What I feel is that it should not be compulsory to do EFA in each case as the factors emerging in EFA may have different sets of items as opposed to the actual conceptual work or theory which can lead to chaos in terms of jumbled factors and their items and nothing else. So, ideally CFA should have the framed items/factors/constructs as inputs because its to refine the data in any case, ie. you go through EFA as input or directly.
If there already is a strong conceptual underpinning to the items then CFA (e.g., via LISREL or AMOS) is a logical path.
If the instrument has limited prior validity evidence, however, using exploratory methods from the "CFA" output may give clues about which items are not working well, and why.
Factor analysis is an iterative procedure - regardless of whether you use "EFA" (e.g., via SPSS common factor analysis) or "CFA" (e.g., via LISREL or AMOS, as above), unless you have a) very strong underlying theory supporting the dimensionality of the construct and b) an instrument with strong prior validity evidence (including rigorous content validity procedures and item-development procedures), a strict confirmatory factor analysis may end up being of limited use.
I put "EFA" and "CFA" within speech marks because they really are philosophical, not analytical, terms. AMOS can be used in an exploratory way (by inspecting modification indices, etc., and making changes), and SPSS can be used in a confirmatory way (by setting the number of factors and specifying a priori which items you expect to load on the same factor). It is how you use the statistical analysis that determines the extent to which your factor analysis is conducted with an exploratory or confirmatory emphasis.
If you have sound validated conceptual framework, you can go straight a way to CFA to confirm the framework. As , the research is new and emerging then need to do EFA at the first stage to explore the factors and it makes easy to do the CFA where you have identified the factors in EFA. The disicion depends on whether you know the priori structure (Validated).
If you already have a known framework and you want to confirm that structure then no you would not need to run an EFA. However, it may be instructive to do this to verify the original structure if your sample differs greatly from the original validation.
yes. It should be done that the researcher feels, either the priori structure is not applied to his samle populations or he wants to re-develop the theory. It may be possible for emerging concepts.
EFA is carried to determine the dimensionality of your measuring items. If you are developing your own items or you customized them from literature, then you need to do EFA using the pilot study data (minimum of 100 sample). Based on the EFA results, you re-arrange the items based on their resulted components. The you carry out the field study - and using your field study data you carry out the CFA.
Thanks a lot to all. Your answers are not only great but apt to the context in which the discussion was going on. Wish to hear from you soon.
Sounds an interesting discussion and helpful for my next analysis...
I would like to know why Patrick said that at least in his fields, "EFA before CFA is usually discouraged" . Please could you tell me more about your point? Many thanks.
Paola
There is no requirement of EFA before applying CFA, because EFA is used to explore the factor structure for the underlying variables. If the factor structure is known to researcher, then he/she should go directly for CFA to confirm that the current data fits into the already proposed factor structure in that particular context (place or industry).
It may be there is a misconception of what was meant by the statement that EFA is not required (or is actively discouraged) before CFA.
In the long-term validation process, EFA would be part of the necessary steps to first determine what is the underlying factor structure. So, in the long term EFA would precede CFA (determining what is the possible factor structure via EFA in early empirical research and in subsequent empirical studies using CFA to confirm the factor structure). But maybe in the short-term (in a single study), the argument that EFA should not precede CFA is more understandable.
Is it compulsory to do EFA (Exploratory Factor Analysis) before applying the CFA (Confirmatory Factor Analysis) ?
Not necessary - if the current research model (with all its constructs & relationships) and its instrument are based on previously tested empirical research works as well as your objective is to confirm the currently collected data fit into the current research model, you can go straight to CFA. If your current research model is rather new & the instrument is new / being developed / adapted on your own, you need to perform EFA to evaluate what will be the actual factors in your current research model.
My present study is in human resource field. Some of the items that I am using are adopted from other research field i.e. relationship marketing field and they are empirically validated items in marketing literature. My question is, do I need to conduct EFA when I am adopting validated items from different research fields?
In my point of view you still need to conduct EFA. Due to each prepective area has its factors (dimensions) and the respondents are different. The purpose of conducting EFA is to confirm the dimensions that have significant effects in your field. @Nor Hisham Ghafar
I agree with Iham Tariq. @Nor Hisha Ghafar, my opinion is more on conservative approach. Conducting CFA is no harm even when you change the context. I think changing the field does not make much difference as long as respondents are same.
Thank you for your response @Ilham Tariq and @Muhammad Saood Akhtar.
I have admi istered the tool with same items but on different age group. Can i go for cfa to confirm already existing structure or run efa. Mine population is adolescents while already its been done with adults
