks in advance
Zöe
|