I'm conducting a study for my PhD thesis. I've got some constructs from the literature, but the items to measure the constructs were modified (some were added and some were removed). In this case, the established constructs will be examined in a new research context. I've read in some discussion that we can't conduct an EFA and then a SEM on the factors we found from EFA. Therefore, I'd like to know if I randomly split my sample into two and first I will conduct an EFA to study the dimensions of the constructs on my first split sample. After obtaining the results from the EFA, I'll examine those structures and their relationship with a SEM with my second split sample. I wonder if this procedure is justifiable.
Yes.
1. If you start from a crash and doing exploratory research. EFA is best. Then, Run EFA and SEM.
2. If you are using already tested questionnaires, then suggest CFA and Run SEM.
Doing EFA is making you on the safe side that what really is cooking in the respondent's mind.
Hi Sathya,
if you have an idea what to expect, go strictly to the CFA and test the structure you assume in a rigid manner. See this thread for an explanation and defense:
https://www.researchgate.net/post/EFA_exploratory_factor_analysis_CFA_confirmatory_factor_analysis_or_both#view=5c9937470f95f1a905604215
Furthermore, cross-validation by means of sample splits are good for some purposes, they are only weakly informative for supporting causal structures (like those involved in CFA or SEM), see again this huge thread.
https://www.researchgate.net/post/Can_we_do_exploratory_and_confirmatory_factor_analysis_in_the_same_data_set2#view=5486170ad4c118047c8b477f
By doing any kind of factor modeling, be careful not to blindly adhere to the "factor reflex":
https://www.researchgate.net/post/How_to_deal_with_Single_and_Multiple_Item_latent_variables_in_Structural_Equation_Modeling#view=5e849fe521137522f2332355
HTH
Holger
Dr. Holger Steinmetz,
Thank you for your answer. My problem here is that while my IV variables and moderating variables are quite well established, my DV variables are not so ad literature still hasn’t shown quite a good agreement among researchers. In this case, what can be a good way to deal with the DVs?
Hi Sathya (please say Holger).
if there is a disagreement among scholars (depending on the kind of disagreement), this would be a good case for creating competitive models (one that expresses "perspective A" and one that represents "perspective B").
However, if you plan to do that don't move into the next trap (although you will find tons of references that does this), namely to compare the models by means of an AIC or other an alternative descriptive index and decide for the "better fitting" model. A car with three wheels is not "better" than a car with only two wheels.
And again: please read the presented threads: you do not need a full reference as a basis for your model specification. If you have some ideas you can test them.
Best,
Holger
CFA is enough for SEM if you clear about the assumptions and indicators.
EFA is commonly used by researchers when developing a questionnaire from scratch. CFA used in case you adopted the items from previous questionnaires.
You have made some changes in the model you adopted, this mean that you now have a new concept which you will validate empirically in your study environment and context. I believe it is not wrong to explore it to determine its validity and reliability. Go ahead to conduct the EFA, after which necessary modifications will be made (if applicable), then confirmatory factor analysis (CFA) to achieve the study's focus.
Hope this helps.
Article Exploratory Structural Equation Modeling
Article Exploratory Structural Equation Modeling: An Integration of ...
Article Confirmatory Factor Analysis (CFA), Exploratory Structural E...
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling