I am trying to validate a questionnaire that has 120 dichotomous items (answer can be either 1 either 2 – yes/no type) and 10 factors. I am trying to perform a CFA in AMOS to be sure that my items load well on factors, but I am not sure that I can do that for dichotomous variables. I know that for categorical variables (Likert scales) this approach is accepted (even if not with a great deal of enthusiasm).
Do you know any reference that I could use in order to find out if it is possible? Or have you done this yourself?
Hi Alexandra,
The answer is yes. I recommend using Robust Weighted Least Squares (WLSMV) as your extraction method vs. Maximum Likelihood. This will estimate a tetrachoric correlation matrix for factor extraction. I'd also recommended this method when using ordered-categorical variables. Please refer to a paper by Flora and Curran (2004) who explain this method as well as pit WLSMV against other extraction methods. Rex Kline also discusses this approach in his book. I do not have experience doing this in Amos, but running an analysis like this is a breeze in Mplus. http://www.statmodel.com/papers.shtml
You interpret your fit indices (Chi Square, RMSEA, TLI, CFI, etc) as you normally would, though you'll need to use something called the DIFFTEST in Mplus to compare alternate models because the Chi Square statistic is estimated differently.
I hope this helps! My best to you and your research!
Thanks a lot Chiara for these great ideas! I will definitely try this approach!
Hi Alexandra,
The answer is yes. I recommend using Robust Weighted Least Squares (WLSMV) as your extraction method vs. Maximum Likelihood. This will estimate a tetrachoric correlation matrix for factor extraction. I'd also recommended this method when using ordered-categorical variables. Please refer to a paper by Flora and Curran (2004) who explain this method as well as pit WLSMV against other extraction methods. Rex Kline also discusses this approach in his book. I do not have experience doing this in Amos, but running an analysis like this is a breeze in Mplus. http://www.statmodel.com/papers.shtml
You interpret your fit indices (Chi Square, RMSEA, TLI, CFI, etc) as you normally would, though you'll need to use something called the DIFFTEST in Mplus to compare alternate models because the Chi Square statistic is estimated differently.
I hope this helps! My best to you and your research!
I would say it depends. There are some cautions.
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&ved=0CEMQFjAB&url=https%3A%2F%2Fbitbucket.org%2Feyecat%2Freadinglists%2Fsrc%2Fd8e8010f0b0d2dbb0863af3050411695254cc6b1%2FReadingList_NotreDame%2FBernsteinTeng1989FactorItemsFacotrScales.pdf&ei=rPtRUdLVGYrXyAGr3YDgBw&usg=AFQjCNG_D3YpnYEeXdvcPkwar6FWvOg_hg&bvm=bv.44342787,d.aWc&cad=rja
This said it is generally possible. You can see the Mplus sire for some information. Muthen offers some guidance in question it seems to me. I would suggest you start with:
http://www.statmodel.com/discussion/messages/9/7231.html?1304097574
Here are some additional source of information:
http://www.psychometrics.cam.ac.uk/uploads/documents/ESRC_RDI_June10/rdicfabrownjune2010b.pdf
http://books.google.com/books?id=iIr5XhE2UMMC&pg=PA69&lpg=PA69&dq=Binary+items+in+CFA&source=bl&ots=LsCmqMT7PB&sig=S_NAb8_rlvUjYDtqLWvkMT6fjnc&hl=en&sa=X&ei=6f5RUZmLJ8P7yAHD8IDQBw&ved=0CFoQ6AEwBDgK#v=onepage&q=Binary%20items%20in%20CFA&f=false
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics