Respected members,
I want to know how to report such result in research where second order relationship comes into existence.
For eg
Variable A, B and C ; all are individual and separate constructs (First order) as per previous studies.
But in my survey, post checking validity parameters in CFA using AVE, MSV , ASV etc values , it is found that Variable A and Variable B are making second order constructs, then how to justify this second order relationship with theory if Previously no such relationship is established.
@Antriksha Forming second-order constructs should be based on the theory and literature support. Merely because of high correlations among first-order constructs, constructing a second-order construct is not advisable. Moreover, you've added your models only in default view without any standardized loadings. Had you shown your model after running it, it would have been better for providing suggestions. However, in your second-order model, paths from first-order variables to second-order factor be reversed and put error terms on both first-order factors and a regression weight on both paths then it will run.
Hello Antriksha,
First, remember that sample to sample variation can often lead to apparently different implied or observed structures. Ideally, you'd like to see affirmation of your more complex structure in new/different data.
You haven't indicated how much better, if at all, the more complex second-order factor model is compared to what we would presume to be just correlated first order factors. If allowing for factor correlation is sufficient, then why stray from parsimony (and, apparently, previous work)?
If you can make a compelling case that: (a) not only is the second order model superior in terms of reproducing observed relationships among the manifest variables; and (b) may be justified by a sensible modification to the underlying theory, then maybe you have good grounds to argue for an amended theoretical framework. Otherwise, you'll have a hard time selling the idea.
Good luck with your work.
What you say is more common than desired. Keep in mind that you are analyzing sentences that will theoretically and logically be grouped in a certain way by the participants. The social or cultural group of the participants can alter this logic (what for some is logical, for others not). This is especially the case when the survey includes elements for ownership quantification. A typical case is the TEIQue survey, which presents highly variable results, unless sums of responses are analyzed. A simpler one that also presents some variation is the DASS-21 that logically groups three individual constructs or factors (Depression, anxiety and stress), but that in many population groups generates four factors (a fourth related to sleep quality), or two....Sometimes this simply indicates that the phrase or its translation does not conform to the intended understanding for its purpose.
- Badges
- Science method