Hey Kishmala Habib

one step at a time: first of all, it is perfectly natural to end up with fewer items than you originally planned when using EFA. Please be so kind as to elaborate on it. In my experience (backed up by meta-analyses), most researchers use the default setting in their software which is usually PCA with Varimax rotation. This is not the way to go in the Social/Behavioural Sciences.

What is more, I do not understand what your reason for eliminating certain items was. I am referring to your statement: "some of the variables do have convergent validity which will make the analysis unmeaningful and difficult."

Best

Marcel

Firstly thanks a lot for your prompt response. Responding to your query, I meant that items of some latent variables poorly loaded resulting in poor validity of the construct. Hence I have to drop those from my analysis.

Also, here I have gathered data on supply chain resilience capabilities ( on a likert scale). If I am not wrong, did you mean PCA with varimax rotation is not appripriate here (though I used maximum likelihood with varimax) ?

Maximum likelihood is a good choice for factor extraction, but you should probably pair that with a correlated factors version of rotation, such as oblimin. Then, you can have your program calculate "factor scores" so that you do not have to worry about cut-off points for your loadings.

Unless I completely misunderstood you, Occam's razor.