I came across a curious case during exploratory factor analysis discussion with a friend, He is suggesting that EFA can also be done by loading items forcibly into certain variables determined prior, just like in CFA (though it blurs the distinction between the two). For example if there are 8 items and 2 variables, and you know priori that it is 4 items for each construct; only the EFA of first 4 items is done to see how much they are loading on the first variable. Similarly, the EFA of the latter 4 items is done to see if they load onto the 2nd variable in SPSS. Loadings greater than 0.5 being the cut-off. Could I obtain more insights on this?
No, this is not how Exploratory Factor Analysis works -- the software itself determines which items load on a given factor and what those loadings are.
However, with Confirmatory Factor Analysis, you have to assign the items to the factors, and you have the option of pre-specifying the loadings as well.
FYI, SPSS itself does not contain a program for Confirmatory Factor Analysis. That requires an additional module called AMOS.
Thank you so much David and I am well aware of AMOS being used for CFA. Issue in my opinion was the selection of only certain variables in EFA using Spss to see how much they load onto one factor. However, the base point that all items must be selected and then see how they load up on factors, has been reinforced. Thanks a lot
EFA cant be splitted as your friend suggested. All the variable/items relationships need to be examined in the statistical program at once. under each factor, the eigenvalue should be calculated from all items, and in front of each variable, the communalities also need to be calcualted.
Then the Statistical program will show the loading for all the varialbes on all the factors. The researcher will decide on what items to consider under each factor.
I think your friend is right, i could imagine that he refers to using "target rotation" in contrast to pure olique rotation techniques (as they are implemented in SPSS). Target rotation is implemented in MPlus exploratory structural equation modeling and allows to define target values, e.g. you expect an item to load zero (or any other value) on one factor, while you might estimate the other loadings freely. However, in the final solution, the value don't have to be exact the predefined value (e.g. zero, as it would be in CFA), but an estimate close to this specific value (.08), e.g. there is still some flexibility as in EFA (Asparouhov and Muthen).
As Asparouhov and Muthen say, its something in the middle between CFA and EFA.
Check this paper of of Asparouhov and Muthen,
https://www.statmodel.com/download/EFACFA810.pdf
or this of Brwon (2001):
http://www.rc.usf.edu/~jdorio/FA/Brown%20(2001)%20An%20Overview%20of%20Analytic%20Rotation%20in%20Exploratory%20Factor%20Analysis.pdf
Best wishes/M
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis